Datasets:

Daniel-P-Gonzalez

/

OCD

like 0

# Introduction Boron (B) is an essential micronutrient required in major physiological functions for the normal growth and development of higher plants. This element participates in cell wall structure formation through the borate-diol bonding of two rhamnogalacturonan II molecules. It is also involved in root elongation, carbohydrate metabolism, phenol accumulation, pollen-tube growth and cell membrane integrity. Boron excess occurs mostly by irrigation with high B level in water, or in arid and semiarid areas where water reaches topsoil by capillarity and then evaporates to cause B accumulation. As boron accumulates in leaves as they age, B toxicity symptoms usually appear on older leaves, first as leaf tip and margin yellowing or mottling, then with a brownish burnt appearance, which finally ends in premature fall at high concentration levels. It can also occasionally appear resinous gum spots on the undersides of leaves and it is as well associated with a shortened distance between leaf nodes. Therefore, severe B toxicity may induce loss of vigour, shorter branches, and even twig dieback. It is widely accepted that *Citrus* is a sensitive crop to B excess, whose toxicity causes major disorders that lead to loss of yield. However, further research is needed to better understand the regulation of the mechanisms involved in the whole response of citrus plants to the presence of high B levels in the media. Thus, after entering roots, boron is transported through xylem vessels, mostly bound to cell wall structures (insoluble B pool) or accumulated in apoplastic fluids (soluble B pool), while only another low soluble portion enters cells. In plants, boron homeostasis seems to be related to the synergic regulation of several genes involved in B uptake, transport and partitioning in the aerial part. Moreover, vacuolar compartmentation might also play a key role in B tolerance of cells. On the molecular basis, the first B transporter identified in a biological system was *Arabidopsis thaliana* BOR1, an essential efflux-type B transporter for efficient loading of B in the xylem. This gene is expressed mainly in root pericycle cells and its overexpression enhances root to shoot B translocation under B-limiting conditions. When analysing the complete *A*. *thaliana* genome, six sequences were found to have a high homology with *AtBOR1*. Among them, only *AtBOR4* has been characterised, and its overexpression markedly improves *A*. *thaliana* growth under high B conditions through B efflux. In fruit crops, *VvBOR1* and *CmBOR1* have been recently characterised as B transporters in *Vitis vinfera* and *Citrus macrophylla*, respectively. Moreover, NIP5 is a boric acid channel that facilitates B influx to root cells in *A*. *thaliana* and *B*. *napus*. It is localised to the plasma membrane on the outer side of epidermal, cortical and endodermal cells in roots. NIP5;1 is required for the uptake of B from the root surface and its accumulation is regulated in response to B deprivation. In *A*. *thaliana*, the overexpression of *AtNIP5;1* results in root elongation under B-deficient conditions, which improves short-term B uptake. Also, some evidences indicate that other PIPs aquaporins are also involved in B transport. Finally, the vacuolar compartmentation of B has been related to the activity of *AtTIP5*, an aquaporin family member localised to the cell tonoplast membrane. Under toxic B conditions, the *AtTIP5;1* overexpression has been reported to lower the cytoplasmic B concentration by accumulating B in the vacuole to, therefore, confer cell tolerance to toxic B levels. Moreover, some proton pumps located in the vacuolar membrane, particularly tonoplast adenosine- triphosphatase subunit A (V-ATPase A) and tonoplast pyrophosphatase (V-PPiase), have been reported to be altered under different environmental stresses, such as B toxicity or salinity. In addition to molecular regulation, some biochemical indicators related with oxidative stress and osmoprotective compounds, notably proline, have been well- defined in studies of plants subjected to different stresses. The possible roles attributed to proline are balance of osmotic pressure, preservation of enzymes in the cytoplasm, detoxification of reactive oxygen species (ROS) and protection of membranes against lipid peroxidation. However, data on proline and B toxicity are still scarce, and the role of this molecule regarding such nutritional stress is still not clear. Citrus species differ in their susceptibility to B concentration. *Citrus macrophylla* stands for a clementine species and is classified as a potential tolerant one to B toxicity. Hence the study of this plant’s global response when cultured under extreme B availability conditions could provide prospective insights into the mechanisms that determine *Citrus* tolerance to B excess. For this purpose, seedlings of *C*. *macrophylla* were cultured in media with normal B and at a high B level to study: A) the expression level of the main putative genes controlling B homeostasis within the plant under the above-mentioned conditions; B) boron distribution in different plant organs and compartmentation inside cells; C) the role of osmoprotective regulation against oxidative stress due to B excess. # Materials and methods ## Ethics statement The experiments were conducted in the Department of Citriculture and Vegetal Production from Valencian Institute of Agrarian Research (Moncada, Spain). Dr. Mary-Rus Martínez-Cuenca was response for experimental analysis in this manuscript and can be contacted in the future. The authors declare that this manuscript does not matter the any ethic issue and it does not involve endangered or protected species. ## Plant material and treatments Seeds of *Citrus macrophylla* W. were sterilised for 5 min in a 2% v/v commercial bleach (0.5 M NaClO) solution prior to seed coat removal, rinsed 3 times with sterilized deionized water and transferred to a media containing distilled water (pH 6.0) with 0.4% agar added (Difco Bacto). Media were previously autoclaved at 120ºC for 20 min and distributed in 150 x 25-mm tubes (40 mL per tube). Seeds (one seed per tube) were germinated in a growth chamber (Sanyo MCR-350H, Sanyo Electric Biochemical Co, Japan) at 20-22/26-28ºC night/day temperatures, 80% relative humidity and 250 µmol m^-2 s^-1 photosynthetic photon flux density for 16 h per day. After 20 days, seedlings with a single shoot were selected for uniformity and transferred individually after removing cotyledons in plastic 50 mL tubes containing a basic nutrient solution \[1.5 mM Ca(NO₃)₂, 1.5 mM KNO₃, 1 mM MgSO₄, 1.2 mM H₃PO₄, 20 µM Fe-EDDHA, 7.65 µM ZnSO₄·7H₂O, 54.4 µM MnSO₄·H₂O, 0.55 µM MoO₃, 0.5 µM CuSO₄·5H₂O\] and 50 or 400 µM H₃BO₃ (Ct or +B, respectively). All the media were supplemented with 0.25% agar (pH 6.0) and sterilized as previously described. Seedlings were kept for 25 days under the same growth chamber conditions as described above. ## Plant growth After 25 days of conditioning, seedlings were removed from culture tubes, rinsed with distilled water and divided into leaves, stems and roots. The length (in cm) of stems and roots was measured. Organs were fresh weighed and dried in a forced draft oven at 70ºC for 48 h until constant weight (DW, in mg). ## Total, soluble and insoluble B fractions Roots and leaves samples were split into two subsamples each (0.5 g) and used for determination of B total (B_t) and B fractions (B_f) concentrations. Stem samples were used only for B_t determination. B_t concentration was measured according to. Dry tissues (0.5 g) were burnt in a muffle furnace at 550ºC for 12 h. B was extracted with 0.1 N hydrochloridric acid, HCl, (Hiperpur Panreac) to a final volume of 5 mL. B_f (soluble in water, soluble in organic solvents and insoluble) concentration was measured according to. Samples were frozen and ground into fine powder in liquid nitrogen (N₂) by a mortar. The powder was homogenized with 10 volumes of ice-cold water and centrifuged at 1,000 rpm for 10 min. The precipitate was washed with 10 volumes of ice-cold water and re- centrifuged. The combined supernatants were defined as water-soluble B fraction, which is B mainly found in the cells free space. The residue was washed 3 times with 10 volumes of 80% ethanol, once with 10 volumes of methanol:chloroform mixture (1:1, v/v) and once with 10 volumes of acetone, to extract B inside the protoplast and apparently linked to organic molecules. The insoluble pellet was used for B bound to cell wall. All fractions were dried and ashed similarly as B_t samples. B_t and B_f concentrations were determined by inductively coupled plasma atomic emission spectroscopy (ICP-AES iCAP 6000, Thermo Scientific). ## Leaf Chl concentration Leaf chlorophyll (Chl) concentration was measured according to Moran and Porath. Samples from the two youngest fully expanded leaves per plant were separately ground with a mortar in N₂. Fresh material (0.05 g) was incubated in 4 mL N,N-dimethylformamide at 4ºC for 72 h and centrifuged at 4,000 rpm and 4ºC for 15 min (Eppendorf Centrifuge 5810R, AG, Hamburg, Germany). Supernatant was left for 1 h in the presence of Na₂SO₄ and the absorbance was measured at 664 and 647 nm (Lambda 25, PerkinElmer, Shelton, CT, USA). The average value of the two leaves was considered representative of each plant. ## Proline and MDA concentration Free-proline concentration in leaves and roots was determined according to Delauney and Verma. Fresh tissue was ground with a mortar in N₂. Sample material (0.1 g) was homogenized (Vortex) in 1.5 mL sulphosalicylic acid (3%) for 1 min, centrifuged at 14,000 rpm for 5 min (Eppendorf Centrifuge 5810R, AG, Hamburg, Germany) and the supernatant was stored at 4ºC. An aliquot (0.2 mL) was incubated with 0.5 mL sulphosalicylic acid (3%), 0.7 mL reactive ninhydrin acid reagent (ninhydrin, phosphoric acid 6 M, glacial acetic acid 60%) and 0.6 mL glacial acetic acid (99%) in a dry bath at 100ºC for 1 h (Thermostatic Bath BD, Bunsen SA, Humanes, Spain). Samples were cooled down in an ice bath for 15 min and absorbance was measured at 520 nm (SmartSpec Plus, Bio-Rad, West Berkeley, California, USA). Lipid peroxidation in leaves and roots was determined by measuring the malondialdehyde (MDA) concentration according to Hodges et al.. Frozen sample material (0.1 g) was homogenized (Polytron PT 3100, Lucerne, Switzerland) in 5 mL of cold ethanol (80%) at 10,000 rpm for 1 min, centrifuged at 3,000 rpm for 30 min and 4ºC (Eppendorf Centrifuge 5810R, AG, Hamburg, Germany) and the supernatant was stored at 4ºC: (a) an aliquot (1 mL) was combined with an equal volume of a 20% trichloroacetic solution (TCA), while (b) another aliquot (1 mL) was combined with an equal volume of a 0.5% thiobarbituric acid (TBA) and 20% TCA solution. Test tubes were covered with glass marbles to avoid evaporation and incubated at 90ºC for 1 h in a water bath (Thermostatic Bath BD, Bunsen SA, Humanes, Spain). Samples were cooled down in an ice bath for 15 min and centrifuged at 1,000 rpm for 15 min and 4ºC (Eppendorf Centrifuge 5810R, AG, Hamburg, Germany). Absorbance of supernatant (a) was measured at 532 and 600 nm (SmartSpec Plus, Bio-Rad, West Berkeley, California, USA) using 20% TCA as a blank. Absorbance of supernatant (b) was measured at 440, 532 and 600 nm using a solution of 0.5% TBA and 20% TCA as a blank. Concentration of MDA was calculated using an extinction coefficient of 155 mM^-1 cm^-1. ## RNA extraction and real-time RT-PCR analysis Total plant RNA from leaf and root tissues was extracted using the RNeasy Plant Mini Kit (Qiagen, Hilden, Germany). Fresh samples (0.1 g) were ground with a mortar in N₂. To remove genomic DNA, RNA samples were treated with RNase free DNase (Qiagen) following the manufacturer’s instructions. RNA quality and concentration were assessed with ND-1000 full spectrum UV-Vis spectrophotometer (Nanodrop Technologies, Thermo Fisher Scientific, Delaware, USA). Quantitative real-time polymerase chain reactions (RT-PCR) were run in a LightCycler 2.0 Instrument (Roche, Diagnostics GmbH, Mannheim, Germany) equipped with version 4.0 of the Light Cycler Software. Reactions contained 2.5 units of MultiScribe Reverse Transcriptase (Applied Biosystems, Roche Molecular Systems, New Yersey, USA), 1 unit of RNase Inhibitor (Applied Biosystems), 2 µL of LC Fast Start DNA Master PLUS SYBR Green I (Roche), 25 ng of total RNA, and 250 nM of the specific forward and reverse primers in a total volume of 10 µL. The PCR programme was run at 48ºC for 30 min, 95ºC for 10 min, followed by 45 cycles at 95ºC for 2 s, 58ºC for 8 s and 72ºC for 8 s. The fluorescent intensity data were acquired during the 72ºC extension step and were transformed into relative mRNA values using a 10-fold dilution series of an RNA sample as a standard curve. The relative mRNA levels were normalized to total RNA amounts as previously described. Actin was used as the reference gene and the specificity of the amplification reactions was assessed by post-amplification dissociation curves and by sequencing the reaction product. At least three independent RNA extractions and real-time reactions with three technical replicates per sample were performed. Putative *BOR4 and TIP5* genes were identified by a homology search in the clementine genome full-length Phytozome v9.0 database. The expression levels of *NIP5*, *PIP1*, *PIP2*, *CmBOR1*, *V-PPiase* and *V-ATPase A* genes were evaluated using the forward and reverse primers described by other authors. Details of the forward and reverse primers used in the RT-PCR are listed in. Gene-predicted products ranged from 75 to 156 bp. ## Statistical analyses For statistical analyses, the values of growth parameters, chlorophyll concentration, proline and MDA concentrations are the mean of six independent plants per treatment. Values from other analyses are the mean and the standard deviation of at least three replicates from three independent groups of ten plants per treatment. Data were submitted to an analysis of variance (ANOVA) with Statgraphics Plus, version 5.1 (Statistical Graphics, Englewood Cliffs, NJ), prior to testing for normality and homogeneity. When the ANOVA showed a statistical effect, means were separated by least significant differences (LSD) at *P \<* 0.05. # Results and Discussion ## Symptoms, plant growth parameters, total B and chlorophyll concentrations Twenty-five days after plants were grown in the presence of 400 μM H₃BO₃, +B plants showed incipient symptoms of B toxicity on a few leaves, consisting in small yellowish zones near their apical end and borders resembling those previously reported for other citrus species. Regarding plant growth, +B plants showed significant reductions in root dry weight and root length of 27.3% and 14.0%, respectively, related to Ct ones, while no significant differences in growth parameters were observed in the shoots. Stunted root growth is a typical symptom in plants grown under B-excess conditions. However, this effect was not evidenced in the shoot, which seems capable of maintaining a relatively high growth index despite accumulating large B concentrations in leaves, as occurs in some *Solanum* and *Puccinellia* plants considered to be B toxicity tolerants. Moreover, +B plants registered a substantial rise in B concentration in the leaves (3.8-fold) when compared with Ct ones and reached a level close to 350 μg g^-1 DW, which is considered to fall within the excess range (\> 250 μg g^-1 DW;). Roots from +B seedlings also showed higher B concentration (1.4-fold increase) than Ct ones, being much lower than that observed in leaves. Finally, we also determined chlorophylls concentration (*a*, *b* and *total*) in leaves as stress indicator to check the degree of tolerance of Cm seedlings to B toxicity in citrus. The results indicated that B excess only led to a significant 26.3% reduction of Chl *b* values in +B seedlings when compared to Ct ones. The data of shoot growth and chlorophyll concentration in leaves, together with the incipient B toxicity symptoms shown by the seedlings cultured at a high external B level, suggest that, despite the elevated B concentration reached by leaves, Cm behaves as a B-excess-tolerant plant. Therefore, the present work studied some of the mechanisms that likely enable Cm seedlings to tolerate B excess in accordance with previous proposals. ## Transcription analysis of candidate genes related with B homeostasis in the plant ### Expression of putative *NIP5*, *PIP1* and *PIP2* genes and its relation with B transport in the plant Acquisition of boron by cells is associated with the activity of some members of the aquaporin family codified by the genes *NIP5*, *PIP1* and *PIP2*. The expression level of putative *NIP5* gene lowered in roots as a result of high B levels. Thus, +B seedlings contained 52% fewer *NIP5* transcripts in roots than Ct ones. It is noteworthy that no effect on the *NIP5* transcript level was recorded when analysed in +B leaves. In *A*. *thaliana*, this gene codified a boric acid channel for efficient B acquisition under limited B supply. This result was also found in Carrizo citrange roots, where *CiNIP5* expression increased under B-deficiency conditions. In contrast, its down-expression under B-excess conditions in the roots of *Citrus* plants has revealed a point of regulation of B uptake efficiency of different citrus rootstocks. Similar behaviour had been previously described by Schnurbusch et al., which showed a relation between *HvNIP2*:*1* gene expression and B toxicity tolerance in barley. Moreover, Tanaka et al. suggested the importance of *AtNIP5*:*1* degradation for plant acclimation to high B conditions. *PIP1* expression level was also reduced when analyzed in +B seedlings related to Ct ones. In particular, +B plants registered a 50% and 40% reduction in *PIP1* transcript levels, in leaves and roots respectively, when compared with Ct ones. This indicates the possibility of reduced cell permeability to B under excess conditions. In this way, the expression of *ZmPIP1* in *Xenopus laevis* oocytes indicates that this aquaporin is involved in B transport within cells. In contrast, no differences in the *PIP2* expression level were found between Ct and +B treated seedlings. ### Expression of *CmBOR1* gene under B toxicity conditions and its relation with B transport in the plant We also monitored the expression level of *CmBOR1* gene, which has been reported to modulate B transport in plants due to its role in loading B into the xylem, and therefore to translocate B from roots to shoots. However, no differences in the *CmBOR1* expression level in the leaves and roots of Ct and +B seedlings were found. This is in accordance with the high capacity of *Citrus* to accumulate B in their leaves at high external B levels, which suggests that citrus plants have an effective system for B transport from roots to leaves, even under B-excess conditions. Therefore, the presented data indicate that B toxicity tolerance is not linked to a repressible B transport mechanism. To support this, it has been reported that the *CmBOR1* transcript level is not affected by high B supply and the activity of this gene does not prevent high B accumulation in citrus leaves. In *Oryza sativa*, OsBOR1 loads B into the xylem and also participates in the absorption of this element in roots. In *Brassica napus*, *BnBOR1;1c* and *BnBOR1;2a* are up-regulated in roots under low B stress, but no differences in their expression have been found between B-efficient and B-inefficient cultivars in low or normal B environments. However, some evidence has indicated that the removal of the BOR1 protein from membranes and its rapid degradation through the endocytic pathway may occur in response to high B levels in *A*. *thaliana*. This suggests that this mechanism in other species could be involved in the regulation of B transport from roots to shoots at high B levels. ### Accumulation of putative *BOR4* transcripts under B toxicity conditions The transcript level of the putative *BOR4* gene was considerably enhanced in the roots and leaves of +B seedlings (2.1- and 2.7-fold increase, respectively) when compared with Ct ones. *BOR4* gene has been reported to codify for a transporter to allow an active efflux of B across the plasma membrane. *BOR4* gene overexpression might lead to B excess to be pumped out of the cell and mitigate B toxicity stress in the cytoplasm. It is likely that a large amount of B exported from cells remains linked to the cell wall. In *Arabidopsis*, GFP fluorescence derived from BOR4-GFP strongly localised BOR4 to the plasma membranes of the distal sides of epidermal cells in the elongation zone. This is of strategic importance for directional export of B to avoid high concentrations in the xylem and growing cells and likely to limit B accumulation in the symplasm. *BOR4* is likely to operate at high B concentrations (low-affinity transporter), but not to mediate B efflux under low B conditions. To support this notion, it has been reported that high B-tolerant barley and wheat cultivars are able to maintain low B concentrations in shoots and roots by active B efflux mediated by the overexpressions of *Hv-BOR2* and *Ta-BOR2*, which suggests a positive correlation between these genes and the degree of tolerance among different cultivars in these species. ### Accumulation of putative *TIP5* transcripts under B toxicity conditions and its relation with B compartmentation in the vacuole The expression of the *TIP5* gene coding for a B transporter related with B sequestration across the tonoplast, and *V-ATPase A* and *V-PPiase* genes coding for two H⁺-ATPases pumps in the vacuole membrane, were checked to know if increased B accumulation in the vacuole could constitute a mechanism of B toxicity tolerance. In this study, boron excess promoted the expression level of putative gene *TIP5*, and the number of transcripts increased in the leaves and roots of +B seedlings by 3.3- and 2.4-fold, respectively, in comparison with Ct ones. Aquaporin TIP5 has been related to a decreasing B concentration in the cytoplasm through its transport across the tonoplast membrane and to compartmenting it into the vacuole complexed with B-polyols. Thus at high B levels in culture media, the up-regulation of putative gene *TIP5* might be responsible for the intracellular B allocation, which could contribute to B tolerance in *Citrus*. Pang et al. pointed out this capacity in *A*. *thaliana*, and provided evidence to support that this gene may be involved in B transport to vacuoles. Vacuolar sequestration has also been widely assumed to play a fundamental role in plant tolerance to excess micronutrients and heavy metals. A raised expression level of *V-PPiase* in +B organs, in comparison to those of Ct plants, suggests that V-PPiase is likely involved in the acidification of the vacuole while no effect was recorded on V-ATPase A. The protonmotive force generated by this vacuolar H⁺-translocating enzyme, enhances antiporters involved in B transport to increase their activity and, consequently, B sequestration into this organelle is more efficient, similarly to what is described for sodium tonoplast antiporters during salt stress conditions. Maybe even the action of putative tonoplast antiporters of polyols (due to their ability to complex with B) might play here as secondary transporters energized by the tonoplast H⁺ gradient and its role should not be discarded. The synergism of both responses in the vacuole will probably help to the active B influx in this organelle through the tonoplast, alleviating from B toxic concentrations in the cytoplasm. ## Cellular B allocation under B toxicity conditions: partitioning in soluble and insoluble B fractions in roots and leaves Plants contain boron in both soluble and insoluble forms. According to Liu et al., B extracted in water is likely to be localized in the plant free space, or apoplast, while the remaining soluble B, extracted using organic solvents, belongs to B located inside cells or the protoplast and linked to sugars, alcohols and polyhydroxycarbolates. Both fractions represent not only mobile B, but also the only form in plant tissues that can be re-translocated in the phloem. Finally, the B-insoluble fraction represents the B bound to cell walls linked to peptic polysaccharides. shows the partitioning of B in soluble (water and organic solvents) and insoluble forms in roots and leaves. Insoluble B was mostly the main fraction of B in both Ct and +B seedlings (more than 65.8% of total B), followed by the water-soluble fraction (ranging from 31.8% to 9% of total B;), and lastly by soluble B in the organic solvents fraction (below 7.6%;). However, the seedlings grown under B-excess conditions increased the concentration and total amount of B located in cell walls and in the soluble fraction from the apoplast when compared with Ct plants, which is likely in equilibrium with the external medium. Nevertheless, the concentration and the amount of B extracted with organic solvents, which is attributed to cytoplasmic B, was similar in the organs from both +B and Ct seedlings. As boron exerts its toxicity inside rather than outside cells, the capability to retain increased B amounts linked to cell walls under B-excess conditions enables this genotype to block excess B in an insoluble form, thus preventing its entry to the cytoplasm to protect cells from B toxicity. Accordingly, Nozawa et al. found a negative correlation between the B concentration inside yeast cells and the degree of tolerance to a high external B level, and concluded that cells with lower protoplasmic B were more tolerant to high B levels. Other authors have also observed a lower B concentration in leaf protoplasts of B-toxicity-tolerant cultivars of barley and wheat, which suggests different B partitioning within the shoot as a tolerance mechanism. Hence it is likely that insolubilisation of B outside the cytoplasm might constitute another mechanism of B toxicity tolerance. ## Accumulation of proline in roots and leaves: possible role in antioxidant response to MDA concentration Boron effects on proline concentration have been studied in organs since this amino acid usually accumulates in response to several abiotic stresses. Boron excess caused a significant rise in the proline concentration recorded in the roots and leaves of +B seedlings (50.7% and 33.6%, respectively) when compared to Ct ones. Among other functions, proline acts as a detoxifier of hydroxyl radicals in plants. Accordingly, proline levels enhanced by high B concentrations have been recorded in tomato and pepper. MDA concentration increased by approximately 18.3% and 20.2% in the roots and leaves, respectively, of +B seedlings when compared with Ct ones. The reduced effect of excess B on MDA concentration detected in both organs of +B seedlings indicates a low response to oxidative stress under B toxicity. Likewise, Gimeno et al. did not find any effect on MDA concentration measured in the roots and leaves of Verna lemon trees when grown at high B levels. It has been reported that proline levels are inversely related with MDA concentration in the leaves of several woody species grown under B-excess conditions. Accordingly, it was described that elevated proline avoids the generation of free radical levels as measured by MDA. This coincides with the data presented herein and suggests that, at high external B concentration, Cm seedlings display a sharp rise in proline levels in both leaves and roots, whereas the increased MDA concentration was much lower. This suggests the existence of an efficient antioxidant system in Cm that is able to cope with ROS species generated as a result of B toxicity. As reported by Xiong and Zhu, proline is capable of detoxifying ROS by forming a stable complex with them to, thus, inhibit the lipid peroxidation process. This points at the role of this amino acid as part of a likely efficient antioxidant system able to cope with ROS species generated as a result of B toxicity in Cm plants. However, the function of other key antioxidant enzymes as catalase, superoxide dismutase and glutathione reductase should not be discarded, as they might also play an important role in the efficiency of the whole antioxidant system of the plant. # Conclusion In *Citrus*, boron exerts its toxicity more markedly in leaves than in root system. However, the low degree of B toxicity symptoms in Cm leaves indicates that this species substantially behaves as a tolerant plant to high B levels. This aptitude is likely based on: A) the down-regulation of the main B transport channels in the root, *NIP5* and *PIP1* genes, which blocks the plant’s B uptake capacity; B) although B transport faculty is not repressed, the overexpression of putative gene *BOR4*, facilitates the export of B excess from cells and avoids cytoplasm damage; C) the capacity to hold B in an insoluble form in the leaves mainly allocated in cell walls; D) the compartmentalization of toxic B from the cytoplasm inside the vacuole due to the up-regulation of aquaporin *TIP5* and the induction of the protonmotive gradient in the tonoplast; and, E) the activation of an antioxidant system against oxidative stress through proline accumulation. We thank Isabel Herrero, Carmen Casamayor, Ramón Pardo and Jesús Asensi for help with the laboratory and field works. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MRMC EPM. Performed the experiments: MRMC. Analyzed the data: MRMC BMA DJI MAFG. Contributed reagents/materials/analysis tools: AQ MR. Wrote the paper: MRMC EPM. Reviewed the manuscript: DJI MAFG.

# Introduction Consumers can access different medicines without a prescription in Australia and New Zealand (NZ) respectively despite similar health systems, populations, and medicines schedules. For example, inhaled salbutamol (for asthma) is available without a prescription in Australia but not NZ, and the opposite is true for some vaccines, and sumatriptan (for migraines). Prescription to non-prescription switch (reclassification or down-scheduling) may expose consumers to risks such as inappropriate use or delayed diagnosis. However, benefits include convenient and timely access to medicines for consumers. Empowering patients to manage minor ailments and use doctors more efficiently can benefit the health system. Such benefits may be significant, given the forecast increased pressure on health resources globally. Australia and NZ have long contributed to an international trend to switch medicines to non-prescription availability. While mid-2000s research found that Australia and NZ were relatively advanced in non-prescription availability of medicines, later research showed switch varied over time. Both countries have three non-prescription categories: general sales (unscheduled); pharmacy-only (Schedule 2 in Australia), and pharmacist-only (Schedule 3 in Australia). In the United Kingdom (UK), Aronson suggested that a pharmacist-only schedule with safeguards would increase control compared with pharmacy-only status, and therefore enable some switches. Researching key opinion leaders’ attitudes, Achanta, *et al*., proposed that the United States (US) would benefit from a pharmacist-controlled class of medicines, but the US Government Accountability Office disagreed. Research in two countries with pharmacist-only classifications could provide insight into the effect of a pharmacist-only category on widening consumer access to medicines through switch. To our knowledge, neither Australia nor NZ has published a government position on switch. Advisory committees in both countries consider switch applications, and make recommendations, but the Minister of Health for NZ, and the Secretary to the Department of Health for Australia, or their respective delegates make the final decisions. The National Drugs and Poisons Scheduling Committee (NDPSC) determined medicines and poisons scheduling in Australia until mid-2010. It included ‘jurisdictional’ representatives (from two states, two territories and, for some of the time, NZ) and appointees (including consumer, industry and pharmacy representatives) and was chaired by the Commonwealth (Therapeutic Goods Administration; TGA). Decisions required a majority vote from jurisdictional representatives (i.e. those representing a jurisdiction such as a state or territory). For switch decisions the NDPSC considered: toxicity and safety; risks and benefits; potential hazards; the extent and patterns of use; dosage and formulation; need for access to a substance; potential for abuse; purposes for which a substance is to be used; and any other matters that the Committee considered necessary to protect public health. These considerations were legislated. From 2010, following a review, medicines and poisons scheduling in Australia were separated. The Advisory Committee on Medicines Scheduling (ACMS) advises on medicines scheduling only. Committee composition and switch considerations for the NDPSC and the ACMS are published elsewhere. In brief, the ACMS comprises nine representatives of States and Territories and the Commonwealth government, and up to six independent expert members. For switch decisions, the ACMS considers: risks and benefits; purposes of use and extent of use; toxicity; dosage, formulation, labelling, packaging and presentation; potential for abuse; and any other matters that the Secretary considers necessary to protect public health. These considerations are legislated. Medicines policy in Australia around the time of the research had four central objectives: timely access to the medicines that Australians need, at an affordable cost; quality, safe and efficacious medicines; quality use of medicines; and maintaining a responsible and viable medicines industry. The concept of access to medicines focuses on cost particularly. Scheduling is mentioned in helping address the risks of medicines. Self-help or self-selection of medicines receive minimal attention. In NZ, the Medicines Classification Committee (MCC), a Ministerial advisory committee, recommends classifications of medicines as prescription medicines, pharmacist-only medicines, pharmacy-only medicines or unscheduled medicines. Medsafe, NZ’s regulator of medicines and medical devices, administers this committee. The MCC comprises six members, two each nominated by the Ministry of Health (including the committee Chair), the Pharmaceutical Society, and Medical Association. Recommendations made by this committee are considered by the Minister of Health’s Delegate, along with advice from Medsafe. The Delegate may support the committee’s recommendation, or accept alternate advice from Medsafe. Following publication of the minutes, the decision will be gazetted unless there are valid objections. At the time of the research, the MCC considered the following when reclassifying medicines: consumer convenience; potency (i.e. effectiveness); current availability, therapeutic index; toxicity; abuse potential; inappropriate use (relevant to the condition being treated); precautions; and communal harm. Medicines policy in NZ around the time of the research aimed to deliver to New Zealanders quality, safe and effective medicines; access to medicines regardless of the ability to pay and within government funding; and optimal use of medicines resulting in optimal outcomes. This policy made no mention of self-care, non-prescription medicines or switch. The Ministry of Health had a “better, sooner, more convenient” strategy for moving services from hospital into primary care, and try to keep people healthy in the community. Since the 1990s, NZ and Australia have been attempting to harmonize their scheduling of medicines, a process called Trans-Tasman Harmonization (TTH). Where the scheduling differs for a medicine, the intent has been to harmonize to the least restrictive schedule while considering public health and safety issues and/or jurisdictional needs. Described in 2003 as a ground-breaking event, to our knowledge such harmonization between two separate countries remains unique. No published literature investigates the TTH scheduling initiative, and other research investigating effects on switch have often focused on a limited range of factors or a single medicine. Trans-Tasman analysis provides a unique opportunity to understand influences on switch in similar countries over time, and reflect on international harmonization policy. This paper therefore compares barriers and enablers to prescription to non-prescription switch in Australia *versus* neighboring NZ. # Methods The University of Auckland Human Participants Ethics Committee provided ethical approval (2008/304) for this study. A modified form of qualitative heuristic research was used. It embraced the personal experience, knowledge and active input of the lead researcher (NG, a doctoral student during the research) who has a history of intense involvement and interest in the topic. Similar to the insider (emic) approach typical of much ethnographic case study research, the heuristic approach promised to deepen our understanding of the nature and meaning of her lived experience of the study phenomenon. summarizes how she modified for this study the heuristic approach described by Moustakas. At the outset of the study, NG believed that, on balance, switch with relative safety is desirable, but was mindful of the need to be constantly reflexive around how this position impacted on this research. NG was, at that time, a member of NZ’s MCC (2004–2009). She had supplied switched medicines as a pharmacist, had researched switched medicines, and had observed an Australian National Drugs and Poisons Schedule Committee (NDPSC) meeting (February 2007), and prepared switch applications for NZ (2010–2014). Effectively becoming a co- participant in this study, she drew upon her professional experience to create a milieu conducive to eliciting rich information. To this end she also shared her professional experience toward the end of interviews. She sought to be reflexive by triangulating field notes, including reflective notes, with data collected in both countries from documents and interviews. She used introspection to examine her thoughts and feelings, and subjected these to further scrutiny through skeptical peer review. Participants checked research outputs to enhance the trustworthiness of the research. NG was an experienced interviewer. The other authors were supervisors or an advisor on this work, and informed the study design, monitored its implementation and provided skeptical peer review of analysis and reporting. FK and LE are academic pharmacists who have worked in pharmacy practice research (including switch-related projects) in Australia and NZ. SB is a social scientist working in primary health care. The flow chart in outlines the method used. We sought to maximize variation in the sample, and identify themes that cut across this variation. Key participants were purposively selected therefore from diverse groups, specifically regulatory authorities, pharmacy organizations, pharmaceutical industry, doctors’ groups, consumers’ groups, academia, and the committees considering switches of medicines, as agreed by the research team. Most participants were selected for their knowledge of or involvement in switch according to pre-existing knowledge held by NG, and sometimes referred by other participants (snow-balling), with assent from the research team. Often only one person in an organization was responsible for switch. NG sought consciously to include people known or likely to hold views contrary to her own. The sample therefore was biased towards efficiently including participants who, within particular occupational niches, had the most experience of switch. As Morse explains, “the sample is biased; it *must* be biased”.\[p734\] Initial contact occurred by telephone, then email or email alone for persons already known. Where no such key person was known, the organization was telephoned with a request to interview the most appropriate spokesperson about switching medicines from prescription to non-prescription, followed by email. For pharmacy organizations, typically the employee who was most involved in switch submissions was interviewed as well as a Board member to gain both a hands-on view and a broader understanding. Data was included from two participants who were interviewed for the international part of this study and had experience of switch in Australia. Following informed consent (written and/or oral), NG conducted semi-structured interviews face-to-face or by telephone from late 2009 to early 2012, at a mutually acceptable time and location. NG had previously communicated with 18 participants. Four participants had long- standing relationships with NG. All participants were informed that the research was doctoral research exploring reasons for variation in switch in different countries. Most participants were aware of NG’s role on the MCC in NZ. The questions were iterative, evolving with successive interviews on the basis of emergent knowledge, depending on the participant’s role and on comments raised in the interview. reports the general topics covered. Field notes were taken. The interviews were audio-recorded and transcribed verbatim. The transcription was offered to participants to review. NG conducted analysis and reporting with input from the co-authors (see). Interview transcripts were carefully read and reread. Concepts identified in the transcripts were labelled (coded). Codes were grouped thematically in context, using the qualitative software package, NVIVO 9 for data management. This iterative process segmented themes by country, or commentary on TTH. Then broad areas and themes that emerged from the data before or during coding provided the next level of coding, with further levels as appropriate, e.g. for commentary about specific medicine switches. Many interview segments fitted into general topics and themes. After preparing a full report for each country and comparing the two countries, transcripts were re-read to ensure quotes were used in context and the overall feel of interviews was reflected. Data saturation was not sought owing to the limited number of persons available with switch experience. A NZ participant reviewed the NZ report, an Australian participant reviewed the Australian report, and a participant with long-standing experience in switch in both countries reviewed both reports. Minor modifications resulted from review. NZ documents analyzed included MCC meeting minutes which were obtained from the Medsafe website (1999–2012) and from Medsafe (1990–1999). Australian documents analyzed included records of NDPSC meetings (2000-June 2010) and ACMS meetings and delegate decisions (August 2010-December 2012). The NDPSC and ACMS documents were sourced from the website of the Therapeutic Goods Administration (TGA; Australia’s medicines regulatory agency). Prescription to non-prescription switch decisions and their reasons were summarized into tables of decisions, and summaries of reasons, coded by medicine. All meeting records were reread and checked against the tables and summaries to ensure accuracy. For medicines that emerged strongly in the interviews, specifically sumatriptan, chloramphenicol, orlistat, potassium chloride, calcipotriol and oseltamivir, relevant meeting records were extracted verbatim, coded by medicine and analyzed with interview data to include in the country reports. One case study has been published elsewhere.. Other documents were obtained as needed for analysis and reporting. For Australia, these documents comprised: media reports of the orlistat advertising and mystery shopping, and the Pan Pharmaceuticals recall; the coroner’s report of the potassium poisoning case; and media statements on switch from medical organizations. For NZ, the documents included switch applications. # Results Analyzed interviews comprised 10 NZ, 15 Australian, and two international participants. Participants were from the medicines regulator (n = 3); pharmaceutical industry (n = 8); pharmacy organizations (n = 7); medical organizations (n = 3); pharmacy academia (n = 3); and a consumer organization (n = 2). A further participant was there solely in their committee capacity. Some participants had experiences of switch in both countries, and many had committee experience: from the NDPSC (n = 6); ACMS (n = 1); and/or MCC (n = 2). Interviews were conducted for 25 minutes to 2.5 hours, with most taking around one hour. One NZ industry person declined to be interviewed. A NZ pharmacy organization interview was discarded after recorder failure on the phone interview. ## Overall themes Dominant themes from the interviews were risk averseness or political conservatism in Australia, and an open, flexible environment in NZ. The Australian environment for switch changed, from relatively progressive in switch in the early 2000s to conservative around 2006–2007 onwards. NZ participants also spoke of their country being used to change or where change could happen easily. ## Enablers and barriers Enablers and barriers differed greatly between the two countries, with NZ data demonstrating more enablers and Australia revealing more barriers. Similar enablers or barriers between the countries commonly appeared stronger in one country. For example, small population size as a barrier received several mentions in Australia, but emerged strongly for the less populated NZ. A barrier could also be an enabler, e.g. small population size limited potential sales in NZ, but helped the open, flexible, proactive attitude and appeared to limit political pressure. ## Themes in medicines switch in NZ ### Pragmatic, proactive, open attitude and change Many participants in NZ alluded to an open attitude to switch and flexibility. Examples of this attitude from the committee (based on interviews, meeting minutes and the lead researcher’s experience) included proactively suggesting switch candidates (evident in meeting minutes), and being open to switches that were firsts for the developed world. The regulator allowed an exemption to prescription instead of the pharmacist-only category to overcome the need for labelling changes, and allowed a switch with mandatory training. > *“I think the Chair and the Chair’s willingness to engage with > industry and say ‘come on let’s put in a submission’, get in, advocate > for change, you know it’s risky but it’s required in the NZ > environment because otherwise we’d have nothing to look at, > basically.”* > > *Regulator participant* Many participants (including some from Australia) believed that the pragmatic approach by the MCC Chair enabled switch. The open attitude has sometimes been less apparent, with a seemingly conservative stance in the late 1990s to 2003. Meeting minutes revealed reverse switches (returning to prescription medicine) under TTH for which safety concerns were not apparent, e.g. with oxybutynin (for incontinence). In 2002, NZ’s MCC rejected switching a topical, moderate-potency corticosteroid (for dermatitis and eczema) despite its switch approvals in the UK and Australia. The MCC later (2005) recommended the switch be approved. Some switches have required many meetings to gain approval, despite switching elsewhere (e.g. chloramphenicol for eye infections and omeprazole for heartburn). Others were not approved (e.g. simvastatin for prevention of cardiovascular events). These findings signify the influence of the overall committee and attitudes of individual members. shows the MCC recommendations for progressive switches from 2000 to 2011 in four-year increments. Committee approvals were more common from 2004, when the committee membership changed significantly. > *“…you had to get the right GPs in the committee to make this happen, > they had to be open-minded enough to say that there are lots of > different ways of treating patients, and that doctors don’t have to be > involved in every consultation…”* > > *Regulator participant* Proactivity was seen in non-sponsors driving switches, strongly contributing to NZ’s recent progress. Meeting minutes show that the regulatory authority drove the emergency hormonal contraceptive switch in 2001, and the committee drove chloramphenicol in 2006–2009, with input from the Pharmaceutical Society including developing training for both switches. From 2010 to 2013, a pharmacy retail group, Pharmacybrands (now Green Cross Health), applied to switch seven medicines, despite not sponsoring any. NG believes that NZ’s flexible, open approach was integral to achieving these switches, as discussed elsewhere for topical calcipotriol. An element of change was evident: NZ as a country was open to change, and, as a small country, NZ could change quickly. For example, the MCC (June 2006 minutes) noted a particular switch “*could easily be reversed*” if necessary, helping the switch to be approved. The committee continued to evolve (as shown in minutes and from an interview), and one participant suggested it had become more evidence-based in its decision-making. Two participants commented that NZ generally sometimes made changes too quickly, but did not relate this to any specific switches. ### Trust Interviews, minutes and the lead author’s experience indicate that committee and regulator trust in pharmacy and the consumer enabled switches in NZ. The regulator participant reported increasing committee trust in pharmacy, helped by pharmacy’s performance after switching oseltamivir (for influenza): “*“Pharmacy policed their profession very, very effectively and self-managed their profession very, very effectively, in a way that certainly makes me more confident in other things can go down in the same way….”*” The Pharmaceutical Society regularly informed pharmacy of switches and their responsibilities around providing switched medicines, and developed training to support some switches. The MCC meeting minutes for trimethoprim in 2012 confirm the confidence in pharmacy, with the MCC removing the proposed need for a previous doctor’s diagnosis stating: “*with the correct training there was no reason why pharmacists could not manage patients who present for the first time with a urinary tract infection*”. Furthermore the meeting minutes reported “*… that pharmacist sale of trimethoprim may be more in line with best practice than the prescribing habits of general practitioners*.” Pharmacy and non-pharmacy participants were generally positive about pharmacy’s abilities, citing improved undergraduate training, and re-professionalization of pharmacists. However, most participants who were pharmacists reported occasional poor practice, publicized in covert shopper studies. Both medical participants worried about pharmacy assistant training, and one lamented that pharmacists are less trained in diagnosis than doctors. Recent switches have seen an increase in the use of screening tools and rise of mandatory training. For example, only pharmacists who have completed approved training can supply trimethoprim or administer vaccinations without prescription, using screening tools. Such tools may help generate trust in pharmacy to supply the medicines correctly. Industry participants trusted the MCC, believing its decisions to be evidence- based. ### Global and regional influences An industry participant suggested NZ could be a test market, noting that “*if it tanks*, *it doesn’t matter… we haven’t lost anything*. *If it goes great*, *we’ve got experience then to provide to the Europeans*, *the FDA…*” NZ potentially influences Australia through TTH, with industry sometimes timing the NZ review before the Australian review. Australia affected switch in NZ. Some switches occurred in Australia before NZ (e.g. orlistat for weight loss and fluconazole for vaginal thrush). Reverse switches (up-scheduling) occurred in NZ to harmonize with Australia, particularly around 1999–2000 (e.g. oxybutynin for urinary incontinence and colestyramine for cholesterol lowering). Influence from the UK (particularly), but also the US, Canada and Denmark was evident. ### Financial influences The financial theme arose in all interviews. Document analysis shows that global pharmaceutical companies (the main drivers of switch internationally) have not driven ‘progressive’ switches in NZ since 2009. ‘Progressive’ switches are prescription to non-prescription switches judged to provide consumer benefit over existing non-prescription medicines. The small market and medicine funding policy reportedly make NZ unattractive for pharmaceutical companies, resulting in little or no NZ presence, and reducing the ability and desire to drive switch. Industry, pharmacy and medical participants suggested that subsidized health care limited self-medication. A medical participant considered subsidized care protected consumers: “*…there’s a connection between being uneducated and their financial set up*, *so they’re less likely to be able to buy OTC \[over-the- counter\] medication…*. *that takes some of the fear away*, *because it’s cheaper for them to come to me*.” Subsidized care was highlighted as a safety net in the switch application for topical calcipotriol (for psoriasis), incentivizing patients using large quantities to seek a doctor’s prescription. The transparency of switch and resulting potential for immediate generic competition reportedly discouraged companies from driving switches. Industry participants desired market exclusivity (an exclusive period selling the switched medicine) to encourage switches, given the cost and effort of driving the switch and training. However, financial pressures on pharmacy were thought to encourage their interest in switch. ### The pharmacist-only schedule Most participants considered the pharmacist-only category enabled switches, as requiring health professional contact was believed to reassure the committee. Many recent switches used an exemption to prescription availability through a pharmacist under strict criteria. Such criteria have sometimes included mandatory pharmacist training, used with the emergency hormonal contraceptive, vaccinations and trimethoprim (for urinary tract infections). This model tightens the pharmacist-only availability, minimizing risk, and/or precludes the need for non-prescription labelling. ## Australia Australia progressed in the early 2000s with multiple innovative switches. The Chair and other committee members were reportedly open to switch, and industry personnel could discuss planned switches with jurisdictional members. provides the level of approvals (or recommended approvals after mid-2010) versus rejections for progressive switches in Australia, showing most progressive switch attempts since 2006 have been rejected. Participants suggested that increased conservatism in the NDPSC around 2005/2006 had multiple causes: changes of committee members; an international increase in risk-averseness affecting the TGA; increasingly complex switches; and local events such as orlistat advertising (see below). Industry personnel could no longer discuss planned switches with jurisdictional members. The key themes from interviews were risk-averseness and distrust, financial influences, global influences, pharmacy, and the effect of individuals. ### Risk-averseness or conservatism and perceived lack of trust The related themes of risk-averseness or conservatism and lack of trust arose in most interviews. Industry, pharmacy, and committee member participants commonly described the committee and regulator as risk-averse or conservative. In contrast, the medical and consumer participants did not suggest committee conservatism. The medical participant worried about effects of some switches that had occurred. Perceived risk-averseness in switch and advertising decisions discouraged industry from applying for switches. Various participants suggested political pressure encouraged the risk- averseness. > *“…their life is to avoid risk…. We’ve got to keep our Minister happy > and all these other stakeholders…”* > > *Pharmacy participant* Many participants, including committee members, suggested that jurisdictional members (considered inherently conservative) impeded switch, sometimes reportedly because of instructions from above. > *“…because of … the committee structure of the NDPSC and the voting > arrangements specifically, it is ultimately about a series of > government or bureaucracy agendas… versus a true evidence-based > assessment.”* > > *Industry participant* Although predominantly viewed negatively, such conservatism was sometimes considered appropriate, usually by committee members. Several participants (including industry) suggested the risk-averseness was a push-back response to industry’s push. > *“…it’s better, in this area of healthcare, to tread warily. And in > fact the resistance of those state jurisdictions to that change is > almost the counterbalance to the arguments made by the sponsors for > the change…”* > > *Committee member* The instructions from above were not always conservative. A former NDPSC member spoke of the progressive effect on switch from a “*liberal*” chief pharmacist who “*certainly gave riding instructions to his people that did represent \[his state\]*.” Participants from industry, pharmacy and medicine described pharmacists as conservative, cautious or under-confident with switched medicines. Some participants reported that this limited sales. > *“…if you don’t prepare the profession, the profession clinically is > very, very conservative, they’re scared, so they’re not going to touch > anything like that.”* > > *Pharmacy academic* Although industry was not described as risk-averse or conservative, one participant commented that fear of medical backlash prevented oral contraceptives switching, and that potential pharmacy backlash delayed nicotine replacement moving into supermarkets. Pharmacy organizations reportedly opposed advertising of most pharmacist-only medicines and some prescription to pharmacist-only switches. Interviews indicated the committee lacked trust in consumers, pharmacy and industry. One participant reported that Australia and NZ viewed the consumer differently. He reported that the MCC considers “*what harm can come to a reasonable NZ consumer if they buy a pack of this medicine from a pharmacy or a supermarket*, *and they read the label?*” He considered that the Australian committee starting position is: “*could the consumer get into harm if we lock them in the room with a bucketful of this medicine in the dark?*” Some committee participants suggested that the committee lacked confidence in pharmacy, citing concerns about variable pharmacy behavior, and pharmacy assistants supplying medicines rather than pharmacists who were often busy. > *“…it’s almost as if that they allow it to go S3 \[pharmacist-only\], > they’re sort of happy well that if the pharmacist doesn’t do his job > properly, it probably wouldn’t be too bad.”* > > *Committee member* Meeting records for oseltamivir (for influenza) indicated concerns about misdiagnosis with pharmacist-supply. Meeting records for orlistat showed that the approval for advertising of this pharmacist-only medicine was revoked with the rationale including that advertising to consumers “*increased pressure on pharmacists to provide orlistat to consumers*. *This in turn had the potential to result in inappropriate patterns of use…”* Various participants suggested that distrust in consumers and/or pharmacists arose from certain events (see “political influence” below), including mystery shopping in pharmacies; a move towards discounting medicines (potentially limiting time for consultations); personal experience of the committee members or their families when shopping in pharmacies; a desire to avoid risk (see “political influence”); and jurisdictional members’ awareness (through their work) of inappropriate pharmacist behavior. However, the recommendation to reclassify chloramphenicol (for red eye) in 2009 showed confidence in pharmacy with the NDPSC records reporting that several committee members contended that the “*rate of misdiagnosis was unlikely to differ significantly whether it was a pharmacist or a GP doing the diagnosis*”. ### Political influence The recent reported risk-aversion is not entirely unexpected, given media- fuelled events cited by participants, including a major recall instigation, and a child’s death through potassium chloride overdose. > *“… we live in a political climate…. And I think where there’s safety > issues, people are more informed, and they’re more likely to take > those issues public. And governments I guess are a little bit > concerned about that… But I don’t necessarily see that as a bad > thing.”* > > *De-identified quote* Orlistat (for weight loss) became the watershed moment in advertising approvals. Orlistat advertising during a television program aimed at teenagers, stimulated an official advertising complaint (partly upheld) and media interest. A subsequent Australian Consumer Association covert shopping study found that 24 of 30 pharmacies sold orlistat to a woman with a lower Body Mass Index than the licensed indication. NDPSC records noted that submissions from professional organizations, pharmacy boards and consumer organizations “*strongly argued*” against advertising because of inappropriate pressure on pharmacists and potential to mislead consumers. The NDPSC considered consumer advertising of orlistat could “*…lead to inappropriate patterns of use…*”, removing the advertising approval to “*…protect public health*”. Since 2007, all applications to advertise pharmacist-only medicines have been rejected noting “*no public benefit*”. > *“…the consumer people jumped on them from a great height and it > became a political thing rather than anything that was based on > evidence…”* > > *Pharmacy participant* Most strongly voiced about the committee, politics also arose from medical and pharmacy organizations, and was used by consumer organizations to highlight pharmacy or advertising concerns. The medical participant interviewed considered that the emergency contraceptive pill and vaginal antifungals (both long- standing switches) should be prescription-only, given that doctors were usually accessible for urgent matters. Conversely, a pharmacy participant considered doctors very inaccessible. ### Committee issues Committee conservatism is considered above. Various participants indicated that NDPSC decision making was sometimes not evidence-based, even with the usual independent evaluation. Participants suggested a large workload, inadequate preparation, the regulator’s influence, and nature of the membership affected decisions. > *“… it seems that no matter what the so-called experts think, it’s all > up to the bureaucratic representatives, and I’m not too sure whether > that’s good…”* > > *Pharmacy participant and former committee member* > *“\[NDPSC members are\] also State employees, so they’ve got a direct > line into the State government, which means they’ve also got that > downwards pressure, that they’re not there as experts, they’re there > as representatives.”* > > *Committee member* Some participants proposed that coal-face experience on the NDPSC with a community pharmacist and practicing general practitioner would be helpful. Several committee participants reported emotive decisions, and sudden turns in meetings. > *“…\[committee members\] wait for one to take the lead … and the rest > would just rush in to support.”* > > *Committee member* > *“…you take one committee, and they’ll make one decision on one day, > and you bring them back two months later, and they might come to a > different decision … it just depends on who’s most persuasive on the > day.”* > > *Committee member* Various participants disagreed with the rejection of sumatriptan (for migraines), and one reported that the sponsor would address the committee’s issues and then new ones would arise, as supported by meeting record analysis across four meetings. A committee participant noted committee concerns about very rare events with sumatriptan, and suggested that trust was an issue: “*“\[The committee\] took the view that people with severe headache would go to the pharmacy and would buy sumatriptan, or pharmacists would want to sell them sumatriptan, so we’re in a trust question again… and then a complete lack of trust in the Australian consumer behaving reasonably.”*” The change in committee from NDPSC to ACMS reduced the committee size and workload, and changed the final vote from jurisdictional members only to all committee members, but jurisdictional representatives remain the majority of the ACMS. Subsequent switch and advertising decisions (Figs.) suggests conservatism remains. A participant added that government health policy initiatives, such as taking more personal responsibility for health, had yet to translate into self- medication, and no specific government policy supports switch. ### Financial effects Pharmaceutical companies were discouraged by financial factors, particularly advertising limitations, lack of market exclusivity, and to a lesser degree a small market. Industry was frustrated at the “*product graveyard*” of switch without advertising (“*S3 non-advertisable*”). While about one-third of participants expressed concern, for example, about “*sensationalist*” advertising, many noted a need for advertising, but preferred advertisements to be informative. > *“…once it falls out of the prescription only category … there’s not > an adequate way of having the consumer realize these products exist > and are available. If the pharmacist fulfils their full professional > responsibility in its sale, then there shouldn’t be a problem…”* > > *Pharmacy participant* The change from the NDPSC to the ACMS in mid-2010 did not appear to change the advertising stance, despite some positive evaluator recommendations, returning to the theme of lack of trust: “*“There is a distrust that the pharmaceutical industry would advertise responsibly.”* *Committee member*” Transparency and lack of market exclusivity allow immediate competition by manufacturers of generic equivalents, compounded by the inability to advertise: “*“…the pharmacist would recommend the in-house \[generic\] brand over the branded products \[whose manufacturer\] does all the training, provision of information, education…”* *Industry participant*” Financial effects also influenced consumers, with some pharmacy participants noting that pensioners with time for doctor visits and low co-payments (AUS\$5.80 per item concession in 2012 \[€4.00\]) would prefer not to pay for non-prescription products. A high co-payment (AUS\$35.60 per item in 2012 \[€24.80\]) by consumers without a concession card was not raised as enabling switch. ### Global or regional effects A participant suggested companies would only launch in Australia to aid switch elsewhere. Certainly, the Australian switch attempts of orlistat (for weight loss), sumatriptan (for migraines) and montelukast (for hayfever) were early, although the last two were rejected. Global effects such as company mergers, OTC structure and politics between the prescription and non-prescription divisions were also said to affect Australian switches. Although participants suggested attempting switch in NZ first to try to influence Australia, this influence seemed minor, and last occurred in 2009. During 2000–2012, one of 15 progressive switches in NZ and one of nine in Australia directly arose from TTH. Third-party switches were exemption to prescription rather than scheduling changes *per se*, so were not considered in Australia under TTH. One committee member noted limited influence from NZ owing to the small market size. ### The pharmacist-only schedule and role of pharmacy Although participants supported having a pharmacist-only schedule and considered it facilitated switch, its benefit was moderated by advertising restrictions and sales of generics instead of the applicant’s branded product. Additionally, a regulator participant observed: “*“…S3 \[pharmacist-only\] is an enabler, but … it has its limitations, because if you stuck everything in S3, it wouldn’t work.”*” A dichotomy appeared around pharmacy even from the same participants. Many participants emphasized pharmacists’ cautious approach, good training, and the Quality Care Pharmacy Program. The medical participant volunteered that pharmacists are generally “*quite conservative*”. However, this participant also espoused that pharmacists cannot examine a patient, take a history, order pathologies, or diagnose. In contrast, pharmacy academics believed pharmacists did diagnose minor ailments. Various participants voiced concerns about privacy, and pharmacy behavior sometimes being substandard (including concerns about pharmacy assistants and pharmacist workload) and the committee sometimes worried about pharmacy’s role. > *“Unfortunately, in reality, it’s not always the pharmacist that you > get to speak to, it’s often the 16 year old pharmacy assistant…”* > > *Committee member* # Discussion Ours is the first research to comprehensively compare two similar countries in factors affecting switch of medicines. Participants in both countries identified some common barriers (small markets, transparency, and no market exclusivity), and enablers (progressive individuals and the pharmacist-only classification), and some differences. Aronson and Gilbert and others thought that the pharmacist-only category may enable non-prescription availability, but the US Government Accountability Office was less convinced. Our research supports that the pharmacist-only category is enabling, particularly with NZ’s flexible approach including exemption to prescription and mandatory pharmacist training. Gauld and others described how NZ’s flexibility and pharmacist-only supply enabled calcipotriol to switch. Current differences between Australia and NZ were suggested to reflect changes in committee membership and increasing risk concerns (Australia) and flexibility, proactivity, and third-party switch (NZ). The committees differ in composition: most members are public servants in Australia (suggested to be potentially influenced by political forces) *versus* primarily health professionals (albeit chaired by the regulator) in NZ. ‘Patch protection’ could arise from health professional organizations nominating MCC members. No participants suggested NZ was too liberal. NZ rejected some switches, and, like NZ, the UK, Germany and Sweden, has reclassified sumatriptan. In 2014, the Associate Minister of Health in NZ and the Chair of the MCC were both positive about switch in NZ, and a NZ Medical Association spokesperson stated that the balance of medications available without a prescription was “*reasonably right*”. In contrast, the Australian Medical Association has publicly opposed reclassifications, and stated that pharmacists are not trained to diagnose, and that general practice nurses should treat minor ailments instead. Duckett opined that Australia had doctor-centered care with a paternalistic focus. Some perceived differences appeared between the two countries in trust in consumers and pharmacy. We found no clear evidence of important differences between the countries in pharmacy or consumers. While Australia has a quality program that most community pharmacies have signed up to and which shows many pharmacies have provided good care, other events and perhaps individual perspectives have seemingly outweighed the gains. One little study found small similarities and differences in behavior between pharmacies in NZ and Queensland, Australia. Pharmacy variability in mystery shopping was evident in both countries. The last robust covert shopping study on reclassified medicines in NZ reported reasonably positive findings, while more recent studies in Australia have continued to show variability between pharmacies. Although *ad hoc* mystery shopping in NZ found all six pharmacies refused inappropriate requests for orlistat, in contrast to the Australian experience, limited conclusions can be drawn given the methodology and small numbers. Research on oseltamivir in NZ suggested pharmacists took their responsibilities seriously and pre-switch concerns did not appear to eventuate. Trust in NZ may arise from a cultural perspective; the Organisation for Economic Co-operation and Development (OECD) described NZ as *“…an open society which engenders trust*.” Attitudes to risk may be similar between NZ and Australia, but NZ is more open and liberal economically than Australia. As both committees varied over the period, a strong national influence seems unlikely. Lack of trust in Australia was possibly more indicative of increasing risk averseness in Australia and media coverage of events in Australia than poorer behavior than NZ. Distrust increases both perceptions of risk and unacceptability of risk. Third-party switch (driven by neither sponsor nor government) accounted for the most recent differences between the countries, occurring often in NZ but not Australia. However, much of this took place after the interviews were conducted. We are unaware of pharmacy retailer-driven switch occurring elsewhere. The UK has switched medicines that Australia and NZ has not. Participants identified important barriers to switch in both countries. If self-care becomes policy to help widen consumer access and reduce pressure on health resources, market exclusivity and advertising (in Australia) could help. In Australia, our findings suggest that confidence in pharmacy and industry might need addressing. Research post-switch is lacking internationally, and possibly contributes to the differing judgments reported here. Such research could inform committees in both countries, but would be costly, so it could be incentivized with market exclusivity, as in Japan. Some participants’ views on the Australian committee resemble commentary about US advisory committee processes by Brass and Hiatt. These authors expressed concerns about US advisory committee processes for medicines, e.g. high workload, inadequate member preparation, insufficient expertise, and not being evidence-based. They suggested carefully selecting committee members, correcting misinformation during the meeting, and feedback and training for members. After reviewing transcripts of US committee meetings for switches, Nguyen, *et al*. recommended member training and structured committee processes. We suggest that a committee of carefully selected experts, with critical appraisal skills and good process, including training, is likely to aid robust decision- making. Literature on the dynamics in health-related committees is limited, but others have shown that individuals can affect groups and ‘groupthink’ can affect judgment, heighten biases and sacrifice quality decision making. Experience affects decision making as alluded to by participants, and NZ’s positive experience with oseltamivir, and Australia’s negative experience with orlistat may have influenced later decisions. ## Limitations and Strengths Participant recall and biases can affect interview findings. Meeting records are also inherently limited, being a summary, and tension, individual dynamics and performance are not documented. Meeting records can sometimes also be inaccurate. We combined interviews and meeting records to minimize these limitations. Two participants sometimes represented the same organization, or the same role (e.g. two pharmacy academics in Australia, or different committee members). Such interviews differed, partly from the semi-structured approach, but also reflecting different involvement and experiences in switch. More interviews might add information, but the number of people involved in switch is limited, and we interviewed the most appropriate person(s) from each organization. Given the wide and long-standing experience of switch from most of our participants, and the long interviews, major barriers and enablers should have arisen for both countries. Adding less-informed participants probably would have provided little extra information and potentially more speculation. Other stakeholders, such as academics in general practice, practicing community pharmacists or health policy experts, could add different viewpoints. However, they may have little experience of reclassification and we sought participants with the most experience, but also including medical and consumer views. We included the key people involved in reclassification from regulators, pharmacy organizations, industry and committees considering switch. Achanta, *et al*. interviewed 18 key opinion leaders from four countries, and used statements from another 18 key opinion leaders in considering how to improve US regulatory processes for non- prescription medicines, including switch. Whether any of their participants were switch committee members was unclear, but participant experience included medicines regulator work, academia, industry and consumer organization work. The timing of interviews would have affected content. Most interviews were conducted in 2010 and 2011. Thus, little recent NZ pharmacy-retailer driven switch activity was reflected in the interviews. Australian interviews preceded extensive experience with the new ACMS, but some concerns expressed about the NDPSC composition may continue with the jurisdictional member majority remaining. Advertising and switch approval data do not indicate the ACMS has become more open to switch than the NDPSC was just before the change. The heuristic approach uses the knowledge, experiences and insights of the lead researcher. Her long-standing involvement in NZ switches from multiple perspectives and awareness of switch in Australia and attendance at an NDPSC meeting facilitated the research, but has potential for bias. Use of documents for triangulation, quotations, participant reviews of country reports, and skeptical peer review helped to limit the bias. A different interviewer might have elicited some different findings, although many of our findings (e.g. conservatism in Australia and evidence-based decisions and flexibility in NZ) came from multiple participants, were supported by meeting records, or both. Participants were selected based on their direct involvement in switch (e.g. consumer and medical voices, Presidents of pharmacy organizations, industry personnel most involved in switch) and not because they were known to NG or known to share her views. The interviewer’s relationship with some participants might have affected their responses, possibly providing a level of trust, encouraging sharing, or shaping answers to please the interviewer. However, participants were highly experienced, senior people, most of whom were not well known to NG, and NG did not share her views until late in the interview to help mitigate this possibility. Participants checking the country reports did not indicate significant concerns. Further research could investigate realized benefits and risks of switch in Australasia to understand consequences of the differences seen. Future research could include interviews of personnel within the States and Territories who influence switch, and explore the influence of the change from the NDPSC to the ACMS. We recommend further research into optimizing committee performance, including the ideal committee constitution for considering switch. Observing meetings and interviewing all committee members may add insight into committee dynamics and differences between the countries. Observing different committees considering switching the same medicine may provide useful insight. # Conclusions Multiple reasons appear to lie behind differences in medicines switch (and hence potential for consumer self-care) between two similar neighboring countries; awareness of these may shape future harmonization policy implementation. Reasons include suggestions of increasing conservatism and political influences in Australia, and flexibility and the rise of ‘third-party switch’ in NZ. Pharmacist-only scheduling appears to aid switch, particularly where a flexible approach is used, but barriers to switch, including perceptions of pharmacy and consumer behavior, lack of market exclusivity and small market size, can moderate such effects. Consumers and the health system in NZ may be benefiting from its environment enabling switch, which allows more opportunities in self-care or shared care than consumers have in Australia. # Supporting Information The contribution by participants is gratefully acknowledged. We are also grateful to Dr Stewart Jessamine (Medsafe, NZ), an advisor on this project, and Dr Linda Bryant, a supervisor of the research. NG had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. [^1]: NG was a member of the Medicines Classification Committee (2004–2009), did project work for Medsafe (2010), and worked on switch for Pharmacybrands (now Green Cross Health) and Pharmacy Retailing. Since the research was conducted and reported in her Ph.D. thesis, NG has received travel reimbursement for speaking at the Australian Self-Medication Industries Conference (2013), and has consulted for Novartis Consumer, Douglas Pharmaceuticals, and the Pharmacy Guild of NZ. Since submitting the first draft of this paper, NG has contracted with the Australian Self-Medication Industries. NG is a member of the Executive of the Pharmaceutical Society of New Zealand (since 2013). The other authors have declared that no competing interests exist. This does not alter the authors’ adherence to PLOS ONE policies on sharing data and materials. For reasons of participant confidentiality, transcripts cannot be made available. Documents used are available from the websites of the Therapeutic Goods Association or Medsafe. [^2]: Conceived and designed the experiments: NG FK SB LE. Performed the experiments: NG. Analyzed the data: NG. Wrote the paper: NG FK LE SB. Supervised the data analysis: FK SB LE.

# Introduction Iconicity refers to the resemblance between the form of a linguistic signal and its meaning. For example, onomatopoetic words such as *screech*, *beep* and *bang* resemble the sounds they describe. Iconicity also involves crossmodal associations, such as when the high front vowel /i/ is associated with the perceptual idea of “smallness”. Similarly, the made-up words *kiki* and *takete* are reliably matched to spiky shapes, whereas *bouba* and *maluma* are matched to round shapes. In stark contrast to the traditional assumption that languages are dominated by the principle of arbitrariness (no direct connection between form and meaning), more and more research is uncovering that languages harbor a considerable degree of iconicity in their vocabularies. Researchers have already noted that iconicity has applications for the development of product and brand names. For example, participants associate the hypothetical name *Nullen* with a thicker type of ketchup than *Nellen*, they associate *Usab* with a darker beer than *Esab*, and they associate *Frosh* with a creamier ice cream than *Frish*. These studies clearly show that consumers use the name of a product to make inferences about its qualities, which is relevant to product valuation as we know that consumers generally like to see their expectations confirmed. In most studies on iconicity in advertising, researchers exploit iconicity in brand names by creating artificial words to see how these are associated with specific product dimensions. Here, we do the reverse: we take a phonological contrast that is attested in an actual advertising campaign, using the advertising campaign to derive predictions for different experiments. Rather than associations between speech sounds and single perceptual dimensions, we are specifically interested in the capacity of iconicity to involve multiple crossmodal associations. Research in the field of metaphor in advertising can illustrate this point. Experts have proposed that one of the advantages conferred by metaphors in advertising is that they allow “multiple, distinct, positive inferences about the advertised brand” (p. 7). That is, metaphors can highlight a product’s qualities in an indirect fashion, with one and the same metaphor having multiple loose interpretations that the consumer is able to select from and/or enrich for themselves. We want to suggest that iconicity might also work like this. For example, the pseudoword *kiki* (as opposed to *bouba*) does not inherently mean “spikey” unless the task highlights this dimension. In other contexts, the high front vowel /i/ is associated with smallness or bitter taste. Thus, we envision iconicity as *latent* and *pluripotential*—the same phonological pattern can have multiple distinct meanings, depending on the context. In this study, we explore this latent and pluripotential quality of iconicity by using an ecologically motivated example from a Korean advertising campaign. Harkness details the multiple semiotic strategies Korean companies use to advertise the “softness” of Korea’s most famous liquor, soju. This liquor takes up 97% of the spirits market in South Korea and is frequently claimed to be one of the most popular spirits of the world (The Guardian, <https://www.theguardian.com/lifeandstyle/wordofmouth/2013/dec/02/soju-popular- booze-world-south-korea>, accessed July 27, 2019). What is relevant for the present discussion is that soju has undergone drastic changes over recent decades. Whereas most sojus in the 1980s were 30–35% in alcohol content, 20% is now considered standard strength, with some fruit-flavored sojus going as far down as 14%. Harkness details how alongside this deliberate diluting of soju to gain mass appeal, particularly from female customers, there were a number of changes in advertising strategy. For example, a particular 1982 commercial for soju of the Bohae brand showed middle-aged men on a construction site eating skewers of meat against the backdrop of a bonfire. A male voice-over with a deep voice advertised the soju with heavy guitar music playing in the background; the entire color palette of the commercial was soaked in brown and red. Compared to this, modern soju commercials emphasizing its “soft” qualities often feature energetic young women, frequently speaking with stylized, high-pitched voices. The color palettes have changed to light green and blue, signaling freshness and building up associations with spring and nature. Amidst these changes, the Jinro company also used a form of what they themselves called “voice marketing” during a 2008 advertising campaign for their Chamisul brand of soju. In Korean culture, after somebody downs a shot of soju, they may utter an interjection that in English can be written as *khu* (in Korean script: 크; in the International Phonetic Alphabet: \[k^hɯ\]). This sound is produced with an aspirated velar stop (a consonant produced at the back of the mouth) and a high back vowel that is unrounded (produced at the same position as \[u\], but without lip rounding). Koreans may utter this sound as a vocal gesture of disgust, or in this case to index the bitterness or intensity of the alcohol they just tasted. However, the Chamisul brand deliberately avoided *khu* in their “voice marketing” and instead used the alternative variant *khya* (캬, \[k^hja\]) on its print ads. *Khya* is a recent innovation which Harkness describes as representing a “softening and brightening” of *khu* (p. 20). This variant form indexes a milder taste, a lighter, less intense drinking experience and, according to Harkness, is preferred by female drinkers and increasingly by younger men too (p. 21). In contrast to *khu*, the “softer” variant *khya* is produced with an open low-back vowel \[a\]. Other soju brands have also made use of these sounds in their advertising, including a 2015 television commercial for the citrus-flavored soju Sunhari Chum Churum produced by Lotte Chilsung Beverage. If the *khya/khu* contrast is indeed iconic, this would fit with various studies that have found reliable associations between speech sounds and tastes. For example, Gallace and colleagues found that potato chips are judged to be more *takete* than brie cheese, whereas *maluma* is reliably associated with sweet tastes. In our experiments, we detach the *khya/khu* sounds from their local Korean context and play them to naïve American English, German, Spanish and Chinese listeners who have relatively little to no exposure to Korean culture. We want to know whether these two sounds have multisensory associations consistent with the “voice marketing” campaign of the Jinro company, and whether these associations are accessible to non-Korean speakers. If listeners from other cultures have the same mental associations as evidenced by Korean marketing campaigns, this would suggest that the *khya/khu* contrast is indeed iconic, since these listeners have no arbitrary language-specific conventions to base their judgments on. Our experiments thus follow the frequently adopted strategy to demonstrate iconicity via testing the cross-cultural transparency of particular sound patterns. In doing so, our study enriches understanding of iconicity as a marketing strategy since the *khya/khu* contrast is an ecologically valid example of a sound distinction being actively used in real- world marketing for major liquor brands. Previous experimental research on iconicity in marketing has tended to use only made-up words, and it has tended to focus only on brand names. In addition, we demonstrate *latent iconicity* and *pluripotentiality* by showing that the same contrast (*khya* versus *khu*) has multiple distinct sensory, emotional, and social associations. For the following experiments, we asked two Korean native speakers (one male, one female) two produce the *khya*/*khu* interjections while simultaneously imagining that they had just downed a shot of soju. These two sounds were then used for a series of experiments where we vary the semantic dimension tested, with predictions for the different dimensions derived from the Korean advertising campaigns. Specifically, we asked American English native speakers which ones of the two sounds is “softer” or “harder” (Experiment 1), based on Harkness’s description of *khya* being a “softening” of *khu*. In Experiment 2, we extended this task from hardness/softness to roughness/smoothness, which is the most salient dimension of touch that is furthermore perceptually associated with the “hardness”/”softness” dimension. In Experiment 3, we extended the task to the social dimension of gender, asking participants which one of the two sounds is more “male” or “female”. Here, following Harkness’s account of the Korean advertising campaigns, we expect *khya* to be perceived as more female than *khu*, which correspondingly should be more strongly associated with masculinity. For Experiment 4, we tested the dimension of alcohol content, asking participants which one of the two sounds fits better with a 15% or 30% liquor. Following the use of these sounds in Korean advertising campaigns, we expect *khu* to be more strongly associated with the 30% liquor than *khya*, which should be more strongly associated with the weaker liquor. Experiment 5 extends the task to the domain of taste by asking participants which on the two sounds is more “bitter” or “sweet”. Finally, Experiment 6 extends the task to the dimension of pleasantness, where we predict *khya* to be more pleasant than *khu*. This prediction is motivated because several of the other dimensions tested in the other experiments have pleasantness (or unpleasantness) as a common denominator. For example, studies found that people find soft and smooth surfaces more pleasant than hard and rough ones, and sweet tastes are preferred over bitter tastes. More generally, Experiment 6 is motivated by the fact that there is an increasing number of studies finding evidence for the presence of affective iconicity in language, i.e., the association between speech sounds and pleasant or unpleasant feelings. After demonstrating latent iconicity and pluripotentiality in this cross- linguistic listening task for American English listeners, we then replicate the association between *khya*/*khu* with softness/hardness tested in Experiment 1 in a listening experiment with Spanish, German, and Chinese listeners. # Experiments: General methods This study was approved by the University of Birmingham Research Humanities and Social Sciences Ethical Review Committee (ERN_17–0040). Consent was obtained electronically. ## Participants For the first series of experiments, all of our participants were native speakers of American English. lists the participants across the different studies and their ages. All participants were recruited via Amazon Mechanical Turk and received 0.25 USD reimbursement. Amazon Mechanical Turk is known to be a valid tool for collecting behavioral data and performing linguistic experiments. Approval rate is a strong factor in data quality, and so we only allowed workers with an approval rate of 93% or above to participate in our study. Since we are interested in the responses of naïve listeners, we excluded participants that had spent time in Korea and/or that had studied Korean as a second language. We had to exclude 10 participants (9%) for Experiment 1, 11 participants (10%) for Experiment 2, 7 participants (7%) for Experiment 3, 8 participants (8%) for Experiment 4, 48 participants (24%) for Experiment 5, and 65 (32%) for Experiment 6 because of these criteria. We additionally assessed whether the results held with even stricter exclusion criteria, where we excluded all participants who indicated to have tried soju or who reported to have at least one Korean friend. All results below hold even if these stricter exclusion criteria are applied. ## Stimuli The *khu* and *khya* utterances were recorded with Praat (created by Paul Boersma and David Weenink, <http://www.fon.hum.uva.nl/praat/>) on a MacBook Air using a head-mounted microphone. Both speakers (one male, one female) were 25 years old and self-reported to be speakers of the Seoul dialect. For each speaker, the two stimuli were approximately equal in length (female *khu* 672ms, *khya* 697ms; male: 687 / 691ms) and amplitude (female: 78dB / 72dB, male: 75dB / 77 dB). For the listening experiment, we added a 300ms silent pause to the beginning of each stimulus. It should be emphasized that the *khu* and *khya* renditions we got from our speakers were, in fact, entirely voiceless. That is, there was no glottal fold vibration and no pitch associated with them. In addition, there was no palatalization of the *khya* sound, i.e., the sound more resembled the transliteration *kha*. Due to the absence of any voicing, participants must make any judgments on these sounds based on the spectral characteristics of the sounds, not pitch. ## Experimental procedure All experiments were run online via Qualtrics. Participants were instructed as follows: > “On the next screen, you will be presented with two audio files. > Listen to both files and answer a question. We prefer that you listen > to each file only once, but you are free to listen multiple times.” After clicking NEXT, one of two trials appeared on the next screen with two sound files that had to be clicked in order to be played. The sound files were labeled “first” and “second.” Below the sound files, we presented the question, “Which one sounds ‘softer’, the first or the second?”, on top of two buttons labelled “First” and “Second”. In Experiment 1, for half of the participants, the question was changed so that participants were asked which sound file sounded “harder” rather than “softer”. Thus, the main condition variable (soft versus hard) was a between-participants variable. Each person was played the male and female pairs back-to-back, with male and female versions presented in a randomized order. Within each pair, the order of *khya* and *khu* was also randomized. The basic setup was used for all other experiments, except for Experiment 4 on alcohol content. In contrast to dichotomies such as “hard” and “soft,” alcohol content has no associated verbal antonyms. We thus asked the question “Which sound do you associate with the 15% liquor?” to half of the participants, and the question “Which sound do you associate with the 30% liquor?” to the other half. However, to make sure that there is some form of binary opposition (consistent with the task), we told participants that they need to compare two liquors: one that was 15% and one that was 30%. The precise percentage values chosen for the alcohol study approximate the shift in the alcohol content of soju from the typical 1980s versions to the low alcohol sojus of today. We also felt the need to contextualize the task to participants with more detail as alcohol content frequently correlates with taste properties (for example, beverages with around 5% are often beers and beverages with around 12% are often wines). To reduce the impact of these secondary associations, we told participants that the 15% and 30% liquor were identical in taste and only differed in alcohol content. In addition, we referred to the drinks as “rice liquor” because we presumed that it would be unfamiliar to our English-speaking participants, who would therefore not have any strong preconception about taste. It should be easier for participants to imagine that unfamiliar liquors at 15% and 30% are similar in taste, compared to, for example, 15% and 30% wines. The reason for conducting a between-participants experiment, and furthermore for splitting the different semantic dimensions across different experiments, was that we wanted responses to be as uninfluenced from each other than possible. If a participant, for example, chose the “soft” option for *khya* on a first trial, and then was asked which one was “smooth,” their response would likely be influenced by the previous choice. Moreover, the task presumably felt quite abstract to participants (who have never heard these sounds and probably thought that they are guessing), which is why we wanted to make the experiment as short as possible. Thus, to avoid order effects and to keep the experiment short, all factors in our experiment (except for the Voice Gender variable) were between- participants factors. ## Statistical analyses For all statistical analyses, we used the R statistical programming environment version 3.3.1. All data and code are available in the following Open Science Framework repository: <https://osf.io/jzqg9> Throughout all analyses reported below, we used logistic regression because our dependent measure was categorical. We furthermore used *mixed* logistic regression because our experiment included repeated measures—two responses per participant (one for each stimulus pair). We included the factor “Condition” (soft versus hard, rough versus smooth, male versus female etc.) and “Listener Gender” (male participant versus female participant) as between-participants fixed effects. We included the factor “Voice Gender” (whether the stimulus set was male or female) as within-participants fixed effect. We additionally fitted Condition \* Listener Gender and Condition \* Voice Gender interactions. Given the role of gender highlighted by Harkness, it is plausible that participants might have different preferences for *khu*/*khya* depending on whether they are male or female, and depending on whether they are listening to a male or female stimulus voice. All categorical fixed effects were sum-coded to facilitate the interpretation of interactions. We included “Listener” as random effect (no random slopes were fitted as the critical effect in question “Condition” was a between-participants variable). All *p*-values are based on likelihood ratio tests (deviance tests) of the model with the effect in question against a null model without the effect in question. Throughout the paper, we report marginal *R*^*2* values for the model (variance attributed to the fixed effects only). # Results ## Experiment 1: *Softness* Participants who were asked “which one sounds softer?”, picked *khya* 73% of the time and *khu* 27% of the time. Participants who were asked the “harder” question showed the reverse pattern: they picked *khu* 79% of the time and *khya* 21% of the time. There was a statistically reliable main effect of Condition (*χ*^*2*(1) = 58.21, *p* \< 0.0001). (left two bars) displays the average proportion of *khya* and *khu* responses in relation to hard and soft questions (averaged over the male and female voices). There were no reliable main effects of Listener Gender (*χ*^*2*(1) = 1.46, *p* = 0.23) or Voice Gender (*χ*^*2*(1) = 1.04, *p* = 0.31). However, there were reliable interactions between Condition and Listener Gender (*χ*^*2*(1) = 8.2, *p* = 0.004) and Condition and Voice Gender (*χ*^*2*(1) = 31.5, *p* \< 0.0001). Specifically, male listeners showed a stronger difference between the two types of questions. Whereas men picked *khu* 85% of the time when asked the “harder” question, women picked *khu* only 64% of the time. For the voices, the gender effect went in the opposite direction: Averaged across both listener genders, participants picked *khu* 93% of the time when asked the “harder” question for the female voice stimuli but only 58% of the time for male voice stimuli. Overall, the logistic mixed effects regression model described 55% of the variance in *khu* versus *khya* choices (marginal *R*^*2*). 54% of the overall variance in *khu*/*khya* responses could be attributed to the effect of Condition and its interactions. The results confirm the central prediction that *khya* is associated with softness, and *khu* with hardness. This was the case even though our listeners had not been exposed to the Korean language or Korean soju-related rituals. ## Experiment 2: *Smoothness* Participants who were asked, “which one sounds smoother?”, picked *khya* 69% of the time and *khu* 31% of the time. On the other hand, participants who were asked the “rougher” question showed the reverse pattern, picking *khu* 66% of the time and *khya* 34% of the time (see). There was a statistically reliable main effect of Condition (*χ*^*2*(1) = 26.02, *p* \< 0.0001), as well as a reliable interaction between Condition and Voice Gender (*χ*^*2*(1) = 10.56, *p* = 0.001). Specifically, there was again a stronger Condition effect for the female voice stimuli than for the male voice stimuli: For the female voice stimuli, 75% of all participants picked *khu* when asked the “rougher?” question. For the male voice stimuli, only 59% of all participants picked *khu* in the rough condition. This time there was no Condition and Listener Gender interaction (*χ*^*2*(1) = 0.53, *p* = 0.47), however, there was a reliable main effect of Listener Gender (*χ*^*2*(1) = 6.85, *p* = 0.009), with female listeners overall picking more *khya* responses (76%). There was no main effect of Voice Gender (*χ*^*2*(1) = 1.74, *p* = 0.19). Overall, the logistic mixed effects regression model described 27% of the variance in *khu* versus *khya* choices, of which 22% could be attributed to Condition and its interactions. Experiment 2 thus shows that the *khya/khu* contrast extends from softness/hardness to smoothness/roughness. ## Experiment 3: *Gender* The next experiment assessed whether the present findings carry over to the social domain of gender, which would show that the *khya/khu* contrast truly has a lot of latent meaning potential; it can not only trigger sensory meaning, but also more abstract social meanings, in line with the gendered cultural meanings of soju discussed in Harkness. Participants who were asked which one sounds “more female?”, picked *khya* 65% of the time and *khu* 35% of the time. On the other hand, participants who were asked the “more male” question showed the reverse pattern: they picked *khu* 70% of the time and *khya* 30% of the time. There was a statistically reliable main effect of Condition (*χ*^*2*(1) = 20.52, *p* \< 0.0001). This time there were no reliable interaction effects of Condition and Voice Gender (*χ*^*2*(1) = 1.16, *p* = 0.28) or Condition and Listener Gender (*χ*^*2*(1) = 1.86, *p* = 0.17). There were no reliable main effects of Listener Gender (*χ*^*2*(1) = 0.28, *p* = 0.60) and Voice Gender (*χ*^*2*(1) = 0.59, *p* = 0.44). Overall, the logistic mixed effects regression model described 15% of the variance in *khu* versus *khya* choices, which was almost entirely due to Condition and its interactions (14.6%). Experiment 3 showed that the *khu/khya* split that we showed for softness/hardness in Experiment 1 and smoothness/roughness in Experiment 2 successfully carried over to a question about gender. American English listeners associated *khya* more strongly with femininity than *khu*. This result is obtained with both the male and the female stimulus voices. That is, even for a male voice *khya* is perceived as more feminine, and conversely, even for a female voice *khu* is perceived as more masculine. ## Experiment 4: *Alcohol content* In Experiment 4, we extend the pattern observed in Experiments 1–3 to the domain of alcohol content. 58% of all participants chose *khya* as opposed to *khu* when asked to match a sound with a 15% liquor. 52% of all participants chose *khu* as opposed to *khya* when asked which sound matched a 30% liquor. In this case, we failed to obtain a main effect of Condition (*χ*^*2*(1) = 1.82, *p* = 0.18). Instead, the Condition effect content only emerged in the form of an interaction with Voice Gender (*χ*^*2*(1) = 10.86, *p* \< 0.001): the *khya/khu* split was stronger for the female stimulus pair. There were no Voice Gender (*χ*^*2*(1) = 1.88, *p* = 0.17) or Listener Gender (*χ*^*2*(1) = 0.05, *p* = 0.83) main effects, as well as no Listener Gender \* Condition interaction (*χ*^*2*(1) = 0.12, *p* = 0.73). Overall, the logistic mixed effects regression model described 8.9% of the variance in responses, which is almost entirely due to the Condition effect and its interactions (7.9%). Experiment 4 confirms our prediction only partially. While Condition did play a role, it was not in terms of being a main effect. The pattern was qualitatively similar to what has been observed in Experiments 1–3, with a stronger effect for the female voice pair than for the male voice pair. ## Experiment 5: *Taste* Experiment 5 tests the association of *khya* and *khu* with the taste dimension of “sweet” and “bitter.” 70% of all participants chose *khya* as opposed to *khu* when asked the question which one sounds “sweeter?”, with only 30% of all participants choosing *khu* in this case. In contrast, only 41% of all participants chose *khya* when asked which one sounds “bitter?”, which conversely showed a higher percentage of *khu* responses (59%). There was a statistically reliable effect of Condition (*χ*^*2*(1) = 24.94, *p* \< 0.0001). There were was no reliable effect of Voice Gender (*χ*^*2*(1) = 1.54, *p* = 0.21) and Listener Gender (*χ*^*2*(1) = 0.98, *p* = 0.32), as well as no Voice Gender \* Condition (*χ*^*2*(1) = 0.04, *p* = 0.83) or Listener Gender \* Condition (*χ*^*2*(1) = 0.63, *p* = 0.43) interactions. Overall, the logistic mixed effects regression described 12.0% of the variance in responses, which was mostly due to the Condition effect and its interactions (10.8%). Experiment 5 thus shows that the *khya/khu* contrast extends to the dimension of taste, particularly, the opposition of “sweet” and “bitter.” ## Experiment 6: *Pleasantness* Experiment 6 tests the extent to which the associations reported so far extend to the pleasantness/unpleasantness dimension. When asked which sound is more “pleasant”, 65% of all participants chose *khya*, as opposed to 35% who chose *khu*. Those participants who were asked which one sounds more “unpleasant”, picked *khya* 43% of the time, and *khu* 57% of the time. There was a statistically reliable effect of Condition (*χ*^*2*(1) = 12.70, *p* = 0.0004), as well as a reliable Condition and Voice Gender interaction (*χ*^*2*(1) = 14.22, *p* = 0.0002). As before, the *khya*/*khu* split was stronger for the female stimulus than the male stimulus. There were no Voice Gender (*χ*^*2*(1) = 0.24, *p* = 0.62) or Listener Gender (*χ*^*2*(1) = 0.38, *p* = 0.54) main effects, as well as no Listener Gender \* Condition interaction (*χ*^*2*(1) = 0.85, *p* = 0.36). Overall, the logistic mixed effects regression described 12.6% of the variance in responses, which was mostly due to the Condition effect and its interactions (12.3%). Experiment 6 thus shows that the *khya/khu* contrast extends to the dimension of pleasantness/unpleasantness. ## Experiment 7: Replication of Experiment 1 with Spanish, German, and Chinese ### Methods So far, Experiments 1 to 6 were conducted with a sample of native-speaking American English listeners that were recruited via Amazon Mechanical Turk. This set-up already demonstrates at least some degree of cross-linguistic and cross- cultural generality to the extent that Korean and American English are two unrelated languages (Korean is a language isolate, American English an Indo- European language) that are spoken in geographically separated areas that have distinctly different cultures. In this section, we extend the present investigation to three more languages, Spanish (a Romance language of the Indo- European family), German (another Germanic language of the Indo-European family), and Chinese (a Sino-Tibetan language). The first author (a native speaker of German) translated the English instructions of Experiment 1to German; the second author (a native speaker of Spanish) translated the the instructions to Spanish. In addition, we consulted a native speaker of Mandarin Chinese to translate the instructions to Chinese (Wangmeng Jiang). The English words *hard* and *soft* were translated into Spanish *duro* and *blando*, German *hart* and *weich*, and Chinese 更强硬 *geng qiangying* and 更柔和 *geng rouhe*. We used a snowballing technique to recruit volunteer participants via social media. Also, additional Spanish and German listeners were recruited via Amazon Mechanical Turk and paid 0.25 USD, as in the other experiments. We ended up with a total of 167 Spanish (mean age: 35, range: 18–73), 118 German (mean age: 34; range: 18–66) and 69 Chinese speakers (mean age: 26; range: 18–36). There was a difference in exposure to Korean culture between the samples: 32% of the Chinese sample have either been to Korea at least once or learning Korean; for the German sample, this figure was 20%; for the Spanish sample this was 7%. As excluding these speakers results in a very small sample size, particularly for the Chinese sample, the results below are reported for the entire sample (no exclusion). However, the main condition effect (hard versus soft) is statistically reliable (*p* \< 0.05) even if everybody who has been to Korea or has experience learning Korean is excluded from the samples. We fitted logistic mixed effects regression models of exactly the same structure as with the American English *hard*/*soft* study. ### Results For the Spanish sample, 71% of all participants chose *khya* when asked to indicate the soft sound (29% chose *khu*), see. When asked to indicate the hard sound, they chose *khya* only 24% of the time (76% chose *khu*). There was a reliable main effect of Condition (*χ*^*2*(1) = 50.45, *p* \< 0.0001), as well as a Voice Gender \* Condition interaction (*χ*^*2*(1) = 9.98, *p* = 0.002). There were no Voice Gender (*χ*^*2*(1) = 2.69, *p* = 0.10) and Listener Gender (*χ*^*2*(1) = 2.47, *p* = 0.12) main effects, as well as no Listener Gender \* Condition interaction (*χ*^*2*(1) = 2.81, *p* = 0.09). For the German sample, 62% of all participants chose *khya* when asked to indicate the soft sound (38% chose *khu*). In the hard condition, only 27% of all participants chose *khya*, and 73% chose *khu*. There was a reliable main effect of Condition (*χ*^*2*(1) = 32.09, *p* \< 0.0001), as well as a Voice Gender \* Condition interaction (*χ*^*2*(1) = 12.84, *p* = 0.0003). There was a reliable Voice Gender main effect (female voices had overall slightly more *khu* responses than male voices, 55% versus 54%, *χ*^*2*(1) = 5.9, *p* = 0.02), but no Listener Gender main effect (*χ*^*2*(1) = 1.47, *p* = 0.22), as well as no Listener Gender \* Condition interaction (*χ*^*2*(1) = 1.06, *p* = 0.30). For the Chinese sample, 59% of all participants chose *khya* when asked to indicate the soft sound (41% chose *khu*). In the hard condition, only 35% chose *khya*, with the remaining participants (65%) chosing *khu* instead. There was a reliable main effect of Condition (*χ*^*2*(1) = 5.56, *p* = 0.02). There were no Voice Gender \* Condition (*χ*^*2*(1) = 1.20, *p* = 0.27) and Listener Gender \* Condition (*χ*^*2*(1) \~ 0, *p* = 0.99) interactions. There were also no Voice Gender (*χ*^*2*(1) = 3.69, *p* = 0.05) and Listener Gender (*χ*^*2*(1) = 0.29, *p* = 0.59) main effects. Overall, these findings support our conclusion. Although the pattern was somewhat weaker for the Chinese sample, we were able to replicate the main finding from Experiment 1 in three additional samples, one of which is from a language unrelated to English (Chinese), and another one which is from another genus of the Indo-European language family (Spanish). # General discussion ## Summary of findings We have found that the Korean *khya/khu* distinction was reliably associated with softness, smoothness, gender, taste (sweet versus bitter), and pleasantness in a group of American English listeners with no knowledge of Korean. We also found partial support for an association with alcohol content. The hardness/softness experiment was replicated for Spanish, German, and Chinese listeners. The results presented in this paper support the following three claims: - The fact that there were associations across multiple different dimensions shows that same sound pattern can be associated with multiple different perceptual and social qualities when presented in the right context. This supports the idea of latent iconicity, as well as the idea that iconicity is pluripotential. - The contrast between the Korean interjections *khya/khu* is understood by American English, Spanish, German, and Chinese listeners, even if these listeners have little to no knowledge of the Korean language or culture. In fact, cultural geographic and cultural proximity did not appear to help the Chinese listeners, who scored the lowest. Overall, the translatability of this contrast suggests that these alcohol-related interjections used by Koreans harbor iconicity. - The semiotic strategies used by Korean advertisers in communicating the softness of soju tap into deeply rooted cognitive associations that are shared across cultures (at least with respect to the cultures tested here). An important aspect of our study was that we used a detailed ethnographic investigation of Korean advertising campaigns by Harkness to generate predictions for our cross-linguistic listening experiments. Moreover, we used an interjection that is actually produced by Koreans in response to downing a shot of soju rather than focusing on artificial stimuli, such as the pseudowords *kiki* and *bouba*. Across the board, we found that there were numerically stronger associations for *khu/khya* for the female voice, compared to the male voice. In some cases, there were statistically reliable interactions between the main condition manipulation and the gender of the stimulus. While this could be taken as a finding about gender (with the *khya/khu* contrast potentially being more pronounced for female voices), we have to be wary about this interpretation as we only tested one male and one female voice. This means that the factors “person” and “gender” are fully confounded. The inclusion of gender was not meant to test for gender effects as such but to show that the effect is generalizable to different kinds of voices even when we control for the voice manipulation statistically. ## Correspondences to existing iconic patterns The finding in this study corresponds to a number of observations that have been made in the literature on iconicity in Korean and other languages. For example, Japanese participants were more likely to associate /a/ with light and smooth textures and pleasant taste characteristics, whereas /i/ is more likely associated with unpleasant characteristics (p. 200). The sound /a/ is what is used to transcribe *khya*, and this vocal gesture is produced with an open mouth similar to the production of this vowel. The sound /i/ is a high vowel, just as the vowel /ɯ/ in *khu*, which is a high back vowel (the unrounded version of /u/). Similarly, Fónagy already noted that the high vowel /i/ may be associated with bitter tastes, compared to /a/ which is more sweet, see also. All of these studies suggest that low, back and open vowels have overall more “positive” sensory associations (smooth, soft, sweet) than high vowels, such as /i/, or as in our study, /ɯ/ (*khu*). It is also noteworthy that in descriptions of Korean ideophones, /ɯ/ is considered a “dark” vowel as opposed to /a/, which is considered a “bright” vowel. Sohn says that bright vowels “connote brightness, sharpness, lightness, smallness, thinness, and quickness”, whereas dark vowels “indicate darkness, heaviness, dullness, slowness, deepness, and thickness” (p. 96). In another characterization, dark vowels such as in *khu* are characterized as big, heavy, ponderous, masculine, clumsy, unwieldy, bulky, dark, and gloomy, whereas bright vowels such as in *khya* are characterized as small, affectionate, cute, feminine, fragile, flimsy, frivolous, bright, and happy. It is specifically noteworthy that the vowels /ɯ/ and /a/ form a pair in the vowel harmonic system of Korean (back unrounded vowels). ## Pluripotentiality: Discussion of potential mechanisms With respect to the pluripotentiality of iconicity, the present experiments are in line with existing evidence showing that iconicity allows limited generalizability across semantic domains. For example, Auracher showed that back vowels are not only associated with the idea of large size, but also with the related idea of social dominance. Our experiments provide a direct test of this pluripotentiality across a more diverse set of semantic domains, ranging from touch-related domains (hardness, roughness), taste and pleasantness, over to alcohol content and even gender. We do not claim that *khya/khu* inherently have all of these meanings but rather that these meanings arise in a relative or contextual manner. The fact that these distinctions are not inherent to *khya/khu* can be underlined by the observation that both *khya* and *khu* are actually very “harsh” sounds, uttered with a lot of aspiration and noisy high- frequency components. Therefore, *khya* cannot communicate soft/smooth/female/low-alcohol content in an absolute sense, but rather does so in a relative or contextual way. Thus, iconicity in spoken language does not involve singular one-on-one correspondences between particular sound patterns and non-sensory meanings. Instead, iconic associations between speech sounds and meanings are multi-layered. This picture of iconicity as something that is context-bound provides a richer description of iconic phenomena than what is sometimes assumed. Rather than being constrained to specific mappings, a resemblance between form and meaning can be *created* when the context makes particular perceptual or social dimensions sufficiently salient. In our experiments, this context was created by forcing an explicit choice between *khya* and *khu*. In Korean advertising, even though *khya* appears on its own without an explicit contrast with *khu*, the context for signaling the abstract quality of “softness” in the new soju is co- created by other semiotic resources, including the appearance of female actors talking in high-pitched voices and the light green and blue color palette. Korean speakers may also associate the innovative form *khya* with softness due to implicit knowledge that this form contrasts with *khu*, which is well- established as a vocal gesture for bitterness or disgust. Our findings therefore support Sidhu and Pexman’s notion of “indirect” iconicity, which is mediated by mental associations. More broadly, the fact that social meanings such as gender are triggered indirectly through contextual usage is consistent with the notion of indexicality in anthropological linguistics. It has to be pointed out, however, that multiple cognitive accounts could be used to explain our results. It could be that the *khya/khu* distinction taps into the same underlying core meaning which would be a highly abstract multisensory one. In this case, different shades of meaning becoming evident in particular contexts. This is akin to Rakova’s “supramodal” concepts, and this idea is also implied by Harkness’s analysis of the Korean ads, who speaks of “qualic transitivity.” According to him, the various semiotic channels employed by Korean advertisers (such as “voice marketing”, color palettes, female protagonists in ads) all index “the overarching abstract quality of softness” (p. 26) of the new soju. Alternatively, it could be that one of the meanings we investigated (such as, for example, the meaning of “softness”) is the primary meaning, and the other meanings are metaphoric extensions, i.e., the concept of softness is mapped onto other domains. At present, our experiment is not designed to differentiate between these cognitive accounts, but these provide fertile grounds for future research. ## Iconicity and sensory marketing By analyzing a case of sound symbolism used in a real-world marketing strategy by major liquor brands, our results also speak to the utility of iconicity in advertising. The field of “sensory marketing” has established the effectiveness of marketing that directly “engages the consumers’ senses”. For example, Elder and Krishna showed that when such food products as potato crisps and popcorn are described not only in terms of a single sense, such as taste, but also in terms of other senses such as sight, sound and touch, participants report to like the food items more. In fact, the advertising campaigns discussed by Harkness can already be understood as being part of a sensory marketing strategy, as the advertisers use multiple semiotic channels to signal the same sensory quality (of abstract softness). Moreover, research on the role of metaphor in advertising has suggested that it may be advantageous to signal product qualities more indirectly, such as via metaphor, rather than explicitly. Given the indirect latent meanings demonstrated here, we surmise that one of the advantages of iconicity in an advertising context is that it allows “multiple, distinct, positive inferences about the advertised brand” (p. 7), just as is the case with metaphor. ## Iconicity and interjections Our study is also important to further our understanding of interjections, which is a cover term used for expressions such as *huh*?, *ouch*, *uh-oh*, or *shh*. Dingemanse and colleagues have demonstrated the universality of *huh*?, with many different languages having the same (or a very similar-sounding) interjection. In our case, it is important that American English (as well as Chinese, German and Spanish) does *not* share the interjections *khya* and *khu* tested in this study. To be sure, speakers of these languages may also produce alcohol-related sounds after downing a shot of liquor. There may also be similarities between the vowel sounds in *khu* and the English *eugh*, which is known to mark disgust. However, the contrast between *khya* and *khu*—as used by Korean advertisers—is not reported for any of the languages used in this study. Nevertheless, American English listeners (as well as Chinese, German and Spanish listeners) interpret this contrast in a manner that is consistent with how the contrast is used by Korean soju advertisers, as reported in. This suggests that interjections may be “motivated” (rather than arbitrary) not only for functional reasons, as suggested by Dingemanse and colleagues, but also via iconicity, even if this is a rather indirect form of iconicity that is mediated by semantic associations. Interestingly, researchers collecting iconicity ratings for English words have repeatedly observed that when English speakers are asked to rate whether a word “sounds like what it means,” interjections receive very high ratings. However, these ratings are subjective and it is not clear what speakers base their iconicity judgments on. In our experiment, we provide a more direct test of the idea that an interjection (in this case, an alcohol-related interjection) can be understood across cultures even if the language in question—in this case American English—does not contain the exact same sound. # Conclusion To conclude, the interjective sounds that Korean speakers produce when downing a shot of soju have a diverse set of latent meanings (relating to hardness, roughness, gender, alcohol content, taste, pleasantness) that can be accessed by naïve listeners from unrelated cultures. This provides direct evidence for the idea that interjections harbor iconicity. Our results furthermore showcase the pluripotentiality of iconicity in spoken language, highlighting how sounds that are produced in the context of Korean alcohol rituals can signal a whole set of related sensory and non-sensory meanings. Finally, our results move the study of iconicity in advertising and marketing from artificial examples that involve constructed pseudowords to testing ecologically valid cases that are used in actual advertising campaigns. This approach also shows how the cross-linguistic experimental study of iconicity can take inspiration from, and subsequently contribute to detailed language- and culture-specific analyses of communicative behaviors. We thank David Sidhu, Marcus Perlman, Daniel Budke and two anonymous reviewers for excellent feedback and suggestions. We thank Sue Yoon in helping with the recordings. [^1]: The authors have declared that no competing interests exist. [^2]: Current address: Bodo Winter, Dept. of English Language, University of Birmingham, Edgbaston, United Kingdom

# Introduction The development of high-throughput technologies has enabled scientists to monitor the gene expression levels in tens of thousands of genes simultaneously in a single experiment. This technology has become a symbol of the post-genomic era. Biomedical research indicates that tumor development is related to the change in gene expression levels and that tumor-related biomarkers are usually associated with a few genes. Thus, identifying tumor tissue or disease-related biomarkers accurately is of great practical significance. However, gene expression profile data are characterized by very high dimensionalities and small sample size. The curse of dimensionality problem makes classification challenging. Some dimensionality reduction methods have recently been proposed to solve the “large, small ” problem. Feature extraction and feature selection are two methods of dimensionality reduction; feature extraction transforms original features (genes) into a set of new features by subspace learning. However, suitable biological interpretation is difficult to obtain from the subspace learning dimensionality reduction results. Feature selection is another commonly used dimensionality reduction method that selects a sub-set of genes that can best predict the response values from the raw data. Although dimensionality reduction can significantly improve computational efficiency, this process can easily lead to over-fitting when a classifier is applied. Sparse representation classification (SRC) was proposed by Wright et al. for face recognition. With sparsity constraint, a testing face can be approximately represented by parts of the training data that are from the same class. Unlike traditional classification methods such as support vector machine and nearest neighbor classifier, SRC is robust to both noise and outliers. However, the orginal training samples may not contain suffiient discriminating information compared with meta-samples. To capture more alternative information from gene expression data, the so-called meta-samples are proposed by. These samples can be regarded as a set of bases, the linear representation of which can represent the training data. In, penalized matrix decomposition is used to extract meta-samples, and clustering is performed on those meta-samples. In, the meta-sample based sparse representation classification (MSRC) method is proposed. This method is robust to over-fitting problem and noise. However, MSRC needs two predefined parameters, namely, the number of meta-samples and the sparse penalty factor. These two parameters are data dependent. Thus, model selection methods, such as cross-validation (CV), significantly affect the classification results. In this study, we propose a non-parametric version of MSRC to address this optimal parameter selection problem. The main contributions of this paper are as follows: 1. The data-dependent sparsity can be automatically adjusted, rather than empirically chosen. Without computationally expensive model selection, our method is scalable and efficient. 2. The existing MSRC method requires the appropriate selection of the number of meta-samples for each sub class, which is a laborious task. We address this problem by introducing a simple weighting strategy for the meta-sample of each category, and the rationality of weighting strategies is mathematically proved. 3. Extensive experiments are performed to evaluate the proposed method. Experimental results show the superiority of the non-parametric version of MSRC compared with some state-of-the-art classifiers. Section 3 presents more details. The remainder of this paper is organized as follows: prior work on sparse representation classification and the fundamentals of the proposed method are described in Section 2. Section 3 presents the experimental results. The proposed method is discussed in Section 4. Section 5 concludes this paper. # Methods This study primarily aims to establish the manner by which to devise an robust classifier for tumor subtype classification. Given a microarray data set and a set of class labels, is a matrix with rows and columns. Each column of denotes a sample, whereas each row of denotes a gene. Let denote the sample, which is a column vector with dimensional. For each element in, denotes the expression level of the gene in the sample. We provide a summary of the abbreviations used in this study in. For clarity, we use boldface and lowercase type letters for vectors and boldface and capital type letters for matrices. Gene expression profile data are high-throughput data with tens of thousands of genes. However, the number of samples is usually very small, which makes classification challenging. To avoid the curse of dimensionality, differential gene expression analysis, is widely used to exclude redundant and irrelevant genes before classification. In our study, we use the Relieff method to select a subset of informative genes for further analysis. In the following subsections, we briefly review meta-sample and sparse representation classification. we then propose weighted meta-sample based parameter free sparse representation classification (PFMSCR). ## Meta-samples versus gene expression samples As illustrated in, meta-samples can be regarded as basis samples that contain the essential information of the original data. A given testing sample can be represented by a linear combination of meta-samples from the same class. Concretely, suppose is associated with the class, where , and the class samples in the training data have meta-samples, namely,. Sample can be formulated as Eq. (1). Mathematically, meta-samples extraction can be regarded as a type of matrix decomposition, including non-negative matrix factorization, singular value decomposition (SVD), and principal component analysis, where matrix, and denote the meta-sample and meta-gene, respectively. In singular value decomposition, is a maximum linearly independent group of column vectors. Biologically, meta-samples are also called eigenarray or basis snapshot for gene expression data. Han et al. used meta-samples to identify tumors from microarray data and found that meta-sample-based classification can effectively avoid over- fitting. Zheng et al., proposed a novel cluster method based on meta-samples, which meta-samples can be regarded as cluster indictors. Prior works revealed that meta-samples preserve some desired discriminant information of samples from the same class. ## Sparse representation classification problem revisited In this subsection, we revisit the sparse representation problem briefly. Sparse representation is one of the most important components of machine learning and data mining community that has wide applications in such fields as text mining, image classification, and bioinformatics. In this work, we interpret the sparse representation problem from the view of linear algebra. From the standpoint of linear equations system, the solution of has three possible states: 1. Linear equation systems have infinitely many solutions if they are underdetermined (i.e.). 2. Linear equation systems have a unique solution if they are well posed. 3. Linear equation systems have no solution if overdetermined (i.e.). In the first scenario, one can pursue the sparse solution by regularization. The problem can be formulated as However, norm is an NP-hard combinational optimization problem, and difficult to solve, fortunately, norm is an appropriate convex approximate to. If the solution is sparse enough, minimization is equivalent to minimization, such that we can reformulate Eq. (2) as For the other two scenarios, the sparsity of cannot be guaranteed. However, one can still obtain a sparse solution by adding a penalty term that shares the same formulation as LASSO Compared with Eq. (3), Eq. (4) is an unconstrained convex problem. Notably, makes a tradeoff between sparsity and regression error and should be empirically chosen. A larger yields a sparser. However, one might run the risk of increasing regression error term. Sparse representation assumes that a signal can be reconstructed by a small number of basis signals within a linear combination. Thus, Eq (3) can be named as basis pursuit. In bioinformatics applications, one can suppose that a testing sample can be well reconstructed by the training data from the same class within a linear combination, which is a very useful assumption for our later work. ## Meta-sample based sparse representation Zheng et al. proposed MSRC method to predict tumor subtypes. In such situations, classes of meta-samples are extracted, denoting as with the same classes being conjoined together, where meta-samples are column vectors (two kinds of meta- sample are proposed). Given a test sample associated with class, MSRC tries to find sparse reconstruct coefficients in terms of all meta-samples using Eq. (4). In particular, tries to solve the sparse representation problem using. In ideal cases, the nonzero entries in will only be associated with the class meta- samples of, as shown in Eq. (5). Notably, the gene expression profile contains data with high dimensionality and small sample size. The sparsity can only be achieved by adding a penalty term. However, the optimal number of meta-samples and penalty factor are essentially important in classification applications. illustrates that if the meta-samples are improperly set, the prediction accuracy of MSRC drops seriously on COLON dataset. Specifically, in the left part of shows that the 10-fold stratified cross validation classification accuracy is achieved by varying the number of meta-samples from 3 to 12 for each subclass. We can observe that the performance is less sensitive to various regularization parameters within the scope of from the right part of. Thus, model selection is essential and laborious work on different data sets. To overcome this weakness, this study proposed a novel parameter free meta- sample based sparse representation classification (PFMSRC) method. ## Parameter free meta-sample sparse representation (PFMSRC) In this subsection, we first propose a heuristic weighted strategy, the reasonableness of which is theoretically proven. We then construct an underdetermined linear equation system, in which the data-dependent sparsity can be self-adaptively tuned by norm regularizer. Let be gene expression profile data, with the same classes being conjoined together, that is, contains all samples associated with the class. We factorize by performing SVD. The singular values are sorted in descending order, where is the column rank of, and denotes diagonal matrix with singular values being diagonal elements. One can extract weighted meta-samples associated with class as, where is a column vector in, and. Alternatively, Eq. (6) can be compactly reformulated as. This weighting scheme can enhance the influence of main singular vector in. That is, larger makes the associated meta-sample more important. Moreover, the weighting scheme works well in the following experiments. Compared with , Zheng et al. extracted meta- samples by performing SVD as well. However, in their algorithm framework, the number of meta-samples used for classification is determined during the cross- validation step. On the contrary, PFMSRC tries to avoid the cross-validation part by weighting the all meta-samples and weakening the influence of minor eigenvectors rather than using several of them for classification. Proposition 1 theoretically proves the reasonableness of the weighting strategy in measuring the importance of each metasample. **Proposition 1.** *Singular value is a reasonable weighting factor for measuring the importance of meta-samples.* *Proof.* Let, where and, considering evaluation metric function, one can conclude that This completes the proof. □ The evaluation metric function is used to measure the meta-sample's contribution of the meta-sample to the raw data reconstruction in terms of. denotes matrix trace. Note that, functions and have the same monotonicity, which makes the weighting strategy reasonable. graph was proposed by Cheng et al. to measure the similarity among samples. Inspired by their work, sparsity can be obtained by regularizer on underdetermined linear equation systems. Concretely, a testing sample can be recovered by weighted meta-samples within a linear combination with a noise term added, formulated as Eq. (7) Let and, where represents the number of meta-samples corresponding to classes, is an identity matrix, and is the noise term. Alternatively, one can solve the following minimization problem: Theorem 1 proves that Eq. (8) is a underdetermined linear system. As stated in Subsection 2.2 the sparsity of underdetermined linear system can be automatically tuned by regularization (the first scenario). Moreover, (8) is a canonical convex problem with equality constraints, which can optimize sparse representation coefficients and noise term simultaneously. The globally optimal solution can be efficiently solved by CVX package in polynomial time. Notably, the package solves the optimization problem by dualization rather than interior point method because the former is significantly faster than the latter. **Theorem 1.** *Linear equation system (8) is underdetermined, and*. *Proof.* We can find a sub matrix in, such as and. This completes the proof. □ Note that is a sparse vector with entries. The first components correspond to linear representation coefficients, whereas the last components characterize model noise or regression error. However, the test sample from one of the classes in training data cannot be well reconstructed by meta-samples associated with the same class in most instances because of the existence of noises. illustrates the flowchart of our PFMSCR scheme, the redundant dictionary is constructed by combining meta-samples and noise term. We define a projection function for each class, which selects the coefficients associated with the class from the first components in, whereas the other entries are appropriately padded with zeros in. The reconstruction relationship is not always holden. However, the minimized reconstruction error criterion is a good approximation to classify testing samples. We summarize the proposed classification method as follows. Step 1. Input training sets, class number, and testing sample ; Step 2. Normalize training set samples and testing sample to obtain unit -norm; Step 3. Extract weighted meta-samples for each class (meta-samples with the same class are conjoint); Step 4. Solve non-parametric sparse representation problem by Eq. (8); Step 5. Compute residuals for each class ; Step 6. Return class label of as ; PFMSRC can be considered as a non-parametric version of MSRC, compared with the former having the following merits: 1. The weighted meta-samples are orthogonal with one another. That is, no redundancy exists among meta-samples, and the weight enhances the influence of the main singular vector, such that discriminant information can be well retained. 2. The data-dependent sparsity can be automatically tuned without human intervention. Thus, PFMSRC has better scalability and robustness. 3. The time complexity of PFMSRC is lower than that of MSRC, since computationally expensive model selection work need not be accomplished for parameter optimization. Time complexity can be estimated as: weighted meta- sample extraction step needs time complexity, minimization needs time complexity, the total complexity for PFMSRC is. In the following section, we will conduct extensive experiments on micoarray data to evaluate the effectiveness of our scheme, and microarray data repository information as well as the accession number is given by. # Experiments In this section, we will evaluate the performance of the proposed PFMSRC algorithm against four state-of-the-art algorithms, namely, linear discriminant analysis (LDA+SVM), independent component analysis (ICA+SVM), SRC, and meta- sample sparse representation (SVD-MSRC). The former two are model based and accompanied by feature extraction. These two algorithms are regarded as baseline. For the model-based method, support vector machine, with radial basis function kernel is employed as a classifier. The experiments are performed on four binary-class classification data sets and four multiclass classification data sets. All experiments are implemented in Matlab environment and run on a personal computer with intel Pentium4 dual core CPU 2.4 GHZ and 4 G RAM. The summarized descriptions of the eight gene expression profile datasets are provided by. - Colon consists of 62 samples with two subclasses including 40 tumor and 22 normal samples. The highest 2000 genes with minimal intensity in the tissues are retained from the original of more than 6500 genes. This dataset can be downloaded from. - Acute leukemia data, consist of 72 samples with two subclasses, including 47 acute lymphoblastic leukemia patients and 25 acute myelogenous leukemia patients. Each sample contains 7129 genes. This dataset can be downloaded from. - DLBCL consists of 77 samples with two subclasses, including 58 diffuse large b-cell lymphoma samples and 19 follicular lymphoma samples. Each sample contains 7129 genes. This dataset can be downloaded from. - Gliomas consist of 50 samples with two subclasses (Glioblastomas and Anaplastic Oligodendrogliomas), and each sample contains 2308 genes. This dataset are available at. - SRBCT consist of 83 samples with four subclasses (Ewings sarcoma, Burkitts, Neuroblastoma and rhabdomyosarcoma). Each sample contains 2308 genes. The datasets are available at - ALL consists of 248 samples with six subclasses. Each sample contains 12626 genes. The datasets are available at. - MLLLeukemia consists of 72 samples with three subclasses. Each sample contains 12582 genes. The datasets are available at. - LukemiaGloub consists of 72 samples with three subclasses. Each sample contains 7129 genes. The datasets are available at. ## Dataset preprocessing and experiment setup Gene expression profiling involves data with high dimensionality and small sample size. The exclusion of redundant and irrelevant data is critical for classification. As suggested by, restaining only the top 400 genes makes a good tradeoff between computational complexity and biological significance. In our experiment, the top 400 genes are selected from each dataset by applying the Relieff algorithm to the training set. For LDA+SVM algorithm, we simply extract new features to train the classifier, as LDA can find at most meaningful projection vectors in the subspace, where denotes the number of classes. SVM kernel parameters are determined by 10-fold cross-validation. In fact, the determination of the number of independent components is also an empirically dependent work. Here, we use the same method as suggested by. SRC and MSRC methods need parameter to control sparsity. MSRC also needs the number of meta-samples of each class as a key parameter. Each dataset is searched from by 10-fold CV on training data, and the number of meta-samples for each class is set as recommended by. ## Experiments on binary classification problem To evaluate the performance of five methods on a balanced split data set, we randomly select to samples per subclass as training set and use the rest for testing to guarantee that at least one sample in each category can be used for test, 20 times training/testing are randomly split, and the average classification accuracies are presented. The best prediction accuracy is in boldface for each gene expression profile dataset. We show the average performance comparison on four binary classification tasks in. PFMSRC exhibited encouraging performance. Although Gliomas was difficult for classification, the proposed approach can still achieve 85% classification accuracy via 20 samples per subclass used for training. Notably, the classification accuracy of LDA+SVM and ICA+SVM dropped quickly as more samples are taken for training; the same observations can be found in. This fluctuation phenomenon can be interpreted as follows: (1) For the binary classification case, the feature extracted by LDA has only one dimension that is insufficient to capture the intrinsic discriminating information. Thus, model-based classification methods have difficulty in preventing the over-fitting phenomenon. (2) When evaluating the performance on the testing set the number of samples changes as more samples are used for training. Classification accuracy, specificity, and sensitivity are some popular evaluation metrics. In this work, we use all three to evaluate performance, and the results are reported in, and, respectively. The three methods can achieve satisfactory performance not only on the specificity metric but also on the sensitivity metric. Compared with SRC and MSCR, PFMSRC outperforms its competitors in most cases. A comprehensive consideration is that PFMSRC achieves the best performance, followed by MSRC and SRC. ## Experiments on multiclass classification problem We investigate multiclass classification performance on four publicly available data sets. The experimental setup is the same as that for the binary classification case. On one hand from and it can be seen that (1) the classification accuracies of SRC, MSRC, and PFMSRC are increased on all multiclass classification datasets as more samples per subclass are taken for training. (2) ALL has six subclasses, and the proposed PFMSRC achieves the highest classification accuracy, which indicates that we have potential superiority on multiclass classification task. (3) LDA can capture more discriminating information on the multiclass classification task, and the over- fitting phenomenon is reduced compared with the binary classification task. On the other hand, sparse representation based classification methods are less sensitive to the number of samples used for training model-based classification methods, which suggests a natural approach to select a classifier when the training sample size is small. provides the performance description of the five classification methods. The proposed PFMSRC method performs consistently well with small standard deviations. On the SRBCT and ALL datasets, PFMSRC achieved 96.98% and 96.73%, respectively. ## Experiments with different number of genes In this subsection, we evaluate the performance of the five methods with different feature dimensions on eight tumor data sets. For the training data, 10 samples per subclass are randomly selected, whereas the remaining samples are used for test. We perform the test with various numbers of genes, starting from 50 to 400 genes in steps of 20. The comparison experiment was performed 20 times, and the average prediction accuracy of our experiments on eight gene expression profile datasets was recorded for evaluation. The balanced training sets for each dataset ensure fair evaluation as stated by. The experimental result in shows that the proposed PFMRSC performs well when only 100 genes are used. We can observe the similar results in the multi- classification case as well. In binary classification case, SRC, MSRC, and PFMSRC share the same curve trend. Compared with SRC and MSRC, PFMSRC performs well by using a smaller number of genes, SRC and MSRC can achieve comparable accuracy by using more genes. Evidently, SRC, MSRC, and PFMSRC consistently outperform LDA+SVM and ICA+SVM in all datasets. In the multiclass classification case, the performance of MSRC, SRC, and PFMSRC is very stable with respect to the number of genes, and all these methods converge fast to the optimal classification rate point. shows that compared with their performance in the binary classification case, SRC, MSRC, and PFMSRC are less influenced by gene dimension. Note that ALL is a multiclass dataset with six subclasses, but PFMSRC can still achieve a higher classification rate of 97% accuracy compared with SRC and MSRC. The same conclusion can be drawn for the SRBCT dataset. In, we report the detailed classification accuracy. PFMSRC outperforms its competitors on most gene expression profile datasets, whereas SRC and MSRC-SVD perform the second best. ## Comparsion of CV performance To evaluate the classification performance on imbalanced split training/testing sets, we perform 10-fold stratified CV on tumor subtype dataset. All samples are randomly divided into 10 subsets based on stratified sampling: nine subsets are used for training, and the remaining samples are used for testing. This evaluation process is repeated 10 times, and the average result is presented. The 10-fold CV results are summarized in. shows that as the training sample size increases, the performance of these five classification methods is significantly improved. Model based methods LDA+SVM and ICA+SVM perform very well, with the classification accuracy increased significantly. In particular, the prediction accuracy of ICA+SVM ranges from 86.5% to 96.57% in all tumor expression profile datasets, which is comparable with those of SRC, MSRC and PFMSRC. We can conclude that model-based approaches are more vulnerable to the small sample size problem, over-fitting should be resolved properly. # Discussion Based on the above experiments, we can draw the following observations: 1. Sparse representation based methods (SRC, MSRC, PFMSRC) consistently outperform the model-based methods (LDA+SVM, ICA+SVM) on all experiments. Especially, in balance splited datasets the prediction accuracy of model- based methods is significantly lower than that of sparse representation methods which may be attributed to the small sample size problem. However, SRC, MSRC, and PFMSRC perform well even when we take 5 samples per subclass for training and the rest for testing. 2. SRC, MSRC and PFMSRC are robust to various sample sizes and feature dimensions, as well as converge fast to the optimal classification rate. The experiments verify the results in, which favors the application of those methods. Note that, model-based methods (LDA+SVM, ICA+SVM) exhibit improved 10-fold CV classification accuracy. A reasonable explanation is that the over-fitting phenomena are dramatically reduced when 90% of original samples are used for training and the remaining 10% are used for evaluation in our experiments. 3. PFMSRC outperforms SRC and MSRC in most cases, which implies that the parameter free sparse representation and weighting strategies can capture more discriminating information, especially in multiclass classification. See. 4. PFMSRC is a parameter-free method, in which the data dependent sparsity can be self-adaptively tuned, compared with SRC and MSRC in which search for a regularization parameter is laborious work. Moreover, the number of meta- samples is a key parameter for MSRC, as shown in, which makes model selection more difficult. # Conclusions In this study, we proposed a novel non-parametric meta-sample-based sparse representation. The algorithm assumes that test samples can be well reconstructed within a linear combination of weighed meta-samples in the same class. We theoretically proved the rationality of the weighting strategy. A simple but efficient projection function is constructed by the sparse representation coefficients to complete the classification work. We also compare the performance of PFMSRC with that of two model-based methods and two sparse representation-based methods on eight tumor expression datasets. Experimental results have shown the superiority of the proposed method. We then drew some conclusions on the effects of both balanced split and imbalanced split testing/training sets on tumor classification problems. PFMSRC exhibits stable performance with respect to different training sample sizes and feature dimensions compared with the other four algorithms. Thus, the extension of the sparse representation with dimensionality reduction (feature selection or feature extraction) in a unified framework is one of our future works. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YJ BL. Performed the experiments: GY ZC. Analyzed the data: LC WZ. Wrote the paper: YJ BL.

# Introduction In everyday life, we frequently encounter situations that require undertaking two tasks simultaneously. Previous research using dual task paradigms has revealed that performing a secondary task may produce either positive or negative effects on the primary task, with the direction of the effect being influenced by the nature of the two tasks. In terms of research reporting detrimental dual task effects, a large body of research suggests performance on various learning and memory tasks can be negatively influenced by simultaneous performance of a secondary task. This body of evidence suggests that the likelihood of the secondary task exerting negative effects on memory performance is dependent upon whether the secondary task occurs during the memory encoding or retrieval stages of the learning task, with a secondary task at the time of encoding more likely to impede memory performance. Instructing participants to focus on the secondary task rather than the learning task produces improved performance on the secondary task while simultaneously impairing performance on the learning task, suggesting that the allocation of attentional resources between the encoding process and the secondary task is under conscious control. It has been proposed that human learning and memory involves multiple cognitive systems, which act in co-operation or competition with each other. One of these proposed systems related to explicit learning has been suggested to involve the frontal and medial temporal lobes, and to be dependent on attention and working memory. In contrast, the implicit/procedural system has been suggested to involve the striatum and to rely on automatic non-conscious processes. Research suggests that modulating the involvement of the explicit learning system (potentially by altering the involvement of attention and executive systems) has the capacity to alter performance on implicit learning tasks. Specifically, evidence suggests that minimising prefrontal cortex and/or medial temporal lobe involvement by introducing a secondary task, a distractor task, or a pharmacological agent has no effect or improves performance on tasks that rely upon the striatal learning system. These findings have implications for dual task situations, as they suggest that the very nature of the tasks may determine the degree to which implicit and/or explicit learning processes are involved, potentially influencing whether the secondary task exerts a positive or negative effect on performance of the primary task. The balance between the two learning systems also has implications for language learning (especially under dual task conditions), as language learning and processing has been proposed to rely in certain conditions on implicit statistical learning mechanisms, with language processing closely linked to implicit sequence learning. Indeed, Nemeth et al (2011) found that a sentence processing task (sentence judgement) interfered with a simultaneous probabilistic implicit sequence learning task (an alternating serial reaction time task). Thus, evidence suggests that the effectiveness of language learning can potentially be altered by introducing a dual task that modulates attention and executive processes (i.e., by manipulating the involvement of explicit learning mechanisms) or by introducing a task that involves implicit sequence learning, with the nature of the task (i.e., implicit or explicit) directly influencing the effect on the language task. To date, the precise mechanisms behind dual task effects in general on memory encoding in healthy adults remain contentious. Naveh-Benjamin and colleagues (2007) proposed that all phases of the memory encoding stage are influenced by the secondary task, with the initial encoding phase (i.e., the initial registration of information) in particular being highly susceptible to dual task interference effects. Evidence suggests that the nature of the dual task may influence potential interference effects. While many studies have used visual or auditory monitoring activities as the simultaneous secondary task,, several studies have also investigated the effects of more motor-based secondary tasks (e.g., button pressing, dowel balancing) as a component of the dual task paradigm. In healthy right-handed individuals, performing non-meaningful movements (i.e., movements that are unrelated to the content of the speaker’s verbal output and do not facilitate the listener’s comprehension) with the right hand can interfere with simultaneous speech production. It has been proposed that the interference effects may originate from shared functional neural systems controlling the execution of the linguistic and motor tasks. Interestingly, Lomas (1980) noted that the interference effects of linguistic activities on motor performance occurred only during specific motor tasks that involve sequenced movements without visual cues. Specifically, the authors found that reciting well-known nursery rhymes produced interference effects on the simultaneous tapping of four keys in a sequence with the right fingers or arm only under conditions where participants were also unable to see the keys. This observation was supported in a meta-analysis by Medland and colleagues (2002) which found that dual task interference effects varied in magnitude between studies according to the type of manual task employed, with studies that used finger tapping as the motor task differing from studies that used other motor tasks. The authors proposed that the variable dual task interference effects observed during the different motor tasks may have reflected task difficulty and the allocation of attention resources, with participants more likely to allocate attention resources to the more difficult task. It must be noted that a competing body of research has found that meaningful gestures may actually produce facilitatory effects in some dual task situations. Specifically, studies have revealed that performing gestures (i.e., meaningful hand movements that accompany speech) assisted participants to explain how to solve a problem while simultaneously remembering a list of unrelated items. In these studies, it was found that meaningful gestures facilitated item recall. The authors proposed that this facilitatory effect was due to the gestures reducing the load on working memory. There are several key elements of these studies that warrant consideration. One key feature is that the gesture was incorporated into the explanation task, rather than being a separate unrelated task. Another important element was the gesture was meaningful (rather than non- meaningful gestures that did not add to the content of the speaker’s explanation). Indeed, Cook and colleagues (2012) found that hand movements in rhythmic synchrony with speech (i.e., non-meaningful gestures) did not produce similar beneficial effects on working memory. As the studies were limited to explanations and remembering lists of familiar items (e.g., letters, numbers), the effect of gesture on new word learning was not examined. There has been some evidence to suggest that meaningful gestures may actually facilitate second language learning, however, it is unknown whether this evidence extends to nonword learning for unfamiliar concepts. Additionally, the research, did not directly examine whether similar effects are observed with non-meaningful hand movements. An additional stream of research into word re-learning after stroke has suggested that dual tasks may actually facilitate language learning and retrieval by acting on intentional mechanisms. Intentional mechanisms refer to the processes involved in the selection and initiation of an action from a number of competing actions, and are believed to be closely linked with frontal action systems of the brain (i.e., the medial frontal cortex, lateral frontal structures, and basal ganglia). Heilman and colleagues (2003) suggested that the side of the brain that the intention mechanisms are activated on depends upon the limb, lateralisation and direction of the movement. In individuals with aphasia, it has been found that complex left hand movements may improve subsequent picture naming in patients with moderate to severe anomic aphasia, and that preactivation of the motor system by standing may facilitate lexical retrieval in patients with chronic aphasia. Another study by Richards and colleagues (2002) found that patients with non-fluent aphasia demonstrated improved picture naming by performing an intentional non-meaningful complex hand action sequence prior to picture naming (For clarity we have referred to the movement as complex as it involved multiple steps as opposed to simple motor movements such as repeated finger tapping). The complex hand movement involved the participant lifting the lid of a small box with his/her left hand and then pressing a button inside the box to initiate the presentation of a picture for naming. It has been proposed that the facilitatory effect may have stemmed from shared functional neural resources between the linguistic and motor systems, such that preactivation of the motor system may serve to prime the linguistic system by activating shared intentional mechanisms. Indeed, research suggests that the pre-supplementary motor areas of the brain may control intention for both complex hand movements and word generation. These findings are at odds with previous simultaneous dual task studies in healthy adults involving non- meaningful motor movements, which have found impeded performance of simple linguistic tasks paired with a simultaneous motor task, suggesting that the shared functional neural resources between the linguistic and motor systems may cause interference. Thus, further research is required to investigate the effects of simultaneous or prior complex hand movements on the complex linguistic tasks of new word learning and picture naming. As most of the previous research into dual task effects with linguistic and non- meaningful motor movements in healthy individuals has used reciting known material (e.g., nursery rhymes), reading aloud, or simple repetition as the verbal component, the effects of simultaneously performing non-meaningful hand movements on new word learning are as yet unknown. New word learning necessitates the mapping of a novel word form to its referent. That is the learner must link the perceptual features of the novel item to its name (phonology) and meaning (semantics). As this process is not required during reading aloud, simpler repetition or reciting known material, it is unknown whether the interference effects observed previously with simple linguistic tasks paired with non-meaningful motor movements will be observed with new word learning. Given that previous research has found that memory encoding is highly vulnerable to interference effects, the reliance of new word learning on the encoding of both new semantic and phonological information may cause new word learning to be particularly susceptible to dual task interference effects, especially when non-meaningful movements are included. Additionally, the established relationship between new word learning and short-term memory may cause new word learning to be further vulnerable to dual task interference effects as a result of the increased cognitive requirements on short-term memory during new word learning. In summary, evidence suggests that in healthy individuals, performing non- meaningful hand movements can interfere with simultaneous verbal tasks. However, to date, most of the research regarding dual task effects and non-meaningful movements has targeted relatively simple verbal tasks such as reciting known material, reading aloud, or simple repetition. As a result, the effect of non- meaningful motor movements on a more complex language task such as new word learning in healthy individuals is unknown. To further complicate the issue, competing evidence suggests that pairing meaningful gestures with an explanation task improves performance on a simultaneous list-based memory task. Furthermore, research in individuals with aphasia suggests that complex non-meaningful motor movements conducted immediately before linguistic tasks may actually enhance linguistic task performance by priming activation of the linguistic system. It is unknown whether similar facilitatory effects are observed in healthy adults when a motor movement is conducted immediately before a linguistic task. In order to address this issue, the present study examined the effects of non- meaningful complex hand movements on subsequent new word learning and word retrieval in healthy individuals. Differences between left and right hand movements were also examined. Given the non-meaningful nature of the hand movements in the present study and that previous research with healthy adults has found that simultaneous non-meaningful motor tasks can impair linguistic performance, it was hypothesized that the hand movements would interfere with lexical acquisition and retrieval. Given these previous findings in healthy individuals, it was expected that the performance of the healthy adults in the present study would differ from the results of previous aphasia research – involving word retrieval and complex non-meaningful hand movements, which would suggest that different possibly compensatory mechanisms are engaged in a neurologically disordered system. Given previous findings of greater interference with right hand movements in right handed individuals, it was expected that poorer new word acquisition and retrieval would be observed with right hand movements compared to left hand or no movements. To establish whether there were differential interference effects between the recall of newly learnt words compared with the recall of known words, the effects of a secondary task on subsequent naming of familiar objects was also investigated. Given the finding that memory encoding is typically more susceptible to dual task interference effects, it was hypothesized that the hand movements would result in greater declines in performance on the new word learning task compared to the familiar picture naming task. # Methods ## Ethics Statement The study was approved by The University of Queensland Ethics Committee and was conducted in accordance with the Declaration of Helsinki. Participants were required to attend one testing session lasting approximately one hour. Participants completed a working memory task, a word learning task, and a familiar picture naming task. A practice task was included for both the word learning and familiar picture naming experiments. ## Participants Twenty-five right-handed adults (12 females) aged between 18 and 32 years (mean age 22.24 years) with English as their first language participated in the study. Exclusion criteria included neurological damage or disease, degenerative disease, mental illness, learning disability, or chronic alcohol or drug abuse. Participants were recruited from the University of Queensland student body and community. Written informed consent was obtained from all participants prior to testing. ## Word Learning Task ### Stimuli The unfamiliar pictures used for this task consisted of 35 Finnish farm tools accessed from a previous study and used with permission of the author. The 35 pictures were divided into seven pictures under each of the three experimental conditions and 14 practice items. The nonword names were obtained from the Australian Research Council (ARC) nonword naming database. The nonword names were all two-syllable words, five or six letters in length, and adhered to the rules of written English. ### Procedure The learning task was administered using E-Prime software for computers (version 1.1). During the learning phase, participants were presented with a series of seven unfamiliar objects on the computer screen. Each unfamiliar object was accompanied with an auditory presentation of a nonword name via headphones. Participants were required to repeat the name of the object aloud immediately after the auditory presentation. The presentation of each object was initiated by the participant opening a wooden box with a lid and pressing a button on the mouse located within the box. This complex non-symbolic movement resembles the intentional movement utilised by Richards et al. (2002) in the aphasia rehabilitation study discussed previously. Three conditions were used for the study: a right hand condition, where the participant used their right hand to open the wooden box located on their right side; a left hand condition, where the participant used their left hand to open the wooden box located on their left side; and a baseline condition, where participants were not required to make any hand movements and the task was initiated by the researcher. The order in which each condition was presented was randomised for each participant. The items used in each of the three experimental conditions were counterbalanced. A recall phase followed the learning phase, where participants were asked to recall the names of the objects by saying the learnt name aloud when presented with the objects on the computer screen. Participants were instructed to respond with ‘don’t know’ if they were unable to recall the object’s name. The learning and recall phases were repeated for a total of five trials using the same stimulus items presented in random order. All participants completed all three conditions during the session. All responses for this task were audio recorded and later transcribed for accuracy. ## Familiar Picture Naming ### Stimuli The familiar objects for naming were obtained from Snodgrass and Vanderwart (1980) and the International Picture-Naming Project (Szekely et al., 2004). The 90 objects were of varying frequency, number of syllables, and object visual complexity as obtained from the International Picture-Naming Project. The objects were split into three matched lists of 30 items for each of the conditions. There were no significant differences between the three lists in terms of frequency, number of syllables, or object visual complexity (all *p*\>0.10). ### Procedure The familiar picture naming task was also administered using E-Prime software for computers (version 1.1). This task utilised the same three conditions (right hand, left hand, and baseline) as in the learning task. Participants used the action of opening the box and pressing the mouse button with the required hand to initiate the presentation of a series of pictures. During the baseline condition, the researcher initiated the presentation of the pictures. Participants were presented with 30 pictures of real objects for each of the three conditions. Participants were required to name the objects as quickly as possible following presentation. Participants’ responses were measured by a serial response box (model number 200A) (Psychology Software Tools) in conjunction with E-Prime software to record the latency of response time. Latency was recorded from presentation of the picture up to 2000 milliseconds. The assignment of lists to hand conditions was counter-balanced across participants. ## Working Memory All working memory tasks were presented verbally and included (a) a wordspan forward task, requiring the participant to repeat a series of word of increasing length; (b) a digit span forward task, requiring the participant to repeat a series of numbers of increasing length; and (c) a digit span backward task, requiring the participant to repeat a series of numbers of increasing length in reverse. The number of items recalled in the correct order was recorded for each participant for all of the working memory tasks. # Results ## Word Learning Task Accuracy data from the learning task were entered into a mixed linear model analysis with *subject* as a random factor, *condition* (baseline, right, left) and *recall trial* (1–5) as fixed factors, and *order* (the order of the three conditions presented to each participant) and *trial within block* (the order of the items within each condition) as covariates. A bonferroni adjustment was used for calculating estimated marginal means and confidence intervals. The intra- class correlation for the model was.103. There was a significant main effect for recall trial (*F*(4, 2600) = 126.730, *p* = .000), with recall of the nonword names improving over the five learning trials, across all conditions (See) (Trial 1: 95% CI = .035 −.174; Trial 2 95% CI = .212 −.351; Trial 3 95% CI = .365 −.504; Trial 4 95% CI = .485 −.624; Trial 5 95% CI = .555 −.694). There was also a significant main effect for condition (*F*(2, 2600) = 3.868, *p* = .021). Overall, the baseline condition revealed the highest mean accuracy (43.1%; 95% CI = 36.5–49.7%), followed by the left hand condition (39.4%; 95% CI = 32.8–46.0%) and right hand condition (37.5%; 95% CI = 30.9–44.1%). Pairwise comparisons (Bonferroni adjusted) between the conditions were carried out, revealing a significant mean difference between the baseline and right hand condition (*p = *.019). No significant difference was found between the baseline and left hand condition (*p* = .221), or between the right hand and left hand condition (*p* = 1.00). There was no significant interaction found between condition and recall trial (*p* = .833). ## Correlations The relationship between performance on the working memory tasks and learning task accuracy was examined using Spearman’s correlation coefficients. Accuracy data for the fifth trial (indicating learning success) was used for this analysis. As participants were required to recall the new word names following each of the five trials, analyses were conducted using the fifth recall trial (i.e., the trial that represented maximal learning) as per. Word span and learning success were significantly positively correlated under all three conditions (Right hand: *r_s = *.437, *p* = .029; Baseline: *r_s = *.405, *p* = .045; Left hand: *r_s* = .621, *p* = .001). Digit span forward was significantly positively correlated with learning success for the right hand (*r_s = *.443, *p* = .027) and baseline (*r_s = *.502, *p* = .011) conditions, but not for the left hand condition (*r_s = *.375, *p* = .065). There was no significant correlation between digit span backward and any of the three conditions (right: *r_s = *.269, *p* = .193; left: *r_s = *.231, *p* = .267; baseline: *r_s = *.370, *p* = .069). When overall accuracy was compared with performance on the working memory tasks, the same pattern of results emerged. There was a significant correlation between word span and overall accuracy (*r_s = *.554, *p* = .004), and between digit span forward and overall accuracy (*r_s = *.532, *p* = .006), but not between digit span backward and overall accuracy (*p*\>.05). ## Familiar Picture Naming Reaction times from the familiar picture naming task were entered into a mixed linear model analysis with *subject* as a random factor, *condition* (baseline, right, left) as a fixed factor, and *order* and *item* as covariates. A bonferroni adjustment was used for calculating estimated marginal means and confidence intervals. The intra-class correlation for the model was.139. Reaction times (RTs) greater than 1500 ms were treated as outliers and removed, resulting in 6.23% of items being excluded. The results revealed a significant main effect for condition (*F*(2, 2047) = 3.292, *p* = 0.037). Participants performed best (i.e., fastest RT) under the baseline condition (mean RT 839.80 ms; 95% CI = 764.86–914.75 ms), followed by the left hand condition (mean RT = 843.06 ms; 95% CI = 768.12–918.03 ms), and then the right hand condition (mean RT = 889.07 ms; 95% CI = 814.12–964.01 ms). Pairwise comparisons between conditions revealed no significant difference between right and left hand conditions (*p* = .097), right hand and baseline conditions (*p* = .066), or between left hand and baseline conditions (*p* = 1.00). # Discussion The present study examined the effect of complex non-meaningful hand movements on subsequent lexical acquisition and retrieval in healthy individuals. The findings suggest that for right-handed, healthy individuals, performing a complex non-meaningful movement immediately before linguistic tasks interferes with both lexical acquisition and retrieval of both newly learnt and previously known material. The present study used a similar experimental design (and complex motor sequence) to the aphasia rehabilitation study by Richards et al (2002). However, contrary to the facilitatory effect of the motor movement observed on later naming by Richards et al (2002), the present study found that the complex motor sequence interfered with subsequent linguistic tasks. Previous reports have suggested that new word learning in healthy individuals can be used as a model for re-learning in aphasia –. However, if this was indeed the case, it would be predicted that the results for healthy individuals in the present study would reflect similar results to the aphasia study by Richards et al (2002). It appears that we cannot simply assume that factors influencing word retrieval in aphasia operate in a similar way to those mechanisms influencing lexical acquisition and retrieval in healthy individuals, which makes sense given the considerable differences between healthy and neurologically impaired systems. The present findings allow for the possibility that the effect seen in individuals with aphasia may arise due to facilitation of motoric aspects of speech production (post retrieval of the phonological form), however, this facilitatory effect may not arise when deficits are not present. Thus, while the present study and Richard et al (2002) employed the same complex non-meaningful hand sequence, it is possible that this task facilitated impaired post-lexical mechanisms in individuals with aphasia but interfered with lexical acquisition and retrieval mechanisms in healthy individuals. In contrast with aphasia research, the interference effects observed in the present study more closely resembled the interference effects observed during dual task experiments with non-meaningful motor movements in young healthy adults. The present study extended the results of the previous dual task research in healthy adults by revealing that interference effects can occur with complex linguistic tasks and when the non-meaningful motor task is performed immediately before the linguistic task (rather than performing the two tasks concurrently). A proposed theory to explain the dual task interference effect in the present study is that of the dual task decrement. Dual task decrement stems from the idea that when carrying out a dual task, such as a motor and verbal task, both tasks need to be held in working memory with attention shared between the tasks. Interestingly, the dual task decrement has been observed to be more pronounced when a right-handed task is paired with a verbal task. This phenomenon may be explained by the interference that may occur as a result of the dual processing role of the left hemisphere for both verbal tasks and right- sided movement. It has been suggested that the interference between a right- sided movement and a verbal task may reflect an overlap in the neural mechanisms responsible for their processing. The present study lends support for this proposal, as both new word learning and familiar picture naming (which may rely upon similar neural networks, especially the inferior parietal cortex, as per Cornelissen et al., 2004) were susceptible to dual task interference effects. The present study also supports previous research demonstrating that modulating the involvement of cognitive processes associated with the explicit or procedural learning systems by introducing a secondary task can influence primary task performance. The idea of shared neural networks between new word learning and familiar picture naming may also provide an explanation for why the two tasks experience similar interference effects despite different memory requirements. Specifically, during the picture naming task in the present study, participants were required to retrieve the name of the item under dual task conditions; no encoding of new phonological or semantic information was required. In contrast, during the new word learning task, participants were required to encode new phonological, semantic and perceptual information about the items under dual task conditions and then recall the newly learnt information. As previous research has revealed that dual task conditions at the time of encoding are more likely to impede memory performance than during other stages of the memory process, it was expected that new word learning would be more susceptible to interference effects than familiar picture naming. However, contrary to this proposal, the present study found interference effects during both new word learning and familiar picture naming. The results of the present study differed from research that has incorporated meaningful gestures in dual task paradigms, by finding that non-meaningful gestures interfered with subsequent new word learning. The present study, however, differed from this previous research on several key points, which may account for the discordant results. One important element was the gesture in the previous studies was meaningful (rather than non-meaningful as in the present study), suggesting that meaningful gesture may enhance language performance while non-meaningful gesture may hinder language performance. It must also be noted that the nature of the tasks differed between the studies. The previous studies involved explaining problems with or without hand movements while simultaneously remembering lists of familiar items (e.g., letters, numbers). In contrast, the present study focused on the effects of hand movements on new word learning and familiar picture naming. Thus, one key feature of the previous studies was that the gesture was incorporated into an explanation task, rather than being a separate unrelated task. The different nature of the experimental tasks (i.e., explanation vs. new word learning) may have contributed to the differing patterns of results. Future research could examine differences between meaningful and non-meaningful gestures on new word learning in more detail. Much of the available research regarding hemispheric asymmetries during dual tasks has utilised relatively simple verbal and motor tasks, which is in contrast with the current study. Previous research has largely focussed on simpler tasks such as reciting known material, reading aloud or simple repetition as the verbal component in dual task investigations. These studies have provided evidence that performing a simple verbal task with concurrent right hand movements can disrupt performance. The present study extends this previous research by demonstrating that hand-related interference occurs for more complex linguistic tasks performed immediately before verbal tasks. The word learning and picture naming tasks used in the current study are also assumed to place higher demands on short-term memory than previous studies which involved simpler linguistic tasks. Given the established relationship between new word learning and short-term memory it could be suggested that new word learning may be more vulnerable to dual task interference effects than simpler linguistic tasks as a result of the increased cognitive requirements on short- term memory during new word learning. As well as the increased complexity of the linguistic tasks in the current study, the motor movements studied here can also be considered more complex than those typically used in previous studies. The motor movements required in previous research have included sequential finger tapping (tapping each finger on the hand as quickly as possible), simple finger tapping (tapping one finger on the spot), arm tapping (tapping a closed fist on a given spot), button pressing, and dowel balancing,. In contrast, the task required for the current study was a complex, non-meaningful movement of opening a box and pressing a button located within it. Medland et al. (2002) observed that the type of motor task performed influences the size of the interference effect. It was proposed that participants allocate more attention to the task that they find most difficult. The current study also supports the previously established link between verbal short-term memory and the capacity to learn new words, by finding a relationship between performance on memory span forward tasks and word learning success. It has been suggested that verbal short-term memory and lexical acquisition share a common cognitive and neural network. Working memory also plays an important role in dual task effects. When carrying out a dual task, such as the motor and linguistic tasks in this study, both tasks need to be held in working memory with attention shared between the tasks. Therefore, it can be assumed that better performance on memory tasks may be predictive of performance on dual tasks given the increased working memory capacity. The results of the current study support this proposal, as demonstrated by a positive correlation between the working memory tasks (word and digit span forward) and performance on the dual tasks utilised in the study (new word learning and familiar picture naming combined with the complex non-symbolic movement). The lack of correlation between digit span backward and word learning in the current study is also of interest. This finding may be explained by the difference in the functions required for memory span forward and backward tasks. The word and digit span forward tasks used in this study involve the maintenance of information only, whereas the digit span backward task requires both the maintenance and active manipulation of information. It may be inferred that working memory tasks involving the simple maintenance of information are more closely related to the processes involved in new word learning studied here. It must be noted, however, in the present study that while right hand movements but not left hand movements differed significantly from baseline during the recall of newly learnt words, there was no significant difference between left and right hand movements on the task. The absence of a significant interference effect with the left hand during the new word learning task may have been an artefact of statistical power. It is possible that with an increased sample size, a statistically significant interference effects might be observed between these conditions or alternatively interference effects may have been observed for both hand movements compared to baseline (no movement). ## Conclusion In conclusion, the present study demonstrated that performing complex, non- meaningful, hand movements interfered with subsequent lexical acquisition and retrieval in healthy individuals. This finding is consistent with previous dual task studies involving simultaneous tasks and non-meaningful movements in healthy individuals. However, the results of the current study appear to be contradictory with findings of intention treatment and facilitation effect found in individuals with aphasia (despite using a similar experimental design). This discrepancy suggests that the facilitation effect found in aphasia may be more due to motor speech facilitation (i.e., the facilitation of body parts involved in speech production) than lexical retrieval or alternatively that normal mechanisms were not being tapped into in this treatment. The present study also conflicts with the results of dual task studies in healthy adults involving meaningful movements, suggesting that the meaningfulness (or lack of meaningfulness) of the motor movements may be an important factor. The present study supports the established link between new word learning and short-term memory and the premise that performance on short-term memory tasks may be predictive of new word learning capacity. Future research is required to fully understand the interactions between intentional non-meaningful movements and language performance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JC DAC. Performed the experiments: JC. Analyzed the data: JC EF DAC. Contributed reagents/materials/analysis tools: JC EF DAC. Wrote the paper: JC EF DAC.

# Introduction Oocyte cryopreservation offers a number of potential benefits for assisted reproductive technologies, the treatment of infertility and biotechnology. For example, it reduces the necessity for women undergoing chemotherapy to have to rely on the highly invasive approach of ovarian freezing and subsequent grafting to rescue oocytes. It also reduces the need for women to undergo repeated hyper- stimulation protocols as oocytes, rather than embryos, could be stored. Furthermore, oocyte cryopreservation is particularly attractive for women in countries where embryo freezing is restricted through legislation and emphasis lies on gamete freezing. Cryopreservation also offers the opportunity to store oocytes from endangered species that can then be either fertilised with fresh or cryopreserved sperm from the same species or sub-species or used as recipients for somatic cell nuclear transfer (SCNT) to propagate the desired genetic material. The resultant embryos could then be transferred to the mother or a surrogate to produce live offspring or cultured in vitro to derive embryonic stem (ES) cells. Characteristically, ES cells derived from the same embryo are genetically identical and self-renew. They are also pluripotent in nature, which confers the potential to differentiate into all cell types of the body and provides an excellent resource for studying cellular differentiation and development, especially when the gametes used to generate these cells are from near extinct or endangered species. Nevertheless, the reestablishment of a species is often dependent on the use of a somatic cell transferred into the enucleated oocyte from a closely related species or subspecies. However, a number of genetic and epigenetic abnormalities arise from SCNT due to the somatic nature and epigenetic memory associated with these cells and, as a result, the process is highly reliant on the reprogramming of the somatic cell by the recipient oocyte's cytoplasm. The use of fertilised cryopreserved oocytes would allay the use of this more genetically divergent approach and promote chromosomal and mitochondrial genetic identity and integrity. In this instance, sperm stored from the same species would be used to fertilise cryopreserved oocytes, which would then be cultured to the blastocyst stage from which ES cells could be harvested. This would provide an unlimited source of pluripotent ES cells for subsequent rounds of nuclear transfer and would ensure that the oocyte would not need to reprogramme the transferred cell. In turn, this would increase nuclear transfer outcome and reduce the number of abnormalities arising from reprogramming. Common approaches to oocyte cryopreservation include: i) slow-freeze and rapid thaw; and ii) rapid freezing and warming, i.e. vitrification. Both approaches are highly dependent upon cryoprotectant agents (CPA) that protect oocytes from damage during the freezing process. Currently, these approaches necessitate the injection of cryoprotectants into the oolemma, laser ablation of the zona pellucida to promote sperm penetration or intracytoplasmic sperm injection. In terms of cryobiology, temperature changes during freezing damage the structural integrity of the oocyte, which can result in oocyte activation and parthenogenesis. Furthermore, cryopreservation can lead to premature extrusion of corticle granules resulting in zona hardening ; microtubular spindle injury and condensed chromosomes resulting in, for example, atypical haplotypes. It is likely that cryopreservation will affect mitochondrial integrity, which will reduce the oocyte's stores of ATP and affect oocyte function. Indeed, mitochondrial function is vital to oocyte competency and ATP, generated through oxidative phosphorylation (OXPHOS) in the mitochondria, rather than cytoplasmic glycolysis, is essential to mediate fertilisation and post-fertilisation events and the first cell division. Furthermore, insufficient numbers of mitochondria in the metaphase II oocyte often result in preimplantation development arrest. Collectively or individually, these outcomes will reduce the oocyte's potential to undergo and complete fertilisation and lead to successful and normal development. In the mouse, modified slow-freezing protocols, with increased sucrose and the replacement of sodium with choline, have resulted in fetal/live birth rates of up to 47.6%. On the other hand, vitrification of mouse oocytes in a G1.2 medium without NaHCO₃ but supplemented with HEPES and using a nylon loop system, resulted in fetal rates of 37.9%. Although the cryoprotective action of sucrose is well-documented, other CPAs such as trehalose and lactose are more able to stabilise membranes in an aqueous environment than sucrose and would thus provide post-warmed cryopreserved oocytes with minimum intra- and extra- cellular disruption. This would most likely reduce the need for intracytoplasmic sperm injection (ICSI) or zona ablation following warming and promote the maintenance of the oocyte's structural integrity. Nevertheless, the inclusion of sucrose allows the toxic effects of DMSO, a key CPA, to be buffered thus adding to the overall protection provided to the oocyte. We have set out to demonstrate that oocyte vitrification using trehalose, in combination with sucrose, is a viable alternative to embryo freezing and ovarian grafting approaches for the generation of ES cells and live offspring. We have demonstrated that vitrified oocytes can be fertilized in vitro where motile sperm can penetrate the zona pellucida and propagate syngamy, instead of relying on invasive processes such as ICSI, injection of cryoprotectants into the oolemma or laser ablation of the zona pellucida. We have demonstrated that embryos obtained from vitrified oocytes were able to generate live offspring. Similarly, when 2-cell embryos from vitrified oocytes were subjected to a further round of vitrification, live offspring were also obtained following warming and embryo transfer. However, the long-term culture of ES cells derived from vitrified oocytes suggests that they are more susceptible to abnormal karyotypes. The latter highlights the importance of the endometrial environment for the maintenance of genetic stability and thus the propagation of specific genetic traits. # Results ## Vitrification of oocytes In order to derive viable ESCs from vitrified oocytes, we generated a series of blastocysts from oocytes treated with a sucrose-DMSO-EG-based vitrification solution containing a range of trehalose concentrations. In this respect, oocytes were vitrified with one of 0 (control), 0.1, 0.2 and 0.3 M trehalose concentrations and following warming, they were either cultured in fertilisation medium or fertilised with fresh epidydimal sperm and cultured in vitro to blastocyst. Parthenogenic activation was not observed when fresh or warmed oocytes were cultured in either mT6 and KSOMaa, as described for the *in vitro* fertilisation (IVF) experiments. Significantly more oocytes treated with 0.1 and 0.3 M trehalose survived post-warming recovery than those treated with 0.2 M trehalose (P\<0.05; see). The 0.1 and 0.3 M trehalose treated oocytes also exhibited recovery rates similar to non-trehalose treated oocytes. Similarly, a higher number of 0, 0.1 or 0.3 M trehalose treated oocytes successfully underwent IVF and progressed to the 2-cell stage than the 0.2 M cohort. Again, significantly more 0.1 and 0.3 M treated cohorts developed to blastocyst than the 0.2 M cohort (P\<0.05). In contrast, blastocyst rates were significantly lower for the vitrified oocytes when compared with embryos produced with fresh oocytes. As significantly fewer oocytes from the 0.2 M trehalose cohort survived post- warming, we determined whether the vitrification process affected the distribution of mitochondria following warming. We stained oocytes from each cohort with MitoTracker Deep Red FM and assessed the distribution of mitochondrial clustering. The clustering was similar for the fresh, 0 and 0.1 M treated groups, although for the 0.3 M treatment it was significantly lower (P\<0.05;). The values for the 0.2 M cohort demonstrated very little clustering relative to the fresh oocyte population (P\<0.05;). We then cultured a cohort (n = 15) of 2-cell embryos from the 0.3 M cohort to the blastocyst stage, to determine whether they could give rise to viable ESCs. Whole blastocyts were plated onto a monolayer of MEFs to promote their expansion and to produce colonies of cells that resembled pluripotent ESCs, which were then further passaged to demonstrate their ability to generate proliferative colonies. In all, 9 lines were established and one was fully characterised, ViO- ES9. ViO-ES9 was analysed for expression of genes associated with pluripotency. They were positive for the expression of *Oct4*, *Dppa3*, *Dppa5* and *Pramel7* but failed to express *Nanog*. Using immunocytochemistry (ICC), we also confirmed that ESC colonies contained a few isolated areas of positive staining for NANOG but no staining for OCT4, SSEA-1 or alkaline phosphatase. In contrast, D3 (control) cells exhibited positive staining for NANOG, OCT4, SSEA-1 and alkaline phosphatase in most cells. Nevertheless, we further tested the pluripotent potential of the ViO-ES9 cells in vivo by generating teratomas. Undifferentiated cells were injected into the hind leg of SCID mice and, after 6 weeks, teratomas were recovered. Heamatoxylin and eosin staining revealed that they contained tissues indicative of the three germ layers including secretory epithelium (endoderm), cartilage (mesoderm) and epidermal tissue (ectoderm). The ViO-ES9 cells were also differentiated using the hanging droplet method to produce embryoid bodies, which were plated onto culture dishes to promote further differentiation. The outgrowths showed clear indications of differentiating cells and they were analysed by RT-PCR and ICC for markers of each of the 3 germ layers. Colonies were positive for endoderm (AFP –), ectoderm (Nestin – ; ; Vimentin –), and mesoderm (GATA4 – ; Flk-1, VE-Cadherin, PECAM –). Close inspection of the *in vitro* growth patterns within the culture dish of late passage cells led us to analyse the karyotype of these cells. This was abnormal (data not shown). Subsequent analysis of early passage cells revealed the same outcome. Cells were commonly trisomies for chromosomes 1, 6 and 11 and contained an extra Y chromosome. In order to determine whether the abnormal karyotype resulted from the vitrification protocol or the in vitro culture environment, we transferred embryos from fresh and the 0.1 and 0.3 M trehalose treated cohorts of oocytes to surrogates. Live offspring were generated from each cohort. Although higher numbers of blastocysts were obtained following IVF with fresh oocytes, no significant differences were observed for offspring rates following embryo transfer. The vast majority of these offspring generated from the 0.1 M and 0.3 M trehalose-treated oocytes (9/10) were karotypically normal. As one of these animals was karyotypically abnormal, we decided to further test whether vitrification was responsible for the chromosomal instability. Here, we again used fertilised fresh and vitrified oocytes from the 0.3 M cohort to produce 2-cell embryos, which were vitrified and stored for a minimum of one month. Following warming, 2-cell embryos were transferred to surrogates who gave birth to live offspring. None of the 3 offspring analysed showed any chromosomal instability (data not shown), although one offspring was larger than the others at birth and remained so during adolescence and adulthood. It subsequently died at 7 months whilst the others appeared healthy up to the age of 12 months at which point they were humanely killed. Furthermore, in order to determine whether the offspring derived from oocyte vitrification had normal reproduction capability, we mated control and 0 and 0.1 M cohort males with C57Bl6/J females. This resulted in the generation of live offspring at a rate of 6.14, 6.25 and 5.5 offspring per mating, respectively. # Discussion Vitrification offers the opportunity for the long-term storage of oocytes and we have demonstrated the cryoprotective potential of administering extracellular trehalose on mouse oocytes during vitrification and warming. Viability was demonstrated by oocyte post-warming survival, cleavage, pre- and post- implantation embryo development and through the generation of live offspring. We have further demonstrated that two-cell embryos from vitrified oocytes can successfully undergo a further round of vitrification and subsequent warming and give rise to live offspring. These offspring appeared to have normal reproductive function as they produced viable offspring. However, vitrified oocytes that are warmed, fertilised and then cultured to the final stages of preimplantation development had a tendency to genetic instability as demonstrated by the chromosomal aberrations observed in the ES cells derived from their blastocysts. There is a substantial body of evidence to suggest that temperature and osmotic changes during the freezing and thawing process can damage the structural integrity of cells. Oocyte cryopreservation has also been reported to damage the mitotic spindle, harden the zona pellucida and to promote oocyte activation, which leads to parthenogenesis. The osmolyte, trehalose, has been used to protect mammalian cells, gametes and embryos following freezing or storage in the dry state. Addition of trehalose to a cryoprotective solution significantly improved the colony forming units of haemotopoeitic cells following thawing and culture. Similarly, extracellular trehalose has been reported to either maintain or increase the post-thaw motility of boar, bovine and ram sperm. Mouse embryo viability was reported to be double when frozen in the presence of trehalose rather than in its absence. Furthermore, Guo et al. showed that human primary fibroblasts, which were engineered to produce this saccharide, were viable following rehydration after storage of up to five days at the dry stage. Likewise, mouse and rhesus macaque sperm have been reported to promote fertilisation and embryo development following ICSI. In our work, the post-warming survival rates were not significantly different when oocytes were vitrified in 0 (85.4%), 0.1 (82%) or 0.3 M (85.9%) trehalose but survival was significantly lower when vitrified in the 0.2 M solution (37.2%). Trehalose is understood to be a non-toxic osmolyte and the non-linear dose response of oocytes to 0.2 M solution observed in our studies was not expected. Consequently, we assessed mitochondrial clustering in each of the cohorts at this stage where we know from previous work that mitochondrial content in oocytes is a key criterion associated with developmental competency. We report here that the addition of 0.1 M trehalose to the CPA improves the distribution of mitochondrial clustering, but 0.2 M trehalose has a harmful effect as evidenced by poor clustering and the increased presence of cytoplasmic-like vacuoles (cf). However, increasing the trehalose concentration to 0.3 M appeared to partially restore mitochondrial clustering. These outcomes suggest an ‘all or none’ effect whereby the lower 0.1 M trehalose concentration is sufficiently low not to be harmful to mitochondria, but 0.3 M is present in sufficient abundance to have a protective effect. Consequently, the 0.2 M trehalose concentration appears to negate this protective effect. Mitochondria are sensitive organelles with their viability dependent on a high mitochondrial membrane potential. Indeed, mitochondrial dysfunction in the oocyte is responsible for the early arrest of preimplantation embryos *in vitro* and when mediated through disruption to the mitochondrial membrane potential can result in apoptosis, which would account for some of the embryo loss during preimplantation development. Furthermore, the failure of metaphase II oocytes to establish mitochondrial developmental competency, and then supplemented with divergent populations of mitochondria, can result in a series of disorders, such as pervasive development disorder, and genetic instability as evidenced by Turner's (XO) syndrome and hence the genomic instability we observe in our ESCs. Cleavage rates for our vitrified oocytes (50–74.4%) following IVF were not significantly different to the control IVF cohort (76.4%). Similar cleavage rates have been reported for oocytes cryopreserved in the presence of 0.5 extracellular trehalose (63%;), or when slowly frozen using a non-programmable freezer in a CJ2 solution containing 18% fetal bovine serum, 1.5 M propanediol and 0.1 M sucrose, and following laser ablation of the zona pellucida prior to IVF (48%;) but lower than those when trehalose was microinjected into the oocyte prior to cryopreservation (83%;). Nevertheless, following warming, sperm were capable of penetrating the zona pellucida of our vitrified oocytes, fertilisation and promoting embryo development to the hatched stage without having to perform ICSI or zona laser ablation. Our developmental rates (number of blastocysts per two-cell stage embryo) were 52.4, 80.5, 29.4 and 78.0% from oocytes vitrified in the presence of 0 M, 0.1 M, 0.2 M and 0.3 M trehalose, respectively. Following embryo transfer, there were no significant differences in offspring survival rates between non-vitrified and 0.1 M trehalose vitrified oocytes (20% v 27.3%). However, oocytes vitrified in 0 or 0.3 M trehalose resulted in 13% and 10% live offspring, respectively. These rates are similar to those reported for oocytes cryopreserved by slow cooling with intra- and extracellular trehalose. Following IVF and embryo transfer, these workers reported the birth of five pups (23%) from non-cryopreserved control oocytes and 4 (19%) from cryopreserved oocytes when embryos were transferred at the two-cell stage. Similar results were reported by these workers when they transferred blastocysts from unfrozen (7 pups; 28%) and frozen oocytes (3 pups; 14%) into surrogates. In contrast, live offspring rates of 26.7% and 46.7% have been reported from oocytes cryopreserved in a choline based solution with 1.5 M 1,2-propanediol and 0.1 M sucrose, and using a programmable or non-programmable freezer, respectively and subjected to laser ablation of the zona pellucida prior to IVF. Nevertheless, the live offspring generated from trehalose-vitrified oocytes maintained genetic stability as evidenced by their normal karyotypes. Our data would clearly suggest that there is a developmental advantage for fertilised vitrified oocytes being transferred at the 2-cell rather than the blastocyst stage. The chromosomal aberrations that we observed in early and later passage ESCs were most likely the result of in vitro culture to the blastocyst stage followed by ESC derivation. This also seemed to affect their ability to maintain pluripotent gene expression although they produced teratomas and representative gene expression of markers of differentiation. It is thus highly likely that the uterus of the surrogate mouse offers considerable metabolic support that the in vitro culture environment is unable to provide. Indeed, in vitro culture can result in a variety of genetic and epigenetic abnormalities. Similar outcomes arise from poor diet prior to conception and during pregnancy, which affect epigenetic integrity, offspring size and development rates. Nevertheless, it is likely that the inter-uterine environment of healthy and reproductive competent mice will be better able to sustain and support the metabolic function of these embryos during early development and until the embryonic genome takes over and mitochondrial replication is initiated at the blastocyst stage. Consequently, key molecular events such as cytokinesis, karyokinesis and *de novo* methylation can be appropriately metabolically supported. The current in vitro culture systems are likely to be unable to provide this degree of metabolic support. This is further substantiated by the generation of live offspring from the vitrification of 2-cell embryos generated from vitrified oocytes, that were able to, on the whole, maintain genetic integrity and give rise to live offspring. This, to the best of our knowledge, is the first report of such an outcome and should eliminate any concerns that vitrification may raise for subsequent offspring survival. However, endeavours to generate pluripotent embryonic stem cells from vitrified oocytes to provide multiple genetically identical donors for nuclear transfer will require refined culture systems to ensure that genetic abnormalities can be avoided and that these cells offer a viable approach to rescue endangered species. A further concern regarding vitrification has been the potential for cross- contamination. Vitrification of oocytes and embryos is traditionally carried out in open systems such as open pulled straws, electron microscope grids, cryoloop, solid-surface vitrification or nylon mesh. The use of these systems subjects the oocytes or embryos to direct contact with liquid nitrogen and to the risk of contamination as it is widely acknowledged that viral and bacterial pathogenic agents survive in liquid nitrogen. Bielanski et al observed that, when samples are stored in sealed straws or closed cryovials, the risk of pathogen contamination during vitrification and storage of samples in liquid nitrogen is removed. They reported that, when samples were stored in unsealed containers, the incidence of viral contamination due to exposure to contaminated liquid nitrogen was 22% whilst no contamination was observed in samples stored in closed containers. Therefore, the method presented here reduces this risk, as straws are sealed with a metal sealing ball before vitrification is carried out. Indeed, use of the method presented here together with the use of sterilised liquid nitrogen either by UV exposure and storage in vapour phase tanks will assist in avoiding the risk of pathogen cross-contamination during sample storage. In conclusion, we have shown that oocytes can be successfully vitrified using a combination of trehalose and sucrose based CPAs and that successful vitrification using trehalose is concentration-dependent. The use of 0.1 and 0.3 M trehalose provides the oocyte with mitochondrial protection and ensures that, when embryos are transferred at early stages of development, live offspring can be successfully produced. Furthermore, this form of oocyte storage can be used as part of a ‘closed’ vitrification system and will prevent cross-contamination. # Materials and Methods Experimental procedures were approved by the Monash Medical Centre Animal Ethics Committee for animal experimentation and were conducted in accordance with the Australian National Health and Medical Research Council (NHMRC) guidelines. Approval numbers were MMCA 2007/1 and MMCA 2007/39. ## Mice F1 hybrid mice (C57Bl×CBA) were used for the present study. Mice were kept under constant environmental conditions of 12L∶12D and a temperature of 25°C. ## Media All media were prepared from analytical grade chemicals (BDH Chemicals Pty., Ltd., Melbourne, Australia or Sigma-Aldrich Co., St. Louis, MO). Oocytes were collected in modified KSOM-HEPES and cultured in modified KSOMaa. Embryos were also cultured in modified KSOMaa. Sperm were collected and capacitated in modified Tyrode's medium (mT6;) without sodium pyruvate or sodium lactate. ## Oocyte collection Six-week-old females were superovulated by an intraperitoneal injection of 5 IU equine chorionic gona-dotrophin (Folligon, Intervet International, Boxmeer, The Netherlands) followed by 5 IU hCG (Chorulon, Intervet International) in 0.2 ml saline 48 h later. At 13 to 14 h after hCG injection, oocyte-cumulus complexes were collected from the oviducts into KSOM-HEPES containing 60 IU/ml hyaluronidase (Sigma type IV-S) for 5 min. Adhering cumulus cells were removed from oocytes by gentle pipetting and washed in KSOM-HEPES. Oocytes were then placed into 20 µl drops of modified KSOMaa previously equilibrated under mineral oil in 5% CO₂ in air at 37°C for at least 30 min before they were vitrified or placed into a 90 µl drop of mT6 under oil in 5% CO₂ in air for 30 min before IVF was carried out. ## Oocyte/Embryo Vitrification and Warming Exposure of oocytes and embryos to cryoprotectants was carried out on the microscope stage set to 37°C. Cohorts of oocytes or two-cell embryos (5–10), as indicated, were placed into 20 µL drops of KSOM-Hepes containing 8% DMSO (1.4 M) and 8% (1.4 M) ethylene glycol for 1 min at 37°C. Oocytes and two-cell embryos were then transferred to KSOM-HEPES containing 16% DMSO (2.8 M) and 16% ethylene glycol (2.8 M), 10 mg/ml Ficoll, 0.65 M sucrose with 0, 0.1, 0.2 or 0.3 M trehalose for 20 s at 37°C. Oocytes were then loaded into 0.25 ml straws sealed with a metal sealing ball (Minitube, Smythesdale, VIC, Australia) slowly pressed against a metal block (Cryologic Vitrification block CVM); Cryologic, Mulgrave, VIC, Australia) for 30–45 s and then transferred into liquid nitrogen and stored for at least a week in a liquid nitrogen vapour phase tank. ## Warming Straws were removed from the storage tank and transferred into a container with liquid nitrogen. Before warming, straws were held for 30–40 sec in the vapour phase of liquid nitrogen, held in the air for 5 sec then transferred into a water bath at 37°C. The straws were then dried, the sealing ball removed and the oocytes expelled into a 20 µl drop of KSOM-Hepes containing 0.25 M sucrose at 37°C for 2 min. The oocytes were then transferred to a 20 µl drop of KSOM-Hepes containing 0.125 M sucrose at 37°C for a further 3 min, then into KSOM-Hepes at 37°C for 5 min and finally into a 90 µl drop of mT6 under oil in 5% CO₂ in air for 30 min before IVF was carried out. ## Oocyte imaging MitoTracker Deep Red FM (Invitrogen) was used to visualize oocyte mitochondria. Zona pellucidae were first removed by incubation in an acidified Tyrode's solution for 3–5 minutes. After recovery, zona-free oocytes were incubated in KSOM-Hepes containing 100 nM of Mitotracker Deep Red for 45 min. Oocytes were then washed twice with PBS containing 0.05% (v/v) Tween-20 (PBS-T) and fixed in 2% (w/v) paraformaldehyde in PBS-T. Following fixation, zona-free oocytes were transferred to an 8-well slide (Ibidi) for analysis with a Nikon C1 laser scanning confocal microscope on a Nikon Ti-E inverted microscope base. 60×1.4NA objective was used to image samples. 633 nm laser was used to specifically excite MitoTracker Deep Red and the fluorescence emission was detected via a 645/75 nm filter. ## Ripley's clustering analysis Ripley's K-function, for three-dimensional data, weighted by the intensity of point:The intensity of a data point x (given by Int(x)) was considered to be the assumed (relative) number of actual agents (or points) in that position. V is the volume of the domain, N is the number of data points, and S is the sum of all intensities:, i.e: the total number of assumed agents. c(i,j,r) is an indicator function, which obtains the value = 1 if the distance between the i'th and j'th points is less or equal to r, and = 0 otherwise. For a data set in a three dimensional space, the benchmark obtained by a homogeneous Poisson process is:. Values significantly different to it will indicate the points are not distributed homogenously, i.e. the points are spatially clustered. In order to avoid underestimation of K due to edge effects we introduced a buffer zone, so that the first sum would only go over points for which the distance from any of the domain's boundaries was larger or equal to the largest value of r. ## In vitro fertilization (IVF) Sperm were collected from 10 to 12-week-old males. Briefly, both caudae epididymides were dissected from the mice and a small slit was made in each cauda before transfer to 1 ml modified mT6 and equilibrated in 5% CO₂ in air at 37°C in a 5 ml (NUNC, Roskilde, Denmark) under oil. The epididymal tissue was then removed from the tube 1 h later, and the sperm concentration was assessed and adjusted to 3 to 5×10⁶ sperm/ml and a 10 µl aliquot was added to the oocytes. Sperm and oocytes were incubated for 6 h, under mineral oil. Oocytes were then removed from the sperm solution, washed in KSOM-Hepes, and cultured at 37°C in modified KSOMaa under mineral oil in 5% CO₂ in air. Six hours later, oocytes were examined for signs of normal fertilization. Oocytes with two pronuclei and a second polar body were regarded as fertilized. These oocytes were separated from unfertilized oocytes and cultured in modified KSOMaa at 37°C under 5% CO in air up to 96 h for development to the blastocyst stage. ## Parthenogenic activation Twenty to twenty five freshly collected MII oocytes and vitrified oocytes per cohort were thawed and transferred into a 90 µl drop of modified Tyrode's medium under oil in 5% CO₂ in air for 6 h without sperm and cultured in modified KSOMaa at 37°C under 5% CO₂ in air for up to 96 h. ## Embryo transfers Two-cell embryos produced following IVF were transferred to the oviducts of Day 1 pseudopregnant F1 females mated to vasectomized F1 males of proven sterility. Between 8 and 12 embryos were transferred to each oviduct. Females were allowed to give birth and the number of pups on the day of birth was recorded. ## Natural mating Two males from the breeding colony (control) and from the 0 and 0.1 cohorts, were bred for two cycles with C57Bl6/J six week old females and the number of live offspring per male/treatment recorded. ## Culture and preparation of feeder layers Murine embryonic fibroblasts (MEFS) were cultured in High Glucose Dulbecco's Modified Eagle's Medium (DMEM; Sigma, Gillingham, UK) with 10% fetal bovine serum (FBS; Sigma), 2 mM L-glutamine (Invitrogen Life Technologies, Paisley, UK), 1% non-essential amino acids (NEAA; Invitrogen Life Technologies) and 1% penicillin/streptomycin solution (Sigma). MEFs were inactivated using mitomycin C (Sigma; 10 µg/ml) for 2 hours at 37°C/5% CO₂, and then washed 3 times with PBS (Sigma) and cultured overnight or frozen before use as feeder cells for undifferentiated mES cells. ## Derivation of mES cells Expanded blastocysts were cultured until they hatched. They were then transferred to 96-well plates containing MEFS and blastocyst culture media and allowed to expand. Once an expanding colony had been identified, the cells were cultured as for D3 mES cells. They were manually passaged onto 48-, 24- and 6-well plates (BD Falcon, North Ryde, NSW, Australia) containing MEFS, and once established, onto Petri-dishes (BD Falcon). ## Culture of mES cells D3 mES cells and ViO- ViO-ES9s were cultured, as described in. Briefly they were placed in High Glucose DMEM (Sigma) with 15% ES cell screened fetal bovine serum (FBS; Hyclone, Utah, US), 1% penicillin/streptomycin solution (Invitrogen Life Technologies), 2 mM L-glutamine (Invitrogen Life Technologies), 1% NEAA (Invitrogen Life Technologies), 0.1 mM β-mercaptoethanol (Sigma) and 1000 U/ml LIF. ## Differentiation of mESCs Undifferentiated ViO-ES9 and D3 mES cells were induced to differentiate using the hanging drop method. Briefly, mES cells (Day 0) were dissociated with 0.25% trypsin-EDTA (Sigma) for 2 minutes, resuspended into mES cell media and plated as 20 µl droplets (approximately 450 cells per drop) on the lid of an inverted Petri dish (Sterilin, Staffordshire, UK) for 48 hours (Days 1 to 2) to promote the formation of EBs. The EBs were then placed into suspension for a further 5 (Days 3 to 7, D3 mES cells) or 6 days (Days 3 to 8, R1 mESCs) at 37°C/5%CO₂. The EBs were then plated onto gelatin (0.1%; Sigma) coated 6 well plates (Nunc, Roskilde, Denmark) and cultured up to Day 20 of differentiation. ## RNA isolation RNA was extracted from cells and contaminating DNA eliminated using the RNAqueous®-4PCR Kit (Applera, Warrington, U.K., <http://www.appliedbiosystems.com>) according to the manufacturer's protocol for mammalian cells. ## Reverse Transcription cDNA was produced using the Bioscript™ system (Bioline, London, U.K., <http://www.bioline.com>) according to the manufacturer's protocol. Each reaction contained 5 µg RNA and 200 U of reverse transcriptase. The reaction was performed at 42°C for 1 h in a MJ Research PTC-200 thermocycler (GRI, Braintree, UK). ## PCR analysis PCR was performed in 50 µl reactions containing 200 ng of cDNA, 1× PCR buffer (Bioline, London, UK), 1.5 mM MgCl₂ (Bioline), 100 µM dNTPs (Bioline), 0.5 µM for each of the forward and reverse primers (Gapdh: GGGAAGCCCATCACCATCTTC and AGAGGGGCCATCCACAGTCT to produce 366 bp; Oct4: AGTATGAGGCTACAGGGACA and CAAAGCTCCAGGTTCTCTTG - 260 bp; Dppa3: CTTTCCCAAGAGAAGGGTCC and TGCAGAGACATCTGAATGGC - 149 bp; Dppa5: GCTTGATCTCGTCTTCCCTG and TCCATTTAGCCCGAATCTTG - 293 bp; Pramel7: AGAGAACCCACATGGCTTTG and GGATTTGGCTTGGCATACAT - 336 bp; Nanog: AGGGTCTGCTACTGAGATGCTCTG and CAACCACTGGTTTTTCTGCCACCG - 364 bp; Flk-1: TCTGTGGTTCTGCGTGGAGA and GTATCATTTCCAACCACCCT - 270 bp; *CD31*: 5′-CGCACCTTGATCTTCCTTTC and AAGGCGAGGAGGGTTAGGTA - 244 bp; and *VE-cadherin*: TCCTCTGCATCCTCACCATCACA and GTAAGTGACCAACTGCTCGTGAAT - 122 bp and 2.5 U *BioTaq* DNA polymerase (Bioline). Reactions were run on a MJ Research PTC-200 machine: initial denaturation at 95°C for 5 min; followed by 35 cycles of 94°C for 30 sec (except for Gapdh 25 cycles); annealing at 60°C for 45 sec (Gapdh; Flk-1), 55°C for 30 sec (Oct4), 57°C for 30 secs (VE-Cadherin), 60°C for 30 sec (Dppa3), 60°C for 30 sec (Dppa5), 60°C for 45 sec (Pramel7), and 62°C for 20 sec (Nanog); 72°C for 45 sec; and a final extension at 72°C for 3 min. PCR products were resolved on 2% agarose gels (Bioline) at 100 V for 1 h. ## Karyotype analysis Karyotype analyses of both mice and ES cells were performed by Southern Cross Pathology, Australia. In brief, for mice primary tail tip fibroblasts were isolated following mechanical dissection and incubated in trypsin for 7 minutes and cultured in MEF media until cell explants were observed. ES cells in their exponential growth phase were analysed. ## Teratoma generation To examine teratoma formation, 1–2×10⁶ cells were injected into the rear leg muscle of 4–6 week old severe combined immunodeficient (SCID) mice. After 4 weeks, the teratomas were excised and fixed in 4% paraformaldehyde, embeded in paraffin, sectioned at 5 uM and stained with Hematoxylin and Eosin using standard methods performed by the Monash Institute of Medical Research Histology Core Facility. ## Statistical analyses Survival, embryo development and offspring data were subjected to contingency tables and Fisher's exact test. For oocyte volume and mitochondrial clustering measurements, the means of each group were tested against the control (Fresh oocytes) by One-Way ANOVA test with Dunnett's Multiple Comparison post-test. We thank Sue Chapman, Southern Cross Pathology, and the Monash Institute of Medical Research, Histology Core Facility, for technical assistance. [^1]: Conceived and designed the experiments: JCSJ LGS-P. Performed the experiments: LGS-P JCSJ RDWK HS CYL RA. Analyzed the data: JCSJ LGS-P RDWK HS MKH MKOB. Contributed reagents/materials/analysis tools: JCSJ MKH MKOB CYL RA. Wrote the paper: JCSJ LGS-P MKOB MKH. [^2]: The authors have declared that no competing interests exist.

# Introduction The severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2) virus that caused the coronavirus disease (COVID-19) pandemic has resulted in public health crises and disrupted normal daily activities. Education is one area that has been adapted to accommodate rapidly changing government policies. According to UNESCO, public health measures to combat the infection have impacted around 365 million students globally. The schools in Asia were shuttered for roughly 40 weeks and replaced with online learning to continue learning. These modifications to online education have resulted in increased exposure time to the digital devices used by school children, which has a negative impact on ocular surface and leads to dry eye. The prevalence of dry eye symptoms has increased among school children over the past two decades, possibly due to a rise in screen exposure when interacting with the internet and social media. Prolonged screen use can cause dry eye symptoms by decreasing the blink rate and drying the ocular surface through evaporation. The resulting symptoms, including dryness, irritation, pain, eye fatigue, and deteriorated vision can have a negative impact on school-related activities. Moreover, the Tear Film and Ocular Surface Society (TFOS) released the Dry Eye Workshop II (DEWS II) Epidemiology report, also pointing out the importance of performing studies in school children or younger populations, which may have a relatively high prevalence to evaluate the potential risk factors, especially the use of digital screens. Thus far, the prevalence and risk factors of dry eye symptoms in school children have been reported in China, Japan, and Mexico, whereas studies in Thailand have been focused primarily on the elderly and university students. Among these studies, the results of well-known risk factors, such as gender and contact lens use, were inconsistent. In addition, smoking and perceived stress as contributing factors of dry eye symptoms remain under investigation. Moreover, the study in school children was conducted before the COVID-19 pandemic, so the result may not accurately reflect the situation digital screen time induced dry eye, which has increased as a result of lifestyle changes. Consequently, it is still necessary to assess the prevalence and identify risk factors for dry eye symptoms among school children. In this regard, the present study aimed to assess the prevalence of dry eye symptoms and digital screen time during the COVID-19 pandemic among students in grades 7 to 12 attending urban secondary schools. # Methods This constituted an online cross-sectional study conducted among grade 7 to 12 students in Samutprakarn Province, Thailand. The Research Ethics Review Committee for Research Involving Human Research Participants Group I at Chulalongkorn University (COA No. 208/201) approved the study protocol on October 7, 2021. Data collection was spanned from 1^st to 30^th November 2021. In section 1 of the online questionnaires, informed consent (the click-if-you-agree type) was obtained from the parents and respondents. ## Population and sampling method The population consisted of respondents studying in Samutprakarn Province, and the expected sample size was calculated as follows: $n = \frac{z_{\frac{\propto}{2}}^{2}Np(1 - p)}{d^{2}({N - 1}) + z_{\frac{\propto}{2}}^{2}p(1 - p)}$, where N is the population size, 50,362, $z_{\frac{\propto}{2}}^{2}$ is the standard normal deviation corresponding to a 95% confidence level, set at 1.96, and p is the prevalence of dry eye, assumed to be approximately 50%. The margin of error, denoted by was set at 5%. The estimated sample size (n) was 382 individuals. Due to the lack of response rate data for online surveys in secondary schools, a response rate of approximately 30% was assumed for the web-based study. Additions of 885 were added to the sample size. Consequently, total of 1,266 participants were recruited for this study. This study examined 25 secondary schools in Samutprakarn Province, clustered into four Secondary Educational Service Areas. Random sampling was applied to randomly select schools from these Secondary Educational Service Areas. Four schools, one from each Educational Service Area, were selected as the study sites. Further, each school was divided into six Grade levels (Grades 7 to 12). Next, two classrooms were selected at randomly from each grade at individual schools, and students were randomly selected using their student numbers from these classrooms. The exact number of samples was chosen at random based on the number of students in each school. When the total number of samples was collected from each school, it was used to calculate the number of available students in each grade level. Subsequently, the research tool was distributed to the selected students via Google classroom or the Line application group for the selected classrooms. The study included students in grades 7 through 12 between the ages of 12 and 18 years who voluntarily consented to participate. Students who were blind (no light perception) in both eyes, had an eye infection during the survey, had an underlying disease, such as allergy, diabetes, or autoimmune disease, or used antihistamines on a regular basis were excluded from the study. ## Translation of 5-item dry eye questionnaire (DEQ-5) The original version of the DEQ-5 was independently translated into the Thai language by a native Thai-optometrists and a native Thai speaker without a medical or clinical background. Next, these two translated versions were combined by the third independent translator, followed by a back translation of the combined version by an individual fluent in the original language. To curate the Thai-DEQ-5, the back-translated version was compared to the original version, and the research team consulted an expert ophthalmologist regarding discrepancies between the versions. The reliability of the Thai-DEQ-5 was examined with 30 students, yielding a Cronbach’s alpha coefficient of 0.847. ## Online questionnaire This study used a Thai online structured questionnaire comprising six sections and a total of 23 questions. **Section 1.** *Brief information*, *consent*, *and exclusion questions*: The first page of the online questionnaire contained brief information related to the meaning and symptoms of dry eye (definition of eye discomfort), followed by the consent and exclusion questions. **Section 2.** *Personal factors*: This section contained a total of two questions, including gender and academic year. **Section 3.** *Behavioral factors*: This section consisted of two questions regarding contact lens wear and smoking. **Section 4.** *Digital device factors*: This section consisted of two questions, including the most commonly used digital device types and digital screen time (the number of hours daily). **Section 5.** *Symptoms of Dry eye*: This section consisted of five questions about dry eye (symptoms/occurrence). The responses to three questions ranged from 0 (never) to 4 (constantly), and the responses to the remaining two questions ranged from 0 (never have it) to 5 (very intense). The possible values ranged from 0 to 22, with a cut-off of ≥6.0 indicating symptoms of dry eye and a DEQ-5 score of ≥12 indicating severe symptoms of dry eye. **Section 6.** *Stress*: This section often consisted of questions that assessed perceived stress. The Thai-Perceived Stress Scale-10 (T-PSS-10) was utilized to assess perceived stress levels. This instrument was a 5-point Likert scale with six negative items ranging from 0 (never) to 4 (the most frequent) and four positive items ranging from 4 (never) to 0 (the most frequent). The possible values ranged from 0 to 40, with higher numbers indicating more stress. Related studies showed that Cronbach’s alpha coefficient was approximately 0.71 to 0.87 among Thai secondary school students, and the present study showed it to be 0.82. ## Data analysis SPSS version 18.0 was utilized for all statistical analyses. Descriptive statistics such as percentage, mean, and standard deviation (SD) were applied for data characterization. The odds ratios (ORs) with 95% confidence intervals (95% CI) were calculated using logistic regression to identify the association, whereas multivariable logistic regression was used to investigate all significant aspects in the univariate analysis. Spearman’s correlation was used to establish the relationship between the DEQ-5 and T-PSS-10 scores. The Mann- Whitney U test was used to compare the continuous variables between the two groups, whereas the chi-square test was applied for non-continuous variable comparison. All statistical significance was based on a *p-*value of \<0.05. The scatter density plot and box plot were reconstructed from Scimago Graphica. # Results ## Respondents’ demographics and prevalence of dry eye 657 students responded to the questionnaire (51.9% response rate). Of these, 621 respondents consented to participate in this study. However, 18 (2.9%) respondents were excluded due to an eye infection or the constant use of antihistamines. Thus, 603 respondents were enrolled in this study, including 182 (30.2%) males and 421 (69.8%) females. 356 (59.0%) of all respondents were students in grades 7 through 9, while 247 (41.0%) respondents were students in grades 10 through 12. Among the respondents, only 25 (4.1%) respondents wore contact lenses. A small fraction (4.1%) of the respondents had experience with smoking. The majority of respondents (86.7%) self-reported that their primary digital devices were smartphones and tablets. More than one-half of the 603 (54.2%) respondents reported daily digital screen time between 6 and 12 hours. The overall mean ± SD of the Perceived Stress Scale was 19.0 ± 5.3 (Minimum = 4, Maximum = 39). All characteristics of the study respondents are summarized in. In this study, the overall prevalence of dry eye symptoms was 62.5%, and the overall mean (SD) DEQ-5 score was 7.2 (4.0). The general demographics and the comparison between the healthy respondents and those with symptoms of dry eye are shown in. The prevalence of dry eye symptoms was significantly higher in females (67.2%) than in males (51.6%). Students in grades 10 to 12 (71.3%) had symptoms of dry eye more frequently than students in grades 7 to 9 (56.2%). Interestingly, the prevalence of dry eye symptoms among contact lens wearers was 80.0%. The respondents with more time on digital screens exhibited an increased prevalence (39.1 to 71.2%). ## Correlation of DEQ-5 score with T-PSS-10 Spearman’s correlation analysis revealed a significant correlation between the DEQ-5 scores and the T-PSS-10 scores (r = 0.38, *p*-value \< 0.01). The density scatter-plot between the DEQ-5 and T-PSS-10 scores is presented in. Moreover, respondents were classified as having healthy eyes (DEQ-5 score \< 6), mild-to- moderate dry eye symptoms (DEQ-5 score ≥ 6 to \<12), or severe dry eye symptoms (DEQ-5 score ≥ 12). The mean T-PSS-10 scores increased significantly in the mild-to-moderate and severe dry eye symptoms categories relative to the healthy status (*p*-value *\<* 0.01). In the univariate analysis, wearing contact lenses (cOR 2.47; 95% CIs 0.92 to 6.69), smoking (cOR 0.64; 95% CIs 0.29 to 1.42) and major type of digital device use (cOR 1.06; 95% CIs 0.65 to 1.73) were not significantly associated with dry eye. However, being female (cOR 1.92; 95% CIs 1.35 to 2.74), a student in grades 10 to 12 (cOR 1.97; 95% CIs 1.39 to 2.79), digital screen time (6 to \<12 hours: OR 2.49; 95% CIs 1.44 to 4.31, ≥12 hours: cOR 3.86; 95% CIs 2.15 to 6.92 when compared to \<6 hours), and higher perceived stress score (cOR 1.13; 95% CIs 1.09 to 1.17) were likely to increase the risk of dry eye. All significant factors from the univariate analysis were further analyzed using multivariable binary logistic regression. Females had a 1.54-fold increased risk of symptomatic dry eye (aOR 1.54; 95% CIs 1.05 to 2.25) compared to males. Students in grades 10 to 12 were 1.77 times more likely to suffer from the dry eye than those in grades 7 to 9 (aOR 1.77; 95% CIs 1.23 to 2.57). Respondents with increased digital screen time (6 to \<12 hours: aOR 2.00; 95% CIs 1.12 3.57, ≥12 hours: aOR 2.54; 95% CIs 1.39 to 4.76) and higher perceived stress scores (aOR 1.12; 95% CIs 1.08 to 1.16) were more likely to experience dry eye symptoms. ## Sensitivity analysis There are no validated questionnaires for dry eye symptoms or cut-off scores currently available for use with children. To increase confidence in the outcomes, a sensitivity analysis using different DEQ-5 cut-off scores ranging from ≥4 to ≥ 8, was performed, which showed that the prevalence ranged from 47.8% to 78.1%. Analyzing the associated factors (sex, grades, digital screen time, and perceived stress) with dry eye produced similar results to the primary findings # Discussion Our results demonstrated a 62.5% prevalence and associated factors of dry eye symptoms during the COVID-19 pandemic, marking the first time school children have been the primary focus (aged 12 to 18 years). In comparison with the survey using dry eye symptom questionnaires in school children pre-COVID-19 pandemic, the present study showed a higher prevalence than studies by *Uchino et al*. (21.6%-2008) and *Zhang et al*., (23.7%-2010) of dry eye symptoms by using a Women’s Health Study questionnaire (WHS questionnaire). However, the prevalence shown in the present study is closer *Garza-Leon et al*.’s study, with 65.3% of respondents in Mexico reporting symptoms of dry eye by using the ocular surface index disease (OSDI) questionnaire. Contrary, the prevalence shown in the present study is slightly lower than the study conducted by *Lin et al*. during the COVID-19 pandemic, indicating 70.5% prevalence by using the OSDI scores in high school children. The present findings also revealed a higher prevalence (62.5%) than other studies conducted among university students in Thailand. *Supiyaphun et al*. Estimated 8.15% had symptomatic dry eye using the WHS questionnaire before the COVID-19 outbreak. Similar to the DEQ-5, the WHS questionnaire is short and simple; however, it includes question pertaining to history of the dry eye diagnosis, whereas the DEQ-5 only evaluates the symptoms. Subsequently, *Tawonkasiwattanakun et al*. demonstrated that 10.8% the dry eye symptoms occurred among open university members, students, and staff aged ≤29 years using the McMonnies Questionnaire at the onset of the COVID-19 pandemic (May to June 2020). This questionnaire differs from the others since the questions evaluate dry eye symptoms as well as known risk factors, such as age, sex, and systemic conditions associated with dry eye. Thus, the McMonneies questionnaire may be suitable for clinical screening, and adult populations. As aforementioned, the difference in prevalence may be a result of the growing popularity of online learning and intensive use of digital screens, which can lead to the development of the dry eye due to public health measures assigning the Thai students learning online until December 1, 2021. In terms of digital screen time, our results were consistent with related reports indicating that the increased use of digital screen time causes dry eye. Notably, the relationship between the duration of digital device use and dry eye varied significantly between studies. *Akib et al*.*’s* study in Indonesia suggested the use of a digital screen for more than three hours triggered symptoms and signs of dry eye, whereas *Muntz et al*. showed that extended screen time use in terms of weekly screen hours was associated with poorer symptomology, elevated blink rates, and proxy tear film stability. These differences may have been caused by additional variables not considered in these studies. *Rossi et al*. found that the number of years and duration of daily breaks from digital screen time varied significantly between groups with and without dry eye. In addition to screen time, perceived stress may also play a role in the symptomatic dry eye during COVID-19. Previous reports suggested that learning online and being physically inactive since implementing public health measures has been linked to increased perceived stress, which has been suggested as a risk factor for symptoms of dry eye in participants aged 15 and older. Although the pathophysiology underlying the relationship is unknown, several hypotheses have been advanced to explain it. First, the relationship may be a vicious cycle in which dry eyes and perceived stress mutually exacerbate one another. The severity of dry eye negatively affects daily life activities and is likely to increase perceived stress, as well as induce cortical production and promote pain perception, resulting in sensitivity to dry eye symptoms. Such conditions were supported by our results, as shown in, in which the DEQ-5 scores were positively correlated with perceived stress, and the severe dry eye symptom group showed the highest T-PSS-10 score. Secondly, perceived stress quite possibly induces pro-inflammatory and inflammatory cytokine secretion, which disrupts tear film homoeostasis and causes ocular surface inflammation. Notwithstanding, our findings suggest that perceived stress is positively associated with dry eye symptoms, which might be prevalent among school children experiencing perceived stress. The present study had a few limitations. First, the study relied solely on self- assessment to evaluate dry eye. The clinical examination may be necessary to confirm the reported cases. However, DEQ-5 is one suggested method to evaluate dry eye symptoms by the Dry Eye Workshop II. Second, none of the validated symptom questionnaires currently available address dry eye in school children. In addition, the Thai-DEQ-5 has not been validated with clinical signs, and a cut-off score ≥ 6 is commonly used in adults older than 18 years. Though, the DEQ-5 has been recommended for use to assess symptoms of dry eye in children. Moreover, sensitivity analysis revealed similar results suggesting the robustness of the outcomes. Thirdly, our study sites were selected from urban schools. Consequently, the results do not represent school children residing in other locations i.e., rural areas, and the study population does not represent the national population. Lastly, contact lens wear and smoking did not show a significant relationship with the dry eye; this could be due to the small sample size of the concerned groups. Regarding smoking, the formal education study environment might contain a small number of smokers. Further research should utilize both clinical examinations and validated questionnaires to confirm the results, and a multicenter study is required to estimate the prevalence of dry eye symptoms and their associated factors in the general population. # Conclusion To conclude, the present study revealed for the first time in Thailand that the prevalence of symptomatic dry eye among school children was high during public health measures, i.e., online education and home isolation, in response to the COVID-19 pandemic. Additionally, the prolonged use of a digital screen, increased perceived stress, older age, and female gender was associated with dry eye symptoms. This information should be shared with stakeholders such as parents, school personnel, and teachers in order to develop preventative measures. Guidelines or campaigns for online education, such as taking breaks during online learning, and encouraging physical activity at home, should be considered. # Supporting information The authors are appreciative of the teachers who approved and distributed the online questionnaire to their students. Additionally, we would like to thank every student who participated in this study. The authors appreciate Professor Dr. Nahathai Wongpakaran for providing permission to utilize the Thai-PSS-10 scale. The authors would like to thank Enago ([www.enago.com](http://www.enago.com/)) for the English language review. 10.1371/journal.pone.0284928.r001 Decision Letter 0 Mimouni Michael Academic Editor 2023 Michael Mimouni This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 6 Dec 2022 PONE-D-22-22320Prevalence and associated factors for self-reported symptoms of dry eye among Thai schoolchildren during COVID-19 outbreakPLOS ONE Dear Dr. Tawonkasiwattanakun, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jan 20 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Michael Mimouni Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please amend your current ethics statement to address the following concerns: a\) Did participants provide their written or verbal informed consent to participate in this study? b\) If consent was verbal, please explain i) why written consent was not obtained, ii) how you documented participant consent, and iii) whether the ethics committees/IRB approved this consent procedure. 3\. You indicated that you had ethical approval for your study. In your Methods section, please ensure you have also stated whether you obtained consent from parents or guardians of the minors (participants below the age of 18 years) included in the study or whether the research ethics committee or IRB specifically waived the need for their consent. 4\. We suggest you thoroughly copyedit your manuscript for language usage, spelling, and grammar. If you do not know anyone who can help you do this, you may wish to consider employing a professional scientific editing service.  Whilst you may use any professional scientific editing service of your choice, PLOS has partnered with both American Journal Experts (AJE) and Editage to provide discounted services to PLOS authors. Both organizations have experience helping authors meet PLOS guidelines and can provide language editing, translation, manuscript formatting, and figure formatting to ensure your manuscript meets our submission guidelines. To take advantage of our partnership with AJE, visit the AJE website (<http://learn.aje.com/plos/>) for a 15% discount off AJE services. To take advantage of our partnership with Editage, visit the Editage website ([www.editage.com](http://www.editage.com)) and enter referral code PLOSEDIT for a 15% discount off Editage services.  If the PLOS editorial team finds any language issues in text that either AJE or Editage has edited, the service provider will re-edit the text for free. Upon resubmission, please provide the following: The name of the colleague or the details of the professional service that edited your manuscript A copy of your manuscript showing your changes by either highlighting them or using track changes (uploaded as a \*supporting information\* file) A clean copy of the edited manuscript (uploaded as the new \*manuscript\* file)” 5\. Thank you for stating the following in the Acknowledgments Section of your manuscript:  This work was financially supported by the 2021 Fund from the Faculty of Optometry, Ramkhamhaeng University. However, funding information should not appear in the Acknowledgments section or other areas of your manuscript. We will only publish funding information present in the Funding Statement section of the online submission form.  Please remove any funding-related text from the manuscript and let us know how you would like to update your Funding Statement. Currently, your Funding Statement reads as follows:  This research was funded by Faculty of Optometry, Ramkhamhaeng University. The funder has no role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. Please include your amended statements within your cover letter; we will change the online submission form on your behalf. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The article addresses an important and relevant subject that might interest the audience of the journal. The study answers the Dry Eye Workshop II (DEWS II) Epidemiology report of the Tear Film and Ocular Surface Society (TFOS), in the context of increased exposure to digital media during the covid-19 pandemic and among younger children. However, the study quotes other studies that examined similar age groups and each contributing factor for dry eye. The novelty of the current research is not well established, besides the context of the accelerated exposure to digital devices during the covid-19 pandemic. The Literature review sets the stage for the study questions, supported by updated and relevant studies. The literature review is based on relevant studies about self-reported symptoms and clinical examinations of young adults and youth, related to exposure-time to digital media. The authors should re-examine that the review relates to all variables examined, e.g., it lacks information regarding gender differences from previous studies. The abstract points out that smoking and contact lens wear were not predisposing factors for dry eye, however the relevant groups were too small to have a significant statistical conclusion (not mentioned in the limitations section). The researchers' assumption of dry eye prevalence greater than 50% should be further established, since all the formula calculations are based on this rate. This rate seems high according to the best of my knowledge of the medical literature, especially among younger populations. The literature review leads to the study aims and the clear research questions but lacks a more specific hypothesis. Method the sampling process and the research population are well described. The research tools are well described, revealing quite high reliability, although they were originally established for adults. Authors should add information regarding translation validation. Results The authors provided rich information accompanied by detailed tables and plots, demonstrating the higher prevalence among girls, higher grades, prolonged exposure to screens, and higher perceived stress. The authors add information regarding the relations between dry eye symptoms and perceived stress that was prevalent during the pandemic. In addition, they identified the daily cutoff exposure time to digital media devices, analyzed by odds ratio. The discussion deals with the findings in relation to previous studies in the literature. The authors demonstrate a critical approach in the study limitations section, especially the lack of clinical evidence to dry eye. It is very relevant in teenagers' populations, which often demonstrate confounding results, especially regarding to self-reports of smoking in the formal educational system context. Conclusions The conclusion repeats the study findings and lack the bottom-line importance and significant contribution of the study, as well as lessons learned and recommendations for the future. Reviewer \#2: Thank you for the Manuscript. I really liked it. I think that DED is not studied enough in this group age and in this set up (covid19 time). please review my remarks I mention and then it's ready to be published. Thank you for allowing me to review this interesting paper. Research article - is suitable for this publish. Remarks abstract 1\. No remarks. Intro 2\. You mention “causing a negative impact on ocular surface” but the reference you bring showed only change in OSDI and not the other Ocular surface findings which showed no correlation. 3\. Line 69-70 you write about increased screen exposure and also mention social media – after reading the reference don’t see how’s it all related. Please elaborate or delete or change reference. Method 4\. Line 120 you mention “blind” do you mean VA of NLP both eyes? 5\. You told the students about dry eye and then asked them questions about dry eye - Did you have any concerns about that issue? (when you mention something and then ask about it – it might raise awareness) Results 6\. If you had almost 70% Females and DED was more prevalent among females – don’t it change all the statistics? Could you elaborate? Discussion 7\. Thank you for the limitation part. Conclusion 8\. Well written I think it is an interesting research article. I would consider to publish. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Adi Segal Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0284928.r002 Author response to Decision Letter 0 13 Jan 2023 Reviewer 1 Comments 1\. The Literature review sets the stage for the study questions, supported by updated and relevant studies. The literature review is based on relevant studies about self-reported symptoms and clinical examinations of young adults and youth, related to exposure-time to digital media. The authors should re-examine that the review relates to all variables examined, e.g., it lacks information regarding gender differences from previous studies. In the revised manuscript, the introduction has been reorganized by adding the information regarding the other factors of dry eye, i.e. gender, smoking, contact lens, and perceived stress, as shown in lines 76 – 79. 2\. The abstract points out that smoking and contact lens wear were not predisposing factors for dry eye, however the relevant groups were too small to have a significant statistical conclusion (not mentioned in the limitations section). Thank you for recommendation. We are truly in agreement with you about the limitation of the small proportion of contact lens wearing and smoking groups. Therefore, we added the limitation regarding the small groups, as shown in lines 322 – 325. 3\. The researchers' assumption of dry eye prevalence greater than 50% should be further established, since all the formula calculations are based on this rate. This rate seems high according to the best of my knowledge of the medical literature, especially among younger populations. We agree with you that this rate may be high among a younger population, but we assumed 50% because: 1\) During the first step of the project, we did not have the previous evidence regarding the dry eye symptoms in schoolchildren in the Thai population. 2\) Changes in lifestyle due to the COVID-19 pandemic may prolong screen time in children, which may result in increased dry eye symptoms in children and adolescents. Therefore, we assumed 50% of prevalence to provide the highest number for the sample size. Presently, there are several studies that show that the prevalence of dry eye is greater than 50% among university students,1 including Thailand.2 Moreover, the study in China indicated approximately 70% of dry eye among senior secondary school students during the COVID-19 pandemic.3 Thus, this rate is possible in this situation. 4\. The literature review leads to the study aims and the clear research questions but lacks a more specific hypothesis. In the revised manuscript, this section was re-written and combined with the previous paragraph as shown in lines 80 – 85. 5\. the sampling process and the research population are well described. The research tools are well described, revealing quite high reliability, although they were originally established for adults. Authors should add information regarding translation validation. Thank you for your recommendation. In the current version, we have added the information regarding the translation process, as shown in lines 122 – 129. 6\. The authors provided rich information accompanied by detailed tables and plots, demonstrating the higher prevalence among girls, higher grades, prolonged exposure to screens, and higher perceived stress. The authors add information regarding the relations between dry eye symptoms and perceived stress that was prevalent during the pandemic. In addition, they identified the daily cutoff exposure time to digital media devices, analyzed by odds ratio. The discussion deals with the findings in relation to previous studies in the literature. The authors demonstrate a critical approach in the study limitations section, especially the lack of clinical evidence to dry eye. It is very relevant in teenagers' populations, which often demonstrate confounding results, especially regarding to self-reports of smoking in the formal educational system context. We are absolutely agres that the survey in formal educational system is a limitation of smoking factor, so we added this limitation as shown in lines 324 – 325. 7\. Conclusions The conclusion repeats the study findings and lack the bottom-line importance and significant contribution of the study, as well as lessons learned and recommendations for the future. Thank you for recommendation. We have revised the conclusion section and provided the significant contribution of the study, as shown in lines 330 – 337. Reviwer 2 Comments 1\. You mention “causing a negative impact on ocular surface” but the reference you bring showed only change in OSDI and not the other Ocular surface findings which showed no correlation. Thank you for recommendation. We deeply agree that OSDI is not associated with the ocular surface findings. Thus, we have removed this reference. We added the study of “Blinking and normal ocular surface in school-aged children and the effect of age and screen time”.4 This study suggested that screen time may be associated with the ocular surface (Tear meniscus height (TMH) and non-invasive tear break-up time (NIBUT)). We have also added the study of “Ocular surface predisposing factors for digital display-induced dry eye”.5 This study suggested that digital screen tasks alter the TMH and NIBUT. 2\. Line 69-70 you write about increased screen exposure and also mention social media – after reading the reference don’t see how’s it all related. Please elaborate or delete or change reference. Thank you for recommendation. In the current version, this reference has been deleted. 3\. Line 120 you mention “blind” do you mean VA of NLP both eyes? Yes, we referred to NLP in both eyes. To clarify this, we have revised the manuscript, as shown in lines 117 – 118. 4\. You told the students about dry eye and then asked them questions about dry eye - Did you have any concerns about that issue? (when you mention something and then ask about it – it might raise awareness) Thank you for pointing this out. We are concerned about satisficing (respondents’ answers to the questionnaire are of low quality).6 Most Thai students don’t know about dry eye. Thus, we tried to motivate the guardians and students to respond to the questionnaire by giving them information about dry eye. We agree with you that providing the information may raise awareness. However, awareness-raising requires multiple processes for successful implementation.7 5\. If you had almost 70% Females and DED was more prevalent among females – don’t it change all the statistics? Could you elaborate? Thank you for your question. Thus, we stratified the analysis of the female group. The results of significant association are still the same in both the univariate and multivariate analyses. References 1\. Garcia-Ayuso D, Di Pierdomenico J, Moya-Rodriguez E, Valiente-Soriano FJ, Galindo-Romero C, Sobrado-Calvo P. Assessment of dry eye symptoms among university students during the COVID-19 pandemic. Clin Exp Optom. 2021:1-7. 2\. Tangmonkongvoragul C, Chokesuwattanaskul S, Khankaeo C, Punyasevee R, Nakkara L, Moolsan S, et al. Prevalence of symptomatic dry eye disease with associated risk factors among medical students at Chiang Mai University due to increased screen time and stress during COVID-19 pandemic. PLoS One. 2022;17(3):e0265733. 3\. Lin F, Cai Y, Fei X, Wang Y, Zhou M, Liu Y. Prevalence of dry eye disease among Chinese high school students during the COVID-19 outbreak. BMC Ophthalmol. 2022;22(1):190. 4\. Chidi-Egboka NC, Jalbert I, Wagner P, Golebiowski B. Blinking and normal ocular surface in school-aged children and the effects of age and screen time. Br J Ophthalmol. 2022. 5\. Talens-Estarelles C, Garcia-Marques JV, Cervino A, Garcia-Lazaro S. Ocular surface predisposing factors for digital display-induced dry eye. Clin Exp Optom. 2022:1-7. 6\. Krosnick JA. Response strategies for coping with the cognitive demands of attitude measures in surveys. 1991;5(3):213-36. 7\. Sayers R. Principles of awareness-raising for information literacy: a case study. Bangkok: Communication and Information (CI) UNESCO Asia and Pacific Regional Bureau for Education 2006. 124 p. 10.1371/journal.pone.0284928.r003 Decision Letter 1 Mimouni Michael Academic Editor 2023 Michael Mimouni This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 20 Feb 2023 PONE-D-22-22320R1Prevalence and associated factors for self-reported symptoms of dry eye among Thai school children during the COVID-19 outbreakPLOS ONE Dear Dr. Tawonkasiwattanakun, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Apr 06 2023 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'. A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'. An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Michael Mimouni Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: (No Response) Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors of the manuscript have definitely shown improvement in the structure, and the improvement is noticeable throughout the entire manuscript. The authors have addressed all previous comments. The sequence of the paper is better organized. The literature review is more robust and up to date. The method and results sections are much clearer and transparent. The authors use the added references in the discussion, making it easier for the reader to understand how the work they have done contributes to current scientific literature. I now find the manuscript publishable, however I suggest four additional minor revisions. • Line 71: The authors mentioned the importance of TFOS epidemiologic report. I recommend adding a few words to summarize its main conclusions rather than its call for additional research. • Line 117: Lack of light perception, the precise professional definition is "No light perception". • Line 254, should read: There are no validated questionnaires for "dry eye symptoms", or cut off scores… • The paragraph starting from 277: The authors added important comparison with other dry eye studies which used different questionnaires. The reader can benefit from some more critical information regarding the different or similar parameters those questionnaires used. Reviewer \#2: Thanks for reading, understanding and integrating our comments. I think this new paper much better and ready for publication. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: **Yes: **Evgeny Gelman \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0284928.r004 Author response to Decision Letter 1 4 Mar 2023 Reviewer \#1: General comment: The authors of the manuscript have definitely shown improvement in the structure, and the improvement is noticeable throughout the entire manuscript. The authors have addressed all previous comments. The sequence of the paper is better organized. The literature review is more robust and up to date. The method and results sections are much clearer and transparent. The authors use the added references in the discussion, making it easier for the reader to understand how the work they have done contributes to current scientific literature. I now find the manuscript publishable, however I suggest four additional minor revisions. 1\. Line 71: The authors mentioned the importance of TFOS epidemiologic report. I recommend adding a few words to summarize its main conclusions rather than its call for additional research. Thank you for you recommendation. We revised this statement and added the summary from the TFOS epidemiology report (line 72-74). 2\. Line 117: Lack of light perception, the precise professional definition is "No light perception". Thank you for your suggestion. In revised manuscript, we used “no light perception” (line 118). 3\. Line 254, should read: There are no validated questionnaires for "dry eye symptoms", or cut off scores. Thank you for pointing this out; we revised this statement (line 254) 4\. The paragraph starting from 277: The authors added important comparison with other dry eye studies which used different questionnaires. The reader can benefit from some more critical information regarding the different or similar parameters those questionnaires used. Thank you for your suggestion. We revised this paragraph and provided details of these questionnaires. We believe that the reader will benefit from this section (in lines 280-294). Reviewer \#2: Thanks for reading, understanding and integrating our comments. I think this new paper much better and ready for publication. We appreciate your kind comment and thank you for reviewing our manuscript. 10.1371/journal.pone.0284928.r005 Decision Letter 2 Mimouni Michael Academic Editor 2023 Michael Mimouni This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 12 Apr 2023 Prevalence and associated factors for self-reported symptoms of dry eye among Thai school children during the COVID-19 outbreak PONE-D-22-22320R2 Dear Dr. Tawonkasiwattanakun, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Michael Mimouni Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors have addressed all my comments. The claims are coherent, and the analysis is simple and clear. I now find the manuscript publishable. Reviewer \#2: Thanks for reading, understanding and integrating our comments. I think this new paper much better and ready for publication. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Adi Segal Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0284928.r006 Acceptance letter Mimouni Michael Academic Editor 2023 Michael Mimouni This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 16 Apr 2023 PONE-D-22-22320R2 Prevalence and associated factors for self-reported symptoms of dry eye among Thai school children during the COVID-19 outbreak Dear Dr. Tawonkasiwattanakun: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Michael Mimouni Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Neglected tropical diseases (NTDs) affect 1.7 billion people worldwide, mostly in tropical and subtropical areas, and they have high morbidity and mortality rates. Major global initiatives, such as the road map for NTDs 2021−2030, have been created to set global targets and milestones to prevent, control, eliminate or eradicate the set of 20 diseases and attain the United Nations Sustainable Development Goals. Despite worldwide efforts to develop new therapies for these diseases, no new chemical entities (NCEs) have been developed for NTDs in the 21^st century. During this period, the few approved products were repurposed compounds, drug combinations, biologicals or new formulations. Remarkably, the vast majority of available drugs present one or more limitations, including low efficacy, toxicity problems, drug resistance, duration of treatment and costs. Therefore, new treatments are urgently needed. Natural products from tropical environments and their structural analogues represent promising sources for the expansion of chemical space for NTD drug discovery. The potential use of unexplored compounds from nature as innovative drugs remains very attractive, especially in a country such as Brazil, which is particularly rich in biodiversity and therefore a source of chemical diversity. In this study, we employed a combination of computational and biological approaches to identify new scaffolds, starting with the screening of natural product libraries from Brazilian organisms. In vitro phenotypic assays and pharmacokinetics evaluation as well as cheminformatics and molecular modelling tools were used to investigate a series of bioactive compounds with antiparasitic properties. # Materials and methods ## Screening a collection of natural products A screening of a natural product collection was performed by using cell assays for infection with *Trypanosoma cruzi* and *Leishmania infantum*. The collection includes 320 compounds from the metabolic classes terpenes, alkaloids, peptides and amides that were isolated by plant species or endophytic fungi. For further evaluation in this study, cytochalasins A-D (CAS numbers: CytA: 14110-64-6, CytB: 14930-96-2, CytC: 22144-76-9, CytD: 22144-77-0) were obtained from Sigma- Aldrich (St. Louis, MO, USA). The purity of these compounds was ≥97%. The stock solution of the compounds was prepared in 100% DMSO (dimethyl sulfoxide—Sigma- Aldrich). ## Cell culture THP-1 (human monocyte cell line) cells were cultivated in RPMI medium supplemented with 10% foetal calf serum (FCS) (Cultilab, Campinas, SP, BR) and 1% penicillin and streptomycin (pen/strep—Sigma-Aldrich) at 37°C under 5% CO₂, and HFF-1 (human fibroblast) and HepG2 (hepatocytes) cells were cultivated in DMEM without phenol red (Cultilab) supplemented with 10% FBS (Sigma-Aldrich) and 1% pen/strep at 37°C under 5% CO₂. ## Parasite cultures The epimastigote forms of the *T*. *cruzi* Tulahuen strain, which expresses the *E*. *coli* β-galactosidase gene lacZ, were grown in liver infusion tryptone (LIT) supplemented with 10% FCS (Cultilab) and 1% pen/strep (Sigma–Aldrich) at 28°C. Metacyclogenesis from epimastigotes to trypomastigotes was induced by the incubation of the epimastigotes in Grace’s insect medium (Sigma-Aldrich, 28°C) every 5 days. After 5 days, the trypomastigote forms were used to infect the HFF-1 cells. The promastigote forms of *L*. *infantum* were maintained in Schneider’s insect medium (Sigma-Aldrich) supplemented with 10% FCS (Cultilab) and 1% pen/strep (Sigma-Aldrich) at 28°C and subcultured every 6 or 7 days. ## *T*. *cruzi* intracellular amastigote assays In vitro assays against *T*. *cruzi* were performed as described previously. HFF-1 human fibroblasts were seeded at 4x10³/well in 96-well tissue culture plates in DMEM (Cultilab) without phenol red and then incubated overnight (37°C, 5% CO₂). Next, trypomastigotes were added at 4x10⁴/well, and the plates were incubated (37°C, 5% CO₂). After 24 h, 2-fold serial dilutions of the test compounds were added at concentrations ranging from 64 to 0.12 μM, and the plates were then incubated (37°C, 5% CO₂). Each compound concentration was evaluated in duplicate. All the plates included BZ (64–0.12 μM) (Sigma-Aldrich) as a positive control and untreated wells (100% growth) as a negative control. After 120 h, chlorophenol red β-D-galactopyranoside (CPRG, Sigma-Aldrich) was added to each well. The absorbance was measured at 570 nm in an automated SpectraMax 384 microplate reader (Sunnyvale, CA, USA), and the data were analysed using GraphPad Prism version 8.0 (San Diego, CA, USA) for IC₅₀ value calculation. ## *L*. *infantum* intracellular amastigote assays In the intracellular amastigote assay, THP-1 cells were seeded at 2x10⁴/well (RPMI-1640, 100 μL) with phorbol 12-myristate 13-acetate (PMA) at 20 ng/mL for differentiation into macrophages. After incubation for 72 h (5% CO2, 37°C), the medium was aspirated, and late-stage *L*. *infantum* promastigotes were added (2x10⁵/well, 100 μL). Following 24 h of incubation, the medium was aspirated to clear extracellular parasites, and the compounds were added in serial dilutions (2-fold) to final concentrations of 64–0.12 μM. The plates included negative controls (100% growth) and MIL (64–0.12 μM) as a positive control. After 120 h of incubation, the medium was removed, and the cells were fixed in methanol and stained with Giemsa. The average number of intracellular amastigotes per THP-1-cell was determined using an inverted microscope and a cell counter. Growth inhibition is expressed as a percentage of the average number of amastigotes per macrophage in the negative control wells. The IC₅₀ values were obtained using GraphPad Prism version 8.0. ## Cytotoxicity in HFF-1 and HepG2 cell lines The cytotoxicity of the compounds against HFF-1 fibroblasts and HepG2 hepatocytes was evaluated using the MTS tetrazolium assay (Promega, Madison, WI, USA). HFF-1 fibroblasts were seeded at 4x10³/well in 96-well culture plates in DMEM without phenol red and incubated overnight (37°C, 5% CO₂). HepG2 hepatocytes were seeded at 7x10³/well in 96-well culture plates in DMEM (Cultilab) with supplements (10% FCS, 1% pen/strep) and incubated overnight (37°C, 5% CO₂). The compounds were added in 2-fold serial dilutions at the same concentrations described before, and the plates were incubated as described before. Each compound concentration was evaluated in duplicate. All the plates included doxorubicin (DOX) (Sigma- Aldrich) as a positive control (10–0.01 μM) and untreated wells as a negative control (100% growth). After 72 h, 15 μL of MTS (Promega) was added, and the plates were incubated for 4 h. The absorbance was measured at 490 nm in an automated SpectraMax 384 microplate reader (Sunnyvale, CA, USA), and the data were analysed using GraphPad Prism version 8.0 for CC₅₀ value calculation. The percentage of nonviable cells was determined and compared to the negative control wells (100% growth). ## Cytotoxicity in the THP-1-cell line THP-1 cells were seeded at 2x10⁴/well in 96-well culture plates in RPMI 1640 supplemented (10% FCS, 1% pen/strep) and incubated overnight (37°C, 5% CO₂). The compounds were added in 2-fold serial dilutions (64–0.12 μM), and the plates were incubated as described before. Each compound concentration was evaluated in duplicate. All the plates included DOX (Sigma- Aldrich) as a positive control (10–0.01 μM) and untreated wells as a negative control (100% growth). After 72 h, 20 μL of resazurin (Sigma-Aldrich) was added, and the plates were incubated for 4 h. The absorbance was measured at 536 and 588 nm in an automated microplate reader SpectraMax Gemini, and the data were analysed using GraphPad Prism version 8.0 for CC₅₀ value calculation. The percentage of nonviable cells was determined and compared to the negative control wells (100% growth). ## Antiphagocytic activity The antiphagocytic assay was performed with THP-1 cells seeded at 2x10⁴/well (RPMI-1640, 100 μL) with phorbol 12-myristate 13-acetate (PMA) at 20 ng/mL for differentiation into macrophages. Following 72 h of incubation (5% CO2, 37°C), the medium was aspirated, and compounds were added in serial dilutions (2-fold) to reach final concentrations of 20–2.5 μM and incubated for 2 h. After incubation, two experiments were performed. 1) The compounds were removed, and late-stage *L*. *infantum* promastigotes were added (4x10⁵/well, 100 μL). 2) The compounds were not removed, and late- stage *L*. *infantum* promastigotes were added (4x10⁵/well, 100 μL). These plates included negative controls (maximum infection without inhibitors). After 24 h of incubation, the medium was removed, and the cells were fixed in methanol and stained with Giemsa. The methodology was adapted from the literature. The average number of infected THP-1 cells was determined using an inverted microscope and a cell counter. Phagocytosis inhibition is expressed as a percentage of infected cells determined by counting 200 random cells. ## Leishmanicidal assay using *L*. *infantum* promastigotes Promastigotes of *L*. *infantum* (1x10⁶ parasites/well) were distributed in 96-well plates, and the test compounds (cytochalasins A-D) were added at concentrations of 20–2.5 μM. The plates were incubated under the same conditions as previously described for 24 h. The leishmanicidal activity was determined by adding 20 μL of resazurin (Sigma-Aldrich), and the plates were incubated for 4 h. The absorbance was measured at 536 and 588 nm in an automated SpectraMax Gemini microplate reader, and the data were analysed using GraphPad Prism version 8.0 for the CC₅₀ value calculation. The percentage of nonviable parasites was determined and compared to the negative control wells. ## Statistical analysis The analyses were performed using GraphPad Prism version 8.0 (GraphPad Software, San Diego, CA, USA). Both the IC₅₀ and CC₅₀ values were calculated by adjusting the sigmoid dose-response curves, and the selectivity index (SI) values were determined by CC₅₀/IC₅₀. Statistical analyses were performed using one-way ANOVA followed by Dunnett’s multiple comparison test. ## Molecular docking of cytochalasins Mol2 files of the docked compounds were generated using the molecular modelling software package Sybyl-X 2.1.1 (Certara, St. Louis, MO, USA), which was run on CentOS Linux workstations. The 3D conformational energy was minimized using the Tripos force field and Powell’s method. Partial atomic charges were calculated using the Gasteiger-Hückel method, and the molecules were considered to be in an implicit aqueous environment (dielectric constant of 80.0). The molecules were docked with the program GOLD 2020 (Cambridge Crystallographic Data Centre, Cambridge, UK) with the actin structure (PDB ID 3EKS, 1.80 Å). The target preparation consisted of removing all crystallographic waters, ligand (CY9), and ATP molecules and adding hydrogens. The binding site was defined as a sphere with a 10 Å radius centred on the crystallographic ligand (CY9). The compounds were docked by applying the default GOLD parameters and using the ChemScore scoring function. An examination of the docking results was performed with PyMOL 1.7.x (Schrödinger, New York, NY). ## In silico ADME property determination The in silico pharmacokinetic properties were calculated with the Swiss ADME Web Tool by considering the following parameters: the clog<u>*P*</u>, size, polarity, unsaturation (fraction of Csp³), flexibility (nRotb), solubility, BBB permeation, CYP enzymes and metabolism, Lipinski’s Rule of 5 and drugability, and PAINS (pan-assay interference compounds) alerts. ## Experimental determination of the distribution coefficient (Log*D*7.4) To analyse cytochalasin lipophilicity, a methodology based on the retention time of molecules in stationary phase was used in LC-MS/MS (liquid chromatography- tandem mass spectrometry) Prominence UFLC (Shimadzu Corporation, Kyoto, Japan) and an interface with an LCMS-8045 triple quadrupole mass spectrometer (Shimadzu Corporation, Kyoto, Japan) with an electrospray ionization source (ESI). The chromatogram was obtained using a Supelco Ascentis express RP amide HPLC column (5 cm x 2.1 mm, 2.7 μM). The mobile phases consisted of 5% methanol in 10 mM ammonium acetate pH 7.4 (A) and 100% methanol (B). The mobile phase was eluted in binary gradient mode, and the gradient was as follows: 0 min, 95% A; 0.3 mins. 100% A; 5.2 mins. 0% A; 5.6 mins. 0% A; 5.8 mins. 100% A; 7.0 mins. And 100% A. The run time was 7 minutes, and the sample injection volume was 5 μL. Test compounds were prepared at 1.0 mg/mL by adding the stock solution at (1:1) mobile phases A:B (internal standard at 200 nM), and the DMSO concentration was lower than 2%. The lipophilicity of the compounds was assessed by individually injecting the test compounds and a series of eight commercial drugs, covering a Log*D* range of -1.86 to 6.1 (acyclovir -1.86, atenolol 0.16, antipyrine 0.38, fluconazole 0.50, metoprolol 1.88, ketoconazole 3.83, tolnaftate 5.40, and amiodarone 6.10). The retention time (in minutes) of each of the eight standards was plotted against their Log*D* values. The resulting equation for the calibration curve (y = mx + b) was used to calculate the Log*D* values for the cytochalasins. ## Human and mouse liver microsomal stability assay The metabolic stability of the cytochalasins was evaluated in human pooled liver human microsomes (20 mg/ml, GIBCO) and pooled liver CD1 mouse microsomes (20 mg/ml, GIBCO). Test cytochalasins were prepared at a concentration of 0.5 μM and incubated with 0.25 mg/mL liver microsomes at pH 7.4), and the DMSO concentration was lower than 1%. The reaction was started by adding the cofactor NADPH at 0.5 μM. Samples were taken at time point 0 (immediately following the addition of the co-factor), 5, 10, 20, 30 and 60 minutes. The reaction was stopped by adding quench solution containing acetonitrile:methanol (1:1) (internal standard tolbutamide at 50 nM). The samples were centrifuged (3500 rpm/30 min.) to obtain the pellet of precipitated microsomal protein. The supernatant fractions were quantified by LC-MS/MS (liquid chromatography-tandem mass spectrometry) Prominence UFLC (Shimadzu Corporation, Kyoto, Japan) and interfaced with an LCMS-8045 triple quadrupole mass spectrometer (Shimadzu Corporation, Kyoto, Japan) with an electrospray ionization source (ESI). The peak area ratios (analyte/internal standard) were converted to the % remaining using the area ratio at time 0 as 100%. The half-life (t1/2 = ln(2)/k) in minutes and intrinsic clearance (Clint = k x 1000/(0.25)) in μL/min/mg were calculated using a nonlinear regression from the % remaining versus the incubation time. From this plot, the slope (k) was determined. The chromatogram for analysis was obtained on a Supelco Ascentis express C18 column (3 cm x 2.1 mm, 5 μM). The mobile phases consisted of water + 0.1% formic acid (A) and acetonitrile + 0.1% formic acid (B). The mobile phase was eluted in binary gradient mode, and the gradient was as follows: 0 min, 95% A; 0.05 mins. 95% A; 0.3 mins: 2% A; 0.7 mins: 2% A; 0.8 mins: 95% A; 1.15 mins 95% A; and 2.0 mins 95% A. The run time was 2 minutes, the sample injection volume was 10 μL, and the flow rate was 0.7 mL/min. ## Parallel artificial membrane permeability assays (PAMPA) A parallel artificial membrane permeability assay was performed in a 96-well precoated PAMPA plate system (Corning Gentest \# 353015). The compound solutions were prepared by diluting the stock solutions (10 mM) in phosphate-buffered saline (PBS) pH 6.5 at a final concentration of 10 μM and then added to the donor portion of the plate (300 μl/well), while PBS pH 7.4 (200 μl/well) was added to the acceptor portion, and the DMSO concentration was lower than 1%. The two portions of the plate were then coupled, and the system was incubated for 5 h at 37°C. Samples of the initial donor solution (T0) were collected and stored at -20°C. At the end of the incubation, samples were collected from the donor and acceptor plates and then added to plates containing quench solution (10% water and 90% methanol:acetonitrile (50:50) + 50 nM tolbutamide). The T0 samples were treated similarly. The final concentrations of compounds in the donor, acceptor and T0 wells were quantified by LC-MS/MS (liquid chromatography-tandem mass spectrometry) Prominence UFLC (Shimadzu Corporation, Kyoto, Japan) and interfaced with an LCMS-8045 triple quadrupole mass spectrometer (Shimadzu Corporation, Kyoto, Japan) with an electrospray ionization source (ESI). The chromatogram for analysis was obtained on a Supelco Ascentis express C18 column (3 cm x 2.1 mm, 5 μM). The mobile phases consisted of water + 0.1% formic acid (A) and acetonitrile + 0.1% formic acid (B). The mobile phase was eluted in binary gradient mode, and the gradient was as follows: 0 min, 95% A; 0.05 mins. 95% A; 0.3 mins: 2% A; 0.7 mins: 2% A; 0.8 mins: 95% A; 1.15 mins 95% A; and 2.0 mins 95% A. The run time was 2 minutes, the sample injection volume was 10 μL, and the flow rate was 0.7 mL/min. The results were used to calculate an effective permeability (Pe) value. ## Kinetic solubility To determine the kinetic solubility, 10 mM samples of each cytochalasin were transferred to a 96-well plate (incubation plate) in duplicate; for each sample on the plate, 195 μL of PBS buffer pH 7.4 and 2.0 (final concentration of 250 μM) was added, and the DMSO concentration was 2.5%. The plate was sealed and shaken for 24 hours (200 rpm, r.t.). The precipitates on the incubation plate were removed by centrifugation (15 min., 3000 rpm, r.t.). The supernatant fractions were quantified by LC-MS/MS (liquid chromatography-tandem mass spectrometry) Prominence UFLC (Shimadzu Corporation, Kyoto, Japan) and interface with an LCMS-8045 triple quadrupole mass spectrometer (Shimadzu Corporation, Kyoto, Japan) with an electrospray ionization source (ESI). From a 10 mM standard solution, an intermediate standard dilution of 0.5 mM in a 1:1 acetonitrile:water solution was prepared. Calibration curves were prepared for each cytochalasin by diluting the intermediate standard several times to reach the desired concentrations of 50, 40, 20, 2, and 1 μM. The resulting equation for the calibration curve (y = mx + b) was used to calculate the experimental concentration values for the cytochalasins. The chromatogram for analysis was obtained on a Supelco Ascentis express C18 column (3 cm x 2.1 mm, 5 μM). The mobile phases consisted of water + 0.05% formic acid (A) and acetonitrile + 0.05% formic acid (B). The mobile phase was eluted in binary gradient mode, and the gradient was as follows: 0 min, 98% A; 1.2 mins. 2% A; 2.0 mins: 2% A; and re-equilibration time: 0.6 min., 98% A. The run time was 2 minutes, the sample injection volume was 5 μL, and the flow rate was 0.6 mL/min. # Results The biological screening of a collection of natural products against *T*. *cruzi* and *L*. *infantum*, consisting of piperamide and piperidine derivatives, cyclopeptides and terpenes, among others, revealed a small series of bioactive cytochalasins, as shown in. Cytochalasins A-D are natural products in the terpene metabolic class produced by fungi in the *genera Penicillium*, *Aspergillus*, *Xylaria* and *Phomopsis* and possess a macrocyclic ring of 14 (cytochalasins A and B) - 11 (cytochalasins C and D) members, a carbonyl (cytochalasin A) or hydroxy (cytochalasin B) substituent at position C20, and a double bond position at C5-C6 (cytochalasin C) or C6-C12 (cytochalasin A, B and D). Based on the preliminary encouraging results from the biological evaluation, these cytochalasins were selected with the goal of investigating and characterizing their potency, pharmacokinetic properties, selectivity index and mechanism of action. The results of the evaluation on the trypanocidal and leishmanicidal properties of cytochalasins A-D are depicted in. Cytochalasin A showed significant potency against the intracellular amastigote forms of *T*. *cruzi* and *L*. *infantum*, with IC_50s of 3.02 μM and 2.76 μM, respectively. Cytochalasins B, C and D exhibited IC₅₀ values of 16.87, 10.24 and 20.15 μM for *T*. *cruzi* and approximately 22–23 μM for *L*. *infantum*. The cytotoxicity was observed in host macrophages, fibroblasts and hepatocytes, and the results are shown in. ## Antiphagocytic effect The potential for cytochalasins to inhibit *L*. *infantum* phagocytosis by THP-1 cells was evaluated. The results in show that cytochalasins A-D were able to inhibit phagocytosis on THP-1 cells at all tested concentrations, confirming the preliminary observations. An even greater decrease in infected cells can be observed in, because the cytochalasins were not removed before adding *L*. *infantum* promastigotes. The results in indicated that cytochalasins A and C, in addition to inhibiting phagocytosis, were toxic to *L*. *infantum* promastigote parasites. ## Molecular modelling of cytochalasins A-D and actin Molecular docking studies were performed to underline the potential interactions of the studied compounds with actin within the cytochalasin binding site. The PDB structure of cytochalasin D crystallized with actin was employed (PDB ID: 3EKS) in these analyses. The predicted binding modes of cytochalasins A-D are shown in. The complexes revealed intermolecular interactions similar to those of the crystallographic complex. The hydrogen bonds of the hydroxyl at position 7 of the ring with the backbone nitrogen of Ile136 were observed for all four cytochalasins. The N-H of the isoindolone ring interacts with either the hydroxyl of Tyr143 or the backbone carbonyl of Gly168. Both the backbone nitrogen and carbonyl of Ala170 interact with the hydroxyl or lactone polar groups of the cytochalasin macrocyclic ring. Considering that the macrocyclic ring of cytochalasins is rather hydrophobic, it interacts with the hydrophobic residues Ile136, Pro172, Phe375, and Ala170. ## Physico-chemical and pharmacokinetic analysis The experimentally determined Log*D* values for cytochalasins A-D were in the range between 3 and 4, showing moderate to high lipophilicity (moderate Log*D* (0–3); high Log*D* (\> 5)). In silico predictions provided lipophilicity values in the range of 2.5 \< log*P* \< 3.22, in accordance with the in vitro data. During the cell assays, no solubility problems were detected. The kinetic measurements of solubility for cytochalasins A-D in PBS at pH 2.0 and 7.4 were medium to high and were in accordance with the predicted Log*S*. Under physiological conditions (pH 7.4, the same used in the cell assays), the solubility was similar among the compounds and similarly observed at pH 2.0. The results of the PAMPA assay showed that the permeability of cytochalasins is considerably high. The predicted bioavailability radar in provides a general glance at the compound’s drug-likeness, in which the background red-coloured area represents the optimal range for the estimated properties. The predicted properties were in a good range for lipophilicity (LIPO), molecular mass (SIZE), polarity (POLAR), fraction of Csp3 (INSATU), number of rotatable bonds (FLEX), and moderate solubility (INSOLU). # Discussion Cytochalasins A-D showed significant potency against the intracellular amastigote forms of *T*. *cruzi* and *L*. *infantum*. The potency of cytochalasin A is comparable to that of the standard drug benznidazole (BZ, IC₅₀^{*T*. *cruzi*} of 5.20 μM) and miltefosine (MIL, IC₅₀^{*L*. *infantum*} of 1.10 μM). A lower cytotoxicity for host cells when compared to parasite cells was observed, which is also a good feature for this class of natural products. Cytochalasins A-D were more than 10-fold less toxic than the standard control doxorubicin. Structure-activity relationship studies explored the structural and physicochemical requirements related to the antiparasitic effects of these compounds. The presence of the 1,4 diketone (C-20 carbonyl), 14-member macrocyclic ring, and the double bond at C6-C12 was an important feature driving the strong potency exhibited by cytochalasin A. The isoindolone, hydroxyl at position C-7, and phenyl at position C-3 are important for the cytotoxic activity of cytochalasins, and the macrocyclic ring was previously regarded as important for bioactivity. The antiphagocytic properties of cytochalasins were demonstrated in this work and are an interesting strategy for leishmaniasis drug discovery. Further from the phenotypic study, we evaluated the interaction of these compounds with actin. This protein is involved in the internalization of *Leishmania* by host cells, and although the actin of the parasite itself is not inhibited, the inhibition of host cell actin is an interesting strategy to hinder parasite infection. The ADME (administration, distribution, metabolism and excretion) properties for cytochalasins A-D were assessed using both in silico and in vitro methods. The PAMPA showed that the permeability of cytochalasins is considerably high, which is an important feature for good bioavailability and an important factor for passive diffusion and gastrointestinal absorption. The metabolic stability results assessed by a test in human and mouse liver microsomes indicated that cytochalasin A is more stable than cytochalasins B-D. The predicted bioavailability radar is useful for a general glance at the compound’s drug- likeness, and the results indicate good oral bioavailability for cytochalasins A-D. Our strategy was to screen a natural product library from Brazilian organisms that is an important source of chemical diversity for developing drugs to treat neglected diseases. This work led to the identification of cytochalasins A-D, natural products presenting antileishmanial and antiphagocytic activity against *L*. *infantum* with good ADME parameters, and these products consisted of promising lead compounds for antiparasitic drug discovery. Similarly, other natural products could be identified in a variety of species to treat a diversity of diseases. Lastly, these results provide further advancements for the understanding of the cytochalasins’ anti-parasitic activity, adding useful information on the drug discovery of new antiparasitic agents. The authors acknowledge the Nuclei of Bioassays, Biosynthesis and Ecophysiology of Natural Products (NuBBE), Department of Organic Chemistry, Institute of Chemistry, São Paulo State University (UNESP) for providing samples for the screening assays. 10.1371/journal.pone.0275002.r001 Decision Letter 0 Saha Bhaskar Academic Editor 2022 Bhaskar Saha This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 9 Sep 2022 Identification of natural cytochalasins as leads for neglected tropical diseases drug discovery PONE-D-22-16784 Dear Dr. Valli, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Bhaskar Saha Academic Editor PLOS ONE Journal requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Thank you for stating the following financial disclosure:     "This work was supported by the Fundação de Amparo à Pesquisa do Estado de São Paulo (FAPESP, <https://fapesp.br/>) grants \#2013/07600-3 (CIBFar-CEPID), \#2014/50926-0 (INCT BioNatCNPq/FAPESP), Conselho Nacional de Desenvolvimento Científico e Tecnológico (CNPq, <https://www.gov.br/cnpq/pt-br>) and Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES, <https://www.gov.br/capes/pt-br>) for grant support and research fellowships. The authors acknowledge the scholarships conferred to MV (Fapesp \#2019/05967-3) and JM (Fapesp \#2019/06034-0)." Please state what role the funders took in the study.  If the funders had no role, please state: "The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript."  If this statement is not correct you must amend it as needed.  Please respond by return e-mail so that we can amend your financial disclosure and competing interests on your behalf. Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Reviewer \#1: Authors have studied anti-protozoan drug activity and bio- availability of cytochalasins suggesting that cytochalasins can be potentially used as drugs to treat infections caused by Trypanosoma and Leishmania. Reviewer \#2: The manuscript entitled “Identification of natural cytochalasins as leads for neglected tropical diseases drug discovery” is an interesting study. Authors in this manuscript using combination of computational and biological approaches identified and characterized four cytochalasins as drug candidate for neglected tropical diseases. Data shown in this manuscript are convincing. \*\*\*\*\*\*\*\*\*\* \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0275002.r002 Acceptance letter Saha Bhaskar Academic Editor 2022 Bhaskar Saha This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 16 Sep 2022 PONE-D-22-16784 Identification of natural cytochalasins as leads for neglected tropical diseases drug discovery Dear Dr. Valli: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Bhaskar Saha Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Seasonal intermittent preventive treatment for malaria in children (IPTc) is a promising new intervention for the prevention of malaria in children. Seasonal IPTc is safe and can reduce the burden of malaria substantially in countries of Sahelian West Africa. Dicko and colleagues in Mali showed that two treatments with suphadoxine-pyrimethamine (SP) given two months apart at the height of the malaria transmission season reduced the incidence of malaria by 40% in children 6 months to 10 years old. Three treatment courses of SP plus a single dose of artesunate given during the high transmission season to children aged less than five years reduced the incidence of malaria by 86% in Senegal. Three courses of SP given bimonthly reduced clinical malaria by 24% and anaemia (Hb\<8.0 g/dl) by 30%, and six courses of artesunate plus amodiaquine (AQ) given monthly reduced clinical malaria by 69% and anaemia by 45% in the Hohoe district of Ghana. A recent study in Senegal showed that IPTc using SP + AQ given for 3 days had the best protective efficacy compared to three other drug regimens. The adjusted hazard ratio for the incidence of malaria was 0.50 (95% CI 0.31,0.81) for SP (1 dose) +AQ given for 3 days, 0.90 (95% CI 0.60, 1.36) for SP (1 dose) + AS given for 3 days 1.13 (95% CI 0.76, 1.67) and AQ+AS given for 3 days when compared to SP +AS each given for one day. Although IPTc with AQ and SP is safe and efficacious, there are concerns over how it can be delivered in an effective and sustainable way. In order to be effective, the delivery system must be accessible to the maximum number of target children, the SP/AQ must be offered to the correct children by the provider, and the correct regimen must be adhered to by the carer of the child. In terms of sustainability, the first consideration is whether it is possible to achieve effective delivery through routine public sector delivery systems. Delivery of IPTc during routine EPI/growth monitoring clinic visits must, therefore, be considered as one possible delivery system. However, there are two reasons to suspect that delivery of IPTc through this system may not reach the maximum target numbers in Ghana and other countries of West Africa. Firstly, the coverage of EPI vaccination is low in several countries of West Africa. Secondly, in Ghana, immunisation and growth monitoring visits are more frequent amongst those under one year old than in those over one year (2004 Jasikan district annual report). It might be possible to increase the effectiveness of delivery through EPI/growth monitoring by using targeted health education campaigns to encourage caretakers to attend. If perceptions of IPTc are positive, then caretakers may be encouraged to use the clinics more regularly. Conversely, it is also plausible that the increased workload on the health services staff due to this integrated delivery of IPTc and growth monitoring could lead to programme fatigue. Another option is a community based, volunteer delivery systems of the kind that has been used successfully for mass treatment of onchocerciasis, lymphatic filariasis, schistosomiasis, and trachoma and to administer treatment for malaria. Due to disappointing coverage with Intermittent Preventive Treatment for pregnant women (IPTp) in some situations, there have been attempts to deliver IPTp through a variety of community based systems. These systems have been successful in delivering IPTp to pregnant women but there is current debate and conflicting evidence on their impact on access to other antenatal services by pregnant women. Ensuring that the correct amount of drugs are delivered to targeted children at the most appropriate time of the year and sustaining adherence to the treatment regimen through any delivery system is a challenge. However, these challenges are not insurmountable and different ways of delivering IPTc need to be explored. In this paper we describe an evaluation of two different approaches to the delivery of IPTc in Ghana. # Materials and Methods The protocol of this trial and supporting CONSORT checklist are available as supporting information see and. ## Study area The study took place in the Jasikan district of Ghana from May to October 2006. The Jasikan District is within the semi-equatorial climatic zone and has an extended rainy season (April – November) with peaks in May-July and September- October. The district has two hospitals, 9 health centres, 17 reproductive and child health static clinics, 3 private maternity homes and 2 private clinics. The district also has 80 Expanded Programme on Immunisation (EPI) outreach clinics. In children under the age of five years resident in the study area in 2006 who were treated for clinical malaria the day28 PCR corrected adequate clinical and parasitological cure rate was 75% (95% CI 69,80) for SP (n = 241) and 85% (95% CI 69,80) for AS+AQ (n = 241) (Kweku M et al in preparation). ## Study design and conduct The study had two arms; IPTc was delivered either at the outpatient department of health centres and at EPI outreach clinics \[health facility-based IPTc\] or by community volunteers \[community-based IPTc\]. Sampling was carried out in two stages. From a sampling frame of all villages in the district with a static reproductive and child health facility (n = 17), six villages were randomly selected using a sampling interval of 3. From this sample of 6 villages, 3 villages were allocated to the facility based arm and 3 villages to the community based arm by ballot. Similarly from a sampling frame of 80 villages that had no static health facility, 6 villages were randomly selected using a sampling interval of 13. From this sample of 6 villages, 3 villages were assigned to the facility based arm (IPTc was delivered at outreach EPI clinics) and 3 to the community based arm by ballot. The study villages were on average within 5 kilometres from the static health facility. None of the health facilities were inaccessible during the study period. Members of the district health management team and the principal investigator (MK) explained the objectives of the study to leaders of the study villages and obtained their consent for the trial. All children aged 3–59 months (1039) resident in the 12 study villages were then enumerated in May 2006. Children whose parents gave written consent (964) were enrolled into the study after a physical examination. Only three children who had a history of drug allergy were excluded. An identification card (ID card) with a unique ID number was issued to all enrolled children. The dates when IPTc was to be administered and the dose required were indicated on the ID card. Parents were advised to refer to the ID cards for information on the dates of IPTc drug administration and on whether they should go to the community-based volunteers or to the health facility to receive the study drugs. A finger prick blood sample for examination of malaria parasite was obtained at enrolment and at the end of the six month observation period. District Health Management Team (DHMT) staff distributed drugs to the health centre staff and community-based volunteers and supervised the administration of drugs to children in May, June, September and October. In the facility-based arm, health workers who were providing outpatient treatment for sick children gave IPTc to the study children on the scheduled IPTc days and EPI nurses administered IPTc during the monthly out-reach vaccination and growth monitoring clinic. In the community based arm, children were invited to a central point in the village on the scheduled IPTc days once a month for four months and the community based volunteers administered the IPTc. Children in each arm received a three day regimen of AQ plus a single dose of SP in May, June, September and October. The first dose of the drug was administered under supervision by a health worker or by a community-based volunteer. Subsequent doses were dispensed by parents/guardians at home after training. Tablets of AQ (150 mg) and (SP) (500 mg sulfadoxine/25 mg pyrimethamine) were given as follows: ¼ tablet of each drug to 3 to 5 month-old infants, 1/2 tablet of each drug to 6 to 11 month-old infants, 3/4 of each tablet to children aged 12 to 23 months and 1 tablet of each drug to those aged 24 months or more. Tablets were crushed and mixed with water and sometimes sweetened with sugar. Drugs were administered based on age not on exact weight. Five days after administration of the first dose of IPTc, field workers visited a randomly selected 20% sample of the children in every village to ask about side effects of drugs, detect any left over medication and to enquire about the use of bednets. The number of children surveyed after each course of IPTc ranged from 73 to 91 in the community based arm and from 68 to 85 in the facility-based arm. Community volunteers received a payment of \$10 and facility-based staff a salary supplement of \$10 every month for undertaking the additional work required to deliver and monitor coverage of IPTc. Further details of the economics of the two approaches to drug delivery will be presented elsewhere (Conteh et al. in preparation). Carers were asked to take their child to the nearest health facility for examination and treatment if he or she developed any adverse event or any illness. Malaria episodes in study children were monitored through a passive surveillance system at the health centres and hospitals within the study area for a period of six months after the last dose of the first course of drug had been given (May - October 2006). A finger prick blood sample was collected from all children who presented with a history of fever or any signs suggestive of malaria before treatment with AS + AQ. Axillary temperature was measured using an electronic thermometer when a child reported at the health facility. Blood slides were read twice at a central laboratory to give a definitive diagnosis. Laboratory assistants examined thick blood films for parasitaemia. A sample was considered negative only after 200 high power fields had been read. Parasite counts were converted to parasites per micro liter (µl), assuming a white blood cell count of 8000 leukocytes per µl of blood. If there was a discrepancy in the findings in a slide between the two initial technicians (positive or negative or a 50% or more difference in parasite density) a third, more senior microscopist read the slide and his reading was deemed to be the correct reading. A senior microscopist from the Noguchi Memorial Institute of Medical Research (NMIMR, Ghana) examined all the positive blood films and a 20% random sample of negative blood slides for quality control. Haemoglobin (Hb) concentration was measured using a Hemocue ® Photometer (Leo Diagnostics, Sweden) in children who attended study health facilities for suspected malaria or anaemia. Children with an Hb\<8.0 g/dl received treatment for anaemia with iron 12.5 mg and folic acid 50 µg tablets daily for 14 days and those with an Hb = \<6.0 g/dl were referred for hospital management. ## Sample size The number of villages required for the trial was based on the following assumptions. It was assumed that coverage of IPTc in the community-based arm would be 80%, that the coefficient of variation in the IPTc coverage between villages would be 0.4 and that the number of children per village would be between 60 and 100 (average 80). On the basis of these assumptions a study with 12 clusters would have 80% power to detect a relative difference in coverage between groups of 35% at a 5% level of significance. It was also assumed that the incidence of malaria in the study area before the intervention would be 19 per 1000 child weeks and it was expected that there would be a 42% reduction in the incidence of malaria post intervention, that the coefficient of variation in the malaria incidence between villages would be 0.4 and that the number of children per village would be between 60 and 100 (average 80). On the basis of these assumptions the study had 80% power to detect a relative difference in incidence before and after the intervention of 42% at a 5% level of significance. The sample size was calculated using the formula suggested by Smith & Morrow. ## Analysis The primary outcome of the trial was coverage with IPTc. Acceptable coverage was defined as the proportion of children who received at least the first dose of 3 or more courses of IPTc. Poor coverage was defined as the proportion of children who received the first dose of two or less courses of IPTc. Secondary outcomes were (1) adherence to treatment (2) the incidence of clinical malaria, (3) the incidence of anaemia, (4) the prevalence of malaria parasitaemia at the beginning and at the end of the rainy season and (5) the incidence of adverse events. Adherence to the IPTc regimen was classified as follows; highly adherent - took first supervised dose of all 4 courses; moderately adherent - took first supervised dose of three courses; poorly adherent - took the first supervised dose of two or less courses. An episode *P. falciparum* clinical malaria was defined as a history of fever during the past 72 hours or an observed temperature ≥37.5° C plus the presence of peripheral parasitaemia of any density. High parasite density malaria was defined as parasitaemia \>7000 per µl in the presence of a history fever during the past 72 hours or a temperature ≥37.5°. Anaemia was defined as an Hb \<8.0 g/dl. For the analysis of coverage of IPTc, an intention to treat analysis was used in which all children who were enrolled and received the first dose of IPTc course one were included in the analysis. Statistical significance of the difference in the coverage of IPTc between the two arms was tested using logistic regression with robust standard errors to take in to account the cluster effects. The secondary outcome of adherence was defined as the proportion of children who received all three doses of each course of IPTc in the 20% random sample of children who were visited at home. Statistical significance of the differences in the adherence between the groups was tested by chi²test. For the analysis of the effect of IPTc on the incidence of malaria and anaemia, we calculated person time at risk from the date of enrolment to date of the final cross sectional survey (6 months after the start of intervention). In addition, if a child received an antimalarial drug (AS+AQ) during the follow up period, we subtracted 28 day post treatment period from the person time at risk. To compare the effect of ITPc on the incidence of malaria between the two arms of the study, we used the random effects Poisson regression model and robust standard errors to allow for intra-cluster correlation and to adjust for the effect of age, and bed net use. To compare the prevalence of malaria parasitaemia between the two arms of the study, we used a logistic regression model to adjust for the effect of clustering in the communities using robust standard errors and for the effect of age, bednet use and the number of IPTc courses administered. All statistical analysis was done using STATA version 10. (©Statacorp LP). ## Ethics The trial was approved by the ethics committees of the London School of Hygiene & Tropical Medicine and of the Ghana Health Service/Ministry of Health (GHS/MOH).The trial was registered on the NIH clinical trials database as an amendment to the number ClinicalTrials.gov NCT00119132. # Results ## Background characteristics At enrolment, children in the two study groups were similar in regard to age, gender, anthropometric indices and malaria parasite prevalence. The number of children per cluster varied but was well balanced between the two groups. It ranged from 40 to 98 children (mean 73.8) in villages where IPTc was given in the community and from 38 to108 (mean 83.2) in villages where the delivery of IPTc was facility-based. There was no statistically significant difference in the proportion of households who owned an insecticide treated net (ITN) between villages in the community based arm and the facility based arm (16.1 vs 19.2%; p = 0.281). Similarly, there was no statistically significant difference in the proportion of children who slept under an ITN on the night before the survey between villages in the two arms (11.4% vs 16.3% p = 0.196). ## Coverage with IPTc and adherence The proportion of children who received 3 or more courses of IPTc ranged from 72.9% to 94.8% in the community-based arm and from 58.5% to 94.9% in the facility-based arm villages. The mean coverage of IPTc was slightly higher in the community based arm than in the facility-based arm (90.5 vs 86.6%; OR 1.47, 95% CI 0.63, 3.45). The proportion of children who received all four courses of IPTc was also slightly higher in the community based arm than in the facility- based arm (69.1 vs 65.5%; OR = 1.18; 95% CI 0.62, 2.22). However these differences in coverage of IPTc between the two arms were not statistically significant. Among the 650 children who were assessed for adherence to IPTc doses, adherence to 3 doses of any course of IPTc ranged from 85% to 100% and there was no difference in the adherence to IPTc between the community based arm and the facility based arm. Forty-four percent (110/248) parents/guardians interviewed reported that they had used the ID card to determine the IPTc drug administration day and where they should take their child to receive treatment. Those who did not use the ID card gave the following reasons: forgot 80 (32.3%), could not read 41(16.5%) and did not understand what was written on the card 17 (6.9%). After the administration of each course of IPTc, drug distributors were asked to record on the ID card that the child had received the first dose of drug under supervision. Eighty-one percent (783/962) of ID cards were available to assess the record of IPTc administration. Recording of IPTc drug administration on ID cards was significantly higher for children who received IPTc from community- based volunteers than for children who received their IPTc from facility-based health workers (65% vs 39% respectively) OR = 2.90 (95% CI 1.57, 5.39 p\<0.001). ## Adverse events reported during the intervention period No serious adverse event attributable to use of the study drugs was recorded. One death occurred during the observation period. This child became unwell before the fourth course of drug administration. The probable cause of death ascertained by verbal autopsy was a brain abscess. Some children spat or vomited drugs immediately after their administration due to the bitter taste of amodiaquine. In the villages where IPTc distribution was community based, 3.5% (23/650) of the children required a repeat dose of IPTc because of vomiting compared to 2.5% (16/650) of children resident in villages where IPTc distribution was facility-based. One hundred and fifty-three of the 650 children (23.5%) visited at home five days after administration of the first dose of IPTc experienced at least one adverse event. Common adverse events noted by parents/guardians were vomiting/spitting drugs (10.9%), drowsiness/general bodily weakness (8.0%). Other adverse events noted were diarrhoea, body itch, abdominal pains, common cold and a swollen face. The reported incidence of adverse events fell with time and by the fourth course of drug administration, less than 2% of children reported any adverse event. ## Incidence and prevalence of malaria and anaemia The incidences of clinical malaria, microscopically confirmed malaria of any parasite density and high parasite density were lower in children in the facility based arm than in the community based arm. There was no statically significant difference in the prevalence of parasitaemia between the two arms at the pre implementation survey (10.2% vs 11.5%; OR 1.14, 95% CI 0.45, 2.87; p = 0.78) or at the post implementation survey (7.5% vs 6.7%; OR 0.88, CI 0.26, 2.96; p = 0.84). # Discussion This study has shown that the coverage achieved for four courses of IPTc was slightly higher employing a community-based rather than a health facility-based delivery system. Both systems achieved more than 60% coverage for all four courses and over 80% coverage for 3 or more courses. However, this required good supervision and support from a District Health Management Team and the study team. In Ghana there are many communities without a static health facility and, during the rainy season which is also the farming season, approximately 25% of the population is not regularly reached by EPI outreach clinics. Thus, although a facility-based delivery system had a relatively high coverage (86.6%) in this study, a substantial proportion of children would not have access to if IPTc is delivered exclusively through the facility-based approach. It is likely that children who would not have access to IPTc through the health facility-based delivery system would be those living in the most inaccessible and deprived areas of the district and would be those most at risk from malaria. Delivering IPTc through the routine health system will require extra effort to reach the deprived population and adequate supervision to ensure that records are kept properly. In this area of Ghana where the study was done, a combination of health facility-based and community-based approach might be needed to maximise the impact of IPTc. In other areas, different approaches may be appropriate to achieve maximum coverage. Although no serious adverse event attributable to the administration of IPTc with AQ + SP was observed, a significant proportion of caretakers (24%) reported mild adverse events within five days of administration of IPTc. However, in the absence of a placebo group it cannot be determined how many of these reported symptoms were due to the drug and how many to associated minor illnesses. Nevertheless, this is a concern if IPTc is to be rolled out on a national scale. The main adverse events observed in this study were bodily weakness and drowsiness and vomiting of drugs (18.9%). The reported incidence of adverse events fell over time after educating caretakers to feed children with food containing sugar at the time of drug administration. Appropriate health education on adverse events and actions to mitigate them should be a priority for any national IPTc programmes. Most children (66/75) who reported at the health facility with a history of fever received treatment for malaria but only 27% (20/75) of the children had malaria parasitaemia. This high use of antimalarial drugs is partly due to the current treatment guidelines in Ghana that all children \<5 years of age who report at the health facility with fever should be treated for malaria unless there is a clear alternative cause for the fever. The reasons for the lower incidence of malaria in the facility based arm than in the community based arm are not clear. We speculate that background incidence of malaria was higher in the rural villages of the community based arm than the villages with a static health facility. However, both arms of the study had rural and semi-rural villages and the prevalence of fever and malaria parasitaemia was not different between the groups at the baseline. We conclude that seasonal IPTc using community-based or facility-based delivery system can achieve high coverage in children aged 3 to 59 months. However, a facility based approach alone would not reach children living in areas without access to formal health care system. Adverse events such as drowsiness, general bodily weakness and vomiting of drugs may be concern for using SP +AQ for IPTc. However, IPTc using four courses of SP+AQ in areas with a prolonged seasonal transmission probably can reduce the burden of malaria. # Supporting Information We are grateful to the Jasikan DHMT staff who delivered drugs and supplies to the communities and monitored drug administration throughout the study period and also to the parents and guardians of the children who participated in the study. We thank the community-based volunteers and health workers who distributed IPTc drugs to children, hospital and health centre staff who managed adverse events and treated sick children in the Jasikan district and laboratory staff from Hohoe who collected blood samples, prepared and read all slides. We thank the entire project staff for putting in their maximum efforts to ensure the successful implementation and running of the trial. [^1]: Conceived and designed the experiments: MK JW BG DC. Performed the experiments: MK SA. Analyzed the data: MK MA. Contributed reagents/materials/analysis tools: MK MA DC. Wrote the paper: MK JW MA BG DC. [^2]: The authors have declared that no competing interests exist.

# Introduction Excessive aggression among pigs is a serious economic and welfare issue and it is also a persistent problem. There are some possible reasons for the occurrence of this behaviour. Prenatal and neonatal environments can modify the behaviour of animals with potential consequences for successive generations. Such characteristics may influence the adaptive response to different challenges. Management strategies focusing on the prenatal and neonatal environments of domestic animals are able to modify immune function, fear, memory and social behaviour, with consequences on the organisation of important coping systems, particularly the central nervous system. There are other factors that may modulate aggressive behaviour, such as the asymmetry in body weight, environmental enrichment, available space per animal and weaning age. Premature weaning, defined by as weaning before four weeks in pigs, caused memory impairment, altered the expression of stress responsive genes in the brain and increased aggression. An appropriate body condition score of the sows is important in order to avoid obesity and excessive leanness, to prevent low reproductive and performance indices. Therefore, sows are subjected to a feed restriction of approximately 50 to 60% of its *ad libitum* intake capacity or motivation. Food restriction is known as one of the most important sources of stress for pregnant sows. Increase in aggression, occurrence of stereotypic behaviour, reduction in weight gain and changes in food motivation are some of the poor welfare outcomes reported in sows kept on restrictive diets ( 26-finisher pigs\]). Alternative diets using fibre have been the subject of studies to mitigate the negative effects of food restriction and reducing the feeling of hunger, which pregnant sows may experience. The inclusion of fibre in the diet of these animals can reduce the expression of stereotypic behaviour, reduce aggression, reduce feed motivation, increase long-term satiety, improve the welfare and improve performance. These cited studies were conducted out in an attempt to understand the effect of a high fibre diet in the behaviour and productivity of the sows themselves. However, as far as we know, no previous study has investigated the effect of a high fibre diet for sows on the behaviour of their piglets. The skin lesion score is significantly influenced by reciprocal fight and bullying, making this measure an indicator of the outcomes of aggression. The score can be obtained by counting the skin lesions, classified as any scratch found in the skin (old, with scab, or recent lesion). Behavioural characteristics associated with emotionality, such as fear and anxiety could be important factors to determine the likelihood of an animal to engage in aggressive interactions. Studies using open field and novel object tests may help to identify susceptible and resilient phenotypes regarding their ability to adjust to their social environment, making optimal decisions, and possibly having better welfare outcomes. The emotionality tests can inform about the individual characteristics of animals. This information is crucial to define the strategy chosen by an animal, such as engaging in a fight or not. The aim of the present study was to investigate the impact of a high fibre diet for pregnant gilts on the levels of skin lesions and behaviour in open field and novel object test in their piglets. # Material and Methods ## Animals The experiment was carried in an experimental pig farm at the Campus Fernando Costa, School of Veterinary Medicine and Animal Science, University of São Paulo, Pirassununga, Brazil, upon approval of the Ethics Committee on the Use of Animals (CEUA) at the School of Veterinary Medicine and Animal Science (FMVZ) of the University of São Paulo (USP), under the number 3606300114. We studied the offspring of 22 sows, distributed by weight, making the groups homogeneous, in high or low fibre diet treatments on the day after they were inseminated with pooled semen. At the beginning of this study, we had 32 sows, 16 in each treatment. However, 7 sows from one and 1 sow from the other treatment returned to oestrus and then, were excluded from analysis, and 2 was excluded for health reasons. The sows were kept in a group-housing system during pregnancy, with individual sows offered either high or low fibre diets during pregnancy, separated during feeding time in closed feeding stalls, in the same pen. Fourteen sows were fed high fibre diet (HFD), (12.86% crude fibre) and eight sows were fed low fibre diet (LFD) (2.53% crude fibre). It was the intention that extra quantities of high fibre diet be sufficient to match the nutrient intake of LFD. LFD animals received 2 kg of diet (3.300 kcal per kg) and HFD animals received 2.4 kg of diet (2.764 kcal per kg) per day, in two equal portions given at 8:00 am and 15:00 pm, 50% in each period. HFD diets had a 35% inclusion of soybean hulls, and were mixed to the diet. High and low fibre diets were offered throughout the pregnancy. The HFD was composed by 57.6% of corn, 4.9% of soybean meal, 35% of soybean hulls, and 2.5% of vitamin-mineral premix. The LFD was composed by 81.7% of corn, 15.8% of soybean meal and 2.5% of vitamin-mineral premix. Parturition occurred in farrowing pens, where the sows were moved on the 107^th day of pregnancy. The lactation diets were the same for both treatments, offered *ad libitum* for all sows, and was composed by 64% of corn, 30% of soybean meal, 3% of soybean oil and 3% of vitamin-mineral premix. For the analysis of the lesions of the offspring of differentially fed sows, we studied 56 piglets born from LFD sows and 100 piglets born from HFD sows. Lesions were counted in the piglets immediately prior to and during weaning. Measures of emotionality were obtained in the open field and novel object tests, where 142 pigs were investigated, 86 piglets from HFD sows and 56 piglets from LFD sows. ## Facilities and handling The gestation pens were 6.7 m wide by 4.4 m long, giving 3.3m²/sow. Each pen had 9 individual feeding stalls (1.8m x 0.55 m) with a nipple drinker in each stall, with *ad libitum* access to water. The feeders were built of concrete. During feeding time, all animals were confined in the stalls, to avoid intermixture among dietary treatments. From the 107^th day of pregnancy, until the 28^th day of lactation, sows were housed in individual farrowing pens measuring 4.3 x 2 m. Connected to the pen, there was a creep area made of concrete (0.97m x 2.2 m), where piglets had unlimited access to solid feed from the first day of life. The sows had access to bedding material composed by dehydrated sugar cane bagasse and hay. The creep area also had bedding material, the same as the pen and a heat source provided by a 60 Watt heat lamp. Farrowing was monitored with IP connected video cameras, with real time internet transmission to the experimenters by computers and smartphones and also through direct observation. Interventions were carried out only when necessary, following a pre-established protocol, enabling standardization of management procedures. The sows were divided in three blocks, according to the gestational period. The first block was composed by 8 animals, the second by 14 and the third by 6 sows, totalizing 28 animals. One-day old piglets were subjected to a common management protocol: they were weighed, received two hundred milligrams of iron dextran, had their teeth grinded and had their ears notched for identification, using local anaesthesia (cream, lidocaine 50 mg/g). In contrast to common farm protocols, male piglets were not castrated and were not tail docked. When piglets reached 27 days of age, they were weighed. At 28 days of age they were weaned and allocated to an experimental pen. Piglets were standardized by weight and gender, at least six piglets (three male and three female) from each sow were used to evaluate skin lesions. The criterion of farrowing order was used in the assignment of piglets in weaning pens, as the sows were not subjected to hormonal protocol for oestrus or farrowing synchronization, and housing by uniform body size in the same group. The strategy for animal allocation resulted in four type of pens, "A”, "B", "C" and "D". Pen "A" had the heaviest animals, followed by "B", "C" and "D”, which housed the lightest piglets, in a descending order considering the differences in piglet weight. Each animal was individually identified, using a marker of non-toxic and non- permanent ink. Piglets were allocated either to pens with only one treatment (HFD N = 21 pens or LFD N = 10 pens), each pen having one male and one female piglet each from 2 different litters (N = 4 piglets per pen), or to pens with two treatments mixed (N = 8 pens). For each litter, 2 piglets were allocated to pens for each of the four weight classes, “A”-“D”, totalling an average of 8 piglets per sow. These nursery pens measured 0.75 m² (1 m x 0.75 m) each, with slatted plastic flooring and no sharp edges. Piglets had access to water and food *ad libitum* and the pens were cleaned daily. Weaning, moving piglets to the new pen and the collection of photos and videos were carried out between 7:00 and 7:30 am. Photos were taken daily to count the lesions, following the methodology of and the novel object and open field tests were held at the end of data collection, three days after weaning. ## Data collection For the counting of lesions, each piglet was photographed and filmed on weaning day (D28, immediately prior to weaning), and during the other two subsequent days, D29 (24 hours after weaning) and D30 (48 hours after weaning). Each piglet was individually restrained and pictures of the body, inner and outer face of the ear, neck and face, on both sides, were captured on 6 photos in total. The counting methodology used was the total amount of identified lesions in the photos. Two independent evaluators analysed the photos, independently, having no knowledge of treatment allocation for each animal. The open field test and the novel object test were performed in combination at the end of the D30, and followed similar methodology to that described by. In each weaning group, all tests began approximately at 15:30 pm and lasted for 10 minutes per animal. The pen used in the test had solid concrete floor, measuring 4.85 by 2.37 m; the floor was marked with permanent yellow ink into quadrants of 83 by 78 centimetres. A digital camera was used for video recording. The pen was always washed with water before the start of each test, to minimize interference of olfactory cues. The test order of the piglets was in accordance with the pens, starting with one piglet from pen "A", followed by a piglet from pen "B", "C" and "D", with an interval of 30 minutes between animals of the same pen. The first animals tested in the same pen were littermates of one litter, followed by their sisters. Each animal was placed in the same starting position. All this strategy was performed to standardize at maximum the test handling. Only when all the littermates were subjected to the test, the litter of the other sow, which was in the same pen, was tested. All piglets were tested in the same day. The open field test was conducted in the first five minutes and the novel object test in the following five minutes. The object used was a polypropylene bucket with a capacity of 20 litres, empty and yellow. The bucket was suspended and always descended in the centre of the pen, 5 minutes after the start of the open field test, using a pulley mechanism, without visual contact between the animal and the experimenter. Because of the position of the camera, a few quadrants was invisible. However, these quadrants were in the same part of the pen and did not influence on our analysis. In the open field test, the following measures were taken: piglet’s activity (time spent walking), amount of time the animal remained in the central and lateral quadrants (quadrants on the edge of the pen), latency to move and vocalization. In the novel object test, we measured latency to interact with the object, time of interacting with the object (close to and with the head toward to the object), time the animal remained close to the object (in quadrants that surround the object), and vocalizations. All types of vocalizations was counted for both tests. All quadrants were evaluated in both tests. The performance data of piglets were obtained during the suckling period. All animals were weighed three times (one day old, 21 days old and 27 days old, see ) and the average daily gain per litter and individual one were calculated according to the piglet´s exact age. ## Statistical analyses Data were analysed with the package Statistical Analysis System (SAS Inst., Inc., Cary, NC). Initially, the data were analysed for the presence of discrepant information (outliers) and to verify residual normality we used the Shapiro-Wilk test. When the normality assumption was not found, the transformation by logarithmic, the square root or arc sine was used. For the open field test, just the data of latency movement was transformed with logarithmic and the vocalization was transformed with arc sine square root. Moreover, for the novel object test, the data of latency to reach the object was logarithmically transformed, and the data from exploratory behaviour and proximity of the object were square-root transformed. The data of skin lesions and open field test behaviour were analysed using mixed models procedure of SAS (PROC MIXED) in a randomized block design. The data of lesion scores were presented as untransformed least square means and SEM of pen averages. The data of open field and novel object test were presented as untransformed least square means and SEM on treatment averages. Performance data were submitted to analysis of variance and the model included the effect of treatment as a fixed effect and the block as a random effect. The means and SEM was obtained from raw data. For all parameters, a 5% level of significance was required. # Results Data on weight gain indicated that there was no treatment effect for the performance parameters evaluated. The performance data of the piglets are presented in. Litter weight at day 27^th of lactation, for LFD piglets, was numerically higher than HFD piglets, approaching significance. Data collected by two independent evaluators revealed that piglets born from HFD sows had fewer skin lesions on the day of weaning than LFD piglets (HFD 4.78 lesions and LFD 7.14 lesions, P = 0.023;, see). No differences in skin lesions were observed on post-weaning days 1 (D29) and 2 (D30). Mixing of the animals, during weaning, induced an eight fold increase in the recorded skin lesions. The two independent observers analysed a total of 936 pictures and the difference between than was 1.6 lesions on D28, 3.8 lesions on D29 and 6.9 lesions on D30. According to the Tukey-Kramer test, HFD and Mixed piglets were equal (P = 0.1236) and LFD piglets differed from HFD piglets (P = 0.0086) and Mixed piglets (P = 0.0016). There was no effect of treatment on the behavioural measures obtained during the open field test and the novel object test (Tables). The number of vocalizations approached significance in the novel object test, being somewhat higher in the HFD piglets. The reason for this tendency could be a different way that piglets cope with challenges, as they were isolated from littermates and adopted different approaches in this stressful situation. # Discussion The aim of the present study was to investigate the impact of a high fibre diet for pregnant gilts on the levels of skin lesions and behaviour in open field and novel object test. Piglets born from HFD sows had fewer skin lesions at the weaning day, day 28, than piglets form LFD sows. This may indicate that HFD piglets avoided engaging in mutual fights, or had less motivation to engage in aggressive interactions before weaning, and therefore had fewer skin lesions at weaning. Previous work demonstrated a correlation between agonistic behaviour and skin lesions in pigs. Our data did not reveal any significant effect of dietary fibre content before weaning on skin lesions after weaning. At days 29 and 30 the skin lesions were considerably higher than at weaning in both groups, probably masking any treatment effect. It is important to highlight the unexpected finding that there was a difference in the mean skin lesions between the dietary fibre treatments at the weaning day, when, as we assumed, the piglets were still part of a stable social group. A likely cause of this difference could be differences between groups in competition to access milk. However, we did not identify any difference in the performance measures, such as total litter weight at birth, average weight at 21 days, average daily gain per litter and individual average individual daily gain. Only the average weight at 27 days was numerically higher for LFD piglets, when compared with HFD piglets, approaching significance (P = 0.077). The insignificantly higher average weights in LFD piglets at all ages may be related to slightly smaller litter sizes in the LFD group. Given the treatment effect on the weaning day, it was surprising to find no treatment effect in skin lesions on days 29 and 30, one and two days after weaning. The choice to house 4 animals per pen, two littermates, may have created an unnaturally less complex, social environment for recently weaned pigs. In natural settings, litters of 4–6 sows interact forming a rather complex social environment after weaning. In commercial settings, the number of piglets housed in weaning pens varies, but it is seldom less the 12 animals. This methodology was adopted because of the importance to systematize the weaning of the piglets. It is plausible to argue that extremely challenging situations, as weaning more animals per pen as in commercial herds, could create a more challenging environment for animals, exacerbating the differences in agonistic behaviour observed before weaning. The choice of open field and novel object tests, aimed at identifying characteristics of individual piglets, which could explain agonistic behaviour, measured by skin lesions. The behaviour of piglets in the open field test and novel object test showed no differences between treatments (Tables). The results found by in rodents showed that a poor environment (without enrichment) before mating, generated more fearful offspring, which spent less time in the central quadrants in the open field test. In this study, the open field test and novel object test did not contribute to explaining the difference found in animal skin lesions before weaning period. Perhaps, with a large number of animals, we could test then in different ages to verify the correlation between the skin lesions prior to weaning and behaviour in open field and novel object tests. In addition, measuring the piglets behaviour, as well as the sow’s behaviour, during the suckling period could contribute to understanding deeply this difference in skin lesions. # Conclusions Our measures of lesions in piglets by weaning at 28 days of age, indicate that feeding pregnant sows with a HFD is effective in generating piglets that had fewer lesions prior to weaning, suggesting less agonistic interactions amongst littermates. This approach of HFD for sows in their piglets is novel and can be a viable alternative to alleviate aggressive behaviour among piglets during lactation. Data on the number of lesions and behavioural responses to the open field test and novel object test after weaning, did not identify differences between piglets of HFD or LFD sows. Further studies with long-term data collection on piglets and their mothers will provide valuable information about the consequences of diets in the prenatal period on the behaviour and welfare of piglets before and after weaning. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** TB AJZ PT BM. **Formal analysis:** TB PHMR PT. **Funding acquisition:** AJZ. **Investigation:** TB PT BM. **Methodology:** AJZ TB. **Project administration:** TB PT BM AJZ. **Resources:** AJZ. **Supervision:** AJZ. **Visualization:** TB. **Writing – original draft:** TB AJZ. **Writing – review & editing:** TB PT BM PHMR AJZ. [^3]: Current address: School of Veterinary Medicine and Animal Science, University of São Paulo, Pirassununga Brazil [^4]: ‡ These authors also contributed equally to this work.

# Introduction Lung cancer is the leading cause of cancer death in the world, responsible for 1,590,000 deaths in 2012, about 4400 per day. Cigarette smoking is the cause of approximately 80% of this mind-boggling death toll in males and at least 50% in females. Decreasing the prevalence of cigarette smoking is one proven approach to lung cancer prevention; a goal would be to return lung cancer to the category of a relatively rare disease, as it was early in the 20^th century. But the world has 1.25 billion smokers, whose nicotine addiction is eagerly fed and supported by tobacco companies with massive financial resources, so it does not appear that this goal will be reached in the near future. In the meantime, it is important to understand factors that dictate susceptibility to lung cancer, so that alternative preventive measures can be devised. One clue to a better understanding of lung cancer susceptibility is different risks among smokers in varied ethnic groups. Thus, investigators in the Multi- ethnic Cohort (MEC) Study found that, for the same quantity of cigarettes smoked, African Americans and Native Hawaiians were at greater risk for lung cancer than Whites while Latinos and Japanese Americans were less susceptible. These differences were evident in men and women and for all histologic types of lung cancer. The differences in susceptibility were most pronounced at lower numbers of cigarettes smoked per day, and were not observed in non-smokers. Many studies comparing lung cancer risk between specific ethnic groups have produced similar results. Our working hypothesis is that differences in the uptake and metabolism of pulmonary carcinogens and toxicants in tobacco smoke are responsible, at least in part, for the observed variation in lung cancer risk. We are exploring this hypothesis by analyzing tobacco smoke constituents and their metabolites in the urine of subjects from the five ethnic groups noted above in tandem with a genome wide association study (GWAS). In studies completed so far, we have reported differences in levels of nicotine and its metabolites as well as metabolites of the tobacco-specific lung carcinogen 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone (NNK) in these five ethnic groups and have examined the relationship of nicotine metabolites to GWAS signals on chromosome 4, specifically variants in *UGT2B10*. In the study presented here, we have explored the possible roles of acrolein and crotonaldehyde by analysis of their metabolites 3-hydroxypropyl mercapturic acid (3-HPMA) and 3-hydroxy-1-methylpropylmercapturic acid (HMPMA), respectively, in urine. Structures of these compounds are shown in. 3-HPMA and HMPMA are formed by conjugation of acrolein and crotonaldehyde respectively with cellular glutathione, followed by metabolism of the glutathione conjugates and excretion in urine. Both acrolein and crotonaldehyde are intense eye and respiratory tract irritants. Consistently, inhalation studies of acrolein in laboratory animals demonstrate a variety of toxic effects including irritation, inflammation, cell proliferation, squamous metaplasia and other effects. The irritant properties of crotonaldehyde are also well established. Both acrolein and crotonaldehyde react with DNA to form cyclic 1,*N*^*2*-deoxyguanosine adducts and related structures. The cyclic adducts have been detected in lung tissue from smokers, and the DNA binding pattern of acrolein in the *p53* tumor suppressor gene is similar to the pattern of mutational hotspots in *p53* found in lung tumors from smokers. There is debate about the role of acrolein in lung carcinogenesis by cigarette smoke, as there is little evidence for its carcinogenicity in laboratory animals. The irritant and inflammatory properties of both of these α,β-unsaturated aldehydes, along with the *p53* data for acrolein, indicate that they may play a role in lung carcinogenesis in smokers. In view of these facts, we have analyzed HPMA and HMPMA in the urine of self-identified African Americans, Native Hawaiians, Whites, Latinos, and Japanese Americans, all of whom were regular cigarette smokers. A GWAS in search of common genetic variants possibly predictive of 3-HPMA and HMPMA levels in these subjects was also conducted. # Materials and Methods ## Subjects Approval for this study, including the consent procedure, was obtained from the Institutional Review Boards of the University of Minnesota, the University of Hawaii, and the University of Southern California. Study participants provided written consent. IRB Code Number: 0912M75654. Subjects took part in the MEC, a prospective cohort study investigating the association of genetic and lifestyle factors with chronic diseases in a population with diverse ethnic backgrounds. The cohort consists of 215,251 men and women, ages 45 to 75 at baseline, belonging mainly to the following five ethnic/racial groups: African Americans, Native Hawaiians, Whites, Latinos, and Japanese Americans. Potential participants were identified between 1993 and 1996 in Hawaii and California (mainly Los Angeles County) through voter registration lists, drivers’ license files, and Health Care Financing Administration data. Each participant completed a self-administered questionnaire which was delivered by mail and inquired about demographic, dietary, lifestyle, and other exposure factors. This specific study was carried out in a subgroup of the MEC participants who were current smokers and were cancer-free at the time of urine collection. Thus, about 10 years after their entry into the cohort, 2,393 of this subgroup participated in the MEC bio-specimen sub-cohort and provided a blood sample and a first morning urine sample (subjects recruited in California) or overnight urine sample (subjects recruited in Hawaii), and completed an epidemiologic questionnaire, smoking history questionnaire and medication record. The overnight urine sample was collected starting between 5 pm and 9 pm (depending on the subject). This sample includes all urine passed during the night as well as the first morning urine. All urine was kept on ice until processing. Aliquots were subsequently stored in a -80°C freezer until analysis. The overnight or first morning urine was used to measure 3-HPMA and HMPMA. ## Analysis of 3-HPMA and HMPMA This was performed using a 96-well high throughput LC-MS/MS method, as described previously. Detection limits were 4.5 pmol/ml for 3-HPMA and 3.5 pmol/ml for HMPMA. Accuracy was 92% for 3-HPMA and 97% for HMPMA. Inter-day precision was 9.1% \[coefficient of variation (CV)\] for 3-HPMA and 11.0% for HMPMA. Among blind duplicates included among the samples, there were inter-batch CVs of 18.9% for 3-HPMA and 19.6% for HMPMA while the intra-batch CVs were 9.2% and 7.7% for the respective metabolites. ## Total nicotine equivalents Total nicotine equivalents, the sum of nicotine and its metabolites nicotine glucuronide, cotinine, cotinine glucuronide, 3′-hydroxycotinine, 3′-hydroxycotinine glucuronide, and nicotine *N*-oxide, were determined as described previously. These data have been published, and were used here for statistical adjustments of the 3-HPMA and HMPMA data. ## Creatinine Analysis Creatinine was analyzed using a colorimetric microplate assay (CRE34-K01) purchased from Eagle Bioscience (<http://stores.eaglebio.com/creatinine- microplate-assay-kit>). ## Statistical methods For this analysis, 2,291 subjects were retained. These subjects had total nicotine equivalents \>1.27 nmol/ml (4-times the limit of quantitation) and had either 3-HPMA or HMPMA measured. Of the 2,291 subjects, nine subjects were missing measures of 3-HPMA and seven subjects were missing measures of HMPMA. Additionally, among the subjects retained for this analysis, 11 participants were missing BMI and 42 participants had missing values for cigarettes per day (CPD) at the time of urine collection. Using the Markov Chain Monte Carlo method and PROC MI statement from the SAS v9.2 software (SAS institute, Cary, NC), the missing values for BMI and CPD were imputed. The imputed values were based on age at cohort entry, race/ethnicity, time between cohort entry and time of urine collection, BMI at baseline or number of CPD at baseline and smoking duration (for missing CPD). Ten datasets were created for the imputed missing values and the mean values across all 10 datasets were used to replace the missing measures. To examine the correlation between 3-HPMA, HMPMA and measures of smoking (CPD and total nicotine equivalents), Pearson’s partial correlation coefficients (r) were adjusted for age, sex and race/ethnicity and creatinine levels (natural log). To compare the rank of 3-HPMA and HMPMA levels across race/ethnicity, the Wilcoxon Mann-Whitney test was employed. Also, the covariate-adjusted geometric means for 3-HPMA and HMPMA were computed for each ethnic/racial group at the mean covariate vector. We used two multivariable linear models. The first adjusted for the following predictors: age at time of urine collection (continuous), sex, race, and creatinine levels (natural log) and the second additionally adjusted for total nicotine equivalents. We also examined whether other factors such as BMI were associated with 3-HPMA and HMPMA. To better meet model assumptions, 3-HPMA and HMPMA were transformed by taking the natural log. For ease of interpretation, the values presented in the tables were back- transformed as geometric means to their natural scale. ## GWAS methods A total of 2,418 current smokers were genotyped using the Illumina Human1M-Duo BeadChip (1,199,187 SNPs), as previously described. Imputation of the variants included in the 1000 Genomes Project (<http://www.1000genomes.org/>) was performed using SHAPEIT and IMPUTE2 using a cosmopolitan reference panel (all groups included). After imputation with IMPUTE2 we used SNPs with an IMPUTE2 info score of ≥0.30 and minor allele frequency (MAF) \>1% in any MEC ethnic group in our association testing. For 3-HPMA and HMPMA, a total of 2,211 and 2,213 study participants, respectively, with complete genotype and phenotype data, and 11,892,802 SNPs/indels (1,131,426 genotyped and 10,761,376 imputed) were included in the analysis. In the genetic analyses we adjusted for principal components, race, sex, age, creatinine and total nicotine equivalents, and enforced criteria of 5x10^-8 for genome wide significance. # Results Characteristics of the subjects in this study are summarized in. Median ages in the 5 ethnic groups ranged from 60–64 years, BMI from 24.4–26.9 kg/m³, creatinine from 54–89 mg/dL, cigarettes per day from 7.1–20, and total nicotine equivalents from 27.2–44.4 nmol/mL. Males comprised 30.6% of the African Americans, 36.5% of the Native Hawaiians, 43.3% of the Whites, 51.5% of the Latinos, and 57.1% of the Japanese Americans. In all groups, males smoked more cigarettes per day than females, and males had higher levels of total nicotine equivalents than females in all five ethnic groups. Correlations among cigarettes per day, total nicotine equivalents, and levels of 3-HPMA and HMPMA are summarized in. All correlations were significant (p\<0.0001). The strongest correlations were between 3-HPMA and HMPMA, r = 0.83–0.86, while correlations between total nicotine equivalents and the mercapturic acid levels were somewhat weaker, r = 0.52–0.6. Similar correlation coefficients were obtained when analyzed by ethnic group or gender, in all cases significant (p\<0.0001). Medians and interquartile ranges for levels of 3-HPMA and HMPMA, expressed as pmol/ml urine, are summarized in. For both 3-HPMA and HMPMA, levels of these metabolites were highest and not significantly different among African Americans, Native Hawaiians, and Whites, whereas Latinos and Japanese Americans had significantly lower levels. The data are stratified by sex in. Within each sex, the same trend was observed when comparing the ethnic groups. The relatively high values of 3-HPMA and HMPMA in Native Hawaiians were not due to outliers because they remained after removing the bottom and top 1% of both the total nicotine equivalents and mercapturic acid values. We also observed that levels of both 3-HPMA and HMPMA were significantly higher in males than in females in all ethnic groups. When expressed per total nicotine equivalents, 3-HPMA levels were highest in Native Hawaiians and Japanese Americans, and HMPMA was highest in Native Hawaiians. When expressed per total nicotine equivalents, 3-HPMA and HMPMA levels were significantly higher in males than females in all ethnic groups (p\<0.01) except HMPMA in Latinos (p = 0.14). Medians and interquartile ranges for levels of 3-HPMA and HMPMA, expressed per mg creatinine, are summarized in. Levels of 3-HPMA and HMPMA were highest in Whites and Native Hawaiians, with significantly lower levels in African Americans, Latinos, and Japanese Americans. The lower levels in African Americans when expressed in this manner are due to the significantly higher levels of creatinine in this group. Geometric means of 3-HPMA and HMPMA in the five ethnic groups are presented in. In Model 1, they have been adjusted for age, sex, and creatinine. For both 3-HPMA and HMPMA, the highest levels were in Whites and Native Hawaiians, with significantly lower levels in African Americans, Japanese Americans, and Latinos. The lowest levels of both mercapturic acids were in Latinos, and these were significantly lower than in all other groups. Model 2 was additionally adjusted for total nicotine equivalents; the results were similar to those in Model 1, except that the differences between the Latinos and African Americans were no longer significant for either mercapturic acid, and the difference in 3-HPMA levels in the Japanese Americans and Whites was no longer significant. These results are illustrated graphically in. In the GWAS analysis of 3-HPMA and HMPMA we observed little evidence of inflation in the test statistic in the overall multiethnic sample (λ = 1.0; and **Figs**) or in any single ethnic group (0.95 ≤ λ’s ≤ 1.0) for either phenotype. There were no globally significant variants for the overall results for 3-HPMA using our genomic threshold of p \<5x10^-8. The overall results for HMPMA showed a total of nine globally significant variants (p-values ranged from 4.3x10^-8 to 9.7x10^-10) on seven different chromosomes; all of these SNPs are common in African Americans, two are also common in Whites and two others in Latinos. The top significant association, rs55922880 (chr12), is located near gene *TBX3*, a gene involved in encoding transcription factors. Together, these nine variants explain only 5.4% of variability in HMPMA overall, but when observed by ethnic group, they explain 15.8% in African Americans, 11.7% in Latinos followed by 2.8% in Native Hawaiians and 2.1% in Whites with the least variability explained in the Japanese Americans at 1.4%. In ethnic specific analyses for both phenotypes there were widely scattered associations that were often difficult to interpret due to very low minor allele frequencies (**Tables**). Of the nine overall significant variants for HMPMA, only one (rs7675915) was also found to be globally significant in the ethnic specific analysis (among Latinos at 2.95 x 10^–8). However, all of the associations tended to be consistent by the different ethnic groups. # Discussion The urine samples in this study have been previously analyzed for total nicotine equivalents and total NNAL. Total nicotine equivalents is an excellent indicator of nicotine dose in smokers. Total NNAL (the sum of free NNAL and its glucuronides) correlates with total nicotine equivalents. Free NNAL and NNAL glucuronides are metabolites of the potent tobacco-specific lung carcinogen NNK. Free NNAL is itself a powerful lung carcinogen. In these studies, median levels of total nicotine equivalents and total NNAL, expressed per ml urine, were highest in African Americans, intermediate in Whites, and lowest in Japanese Americans, and these differences were significant. This order of total nicotine equivalents and total NNAL concentrations is consistent with the lung cancer risk for African Americans (highest), Whites (intermediate), and Japanese Americans (lowest). But the data for Native Hawaiians and Latinos in those two studies did not reflect their lung cancer risk. Native Hawaiians had significantly lower levels of total nicotine equivalents and total NNAL than Whites, when expressed as medians per ml urine, and Latinos had levels the same as Whites. The results of the study reported here, in Model 2 ( **and**), demonstrate that Native Hawaiians had the highest levels of the acrolein and crotonaldehyde metabolites 3-HPMA and HMPMA among the five ethnic groups, and these were statistically indistinguishable from those of Whites, and significantly higher than those of the other groups. Collectively, our results to date thus suggest that acrolein and crotonaldehyde may play some role in the relatively high lung cancer risk of Native Hawaiians. Furthermore, the relatively low levels of 3-HPMA and HMPMA in Latinos are also consistent with their lower lung cancer risk. All humans have 3-HPMA and HMPMA in their urine because acrolein and crotonaldehyde are ubiquitous environmental and dietary constituents as well as being products of endogenous metabolism. Levels of 3-HPMA are typically about 4–10 times higher in the urine of smokers than non-smokers and they decrease rapidly and significantly when people stop smoking. Similar findings pertain to HMPMA but fewer studies have been reported. Cigarette mainstream smoke typically contains 5–60 μg of acrolein per cigarette, and these levels as well as those of nicotine correlate with “tar” in the same brands; less data are available for crotonaldehyde. “Mouth level exposure” to acrolein in cigarette smokers, as determined in cigarettes with increasing deliveries of acrolein, is highly correlated with 3-HPMA in urine. Similarly, mouth level exposure to nicotine is highly correlated with nicotine equivalents in urine. Furthermore, in our study and in a large population based study, total nicotine equivalents in urine correlated with 3-HPMA in urine. Collectively, these observations indicate that acrolein and nicotine in cigarette smoke as well as total nicotine equivalents in urine are strong determinants of 3-HPMA in urine. The Native Hawaiians in our study seem to represent an exception to this generality as their total nicotine equivalents were significantly lower than those of African Americans (p\<0.0001) or Whites (p = 0.0145), yet their 3-HPMA levels were as high as those of African Americans and Whites. This suggests that there is an important source of acrolein exposure in Native Hawaiians, either exogenous or endogenous, other than cigarette smoke, that contributes to their relatively high levels of urinary 3-HPMA and possibly to lung cancer risk. Similar considerations would presumably apply to HMPMA. We do not have data on the types of cigarettes smoked by the Native Hawaiians in this study, nor do we have data on 3-HPMA or HMPMA in Native Hawaiian non-smokers compared to non-smokers from other ethnic groups. Such data could possibly provide further insight into their 3-HPMA levels. Previous studies have clearly established the presence of 3-HPMA in the urine of all non-smokers, but the levels can be variable. In one study, a non-smoker group had levels of 3-HPMA which were more than 5 times higher than those reported in some large studies of smokers, indicating important sources of acrolein exposure other than tobacco smoke, but these were not identified. Forest fires, urban fires, automobile exhaust, cooking fumes, and other sources of incomplete combustion including industrial emissions are among the environmental sources of acrolein. It has also been detected in certain foods and beverages such as coffee and tea and is produced in the body during lipid peroxidation, amino acid metabolism, and polyamine metabolism. Sources of exposure to crotonaldehyde are similar to those of acrolein, and there is convincing evidence from studies of DNA adducts that crotonaldehyde is formed endogenously in humans. The toxic effects of acrolein, an intensely irritating compound with a disagreeable acrid and pungent odor, are well documented. Inhalation of acrolein causes severe respiratory distress and a wide variety of toxic effects in laboratory animals including toxicity to cilia, depressed respiratory rate, weight loss, inflammation, immunosuppression, cell proliferation, and various histopathological changes in the respiratory tract. Acrolein reacts easily with critical proteins such as thioredoxin reductase in bronchial epithelial cells, resulting in dysregulation of cellular oxidative balance and related effects. It also inhibits acetylation of aromatic amines and nucleotide excision repair. Carcinogenicity studies of acrolein alone have been uniformly negative. One study demonstrated a significantly increased incidence of bladder tumors in rats treated by i.p. injection with acrolein followed by dietary uracil. In spite of these relatively negative carcinogenicity data in laboratory animals, which might be partly a consequence of acrolein’s toxicity, a possible role of acrolein in cigarette smoke-induced lung cancer has emerged from studies of its interactions with the *p53* tumor suppressor gene which mirror those found in lung tumors from smokers. Acrolein is known to form exocyclic 1,*N*^*2*-deoxyguanosine adducts in DNA. The mutagenicity of these adducts varies from none to moderate. Multiple studies have detected acrolein- DNA adducts in human tissues or cells, including oral cells, colon cells, leukocytes, bladder mucosa, and lung, but there is presently no evidence that these adduct levels are higher in the lungs of smokers than non-smokers. Collectively, these studies suggest a possible genotoxic role for acrolein in lung cancer induced by cigarette smoke, but this hypothesis has gaps. Acrolein could also contribute to lung cancer etiology by increasing inflammation in the lung, thus acting as a co-carcinogen by enhancing the consequences of DNA damage by carcinogens in tobacco smoke. Crotonaldehyde, like acrolein, is a potent irritant to the respiratory tract and eyes. It caused preneoplastic lesions and a low incidence of neoplastic nodules and hepatocellular carcinoma in rats when administered in the drinking water. Crotonaldehyde forms mutagenic cyclic 1,*N*^*2*-deoxyguanosine adducts in DNA, similar to those produced from acrolein, and these have been detected in human lung as well as other tissues. Presumably, inhalation would be the relevant route of exposure for most of the effects discussed here, although it is possible that endogenous processes in the lung associated with the toxic effects of smoking could result in local generation of acrolein or crotonaldehyde. For example, oxidation of ω -3 fatty acids produces acrolein; it is possible that oxidants in cigarette smoke interact with ω -3 fatty acids in the lung resulting in local formation of acrolein. There could also be unrecognized dietary or endogenous sources of acrolein and crotonaldehyde. We observed higher levels of 3-HPMA and HMPMA in the urine of male than in female smokers, consistent with previously reported results of some, but not all studies of 3-HPMA. Higher levels of a number of biomarkers including total nicotine equivalents in the urine of male smokers compared to female smokers have been observed in multiple previous studies, reflecting differences in smoking topography. However, in our study, levels of 3-HPMA and HMPMA were higher in males than in females even after correcting for total nicotine equivalents. Consistent with the discussion above, this suggests that there is another source of acrolein and crotonaldehyde exposure which is greater in males than in females. The relatively low levels of 3-HPMA and HMPMA in Latinos is also worth noting, as it is consistent with their relatively low lung cancer risk, in contrast to our previous observations of total nicotine equivalents and total NNAL in this group which were not significantly different from those of Whites. However, we also note that the Japanese-Americans levels of 3-HPMA in Model 2 (**,**) are statistically indistinguishable from those of Whites, and this does not correlate with their lung cancer risk. We did not observe a significant signal in the GWAS in search of common genetic variants associated with urinary levels of 3-HPMA in our population. Acrolein is known to be an excellent substrate for GSTP1 with K_cat/K_m values of 92–350 mM^-1.s-1 while GSTM1-1 and GSTA1-1 have less catalytic activity. The K_cat/K_m values for catalysis of glutathione conjugation of crotonaldehyde by these enzymes are less than 1/10^th those of acrolein. The *GSTP1* gene, located on chromosome 11q13, is a polymorphic gene encoding variant proteins with moderate activities for catalysis of acrolein conjugation. *GSTP*-null mice suffered increased bladder damage upon treatment with cyclophosphamide, which releases acrolein as a metabolite, although the overall excretion of 3-HPMA in these mice was unchanged. Based on these data collectively, we might have expected to see a signal on chromosome 11, particularly in association with 3-HPMA, but this was not observed. We did see significant variants associated with HMPMA, with the strongest association on chromosome 12 (rs 55922880), located near the gene *TBX3*, but the relationship of this gene to crotonaldehyde metabolism is not clear at present. Overall, the results suggest that the formation of the glutathione conjugates of acrolein and crotonaldehyde that are metabolized to urinary 3-HPMA and HMPMA is mainly a non-enzyme catalyzed process. The uncatalyzed reactions of acrolein and crotonaldehyde with glutathione and related sulfhydryl compounds are well established. In summary, the results of this study provide some intriguing new leads with respect to the possible role of the α,β-unsaturated toxicants acrolein and crotonaldehyde in lung cancer etiology in smokers. The relatively high levels of the acrolein and crotonaldehyde biomarkers 3-HPMA and HMPMA in the urine of Native Hawaiians, compared to the other groups in the MEC, is particularly interesting. Further studies are required to investigate endogenous or exogenous exposures to acrolein and crotonaldehyde that might account for these results. # Supporting Information This study was supported by grant no. CA-138338 and in part by grant no. CA-14089 from the U.S. National Cancer Institute ([http://www.cancer.gov](http://www.cancer.gov/)). Mass spectrometry was carried out in the Analytical Biochemistry Shared Resource of the Masonic Cancer Center, supported in part by National Cancer Institute Cancer Center Support grant no. CA-77598. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SLP DOS CAH LLM SSH. Performed the experiments: SGC MC YP. Analyzed the data: SLP SSH DOS YP. Contributed reagents/materials/analysis tools: SGC MC SSH. Wrote the paper: SLP DOS YP LLM CAH SSH.

# Introduction Oxaliplatin is a third-generation platinum derivative that has become the standard treatment for advanced colorectal cancers. However, oxaliplatin also has a unique pattern of neurotoxicity. In more than 90% of patients, oxaliplatin induces an acute and transient thermal hypersensitivity, with a rapid onset of distal paresthesias and cold-triggered dysesthesias. Thereafter, with the repetition of chemotherapy cycles, 50–80% of patients experience a severe cumulative peripheral neuropathy that is poorly reversible. As a consequence, oxaliplatin-induced neurotoxicity is strongly incapacitating for patients and affects their quality of life. Moreover, the lack of neuroprotective strategies to prevent these peripheral neuropathies often compels oncologists to reduce or discontinue the chemotherapy cycles, thus decreasing the chances of survival. The exact cause of this neurotoxicity is yet to be conclusively identified. Mihara *et al.* (2011) recently showed that repeated administrations of oxaliplatin induced N-methyl-D-aspartate receptor subtype 2B (NR2B) up- regulation in the rat spinal cord, which is known to be involved in the development and maintenance of chronic pain resulting from peripheral nerve injury. However, the clinical use of selective NR2B antagonists remains limited by severe adverse effects, despite their analgesic efficacy in neuropathic pain. It has recently been shown that a polyamine-deficient diet (PDD) relieved pain hypersensitivity in animal models of chronic mononeuropathy and monoarthritis and decreased tyrosine phosphorylation of the NR2B subunit in the rat spinal cord after intraplantar carrageenan injection. Polyamines are biogenic polycations ubiquitously distributed in eukaryotic cells. The main source of polyamines is exogenous, through dietary intake and absorption of the gut flora metabolites. An endogenous source is also provided by the decarboxylation of ornithine *via* ornithine decarboxylase (ODC) and by interconversion pathways. Polyamines (including spermine, spermidine and putrescine) have been extensively studied as positive modulators of NR2B-containing NMDA receptor activity, thus facilitating pain sensitization. The aim of the present study was to validate the concept that a PDD could be a beneficial strategy to reduce oxaliplatin-induced acute pain hypersensitivity by inhibiting glutamate neurotransmission in the spinal cord. First, the preventive effect of a PDD on oxaliplatin-induced acute pain hypersensitivity was investigated in an animal model of oxaliplatin-induced acute hypersensitivity using behavioral tests. Secondly, the involvement of spinal glutamate neurotransmission in this animal model was monitored by using a ¹H-NMR-based metabolomic approach and assessing the expression and phosphorylation of the NR2B subunit. Thereafter, the impact of PDD on oxaliplatin-induced increase of spinal glutamate levels was explored. Finally, the implication of the NR2B subunit was confirmed by evaluating the antihyperalgesic effect of ifenprodil, a NR2B-selective antagonist, on oxaliplatin-induced acute cold and mechanical pain hypersensitivity. # Materials and Methods ## Ethics Statement All experiments involving animals were performed in compliance with EU legislation (directive 86/609/EEC of 24 November 1986), and all procedures were approved by the local ethical review board for animal experiments (CEMEA - *Comité d’Ethique en Matière d’Expérimentation Animale Auvergne*) under agreement numbers CE14-11, CE26-10 and CE16-10. ## Animals The experiments used adult male Sprague-Dawley rats (Charles River, France) weighing 150 to 175 g at the beginning of the study. The rats were housed in a controlled environment (22°C ±2°C; 50% relative humidity; 12/12 light-dark cycle) in groups of 4 rats per cage. To minimize stress during the behavioral tests, the rats were given an acclimatization period of 7 days to habituate them to the handling and testing procedures. ## Diet The synthetic diet was prepared with products from Sigma Chemical Co. (Sigma, France) according to the formulation of Kergozien *et al.*. Two different synthetic diets were used in this study: (1) a normal diet providing an equivalent amount of polyamines to normal rat chow (54 mg/kg putrescine, 37 mg/kg cadaverine, 27 mg/kg spermidine and 7 mg/kg spermine) and fulfilling all rodent nutritional needs ; (2) a polyamine-deficient diet (PDD) equivalent to normal rodent chow but containing less than 10 µg/kg of polyamines. Dietary polyamine level was determined by atmospheric-pressure chemical ionization as previously described. On reception at the animal housing facility, *i.e*. 7 days before the beginning of the experiments, the rats were assigned to a specific diet (normal diet or PDD) and left to become habituated to the experimental environment. Food and water were provided *ad libitum.* ## Animal Model of Oxaliplatin-induced Acute Pain Hypersensitivity Acute neurotoxicity was induced in rats by a single intraperitoneal (i.p.) oxaliplatin injection (6 mg/kg), as described by Ling *et al.*. Oxaliplatin (Debiopharm, Switzerland) was dissolved in a 5% glucose solution (Freeflex®, Fresenius Kabi, France) at a final concentration of 2 mg/ml. Control rats received an equal i.p. injection of vehicle. ## Behavioral Testing Procedures of Hypersensitivity Behavioral tests were chosen according to clinical symptoms (mechanical allodynia/hyperalgesia and cold hypersensitivity) observed in oxaliplatin- treated patients. The electronic von Frey test was used to assess mechanical allodynia/hyperalgesia and the thermal place preference test was used to measure cold avoidance behavior. ### Mechanical allodynia/hyperalgesia The electronic von Frey test (Bioseb®, France) consists of a hand-held force transducer fitted with a plastic tip. Each rat was placed in a suspended plastic cage (17×11×13 cm) with a wire grid floor 15 min before the beginning of the test. The plastic tip was applied perpendicularly to the medial plantar surface of the hindpaw from below the wire grid floor. The force applied was gradually increased until withdrawal of the hindpaw. Mechanical pressure thresholds (g) were automatically registered by the electronic device and used as a pain parameter. ### Cold hypersensitivity The thermal place preference test (Bioseb®, France) was used to assess a cold avoidance behavior. This test consists of two contiguous metal plates surrounded by a plastic enclosure. The first plate was kept at neutral temperature (25°C) whereas the second plate was kept at cold temperature (12°C, determined by preliminary experiments). The test was performed in the morning, and in darkness. Each session lasted 3 minutes, during which the animals were left free to explore both plates. The time spent on the cold plate during the entire session was recorded using an infrared camera connected to a computer in order to determine cold avoidance behavior. To improve the explorative behavior of the rats used for this test, the light-dark cycle was inverted immediately after the animal arrived at the animal housing facility. A few days before the beginning of the experiment, the rats were repeatedly placed on the apparatus with both plates held at room temperature (25°C) during 3 minutes. In these conditions, rats spent an equal amount of time on each plate, thus showing no place preference. To avoid learning or any place preference unrelated to cold, the temperature of the plates were inverted between two consecutive sessions. ## Immunoblotting Dorsal spinal cord samples were homogenized in 400 µL of ice-cold lysis buffer (50 mM HEPES, pH 7.5, 150 mM NaCl, 10 mM EDTA, 10 mM Na₄P₂O₇, 2 mM orthovanadate, 100 mM NaF, 1% Triton X-100, 0.5 mM PMSF, 20 µM leupeptin, and 100 IU/mL aprotinin; Sigma, St- Quentin-Fallavier, France), as described in. After 1–2 min sonication at 4°C, samples were centrifuged (16,000 *g*, 15 min, 4°C), and the pellets were discarded. Protein concentration was determined using a bicinchoninic acid protein kit assay (BC Assay UP40840A®, Interchim, France). Protein content was then separated in 7.5% SDS-PAGE and transferred to a nitrocellulose membrane (Millipore, France). Membranes were blotted overnight at 4°C with the antibodies indicated according to the manufacturer’s recommendations. Tested antibodies were total NR2B (1∶500, ref: 06–600, Upstate Biotechnology, Millipore, France), phospho-Tyr1472 (1∶500, ref: AB5403, Millipore, France), phospho-Tyr1336 (1∶500, ref: AB9690, Millipore, France) and phospho-Ser1303 (1∶500, ref: 07–398, Millipore, France). The next day, the membranes were blotted with horseradish peroxidase-conjugated secondary antibodies (1∶50,000, goat anti-rabbit, Pierce, France). Immunoreactivity was detected using a chemiluminescence kit (Immun- star™ WesternC™ Kit, Bio-rad®, France). Representative western blots were scanned using a ChemiDoc™ XRS system (Bio-rad®, France). The density of each band was quantified using Image Lab™ software (Biorad®, France) and normalized against the corresponding beta-actin signal used as a loading control (1∶5000, A5441, Sigma-Aldrich, France). The resulting ratios were expressed as percentage of control (vehicle). ## Proton NMR Spectroscopy-based Metabolomic Analysis ### Sample preparation Spinal dorsal horn from 2 rats receiving the same regimen were pooled to reach a minimum of 80 mg of tissue (n = 8 to 10 pooled by two, i.e. n = 4 to 5 for each group). Aqueous metabolite extraction was performed as described by Angenstein *et al.*, with minor changes. Briefly, tissue samples were homogenized in a 6% perchloric acid solution (5 ml/g of tissue) and centrifuged (12,000 *g*, 10 min, 4°C). The supernatant was adjusted to pH 11 using a 10% potassium hydroxide solution. Potassium perchlorate precipitate was then removed by centrifugation (4,000 *g*, 10 min, 4°C). Samples were lyophilized overnight and stored at −80°C until NMR spectroscopy analysis. Dried samples were reconstituted in 600 µl D₂O containing 1 mM of 3-(trimethylsilyl)-2,2,3,3-tetradeuteropropionic acid (TSP, Sigma, France) used as internal standard for concentration and chemical shift. ### Data acquisition and processing Experiments were performed on a Bruker Avance 400MHz spectrometer (Bruker Biospin, Germany) equipped with a quad-nuclear probe. Spectral acquisitions were automated using ICON-NMR software (Bruker BioSpin GmbH, Germany). 1D ¹H-NMR spectra were obtained using a Nuclear Overhauser Enhancement SpectroscopY sequence with low-power water signal PResaturation (NOESYPR), a 2.5-s presaturation delay, and a 1-s relaxation delay. Spectral width was 12 ppm with 16K complex datapoints and 128 repetitions. After Fourier transformation, low-order phase correction and spline baseline correction were applied in a standardized way. 1D ¹H-NMR spectra were processed using TopSpin 3.1 (Bruker BioSpin GmbH, Germany). ### Metabolite identification and quantification Representative high-resolution 1D ¹H-NMR spectra of the metabolite extract from control and oxaliplatin-treated rats fed with a normal diet and oxaliplatin-treated rats fed with a PDD are presented in. Four to 5 ¹H-NMR spectra were recorded for each treatment condition. Eighteen metabolites were identified and quantified in each ¹H-NMR spectrum. Metabolite concentrations were measured by peak integration and compared to the integral of the reference compound TSP 1 mM in each spectrum. Metabolite concentration, expressed as a percentage of control fed with a normal diet, was calculated as follows: \[mean metabolite concentration in a treated group/mean metabolite concentration in the control group fed with a normal diet\]×100. ## Experimental Design For *in vivo* studies, four different groups of diet and treatment were defined to determine the effect of a PDD on oxaliplatin-induced acute nociceptive disorders. They included control rats fed with a normal diet or PDD, and oxaliplatin-treated rats fed with a normal diet or PDD. The inhibition of NR2B-containing NMDA receptors was evaluated in oxaliplatin-treated rats fed with a normal diet with the specific inhibitor ifenprodil. Ifenprodil tartrate (Sigma, France) was dissolved in a saline solution (0.9% NaCl) just before administration. At 3 days post-oxaliplatin injection, neuropathic rats were randomly assigned to 4 groups (n = 8 per group) according to ifenprodil dose administered: 0 µg (NaCl 0.9%), 2.5 µg/rat, 5 µg/rat or 10 µg/rat. Ifenprodil was intrathecally administered under isoflurane anesthesia at a final volume of 10 µL/rat. Experiments were performed blind with randomization of treatments and according to the method of equal blocks to avoid any uncontrollable environmental changes. For *ex vivo* analyses, rats were sacrificed by decapitation at day 3 (peak pain behavior) post-oxaliplatin injection (Ling *et al.*, 2007). Sacrifices were performed by a trained professional, without any anesthetic that could modify spinal metabolism. The spinal cords (L4–L6) were quickly removed, and the dorsal part of the spinal cord were isolated on an ice pack, immediately frozen in liquid nitrogen, and stored at −80°C until analysis. The total procedure lasted less than 5 min for each sample. ## Statistical Analysis In behavioral tests, means and standard error of the mean (SEM) were calculated for quantitative variables. Normality of the distribution was checked with a Shapiro-Wilk test. To compare the time-course evolution of different parameters, a repeated-measure ANOVA (or a non-parametric Friedman test if necessary) was followed by a Tukey-Kramer test to compare differences between and within groups. If a significant interaction between time and group was observed, a one- way ANOVA was performed for all time points. These analyses were completed by random-effects models (REM), considered more robust to missing data. The REM were able to take into account 1) fixed effects as treatment, time and interaction between time and group, and 2) random subject effects as random intercept and slope. Residual normality was checked for all models presented in this article. Results from immunoblotting were analyzed using a nonparametric Mann-Whitney-Wilcoxon test to compare differences between groups. Differences were considered statistically significant at p\<0.05. Statistical analysis was performed using STATA® v.10 software (StataCorp, College Station, TX, USA). In the metabolomic analysis, statistical comparisons between control and treated groups were performed using the nonparametric Mann-Whitney test. Differences were considered statistically significant at p\<0.05. To identify the most discriminating metabolites of response to oxaliplatin, PDD or the combination of both, partial least squares discriminant analysis (PLS-DA) was applied to concentrations of the 18 identified metabolites using XLSTAT v2012.6.08 software (Addinsoft). Coefficients of PLS-DA loading scores were used to express the level of correlation between original variables and components of the PLS-DA model. Cross-validation of each model led to an evaluation of the R₂Y and Q₂ factors. R₂Y provides an estimate of how well the model fits the Y data, and Q₂ provides an estimate of how well the model predicts the Y data. Typically, a robust model will have R₂Y\>0.5 and Q₂\>0.4. In PLS-DA, the variable influence on the projection (VIP), a weighted sum of squares of the PLS weights, was calculated to indicate the importance of a metabolite in the model. A VIP value clearly above 1.2 was required to retain the discriminating power of a metabolite. # Results No rat displayed any loss of weight or altered clinical condition regardless of the treatment or diet received (data not shown). There was no significant difference in daily food intake between rats fed with a normal diet (6.46±0.33 g/day/100 g body weight) and rats fed with a PDD (6.32±0.32 g/day/100 g body weight) (p = 0.56). Before oxaliplatin injection, there was no statistical difference between mean thresholds of the different groups in behavioral tests. ## The PDD Prevents Oxaliplatin-induced Mechanical and Cold Hypersensitivity The PDD for 7 days did not modify sensitive thresholds compared to control rats fed with a normal diet. While, with a normal diet, oxaliplatin treated rats showed a significant decrease in paw withdrawal thresholds from day 2 (−22%, p\<0.001) to day 4 (−15%, p = 0.016), compared to control rats, with a maximum reduction at day 3 (−23%, p\<0.001), oxaliplatine-treated rats fed with a PDD exhibited no significant change in mechanical thresholds. Analysis of REM showed a significant interaction between treatments and diets (p = 0.002). Regarding oxaliplatin-induced cold hypersensitivity, while oxaliplatin treated rats fed with a normal diet showed a significant decrease in the time spent on the cold plate from day 2 (−50%, p\<0.008) to day 4 (−45%, p\<0.002), compared to controls, oxaliplatin-treated rats fed with a PDD showed no difference compared to control rats regardless of the diet received (normal diet or PDD). Analysis of REM showed a significant interaction between treatments and diets (p = 0.006). ## The PDD Suppresses Oxaliplatin-induced Increase of Glutamate Level in Spinal Dorsal Horn ### Oxaliplatin increases spinal glutamate concentration without modifying NR2B expression or phosphorylation Three days after oxaliplatin injection, there was a significant increase in glutamate (Glu, +85%, p = 0.0253) and a decrease in total creatine (tCr) (−21%, p = 0.0339) and adenosine phosphate (AXP) (−20%, p = 0.0143) concentrations in oxaliplatin treated rats compared to control rats fed with a normal diet. PLS- Discriminant Analyses (PLS-DA) were performed on samples taking into account the treatment and diets received to identify relevant potential metabolites involved in oxaliplatin-induced metabolic changes. For animals fed with a normal diet, PLS-DA showed a clear separation between the samples of oxaliplatin treated and control animals (R₂Y = 0.721, Q₂ = 0.527, validated model), indicating that the metabolic fingerprints of these two groups are significantly distinct. The most discriminant metabolites with a VIP value above 1.2 were acetate (Ace), AXP, Glu, glycerophosphocholine (GPC), phosphatidylcholine (PtC) and tCr. At the same time after oxaliplatin injection, as shown in, immunoblot analysis revealed no change in NR2B subunit expression compared to control rats. Furthermore, levels of p-Tyr1472, p-Tyr1336 and p-Ser1303 NR2B subunits remained unchanged after oxaliplatin injection compared to control rats, at day 3. ### The PDD suppresses oxaliplatin-induced increase of glutamate level The PDD, devoid of any effect by itself on glutamate level, suppressed the oxaliplatin-induced increase in Glu concentration at day 3 post-oxaliplatin injection and the decrease in tCr. For oxaliplatin-treated rats, spinal Glu concentration was significantly lower in rats fed with a PDD compared to a normal diet (−59%, p = 0.0143), as well as concentrations of glutamate precursors: glutamine (Gln) (−16% p = 0.0143) and N-acetylaspartylglutamate (NAAG) (−24%, p = 0.0283). A significant decrease in aspartate (Asp) (−36%, p = 0.0283), another excitatory amino-acid, was also observed in the oxaliplatin-treated rats fed with a PDD compared to a normal diet, as well as a decrease of its precursor N-acetylaspartate (NAA) (−24%, p = 0.0143). ## The PDD Induces Other Metabolic Variations in the Spinal Cord At day 3, in control rats fed with a PDD, there was a significant decrease in PtC (−25%, p = 0.0143), GPC (−22%, p = 0.0143), AXP (−22%, p = 0.0143), alanine (Ala) (−19%, p = 0.0283), lactate (Lac) (−15%, p = 0.0209) and myo-inositol (MyI) (−15%, p = 0.0275) concentrations compared to control rats fed with a normal diet. Inversely Ace levels increased (+25%, p = 0.0339) in response to PDD compared to control rats fed with a normal diet. Similarly, PLS-DA evidenced a marked separation of control rats according to diet (R2Y = 0.646, Q2 = 0.547, validated model) and the most discriminant metabolites explaining this separation were AXP, beta-hydroxybutyrate (BHB), Lac, PtC and tCr. At the same time, in oxaliplatin-treated rats fed with a PDD, the concentrations of succinate (Succ) (−59%, p = 0.0472) and choline-derivatives metabolites phosphocholine (PC) (−13%, p = 0.0472), PtC (−19%, p = 0.0163) and GPC (−19%, p = 0.0163) decreased significantly compared to oxaliplatin-treated rats fed with a normal diet. PLS-DA also showed a clear separation between the diets for oxaliplatin-treated rats (R₂Y = 0.649, Q₂ = 0.504, validated model), indicating that the metabolic phenotype of these two groups is significantly distinct. This separation was attributed to Ala, Gln, Lac, MyI and NAA. ## Ifenprodil, a NR2B-selective Antagonist, Relieves Oxaliplatin-induced Mechanical and Cold Hypersensitivity As a PDD reduced the oxaliplatin-induced increase in spinal Glu concentration, we suspected that a direct inhibition of the spinal glutamate/NR2B-containing NMDA receptors pathway would be effective to reduce oxaliplatin-induced hypersensitivity. Thus we assessed the effect of an intrathecal injection of NR2B-specific antagonist, ifenprodil, on oxaliplatin-induced mechanical and cold hypersensitivity. In the electronic von Frey test, while 2.5 µg ifenprodil/rat was inactive, 5 µg/rat induced a significant increase in paw withdrawal thresholds from 4 h to 8 h compared to controls (+16%, p = 0.006 and +21%, p = 0.003, respectively). Rats treated with 10 µg ifenprodil/rat showed a significant increase in paw withdrawal thresholds at 2 h (+12%, p = 0.005), 4 h (+31%, p\<0.001) and 8 h (+43%, p = 0.004) compared to saline-treated rats and oxaliplatin-induced mechanical hypersensitivity was totally reversed, at 4 h and 8 h (mechanical thresholds were not significantly different from pre-oxaliplatin values, p = 0.90 and p = 0.24, respectively). At 24 h post-ifenprodil administration, paw withdrawal thresholds were not significantly different from the saline group regardless of the dose. Similarly, in the thermal place preference test, the first active dose of ifenprodil was 5 µg/rat with a significant increase (+140%, p = 0.02) in the time spent on the cold plate, at 4 h, compared to control rats. Rats treated with 10 µg ifenprodil/rat showed a significant increase in the time spent on the cold plate at 4 h (+120%, p = 0.001) and 8 h (+114%, p = 0.01) compared to control rats. Moreover, the time spent on the cold plate was significantly higher than pre-oxaliplatin values in rats treated with 10 µg/rat at 4 h (+35%, p = 0.017) and 8 h (+28%, p = 0.045). At 24 h post-ifenprodil administration, time spent on the cold plate was not significantly different from the saline control group regardless of the dose. # Discussion Up to 96% of patients experience acute cold hypersensitivity after oxaliplatin administration. Even if oxaliplatin-based chemotherapy remains the reference treatment for advanced colorectal cancers, its neurotoxicity is often a cause for early chemotherapy discontinuation and greatly impairs patient quality of life. This study reports that a 7-day PDD was able to prevent oxaliplatin-induced acute cold and mechanical hypersensitivity in rats and brings novel findings on the molecular mechanisms enabling PDD to prevent oxaliplatin-induced neurotoxicity. The antinociceptive effect of the PDD might result from a decrease in spinal NMDA receptor activity as demonstrated by Rivat *et al.*. Increased spinal NR2B subunit expression and/or phosphorylation has been associated with the development of pain behavior in a wide range of animal pain models, including inflammation, traumatic neuropathy, toxic neuropathy and cancer pain. However, in the present study, there was no variation in NR2B protein expression or tyrosine and serine phosphorylation in response to oxaliplatin injection. These results are in accordance with Mihara *et al.* (2011), who did not find any variation in NR2B expression (gene and protein) in the rat spinal dorsal horn 5 days after oxaliplatin injections (4 mg/kg i.p., day 1 and 2). An increase of NR2B expression was only found following repeated administration of oxaliplatin over 25 days (4 mg/kg i.p. twice a week, cumulative dose: 32 mg/kg). On the other hand, even though we failed to show an increase in NR2B activity, we demonstrated a hyperactivity of the glutamate pathway in the spinal dorsal horn following an acute injection of oxaliplatin. This hyperactivity, illustrated by the increased concentration of glutamate in the spinal dorsal horn, was reversed by the PDD simultaneously with the induced hyperalgesia, which suggests a causal link between the two changes. Indeed, glutamate is long known to be a major neurotransmitter of sensory input between primary afferent fibers and nociceptive ascending pathways. Furthermore, numerous data from literature establish a link between increase in spinal glutamate levels and pain hypersensitivity in chronic pain models (e.g. an increase of spinal glutamate was shown in a rat model of trauma-induced painful neuropathy. Moreover, antagonists of NMDA receptors are effective in patients with neuropathic pain. The metabolic analysis performed here allows the evaluation of the biochemical impact of PDD on glutamate biosynthesis. Glutamate is synthetized in neurons from 2-oxoglutarate originating from the tricarboxylic acid cycle. In oxaliplatin-treated rats fed with a PDD, the spinal level of succinate (a tricarboxylic acid cycle intermediate) is significantly lower than in oxaliplatin-treated rats fed with a normal diet, suggesting that the absence of increased glutamate concentration in oxaliplatin-treated rats fed with a PDD could result from a decreased activity of the tricarboxylic acid cycle in spinal neurons. This hypothesis is supported by the significant decrease in adenosine phosphate concentration in control rats fed with a PDD. Interestingly, significant decreases in glutamine and NAAG (respectively the astrocyte and neuron storage form of glutamate) were also observed, suggesting that the PDD may interact with several metabolic pathways to prevent oxaliplatin-induced increase in spinal glutamate biosynthesis. Another potential precursor of glutamate is ornithine (through the concerted action of ornithine aminotransferase and glutamate-5-semialdehyde dehydrogenase or the 1-pyrroline-5-carboxylate dehydrogenase), which is also a precursor of polyamines. It is possible that the polyamine deficiency induces an increased *de novo* synthesis of polyamines (through activation of ODC which is dependent upon polyamine uptake), thus depleting ornithine from neurons and preventing glutamate synthesis from this metabolite. The decreased lactate concentration observed in rats fed with a PDD (control and oxaliplatin-treated) could indicate an alteration of aerobic glycolysis in astrocytes, which is directly dependent on glutamate uptake. Finally, we observed a significant decrease of glycerophosphocholine and phosphatidylcholine in rats fed with a PDD (regardless of the treatment received) compared to rats fed with a normal diet (regardless of the treatment received). Neurodegenerative diseases are commonly associated with an increase of these metabolites in the nervous system. Moreover, polyamines have been shown to interact with phospholipids and to stabilize membrane formation. We could suspect that the PDD might play a role of neuroprotection which, in combination with the decrease in glutamate levels, could participate to the prevention of acute pain hypersensitivity induced by oxaliplatin. Intrathecal injection of ifenprodil induced an antihyperalgesic effect in conditions of oxaliplatin-induced mechanical and cold hypersensitivity. This result, with recent previous ones obtained with ifenprodil (i.e. an antihyperalgesia in a rat model of chronic dorsal root ganglia compression or in oxaliplatin-induced mechanical allodynia confirms the involvement of an increased, NR2B-dependent, activity of the spinal glutamate system in the pathophysiology of oxaliplatin-induced neuropathic pain symptoms (cold and mechanical hypersensitivity). Together with the metabolic and antihyperalgesic effects of the PDD, these results show that oxaliplatin-induced hypersensitivity can be reduced by inhibiting the glutamate pathway either upstream through a decrease in glutamate level (PDD effect) or downstream by an antagonism of the glutamate target (ifenprodil action). Whilst a PDD is able to prevent the acute oxaliplatin induced cold and mechanical hypersensitivity, we could also put forward the hypothesis that a PDD could also prevent oxaliplatin-induced chronic neuropathy, associated with repeated oxaliplatin administrations. Considering that the thermal hyperalgesia is a predictive marker of the chronic neuropathy, this PDD could ultimately provide neuroprotection against oxaliplatin-induced chronic neuropathy. However, the hypothetic preventive effect of a PDD on the oxaliplatin-induced chronic neuropathy must be assessed in the same experimental conditions in an animal model reproducing the symptomatology of this chronic neuropathy such as described by Ling et al.. In this chronic neuropathy, NR2B antagonists (ifenprodil and Ro25-6981) decreased mechanical allodynia. Moreover, excessive release of glutamate (which could be induced by repeated cycles of oxaliplatin) in CNS is known to be responsible for the excitotoxicity associated with neuronal damage (e.g. in spinal inhibitory neurons) and death. The pharmacological inhibition of the glutamate carboxypeptidase, which hydrolyses N-acetylaspartylglutamate (NAAG) into N-acetylaspartate (NAA) and glutamate, decreased glutamate level and improved nerve conduction velocity as well as morphological alterations in dorsal root ganglia of animal models of cisplatin- and bortezomib-induced neuropathy. To our knowledge, this is the first study to report an increased glutamate concentration in the spinal dorsal horn of rats after one injection of oxaliplatin and to report that a PDD prevented both this increase and the mechanical and thermal hypersensitivity. Managing oxaliplatin-induced neurotoxicity is of crucial importance in oncology, as this toxicity is often a cause of dose reduction or treatment discontinuation in cancer patients. Thus, reducing polyamine dietary intake for these patients could represent a safe and effective strategy to prevent the acute sensory disturbances associated with oxaliplatin infusions. A clinical trial (NCT01775449) is currently recruiting patients to assess this new hypothesis. The authors thank Dr. Mounir Traïkia (CNRS U6504, Université Blaise Pascal, Clermont-Ferrand, France) for helpful advice on proton NMR analysis, and René Havouis and Eric Chapuy for their technical assistance. [^1]: JPM is a co-founder of Nutrialys, a R&D company commercializing low polyamine content nutritional supplements. The authors declare a patent on the concept on polyamine deficient diet to prevent oxaliplatin-induced neuropathy (FR 2967868 (A1) - Utilisation d’une composition alimentaire dans le traitement et/ou la prévention des douleurs neuropathiques induites par un agent anticancereux). There are no products in development or marketed products to declare. This does not alter the authors’ adherence to all the PLOS ONE policies on sharing data and materials, as detailed online in the guide for authors. [^2]: Conceived and designed the experiments: JF MBR LD JPM DP DB. Performed the experiments: JF MBR DB. Analyzed the data: JF MBR BP. Contributed reagents/materials/analysis tools: JF MBR JPM DP LD. Wrote the paper: JF MBR JPM BP AE DB.

# Introduction Two month culture conversion has been used as a measure of efficacy in the assessment of drugs used to treat TB in phase 2 trials. The timing of culture conversion is influenced by bacillary quantitation on smear and extent of disease on chest radiograph (CXR), particularly cavitation. More recently, lower rates of culture conversion have been associated with diabetes, lack of directly observed therapy, HIV infection and smoking. The TB Trials Consortium (TBTC) recently completed a double-blind, randomized, placebo-controlled, phase 2 trial examining the substitution of moxifloxacin for isoniazid (INH) during the first 2 months (intensive phase) of therapy of drug- susceptible, AFB smear-positive pulmonary tuberculosis (TB). The study was conducted at urban sites in Brazil, Canada, South Africa, Spain, Uganda and the United States. In this study, the proportion of patients with negative cultures at 8 weeks of therapy did not differ significantly between patients in the INH and moxifloxacin arms. In multivariate analysis, positive sputum culture at 8-weeks was associated with enrollment in Africa, cavitation on baseline CXR, higher bacillary load on baseline smear and increasing age, but not HIV co- infection. A difference in the rate of culture conversion between patients at African and non-African sites was also observed in TBTC Study 27, a phase 2 clinical trial in which moxifloxacin was substituted for ethambutol during the first 2 months of therapy of drug-susceptible smear positive TB. In both TBTC Studies 27 and 28, the definition of sputum culture conversion was based upon the combined results of both liquid and solid media such that a positive result on either media type was considered positive. Since liquid media are more sensitive than solid media in detecting the growth of *M. tuberculosis*, culture status in Studies 27 and 28 was most influenced by the liquid media results. We explored potential contributions of the following factors to explain lower culture conversion rates in Africa: 1) baseline severity of disease, 2) microbiological procedures, and 3) pharmacokinetic characteristics (published elsewhere). # Methods ## Epidemiologic Investigation This analysis included 328 protocol-correct participants in Study 28. Patients provided spot sputum specimens for AFB smear microscopy and culture, on both solid and liquid media, at baseline and weeks 2, 4, 6, and 8 of study therapy. Thereafter spot sputa were obtained monthly until 2 consecutive specimens were culture-negative. Mycobacterial isolates from protocol-correct participants were confirmed to be *M. tuberculosis* complex and susceptible to isoniazid, fluoroquinolones, rifampin, and pyrazinamide. We defined culture conversion as the first negative sputum culture with at least one subsequent negative culture and no subsequent positive results. We examined the probability of sputum culture conversion in liquid media using smoothed Kaplan-Meier plots, calculated at days 15, 29, 43, 57 after the start of therapy corresponding to weeks 2, 4, 6, 8, respectively. We investigated four stratified measures of disease extent at enrollment for African v. non-African patients. These stratified measures included: 1) Smear status normalized to the WHO standard (0–1+, 2+, 3+), 2) Radiographic extent of pulmonary involvement (\<25%, 25%–50%, \>50%) on baseline CXR, 3) aggregate cavity size on CXR (absent, \<4 cm, ≥4 cm), and 4) the number of days-to-detection (DTD) of growth in liquid media for *M. tuberculosis* positive specimens (0–5 days, 6–9 days, 10+ days). The Wilcoxon test for paired samples was used to determine whether culture conversion curves were significantly different between African and non- African patients with equivalent measures of disease severity. Information about smoking, diabetes, and self-designated race was obtained from participants through interview by study staff. HIV testing was performed on all study participants. This work is a secondary analysis of data from TBTC Study 28 which was registered on ClinicalTrails.gov with the identifier NCT00144417. The protocol for this trial and supporting CONSORT checklist are available as supporting information; see and. The protocol for TBTC Study 28 was approved by the CDC Human Research Protection Office and the 50 institutional review boards of the 27 sites enrolling patients. Patients enrolled in TBTC Study 28 provided written informed consent to participate in the study. ## Laboratory Methods and Investigation Each enrolling site submitted sputum specimens to a local microbiology laboratory that participates in a quality assurance testing program for both smear and culture. Twenty-nine laboratories processed and cultured sputum for *M. tuberculosis* from TBTC Study 28 patients. Sputum smears were prepared from concentrated, decontaminated specimens and read using either the ATS/CDC or WHO smear grading systems and then harmonized to the WHO grading system for analysis. We reviewed laboratory practices and processes at these laboratories including media type used, decontamination procedures and inoculation volume. All laboratories used N-acetyl-L-cysteine (NALC)/sodium hydroxide (NaOH) for specimen decontamination. The TBTC site in Uganda enrolled 55% of protocol-correct Study 28 patients of which many had delayed culture conversion so further investigation of the *M. tuberculosis* isolates from Uganda was undertaken. As a routine practice at this laboratory, all positive liquid culture tubes were subcultured on blood agar plates and stained using the Ziehl-Neelson method. Nucleic acid amplification of IS6110 was performed routinely on all AFB-positive culture tubes at enrollment to assess for *M. tuberculosis* complex. To assess for cross-contamination in the Ugandan laboratory and for potential re-infection with a different strain of *M. tuberculosis* a convenience sample of 21 pairs of baseline and 8-week isolates from Ugandan patients underwent spoligotyping, MIRU and/or RFLP analysis at CDC. # Results Of the 328 protocol-correct patients in TBTC Study 28, 213 (65%) were enrolled in Africa, including 182 (55%) from Uganda and 31 (9%) from South Africa. Among the remainder, 19 (6%) were enrolled in Brazil, 12 (4%) in Spain, and 84 (26%) in North America. After 8 weeks of study therapy, sputa cultured in liquid media from African site patients were 29% less likely to have converted to negative than non-African patients (p\<0.001; Wilcoxon test comparing the curves). Based on cultures on solid media, African site patients were 10% more likely to have converted than from non-African patients (p = 0.6; Wilcoxon test comparing the curves). In Africa there was a 45% difference in the rate of conversion between solid and liquid media while there was only a 9% difference in non-African patients. ## Baseline Extent of Disease At baseline African patients had a higher bacillary load based on sputum smear and a smaller number of days required for detection of *M. tuberculosis* growth in liquid media, compared to non-African patients. Compared to non-African patients, a higher proportion of African patients had cavitary disease, and African patients had larger cavities and more lung fields involved by tuberculous lesions. To evaluate whether more severe disease in African patients explained the lower culture conversion rates in liquid media in African patients, we evaluated the probability of culture conversion by week of therapy for African and non-African patients stratified by severity of disease for baseline smear status, cavitary size, radiographic extent of disease and the number of days to detect growth in liquid media. Within each region (i.e., African, non-Africa) for each of these measures of extent of disease at enrollment, the relationship between severity and time to conversion was consistent and generally as expected. Patients with the most severe measures of disease had slower sputum culture conversion in liquid media compared to patients with moderate disease, who had slower conversion than patients with the least severe measures of disease. However, for all four baseline measures of disease examined, when compared at equivalent levels of severity, African patients had substantially lower conversion rates in liquid media than non-African patients. In fact, African patients with the least severe measures of disease consistently had conversion rates similar to or lower than non-African patients with the most severe measure of disease. The curves for all 12 comparisons of the time to sputum culture conversion for equivalent measures of disease severity shown in, were significantly different between African and non-African patients (See). Compared with patients from non-African sites, those from African sites had lower prevalences of smoking (63% versus 27%, p\>0.001) and of prior diagnosis of diabetes mellitus (17% versus 0%, p\>0.001). African patients were significantly younger than non-African patients (29.4 years vs. 40.5 years, p\<0.001). Of the 35 HIV-infected, protocol-correct patients, 32 were from African sites. Among African patients, HIV-positive patients had slightly higher conversion rates in liquid media compared to HIV-negative patients. Culture- conversion rates by race (black and non-black) were examined to determine if non-African patients of black race had lower conversion rates. At non-African sites the time to conversion in liquid media did not differ substantially between black (n = 26) and non-black patients (n = 80), but both were significantly lower than that in African patients (p\<0.001, log-rank test). Therefore, smoking, diabetes, age, HIV status, and race did not explain lower sputum culture conversion rates in Africa. ## Laboratory Investigation Among the 29 laboratories processing mycobacterial cultures for TBTC Study 28: 1) the final concentration of NaOH used for decontamination ranged from 1% to 3% (mean 1.6%, median 1.25%), Most of the non-African laboratories decontaminated with a final NaOH concentration of 2% while the two African laboratories processed specimens using a final NaOH concentration of 1% and 1.5%, 2) the median duration of decontamination was 15 minutes (range 15–20 min), 3) inoculation volume was 0.5 ml for liquid media for all but one site (See). To assess whether differences in the final concentration of NaOH used in specimen decontamination might explain delayed conversion among African patients we examined the probability of culture conversion stratified by NaOH final concentration used in decontamination and geographic region of enrollment. We found that through week 6 of TB treatment non-African patients whose sputum specimens were decontaminated with NaOH concentrations \<2.0% (range 1.0% –1.5%) had almost identical times to conversion as non-African patients with specimens decontaminated with NaOH concentrations ≥2.0% (range 2.0%–3.0%). However, at week 8 it was surprising to find that non-African patients whose sputum specimens were decontaminated with NaOH concentrations \<2.0% had a higher rate of conversion than non-African patients with specimens decontaminated with NaOH concentrations ≥2.0% suggesting that the NaOH concentration does not explain the observed delayed culture conversion for patients at African sites. All isolates from Ugandan Study 28 protocol-correct patients were confirmed at enrollment and week-8 to be *M. tuberculosis* complex. All of the baseline and week 8 isolate pairs from 21 patients matched genotypically using RFLP, MIRU, and spoligotyping methods, suggesting that cross-contamination or re-infection did not explain the low culture negative conversion rate in liquid media among African patients. # Discussion Interest in geographic differences in response to TB chemotherapy dates back to 1956 when Fox et al. compared clinical, radiographic, and microbiologic outcomes in patients from Britain and Uganda. In TBTC study 28 during the first 8 weeks of therapy we found later sputum culture conversion and lower rates of conversion in liquid media in African patients compared to non-African patients, despite African patients having comparatively higher conversion rates on solid media. African patients had more severe disease when their therapy was begun. However when we evaluated conversion rates stratified by varying levels of disease severity it became clear that our measures of baseline severity of disease did not explain the lower and delayed culture conversion in liquid media in African patients. In fact, African patients with the lowest severity of disease had conversion rates in liquid media that were similar to or lower than non-African patients with the highest severity of disease. Some studies suggest that smoking, diabetes, increasing age and HIV positive status are associated with delayed conversion. In Study 28, African patients had a lower prevalence of smoking and diabetes and were younger compared to non- African patients, so these conditions did not appear to contribute to the lower conversion rates in Africa. Patients with HIV infection actually had slightly higher conversion rates in liquid media compared to HIV-negative patients. In Study 28 the administration of all doses of anti-TB drugs were directly observed. In an intensive pharmacokinetic sub-study performed on a convenience sample of 72 patients enrolled in TBTC Studies 27 and 28 (37 African and 35 non- African), the mean AUC_0-24 for rifampin and moxifloxacin did not differ between African and non-African patients. Isoniazid levels were not examined. We did not find a pharmacokinetic reason for lower conversion rates among Africans. A panel of experienced mycobacteriologists, clinicians, and epidemiologists reviewed laboratory processes in the context of the site-specific conversion rates and the number of patients represented. They determined that the laboratories were operating within accepted guidelines for conducting mycobacterial cultures. We found some variation in laboratory processes (e.g., decontamination procedures) that might have led to differences in the rates of culture conversion. In particular, the concentration of NaOH used during specimen decontamination was generally lower at African sites (albeit with longer time for decontamination). Laboratory studies done on sputum seeded with bacteria and *M. tuberculosis* H37Ra revealed that decontamination with NALC- NaOH with final concentrations of 1–2% NaOH and decontamination treatment times of up to 30 minutes did not affect the viability of *M. tuberculosis* when grown on solid media. However a study of patient sputum specimens collected prior to initiation of TB therapy suggested that relatively small increases in the final NaOH concentration used in decontamination from 1% to 1.25% significantly decreased the recovery of *M. tuberculosis* on a solid media. When we evaluated the effect of the NaOH concentration on liquid culture results after the initiation of TB therapy, stratified by geographic region, we found that the NaOH concentration did not explain delayed conversion in African patients compared to those at non-African sites. At baseline, prior to the initiation of therapy, solid and liquid media performed equally well for isolating *M. tuberculosis* from the sputa of the smear-positive patients enrolled in Study 28. As therapy progressed, at both African and non-African sites, patients had lower conversion rates in liquid media compared to solid media. This is expected, as liquid media allow for growth of some *M. tuberculosis* organisms that are unable to grow on solid media. Unexpectedly, the difference in conversion rates between solid and liquid media was much greater in Africa. This finding of large differences in the yield of solid and liquid media for *M. tuberculosis* for African patients receiving therapy is not unique to TBTC Study 28. A similar clinical trial conducted in South Africa by the OFLOTUB group found that at 8 weeks of TB therapy the proportion of sputum cultures that were negative on solid media (7H11) was twice that found on liquid media (MGIT). Joloba et. al. studied sputum culture conversion on both solid (7H10 selective) and liquid media (BACTEC 460) during the course of TB therapy for both HIV infected and uninfected populations. They also found that conversion in liquid media was lower and substantially delayed compared to that of solid media for both populations. Regional differences in outcomes of TB therapy are not unique to TBTC studies. In a recent phase 3 trial conducted in Brazil, the Philippines and Uganda examining treatment shortening for HIV negative adults with non-cavitary TB and negative 2-month sputum culture on solid media, patients from Uganda were more likely to relapse than participants from Brazil or the Philippines. This study was stopped early because patients in the 4-month treatment arm had significantly more relapse (13 relapses among 196 patients) than those in the 6-month treatment arm (3 of 198). Among the 16 relapses, 12 (75%) occurred in the Ugandan patients and 4 (25%) in Brazilian patients. Initial sputum smear grade and the number of lung zones involved by tuberculous lesions were greater among patients enrolled in Uganda. In the multivariate analysis, enrollment at the Uganda site was an independent risk factor for relapse after accounting for baseline sputum smear and radiographic extent of disease. Two-month sputum culture results in liquid media were not reported in that study. Recent TB trials illustrate important advantages and disadvantages of different approaches to phase 2 clinical trials. Multicenter international studies are likely to enroll more rapidly and to be more representative and easier to generalize, they also allow examination for regional differences in response to therapy. However, the potential influence that variability in laboratory or other procedures might have on the interpretation of clinical trial results must be considered. While laboratories engaged in clinical trials should meet recognized clinical diagnostic standards for culturing mycobacteria, the degree of heterogeneity of procedures allowed to meet these routine clinical diagnostic standards may not be sufficient for early phase controlled clinical trials. Further research is needed to determine whether and how modest differences in processing and culturing of sputa influence microbiological outcomes over the course of TB therapy in phase 2 studies. Such studies should focus on determining the principle processes that contribute to variability in microbiological outcomes measured during clinical trials. Laboratory processes that diminish the sensitivity or specificity of microbial biomarkers of treatment outcomes (e.g., 8-week culture conversion) will influence the ability of both single site and multi-site studies to determine the relative efficacy of drug regimens. Efforts to optimize the performance characteristics of microbiological tests in predicting treatment failure and relapse are needed if we hope to make good decisions regarding which drugs and regimens to move forward into phase 3 clinical trials. If, after investigation, variations in laboratory practices are not found to explain geographic difference in time to culture conversion, then alternative explanations (e.g., differences in socioeconomic/nutritional status, host immunity, or microbial pathogenesis) should be explored. # Supporting Information Andy Vernon, Michael Iademarco, Karen Morgan, Richard Chaisson, Moses Joloba, Erin Bliven, Andrey Borisov, Elsa Villarino, Bill Burman, Debra Benator, Robert Belknap, Carol D. Hamilton, Kathleen Robergeau, Salman Siddiqi, Susan Kayes, Pei-Jean Feng, Alfred Etowah. [^1]: Data Interpretation: WRM LB JLJ DD KCJ BM KE SD SG. Critical Revision of Manuscript: SD LB JLJ SG GM DD KCJ LD BM KE. Conceived and designed the experiments: ARM CMH SG. Performed the experiments: LB JLJ GM DD KCJ LD BM KE. Analyzed the data: CMH. Wrote the paper: WRM CMH. [^2]: The authors have declared that no competing interests exist.

# Introduction Colorectal cancer (CRC) is a major cause of morbidity and mortality worldwide and is the second most common cause of cancer death in Europe and the United States. The overall five-year survival rate is less than 65% and when the cancer is already at an advanced stage when diagnosed treatment has very limited success. Currently, 25% of patients have metastasis at the time of diagnosis. Detection of cancer at early stages is critical for curative treatment interventions. However, the invasive, unpleasant and inconvenient nature of current diagnostic procedures limits their applicability. Noninvasive molecular approaches would enable detection of relevant lesions at the population level and could be enhanced by widely distributable screening tools that are accurate, user-friendly, safe and affordable. Nonconventional methods could potentially meet all of the desired criteria of an ideal screening tool, so there is good reason to pursue rationally designed alternative approaches. In particular, molecular tests of blood could potentially provide this ideal screening tool. However, the use of single or a combination of serum markers, including carcinoembryonic antigen (CEA), has so far failed to deliver diagnostic tests of high sensitivity and specificity for CRC. There is reasonable hope and further evidence emerging that the presence of CRC could be detected by specific changes in the composition of serum proteins or mRNA expression levels. Different markers are overexpressed in CRC compared with matched normal adjacent mucosa. These observations suggest a model wherein as a key regulator of epithelial cell proliferation along the associated with neoplastic transformation. The targeting of RNA makes use of the fact that tumour phenotypic variation is associated with changes in the mRNA levels of genes regulating or effecting this variation. This has led to the widespread use of real-time reverse transcription polymerase chain reaction (RT-PCR) assays for the study of tumour dissemination as well as many reports describing its use in clinical diagnostics. In addition, the levels of mRNA expression in blood for various proteins, in particular cyclooxygenase 2 (COX-2), vascular endothelial growth factor (VEGF) and others including metalloproteinase, have been proposed for CRC diagnosis and prognosis. Various different serum markers such as the complement protein C3a and alpha 1-antitrypsin have been judged to be useful in the diagnosis of CRC. Some of them, such as VEGF or urokinase-type plasminogen activator receptor (uPAR) have been studied in other intestinal diseases such as inflammatory bowel disease and have not found elevated. Other markers, such as tetranectin were found elevated in other gastrointestinal and gynecological tumors. Our objective in this study was to assess the usefulness of different serum and mRNA expression markers for the diagnosis of CRC. # Materials and Methods Sixteen serum markers were studied prospectively in 42 patients with CRC and 33 healthy controls. Individuals with CRC who had inflammatory bowel disease and polyps were excluded. Participants with CRC were staged according to the 5^th edition of the TNM classification. The controls were individuals with no signs or symptoms of chronic disease and who had undergone a colonoscopy to the caecum with normal results. These controls were not taking drugs or NSAIDs. For all participants the following data was collected: age, weight, height, body mass index (BMI), associated medical conditions and medications taken. The serum markers measured were alpha 1-antitrypsin, complement protein C3a and CEA. The markers in the blood analysed by RNA extraction, cDNA synthesis and quantitative PCR (qPCR) were carbonic anhydrase, guanylyl cyclase C, plasminogen activator inhibitor (PAI), matrix metalloproteinase 7 (MMP7), uPAR, urokinase- type plasminogen activator (uPA), survivin, tetranectin, VEGF, cytokeratin 20, thymidylate synthase, COX-2 and CD44. These last markers were determined in tumoral tissue and healthy tissue obtained after surgical resection of CRC. All markers were selected based on their relationship to diagnosis or prognosis in CRC after a review of the literature. For each participant, two fasting blood samples were taken from the antecubital vein. These blood tests were carried out in the 24 hours prior to surgery in the case of the patients diagnosed with CRC and 24 hours following the colonoscopy in the controls. One of these samples was used to determine three proteins in serum (alpha 1-antitrypsin, activated C3 and CEA levels), while mRNA was extracted from the other. Blood samples from all participants were collected in PAXgene RNA blood tubes (QIAGEN, K2E 18.0 mg – BD, Plymouth, UK) and kept frozen at −80°C until RNA extractions were carried out. ## Measurement of Alpha 1-antitrypsin, Activated C3 and CEA Levels The level of CEA in the serum was determined using an electrochemiluminescence immunoassay. Pre-operative levels of serum CEA of ≤5.0 ng/mL were considered normal. Measurements of the serum concentration of complement C3a were performed using the OptEIA Human C3a ELISA kit (BD Biosciences Pharmingen, San Diego, CA, USA). Normal levels for serum complement C3a was taken to be between 10 mg/dl and 40 mg/dl. Alpha 1-antitrypsin (g/l) concentrations were determined using an enzyme immunoassay kit (Dade Behring, Newark, NJ, USA) on a BN II ProSpec nephelometer. A cut-off level of 1.5 g/L was chosen as the upper limit of normal. ## RNA Extraction and cDNA Synthesis Total RNA from 2.5 ml whole blood collected in the PAXgene blood RNA tubes was extracted using PAXgene Blood Minikit (QIAGEN, PreAnalytix cat: 762164) according to manufacturer’s instructions. Each RNA sample was quantified and quality assessed using an Agilent 2100 bioanalyzer system to obtain an RIN value. The RIN algorithm allows RNA integrity to be calculated using a trained artificial neural network based on the determination of the most informative features that can be extracted from the electrophoretic traces out of 100 features identified through signal analysis {IMBEAUD, 2005 \#42}. Total RNA from 20–30 mg of tissue (tumoral and healthy) was extracted using the RNeasy Plus Mini Kit (QIAGEN) following manufactureŕs instructions. This kit includes a DNase step which removes contaminant genomic material from RNA extractions. Using µg of RNA. cDNA was synthesized by reverse transcription using a Superscript III kit (Invitrogen Ltd., Paisley, UK) following manufacturer’s instructions in a total volume of 20 µl. Subsequently cDNA was prepared from 1 µg of RNA using the Superscript III RT PCR kit (Invitrogen). ## Quantitative PCR (qPCR) The qPCR was performed using a Bio-Rad iQ5 system with a total of 13 CRC potential markers. Each qPCR reaction contained a gene-specific oligonucleotide primer set (25 uM), 12.5 µl Platinum qPCR supermix (Invitrogen) and 1/20 of the original cDNA reaction mix. Each reaction was carried out in duplicate, together with a negative control (H₂O as template). PCR efficiency was assessed by performing triplicates of 5-fold serial dilutions from a pool of cDNA derived from 12 healthy individuals to obtain an R² value of \>0.98. To compensate for RNA degradation and variability in the starting amounts of RNA, four housekeeping genes, chosen with the GenEx Light Software (TATAA Biocenter, Sweden) from a panel of 13 genes, were used as endogenous controls (GAPDH, HPRT1, PPIA and TBP). The amplification values obtained from each of the 13 markers used in this analysis were divided by the corresponding amplification values for each of the housekeeping genes to produce an expression index. Of these, GAPDH produced the most stable results and was used in further analysis. To determine changes in levels of expression of all the 13 markers in relation to the housekeeping genes we used the comparative ΔC_T method where the copy number of the marker gene (target) in the samples from patients with CRC (tumour) is compared to the relative amount of the target gene in normal cDNA from either its corresponding paired-sample or pooled blood samples and relative to expression of the housekeeping gene, which was used as an endogenous control (reference). All CRC markers used in this study and their corresponding oligonucleotide primer sequences are listed in. The committee for clinical research ethics committee of Country Basque approved this study and all participants gave written informed consent. ## Statistical Analysis The comparison of the parameters analysed between CRC cases and controls were carried out using the Student’s t-test. In addition, ROC (Receiving Operating Characteristic) curve analysis, with calculation of both the area under the curve and the corresponding 95% confidence intervals, was used to assess the accuracy with which the parameters diagnosed CRC, that is, whether they could be used to discriminate between patients with CRC and controls. To evaluate the differences in some markers according to the stage has been applied univariate generalized linear model which has assessed linear trend by polynomial contrasts and multiple comparisons (post-hoc) pairwise Tukey’s test. Analyses were performed by using the SPSS statistical software, version 19 and MedCalc version 6. # Results ## Subjects Of the 42 patients with CRC 25 (59%) were men and 17 (41%) women. Their mean age was 72.33 years old and their BMI was 26.57. The tumours were classed as Stage I in 8 patients (20%), Stage II in 14 (33%), Stage III in 17 (40%) and Stage IV in 3 patients (7%). Of the 33 controls, 20 (60%) were men and 13 (40%) were women, their mean age being 75 years old and BMI 26. There were no significant differences between the cases and controls in terms of age, sex or BMI. When serum data were analysed, there were significant differences between cases and controls in the levels of alpha 1-antitrypsin. The alpha 1-antitrypsin levels were 1.79±0.25 in the patients with CRC compared to 1.27±0.4 in the controls (p\<0.0005). CEA in serum was 1.5±1 in controls and 8.5±15 in cases (p = 0.01). However, differences in complement protein C3a concentrations between cases and controls were not significant (14.56±4.08 vs 12.89±6.5; p = 0.28). mRNA expression levels of survivin, MMP7, uPA, COX-2 and CD44 were increased statistically significant in tissue with CCR versus normal tissue whereas carbonic anhydrase, tetranectin, and cytokeratin 20 were significantly decreased. The mRNA expression levels in blood are reported in. Among these, there were five markers in which the differences between cases and controls were significant, namely, carbonic anhydrase, MMP7, uPAR, tetranectin, and COX-2. There were no significant differences in any of the 16 markers studied according to sex. In the case of the two serum markers that showed significant differences, the areas under the ROC curve were 0.88 (0.7–0.9) for alpha 1-antitrypsin and 0.84 (0.6–0.9) for CEA. For the mRNA expression, the areas under the ROC curve for the different markers are listed in. The highest diagnostic accuracies were found for MMP7 (0.81) and tetranectin (0.80), COX-2 (0.78), uPAR (0.78) and carbonic anhydrase (0.77). For screening, a context in which the priority is sensitivity and minimisation of false negatives, a suitable cut-off was 1.43 for alpha 1 antitrypsin, giving a sensitivity of 87% and a specificity of 73%. A value for MMP7 of 15.4 was the cut-off that minimised the percentage of false negatives, giving a sensitivity of 58% and specificity of 100%. In compares the area under the ROC curve between serum CEA and mRNA expression of MMP7. When the values of the markers were compared between controls and patients with Stage I or II CRC, it was observed that alpha 1-antitrypsin, uPAR, COX-2 and MMP7 levels were significantly different, indicating that these compounds could be useful for the early detection of CRC. By contrast, differences between groups in CEA and tetranectin were only significant for advanced tumours (Stages III and IV), indicating that are not suitable for early diagnosis of CRC. # Discussion CRC has several the characteristics to justify population-wide screening: high prevalence, the occurrence of precancerous lesions, and the existence of treatment that is effective when applied in the early stages. The heterogeneity of colorectal neoplasms must be taken into account in selecting markers for molecular detection and determining the frequency of screening tests. Detection of these markers in blood depends on a cascade of biological events. Markers must be regularly released from tumours or precancerous lesions, dispersed into the medium that is collected for the assay, survive metabolic degradation, and be measurable. In our study we analyzed 16 potential blood markers and watched as three may be useful in identifying early-stage CRC (uPAR, MMP7, COX-2 and alfa 1 antitrypsin). The combination of them could be useful for screening of CRC. Two of the 3 markers are mRNA expression determined by real-time or quantitative PCR (qPCR). This simply method is a variation of the standard PCR technique used to quantify mRNA in a sample. Using specific sequence primers, it is possible to determine the number of copies or relative quantity of given sequences of RNA. When real-time PCR is combined with reverse-transcription (sometimes called qRT-PCR), the quantity of mRNA in a sample can be ascertained for different proteins. With this approach, it has been observed, for instance, that CEA mRNA expression in blood can predict recurrence in gastric cancer, esophageal squamous cell carcinoma and CRC. Soluble tetranectin was measured preoperatively in serum from patients with primary colorectal cancer. Tetranectin is a recently described human plasma protein, which is found in most secretory cells throughout the body and has been found to be a strong prognostic factor in patients with CRC in others studies. Specifically, significant correlations were found between tetranectin and Dukes’ stages and other biochemical markers, such as PAI, uPAR and CEA. However, there are no previous reports of studies examining the tetranectin mRNA expression for CRC diagnosis. In our study, it was found that the expression of tetranectin mRNA had a good sensitivity and specificity (78% and 86%, respectively) to a cutoff of 12,8. There is considerable experimental evidence that uPAR is functionally involved in cancer invasion, consistent with its ability to concentrate and enhance uPAR activity on the cell surface. The level of the uPAR is elevated in tumour tissue from several forms of cancer. It is known that uPAR is shed from the cell surface and the soluble form uPAR has been detected in several body fluids. The preoperative plasma uPAR level independently was found to predict survival of patients with CRC. In that study, the plasma concentration of uPAR was measured by a kinetic ELISA method in EDTA-anticoagulated plasma and not by level of expression of mRNA in serum as in our study. In colon cancer uPAR mRNA expression is enhanced in tumour cells. In our study the expression on blood showed significant differences with controls y entre controles y pacientes con estadios I-II de CCR. Other studies have demonstrated how mRNA expression levels of angiogenic factors, such as VEGF, in liver metastasis can be predicted from those in primary colorectal cancer. Further, the measurement of thymidylate synthetase mRNA in primary colorectal cancer can reflect those in synchronous liver metastases. However, most studies have investigated the levels of mRNA expression in tissue but not in blood as we have. We did not observe differences in thymidylate synthetase and VEGF in tissue or blood of patients with CRC. Is possible that there were not differences due to the small number of patients with stage IV. Proteomic profiling of serum from patients with CRC and controls combined with the use of artificial neural networks was able to diagnose CRC with 94% sensitivity and 96% specificity in a cohort of patients. Using surface-enhanced laser desorption/ionisation (SELDI), Ward et al. in 2006 identified four proteins (alpha 1-antitrypsin, complement C3a, transferrin and apolipoprotein C1) potentially useful for the diagnosis of CRC. The proteins identified are common serum proteins and changes in their concentrations most likely reflect epiphenomena rather than secretion by cancer cells. In that study (24), CEA was measured in all of the samples (62 CRC patients and 31 subjects not diagnosed with cancer), and a using the manufacture’s recommended cut-off level of 4 ng/ml, the sensitivity and specificity obtained was 53% and 93%, respectively. By contrast, in our study, though we also studied the levels of transferrin we did not find significant differences between patients with CRC and controls, (225.4±53.69 vs 231.1±37.1; p = 0.60) with an area under the ROC curve of 0.47 (0.34 to 0.61). Levels of apolipoprotein C1, on the other hand, were not investigated. CEA is clinically used as marker of CRC progression, but alone is not useful as a marker diagnostic. In our study, the ROC values was higher to others studies, but was not useful in differentiating early stages and a control group. Unlike other studies, the combination with other markers did not improve the ROC curve. Elevated serum alpha 1-antitrypsin levels have been observed in association with malignancy and inflammation. Some authors, have shown that serum alpha 1-antitrypsin levels correlated with CRC and with clinical staging. In these studies, the correlation of CEA and alpha 1-antitrypsin to the stage of disease had almost the same statistical weight (p\<0.004 and p\<0.003)(26). The other three markers (C-reactive protein, alpha 1-acid glycoprotein and the percentage of serum protein electrophoretic components) in the series of Stamatiadis et al. had a less significant correlation to the spread of the disease. Nevertheless, their potential role in the diagnosis of CRC, especially in comparison with other mRNA expression markers had not been established until now. Interestingly, high levels of alpha 1-antitrypsin in gastric juice have been associated with gastric cancer. In our study we observed that serum levels of alpha 1-antitrypsin were increased compared to controls. Serum levels were increased by stage and discriminated between early and advanced stages. Complement C3a is highly biologically active, binding to mast cells and basophils and triggering the release of their vasoactive contents. The elevated level of complement C3a in the serum of CRC patients may reflect an immune response to the tumour, or possibly in vitro complement activation. Based on a specific ELISA, serum levels of complement C3a predicted the presence of CRC in a blinded validation set with a sensitivity of 96.8% and a specificity of 96.2%. Increased serum levels were also detected in 86.1% of independently collected sera from patients with colorectal adenomas, while only 5.6% were classified as normal. In our study it was not useful to discriminate cases and controls. The discrepancies observed in some markers in our study between tissue and blood may be due to different reasons, first, not all mRNA is overexpressed in tumor tissue enter in blood. Second, the mRNA extracted from whole blood comes from blood cells (leukocytes, platelets …), very different from mRNA extracted from CCR. Third, the blood carries RNases (enzymes that degrade RNA), so that RNA circulating in the blood is more likely to be degraded. This is more difficult to occur in the RNA extracted from tumor cells. Among the limitations of the study are; first, the small sample size; second, unknown if the results well be reproduced in asymptomatic individuals over 50 years (screening); finally, we also unknown whether the expression of these markers in patients with other gastrointestinal diseases such as inflammatory bowel disease, may be altered. Other studies are needed in other diseases and other gastrointestinal tumors to determine the value of these markers in these diseases. For example, tetranectin has not been studied in inflammatory bowel disease but has been elevated in other tumors such as gynecological, gastric, etc. In summary, we report that alpha 1-antitrypsin and uPAR, COX-2 and MMP7 mRNA expression in blood could be useful for the early diagnosis and/or screening of CRC. Further studies in other cohorts the patients and screening programs are required involving larger numbers of subjects to confirm these results. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Performed the experiments: A. Cosme EH JMEN CP EV MHV. Contributed reagents/materials/analysis tools: A. Cosme EH JMEN CP EV MHV. Wrote the paper: LB MG CS. Acquired, analyzed and interpreted the data: CS LB MG MHV EH MDG FB A. Castells. Critically revised the manuscript for important intellectual content: LB.

# Introduction The exchange of genetic material between homologous chromosomes via crossing over is a key and defining step of the meiotic cell cycle. Crossing over is essential for proper chromosome segregation, with too few or poorly positioned crossovers posing leading risk factors for aneuploidy and infertility. At the same time, crossing over (and, more broadly, recombination) is an important mechanism generating DNA diversity. Recombination controls the rate at which new haplotypes are created and influences their frequencies within populations. In spite of its significance for the maintenance of genome integrity and evolution, there is tremendous variation for recombination rate between species, among individuals, and within single genomes. Some of this variation is under genetic control. Classical genetic experiments have demonstrated a clear heritable component to population variation in crossover frequency. More recently, quantitative genetic, association analyses, and candidate gene-driven approaches have identified specific loci, including single genes, contributing to natural variation in recombination rates. As our understanding of the genetic control of recombination rates continues to grow, it has become clear that genes do not account for all observable variation in this phenotype. In fact, narrow-sense heritability estimates indicate that most recombination rate variation cannot be explained by additive genetic factors alone, suggesting an important contribution from the environment. Changes in diet, temperature, age, and social stress have been previously shown to elicit plastic, within generation responses in recombination rate, consistent with the possibility of a general stress-induced recombination response. Recent studies have demonstrated that exposure to xenobiotic estrogens can also trigger changes in global recombination frequency in house mice, including an increase in the fraction of aneuploid gametes. Despite these important contributions, there are many unanswered questions regarding the interplay of environment and recombination. For one, most studies have been restricted to *Drosophila*, plant, and house mouse model systems. Thus, the universality of these observations across the eukaryotic kingdom remains largely untested. Additionally, a limited number of environmental agents have been specifically tested for an effect on recombination. It is unclear how sensitive (or resilient) recombination rates are to the range of abiotic variables organisms encounter in their environment. Pathogens, including infectious bacteria, are a ubiquitous feature of an organism’s environment and present a persistent source of physiological stress. Theoretical models indicate that increased genetic mixing via recombination can allow organisms to more rapidly evolve to shifting parasitic pressures in their environment, a consideration that may pose a major evolutionary advantage to sex and recombination. Indeed, sex and meiotic recombination have been shown to evolve and plastically increase in response to pathogen pressures in species with facultative sexual reproduction. However, evidence that pathogen pressures can trigger increases in meiotic recombination rates in host species with obligate sexual reproduction is currently limited to *Drosophila*. Here, we build on this theoretical foundation to test for an effect of bacterial infection on global meiotic crossover rates in house mice (*Mus musculus*), a species with multiple genetic modifiers and established non-genetic determinants of recombination rate variation. In contrast to theoretical predictions and the hypothesized link between stress and recombination, we find no evidence in support of a pathogen-associated recombination rate response in house mice. We discuss potential explanations for this negative result, as well as the biological implications of this finding. # Methods ## Animal Husbandry and Ethics Statement All aspects of this project were carried out in strict accordance with protocols approved by the North Carolina State University Institutional Animal Care and Use Committee (Protocols 13-095-B and 13-066-B). Throughout the course of this experiment, animal health was monitored at least once per day by trained animal technicians, veterinary staff, or one of the authors. Mating pairs of strains C57BL/6J (B6) and PWD/PhJ (PWD) were purchased from The Jackson Laboratory and housed at the Biological Resources Facility at North Carolina State University under specific pathogen free conditions. Animals were provided with food (PicoLab^® Mouse Diet 20 5058\*) and water *ad libitum*. Mice were purpose-bred by intra-strain crosses to generate progeny reared under controlled laboratory conditions. Male pups were weaned into same sex groups at approximately 3 weeks of age and subsequently isolated into individual cages at 8 weeks. ## Bacterial Cultures and Experimental Design An infectious clonal isolate of *B*. *burgdorferi* (B31-MI-16) was cultured at 34C in Barbour-Stoenner-Kelly II (BSKII) medium supplemented with 6% rabbit serum. After reaching mid-logarithmic phase (5x10⁷ bacteria/mL) the culture was diluted to a concentration of 50,000 cells/50μL for intradermal injection. Eight-week-old adult male mice were treated with Nair™ hair removal product to expose a patch of bare skin and allowed to recover for 24 hours. Animals were then subjected to one of three alternative treatments. Mice in the first group were injected intradermally with 50,000 cells/50 μL of *B*. *burgdorferi*. The second treatment group was injected with sterile BSKII media (media control). The third group was not subjected to any further treatment (Nair-only). Animals were then aged 10–14 days under sterile conditions in a BSL-2 environment. During this time, one PWD male in the Nair-only treatment group was found dead of apparent natural causes. Mice were sacrificed by over-exposure to isoflurane gas in a sealed container at approximately 10 weeks of age. The left testis was dissected and immediately disaggregated with a handheld blender. Testis cells were cultured in BSKII media supplemented with phosphomycin and rifampicin for 14 days, and then examined by dark field microscopy to confirm the presence of *B*. *burgdorferi*. ## Spermatoctye Spreads and Immunostaining Spermatocyte cell spreads were made from the right testis as previously described and subjected to immunostaining according to published protocols. Slides were blocked and antibodies were diluted in 1x antibody dilution buffer \[10x ADB: 2.5 mL normal donkey serum (Jackson ImmunoResearch), 22.5 mL 1x PBS, 0.75 g bovine serum albumin (Fraction V; Fisher Scientific), and 12.5 uL Triton X-100\]. The following primary antibodies were used at 1:100 dilution: mouse anti-MLH1 (BD), goat anti-SCP3 (Santa Cruz Biotechnology), and human anti- centromere polyclonal (Antibodies, Inc). The following secondary antibodies were used at 1:200 dilution: donkey anti-mouse Alexa Fluor 488, donkey anti-goat Rhodamine Red-X, and donkey anti-human Coumarin AMCA (Jackson Immunoresearch). Slides were mounted in ProLongGold antifade (Promega) prior to microscopic analysis. ## Microscopy and Image Analysis Slides were analyzed with a Leica DM5500 B microscope equipped with a Photometrics CoolSNAP HQ² CCD camera and a 63x oil-immersion objective lens. Images were captured as RGB stacks in Leica Application Suite (v. 2.3.5) software and stored as high-resolution tiff files. Images were subsequently cropped and the fluorescent intensity adjusted using ImageJ software. We aimed to capture approximately 25 well-stained late pachytene stage spermatocytes per animal. Cells in this meiotic sub-stage were defined by two key criteria: (1) the complete co-localization of SCP3 signals along the paired homologous chromosome axes and (2) a minimum of one MLH1 focus per autosome, excepting the possibility of one achiasmate bivalent per cell. Spermatocytes that appeared damaged during preparation, exhibited synaptic defects, or with bulbous chromosome termini (indicative of transition into diplotene) were not imaged. For each cell, the number of autosomal MLH1 foci was scored. Given that the dynamics of the X and Y chromosomes are temporally decoupled from those of the autosomes during early meiosis, MLH1 foci on the heterogametic sex chromosomes were not included in this total. ## Statistical Analyses All statistical analyses were carried out in the R environment for statistical computing (v 2.14.1) using base packages. Our MLH1 dataset consists of bounded, ordinal data that do not comply with the standard assumption of normality. Consequently, we used non-parametric Mann-Whitney U-tests to compare MLH foci counts between treatment groups. # Results and Discussion ## *B*. *burgdorferi* invade testis tissue We used *Borrelia burgdorferi*, the causative agent of Lyme disease, as a model bacterium to test for a plastic, within generation global recombination rate response to infection in male house mice. Rodents, including house mice, are important biological reservoirs for this bacterium and play an integral role in its lifecycle. *B*. *burgdorferi* infected house mice show a well-characterized and stereotyped response to infection, with an antibody-mounted immune response initiated at approximately two weeks post-infection. Prior to this time point, animals display no overt symptoms of infection, even as the bacteria multiply and infiltrate tissues distant from the site of initial infection. One possible mechanism by which *B*. *burgdorferi* infection could induce a plastic recombination response is via direct interaction of the bacterium (or bacterial secretions) with meiotic cells. To confirm that our experimental design could detect such an effect, we first tested whether intradermal injection with *B*. *burgdorferi* resulted in bacterial invasion of the testis tissue in infected animals. Importantly, dark field microscopy indicated that all testis cell cultures initiated from uninfected control animals were sterile. In contrast, *B*. *burgdorferi* were identified in cultured testis cell extracts from all infected animals, indicating bacterial colonization of the testis. To the best of our knowledge, this is the first evidence that *B*. *burgdorferi* invade testis tissue in infected animals. ## Bacterial infection elicits no change in global crossover rates We used MLH1 mapping to estimate global crossover counts in 16 male house mice representing 2 genetically distinct inbred strains. We analyzed a total of 450 spermatocytes, corresponding to an average of 28 cells per animal (range: 8–51 cells). Animals were reared under one of three experimental treatments: (1) intradermal injection with *B*. *burgdorferi*, (2) injection with sterile media, or (3) Nair-only control. We observed a significant difference in mean MLH1 foci count between B6 and PWD mice irrespective of treatment, revealing a large difference in global crossover rate between these two genetically distinct strains (24.05 versus 29.75, respectively; Mann-Whitney U-Test *P* \< 2.2x10^-16). The numbers reported here are consistent with prior measurements in these strains and confirm the robustness of MLH1 measurements from animals reared in distinct laboratories. In contrast to the significant recombination rate difference between strains, we found no difference in mean MLH1 counts between treatment groups within a strain. This result is insensitive to the control group used for comparison. Although there are minor fluctuations in mean MLH1 counts between B6 individuals within a treatment group, these slight differences do not appear to mask an effect of infection status on global crossover frequency. One-way *ANOVA* tests performed on replicate animals within a treatment group, modeling animal replicate as a factor and the total autosomal MLH1 focus count of single spermatocytes as the response variable, are not significant *(P* \> 0.05 for all tests). We conclude that *B*. *burgdorferi* infection does not alter global meiotic crossover rates in male house mice, at least at the dosage and for the exposure time considered in this experiment. ## Determining power to find statistical differences in global crossover frequency One interpretation for the absence of an effect of infection on global crossover frequency is a lack of statistical power to find differences between treatment groups. To calculate statistical power over a range of effect sizes, we simulated datasets that replicated the sample structure of our data, assuming that MLH1 counts follow a normal distribution with mean and standard deviation following the observed B6 and PWD values. In reality, MLH1 count data are ordinal, but in practice, the use of random numbers sampled from a continuous distribution should have little effect on the qualitative conclusions of our simulation study. We further allowed for the possibility of inter-individual variation in average MLH1 focus counts by sampling individual values from a random normal distribution centered on the strain mean with a standard deviation of 0.5. The selection of this latter value was guided by observed differences in mean MLH1 counts between replicate animals. With our B6 sample structure, there is excellent statistical power to detect a mean difference of ≥1 MLH1 focus (Power \> 0.8 at α = 0.05 using a Mann-Whitney U-Test;). Our power to detect differences in mean MLH1 foci counts for PWD individuals is reduced owing to a single biological replicate per treatment. However, our analysis is still well powered to find differences ≥1.5 MLH1 foci for sample sizes mirroring the collected data (Power \> 0.7 at α = 0.05;). For both the B6 and PWD strains, approximately 20% of simulations uncover a significant difference between treatment groups in the absence of a simulated effect. These false positives are explained by chance differences between treatments that arise from random sampling of individual means. ## Infection induces no changes in the distribution of meiotic crossovers Although there is no detectable difference in global MLH1 frequency between infected and uninfected animals of either tested strain background, there may be shifts in the distribution of crossover events, including differences in the frequency of achiasmate chromosomes or chromosomes bearing multiple MLH1 foci, that are not captured by this overall measure. We tested for a significant infection-driven response in the number of achiasmate bivalents, the number of bivalents with a single MLH1 focus, and the number of bivalents with two MLH1 foci. None of these meiotic sub-phenotypes was plastically altered in infected animals (Mann-Whitney U-Test *P* \> 0.15 in all comparisons;). Similarly, the cell-to-cell variance in mean MLH1 foci number is independent of infection status (Fligner-Killeen Test *P* \> 0.35 for both strains). Although these meiotic phenotypes differ between B6 and PWD, infection status has no discernable impact on any of these sub-phenotypes. The heterogametic X and Y sex chromosomes pair and undergo recombination across a narrow region of homology on their telomere-proximal end. Although the dynamics of pairing and recombination in this pseudoautosomal region are de- coupled from the activities of the autosomes at meiosis, the fraction of paired and recombining sex chromosomes at late-pachytene provides a snapshot of the temporal progression of these processes. Recent studies have highlighted several unique aspects of sex chromosome behavior during meiosis, raising the distinct possibility that infection stress could impose unique pressures on the meiotic behavior of the X and Y, independent of the autosomes. However, as above, we find no difference in the frequency of paired XY chromosomes or XY chromosomes with an MLH1 focus in the paired pseudoautosomal region at late pachytene between infected and uninfected animals (Mann-Whitney U-Test *P* \> 0.4 in all tests;). ## Reconciling current observations with previous work Despite considerable evidence for a plastic recombination rate response to diverse environmental stressors, here we have shown that exposure of adult male house mice to the pathogen *B*. *burgdorferi* does not alter global crossover rates or gross features of crossover distribution. Below, we summarize several distinguishing features of our study that could account for the apparent discrepancy between our findings and previous work. First, although certain environmental variables have direct effects on recombination frequency in house mice, it is possible that other factors, including infection, could impact recombination rate by fundamentally distinct mechanisms. For example, *D*. *melanogaster* exhibit a strong pathogen- associated recombination response, but the observed plastic increase in recombination rates is driven, at least in part, by non-Mendelian transmission of recombinant chromatids. Notably, our current study is not designed to detect differences in fertilization efficiency, differences in sperm viability, zygotic viability differences between recombinant and non-recombinant gametes, or asymmetries in meiosis that could distort transmission and lead to biological increases in recombination rate. Second, our analysis tested the effect of exposure to one infectious pathogen, *B*. *burgdorferi*, on global crossover rates. A recombinational response to bacterial infection in house mice could be pathogen-specific, and potentially related to the nature of the mounted immune response. Although *B*. *burgdorferi* are present in the testis tissue of infected mice, the bacterium is only mildly morbific to most inbred mouse strains, producing no chronic, outward signs of disease. Thus, *B*. *burgdorferi* infection alone may not impose sufficient stress to elicit a change in the frequency of recombination. Consistent with this possibility, exposure to heat-killed bacteria does not induce a recombinational response in *D*. *melanogaster*, even though the immune system is activated, whereas infection with live bacteria significantly increases recombination rates. Future experiments that test for a recombination response to more virulent pathogens could yield results distinct from our findings. Third, our study also lacks power to find small changes in global crossover rate that may arise as a result of infection. However, documented changes in recombination frequency in response to pathogen infection in *Drosophila* are sizeable, corresponding to \~3% change in the estimated recombination fraction, or a \~15–20% relative increase in recombination rate. Similarly, exposure to environmentally relevant doses of BPA induces differences of \>1 MLH1 focus in mice. A plastic phenotypic response of these magnitudes could be reliably detected with our sample size. Finally, our analysis relies on the immunodetection of MLH1 foci in late pachytene spermatocytes to estimate global crossover rates in males. Previous studies have established the accuracy and power of this immunofluorescence approach for approximating global crossover rates in single animals. However, a minor subset of crossovers (\<10%) is resolved by an MLH1-independent pathway in house mice; any effect of infection limited to crossovers resolved via this alternative pathway would obviously go undetected by our study. In addition, the MLH1 immunofluorescence assay lacks the resolution to identify compensatory finer-scale changes in crossover rates. Finally, owing to the challenge of obtaining early meiotic cells in females \[e.g.,\], we did not address the possibility of a female-specific response to *B*. *burgdorferi* infection in the current experiment. These limitations present clear opportunities for future investigations. ## Evolutionary implications of the absence of a plastic recombination rate response The Red Queen theory for the evolution and maintenance of sex posits that increased recombination may be selectively favored to allow species to rapidly adapt to changing parasitic pressures in their environment. Consistent with the predictions of this hypothesis, species with both sexual and asexual modes of reproduction can shift the relative population ratio of sexual versus asexual individuals to favor sex when confronted by pathogens. Similarly, recombination rates can plastically increase in response to infection in species with both obligate and facultative sexual reproduction. House mice are exposed to a barrage of pathogens in their wild habitats, conditions that set the stage for potential host-pathogen arms races. Despite this clear ecological opportunity, our data do not suggest that house mice attempt to genetically outwit one common bacterium in their environment (*B*. *burgdorferi*) via an increased rate of global meiotic recombination. It is possible that a pathogen-driven plastic recombination response is a biological mechanism present in only some taxa. Indeed, mathematical models have demonstrated that the conditions under which plastic recombination can evolve in diploid organisms are quite restrictive. Moreover, for mammalian species with small effective population sizes, the magnitude of the selective advantage associated with moderate plastic changes in recombination rates may not be sufficient to overwhelm the power of random genetic drift. However, as is the case in *Drosophila*, an infection-driven recombination response in house mice could be mediated by transmission distortion rather than an overt increase in recombination rate. Future work will be aimed at specifically testing this hypothesis. Increased recombination–whether via increased sexual reproduction, increased recombination rates, or biased transmission of recombinant gametes–is one effective strategy for evading parasites, but additional biological mechanisms are integral to host defense. In particular, mammals possess a sophisticated immune system with both innate and adaptive components. The mammalian adaptive immune system may provide an effective biological barrier to rapidly evolving biotic stressors, potentially mitigating the effect of a plastic recombination response. Indeed, plastic recombination responses have, to date, only been documented in invertebrates with comparatively simple innate immune systems. However, without first ruling out transmission distortion as a potential mechanism for pathogen-associated increased recombination in house mice and in the absence of data from other mammalian species, this possible explanation remains speculative. Vertebrates, including mammals, also have higher per base mutation rates than invertebrates. The higher rate of input of new variants may effectively match the challenge of adapting to new parasitic pressures, independently of changes in recombination rate. Clearly, investigations that explicitly test predictions of the Red Queen hypothesis in diverse organisms, including other mammals, are needed. Our findings prompt us to speculate that the relative contributions of genetic and non-genetic factors to recombination rate variation may differ between species. Although heritability estimates are not strictly comparable between studies, it is noteworthy that estimates from house mice are consistently higher than those reported in other species. Thus, relative to other taxa, recombination rate variation in house mice may be more strongly tied to the effects of segregating variation in recombination modifying genes and more weakly influenced by environment. If true, this possibility would add an additional layer of complexity to our nascent understanding of the mechanisms contributing to recombination rate variation. # Supporting Information This work was supported by North Carolina State University (NCSU) start-up funds to NDS and JCM. BLD is supported by a distinguished postdoctoral research fellowship with the Initiative in Biological Complexity at NCSU and a K99/R00 Pathway to Independence Award from the National Institute of General Medical Sciences (1K99GM110332). [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BLD JCM NDS. Performed the experiments: BLD AAD DMT JCM. Analyzed the data: BLD. Wrote the paper: BLD NDS.

# Introduction Ischemic cardiac events, such as acute myocardial infarction (AMI), cause a decline in renal function. Furthermore, in patients after AMI, combined renal dysfunction increases subsequent total mortality and cardiovascular death. The progression of chronic kidney disease (CKD) worsens the success rate of percutaneous coronary intervention (PCI) and prognosis. Accordingly, establishing a therapeutic modality to maintain or improve renal function in patients after AMI is important. In recent years, renal function has been recognized as a new target for exercise-based cardiac rehabilitation, and we previously demonstrated the renal-protective effects of chronic exercise in an experimental animal model. Additionally, cardiac rehabilitation focusing mainly on supervised exercise therapy for AMI patients, or a group of patients with heart disease including those with AMI, was reported to be associated with maintaining and improving renal function. These reports suggest the possibility of kidney protection through exercise in AMI patients. However, in Japan, the rate of cardiac rehabilitation implementation for outpatients after AMI is very low. In many cases, patients select remote exercise management, such as maintaining physical activity level in daily life and walking based on education by medical staff, such as doctors or physiotherapists. Therefore, elucidating whether physical activity level in the daily life of AMI patients has an effect on the changes in renal function is necessary, but there are no previous reports. Furthermore, most previous studies that examined the effect of exercise on renal function in AMI patients estimated renal function with serum creatinine, which depends on skeletal muscle mass. Physical activity, including exercise, can change serum creatinine levels via changes in skeletal muscle mass. Therefore, the use of cystatin C, which is independent of skeletal muscle mass, is recommended when examining the effects of physical activity on renal function. Accordingly, the aim of this study was to elucidate the association between physical activity level and changes in renal function in patients after AMI using cystatin C. # Materials and methods ## Study design and patients This study was a prospective observational study, as shown in. In this study, the follow-up period was decided to be 3 months with reference to a previous study that examined the effect of exercise on renal function in AMI patients. Forty-one patients who were admitted to Southern Tohoku General Hospital from May 2017 to June 2018 due to AMI onset and who underwent PCI and cardiac rehabilitation during hospitalization were enrolled in the study. Exclusion criteria were as follows: refusal or inability to provide informed consent; follow-up not possible for 3 months after discharge; dependence on others for activities of daily living; receiving maintenance hemodialysis therapy; complicated by other acute diseases; underwent invasive treatments such as surgical operation during hospitalization or follow-up; and a diagnosis of dementia. This study was conducted in accordance with the Helsinki Declaration. It was approved by the Ethics Committee of Tohoku University Graduate School of Medicine (Approval No. 2016‐1‐683) and the Ethics Committee of Southern Tohoku General Hospital (Approval No. D16‐14). All patients provided written informed consent after receiving an explanation about the purpose and protocol of the study. ## Baseline characteristics Clinical information on age; sex; height; weight; body mass index; smoking history; employment status; comorbidities of CKD, diabetes mellitus, hypertension, and dyslipidemia; use of medications such as angiotensin- converting enzyme inhibitors, angiotensin receptor blockers, calcium-channel blockers, beta-blockers, and statins; Killip classification, peak value of serum creatine kinase; left ventricular ejection fraction on an echocardiogram; and Geriatric Depression Scale-15 score, in which a score of 5 or higher was defined as the presence of depressive symptoms, were collected from medical records or by interview at registration. ## Blood biochemistry and urinalysis Blood biochemistry and urinalysis were performed at discharge and at 3 months after discharge. For blood biochemistry, B-type natriuretic peptide, hemoglobin, glucose, triglyceride, high-density lipoprotein (HDL) cholesterol, low-density lipoprotein (LDL) cholesterol, LDL/HDL ratio, cystatin C, creatinine, urea nitrogen, cystatin C-based estimated glomerular filtration rate (eGFRcys), and creatinine-based estimated glomerular filtration rate (eGFRcreat) were collected. For urinalysis, urine protein/creatinine ratio (UPCR) was collected. In this study, eGFRcys was used for the primary evaluation of renal function because serum creatinine levels, which depend on skeletal muscle mass, may be influenced by physical activity. GFR was estimated using the Japanese Society of [Nephrology](https://www.sciencedirect.com/topics/medicine-and- dentistry/nephrology) equation as follows: eGFRcys (mL/min/1.73 m²) = 104 × Cystatin C^−1.019 × 0.996^Age × 0.929 (if female)– 8 eGFRcreat (mL/min/1.73 m²) = 194 × Cre^−1.094 × age^−0.287 × 0.739 (if female) ## Physical function tests and blood pressure measurement Physical function tests, including the 6-min walk test (6MWT), handgrip strength test, and 30-second chair-stand test (CS30), were conducted at discharge and at 3 months after discharge. The 6MWT was conducted according to the American Thoracic Society recommendations. The patients had to walk as far as possible. If they experienced shortness of breath and fatigue, they were allowed to reduce their walking speed and take a break as needed. After the break, they had to resume walking as soon as possible. Patients were informed of the elapsed time every minute by a physiotherapist. The test was conducted after a 10-min rest. Handgrip strength test was measured using a digital dynamometer (TKK 5101 Grip-D; Takei, Tokyo, Japan). This test was conducted according to the methods of a previous study. It was performed with the right and left hands in turns. The highest strength values on the right and left sides were averaged and expressed as the absolute value (kg). The CS30 consists of standing up and sitting down from a chair as many times as possible within 30 s. A standard chair (with a seat height of 40 cm) without a backrest was used. The patients were initially seated on the chair with their back in an upright position. They were instructed to look straight forward and to rise after the “1, 2, 3, go” command at their own preferred speed with their arms folded across their chest. Blood pressure measurements were conducted at discharge and 3 months after discharge. All tests and measurements were performed after a 10-min rest. ## Physical activity To evaluate physical activity, we used an accelerometer with a tri-axial acceleration sensor (AM500N; Acos Co., Ltd, Nagano, Japan). The accelerometer was distributed to the patients at discharge. The patients were instructed to wear the accelerometer consistently for 3 months, except when bathing and sleeping. The patients were also instructed to record their daily number of steps in the recording notes for 3 months. The accelerometer can record the number of steps and duration of activity when the instrument detects acceleration greater than the preset threshold value of activity (metabolic equivalents \[METs\]). We set the value to 3.0 METs as the threshold of moderate intensity of physical activity. Three months after discharge, the patients came to the hospital and physical activity data were transferred into a computer from the accelerometer via a reader (RC-S360/S; Acos Co., Ltd, Nagano, Japan). Subsequently, the average number of steps (steps/day) and duration of activity \> 3.0 METs (min/day) for 7 consecutive days at 1 month after discharge and 3 months after discharge (total of 14 days) were calculated and recorded as the absolute values of both parameters. The 14 days were shown in the research manual and recording notes in advance, and patients were able to confirm the 14 days. When the patients performed activities that could not be measured by the accelerometer, such as cycling or swimming, or forgot to wear the accelerometer, these had to be indicated in the recording notes, and data on the relevant days had to be excluded from analysis. However, none of the patients fulfilled the exclusion provision during the 14 days. Then, we examined the maximum number of steps (steps/day), minimum number of steps (steps/day), and coefficient of variation of number of steps for each patient to confirm variations in physical activity in 14 days. We also examined the adherence rate of monitoring physical activity in 3 months and the main seasons of the monitoring from the recording notes. ## Statistical analysis We calculated the median number of steps in all patients (4113 steps/day). Patients who walked \>4113 steps/day and ≤4113 steps/day were classified into the high-activity group (high group) and low-activity groups (low group), respectively. Changes in eGFRcys (ΔeGFRcys) (mL/min/1.73 m²) were defined as the value obtained by subtracting the eGFRcys at discharge from the eGFRcys at 3 months after discharge and change in eGFRcreat (ΔeGFRcreat) was also calculated in the same manner. Descriptive statistics were used to present baseline characteristics and several parameters related to physical activity in the high and low group. Clinical data including blood biochemistry, urinalysis, and physical function tests at baseline and both ΔeGFRcys and ΔeGFRcreat were compared between the two groups using an unpaired t-test or Wilcoxon signed-rank test. Clinical data at baseline and at 3 months after were compared within each group using a paired t-test or Wilcoxon signed-rank test. As a main analysis, generalized estimating equations (GEE), which account for within-participant clustering, was used to test the longitudinal association between physical activity level and within-patient changes in eGFRcys adjusting for potential confounding variables. An unstructured working correlation structure was applied to adjust for within-patient correlation. As age, renal function, diabetes mellitus, hypertension, and dyslipidemia have been reported as factors that can affect the progression of kidney disease, age, glucose, triglyceride, and systolic blood pressure at baseline were determined as covariates to assess the association. Interaction terms between physical activity level (high group, low group) and time (baseline, 3 months after) were included in a model to investigate the effects of physical activity level on change in eGFRcys during follow-up. As a secondary analysis, to verify the difference between ΔeGFRcys and ΔeGFRcreat when verifying the continuous association between physical activity and change in renal function, we analyzed the correlations between the number of steps and ΔeGFRcys, or ΔeGFRcreat in all patients using Pearson’s correlation, and compared their correlations. All analyses were performed using SPSS 21.0 for Windows (SPSS Inc. Chicago, IL). An alpha-level of 0.05 was considered for statistical significance. To reduce the risk of type I error in multiple comparisons in unpaired t-tests, paired t-tests, or Wilcoxon signed-rank tests on the clinical data at baseline and after 3 months in the low and high groups, we corrected the p-values to q-values for each of the tests, according to the false discovery rate (FDR) procedure. The FDR threshold was set at q = 0.05. # Results ## Physical activity Forty-one patients were stratified as 21 patients in the low group (number of steps, 2335 ± 1219 steps/day; duration of activity \>3.0 METs, 18.3 ± 11.0 min/day) and 20 patients in the high group (number of steps, 7102 ± 2365 steps/day; duration of activity \>3.0 METs, 56.8 ± 24.1 min/day), as shown in. The adherence rate of monitoring physical activity in 3 months in the low group was 76.3% ± 31.7% and that in the high group was 91.3% ± 13.3%. ## Baseline characteristics shows the baseline characteristics of the low and high groups. ## Clinical data at baseline and after 3 months shows the clinical data at baseline and after 3 months. At baseline, the high group had higher values of 6-min walk distance (6MWD), grip strength, and CS30, and lower value of BNP than the low group. In the high group, eGFRcys (76.5 ± 13.8 to 83.2 ± 16.0 mL/min/1.73 m², q = 0.013), glucose, HDL cholesterol, 6MWD, systolic blood pressure, and diastolic blood pressure increased, and BNP, LDL cholesterol, LDL/HDL ratio, and cystatin C decreased from baseline to 3 months after. In contrast, in the low group, hemoglobin, glucose, triglyceride, HDL cholesterol, 6MWD, CS30, and systolic blood pressure increased, and BNP and LDL/HDL ratio decreased without significant changes in eGFRcys (65.1 ± 15.9 to 62.2 ± 20.2 mL/min/1.73 m², q = 0.200). Based on the results of the comparison of changes in eGFR between the two groups, ΔeGFRcys in the high group was significantly higher than that in the low group (+6.7 ± 9.2 vs -2.9 ± 8.2, q = 0.009). ## Associations between physical activity and changes in eGFR shows the results of the GEE analysis testing the longitudinal association between physical activity level and within-patient changes in eGFRcys. Both results of the crude model (p \< 0.001) and adjusted model (p = 0.003) showed significant positive associations. Change in eGFRcys was -2.9 mL/min/1.73 m² in the low group versus +6.7 mL/min/1.73 m² in the high group. After adjustment for potential confounding variables, similar results were shown. shows the associations between the number of steps and ΔeGFRcys or ΔeGFRcreat in all patients. The results of Pearson’s correlation analysis showed significant correlations between the number of steps and both parameters. Furthermore, the correlation coefficient between ΔeGFRcreat and the number of steps (r = 0.38, p = 0.015) was lower compared to the correlation coefficient between ΔeGFRcys (r = 0.55, p \< 0.001) and the number of steps. # Discussion The primary findings of this study are as follows: (1) physical activity level in daily life was positively associated with changes in renal function in patients after AMI, and (2) the number of steps correlated significantly and positively with both ΔeGFRcys and eGFRcreat in all patients, although the correlation coefficient using ΔeGFRcreat as an indicator for changes in renal function was lower than that using ΔeGFRcys. In recent years, cardiac rehabilitation focusing mainly on supervised exercise therapy for patients with AMI or a group of patients with cardiovascular disease was found to maintain or improve renal function. According to Takaya et al., in a phase II cardiac rehabilitation program for 3 months for AMI patients, eGFRcreat improved significantly in active patients with a participation frequency of 1.1 times or more per week, but no significant improvement was observed in inactive patients with a participation frequency of less than 1.1 times per week. There is a possibility that renal functions in AMI patients will improve with exercise therapy, and in such cases, maintaining the frequency and amount of exercise is especially important. However, most previous studies that examined the effect of exercise on renal function in AMI or CVD patients estimated renal function with serum creatinine, which depends on skeletal muscle mass. Only one study demonstrated that the renal function assessed by cystatin C improved in CVD patients with CKD after cardiac rehabilitation, but the study did not have a control group and was unable to verify the effect of amount of exercise on renal function. In the present study, the physical activity of AMI patients was recorded from the time they woke up until bedtime for over 3 months using an accelerometer to verify the association with changes in the renal function over time. Additionally, we considered eGFRcys as a valid indicator of renal function in this study. The results indicated significant longitudinal associations between physical activity level and within-patient changes in eGFRcys, and in the analysis adjusted for potential confounding variables, similar significant association was confirmed. Additionally, exacerbation in the urine analysis data was not confirmed within each group as well. These results support and supplement prior studies mentioned above that indicated the improving effect of exercise on renal function in AMI patients. Significant correlations between changes in renal function and number of steps were found in all patients in the secondary analysis. However, the correlation coefficient using ΔeGFRcreat as an indicator for the changes in renal function was lower than that using ΔeGFRcys. Based on the scatter diagram, an increase in the variation of ΔeGFRcreat with the increase in the number of steps was confirmed. As previously pointed out, changes in the serum creatinine level through movement of the skeletal muscles is one of the causes, and the significance of using eGFRcys as an indicator for renal function in this study was confirmed. In recent years, a prior prospective study verified the association between physical activity level and renal function in CKD patients. The results of this study are similar to our present findings and indicate that maintaining a high level of physical activity in daily life leads to suppression of renal function deterioration. In contrast, studies conducted on CKD patients have shown conflicting findings. A study reported the same results as found in CKD patients, but a study also reported that physical activity level does not affect renal function. Thus, the relationship between physical activity level and renal function is not clear for persons without CKD. Not being able to sufficiently grasp the level of physical activities using the previous methods may be one of the reasons. These reports were based on the evaluation of physical activity level, and measurements were conducted using a recollection type questionnaire to simplify the evaluation process. However, when elderly people are included in such studies, the evaluation results are possibly biased due to poor memory or inability to understand questions, and this may be one of the factors for conflicting results between studies. In this study, reliability of the evaluation was ensured by measuring the amount of physical activity using an accelerometer. The adherence rate for the physical activity monitoring was 83.6% during the 3 months follow-up, and there were no missing data during the 14 days extracted for analysis. Additionally, the target patients of this study did not engage in exercises such as cycling or swimming, which cannot be measured using an accelerometer, during the observation period. The physical activity level measured in this study was a better reflection of the actual physical activity level in daily life. This study is the first to show the association between physical activity level and changes in renal function after the onset of AMI, and it demonstrated that maintaining a high physical activity level is essential for protecting the renal function of AMI patients. The strengths of this study include the reliability of evaluating renal function and physical activity. In this study, the primary evaluation for estimating renal function utilized the cystatin C level, which has been reported as a more reliable marker of renal function than creatinine. Furthermore, we analyzed the actual amount of physical activity by measuring the daily number of steps for 3 months using an accelerometer rather than a recollection type questionnaire, which may be associated with recall and other biases. In the clinical setting, the therapeutic strategy of using the number of steps in daily life, which is simple, cost-effective, and has less time constraints, may be easily accepted as one of the strategies to manage renal disease in outpatients after AMI. On the contrary, in this study, the clinical data at baseline indicated that the low group had significantly lower physical function than the high group. Furthermore, the differences similar to those observed at baseline between groups were observed even after 3 months. These results indicate that deterioration in the physical activity level was likely to occur continuously in patients with low physical function at baseline in a non- interventional situation. To increase physical activity and function of such patients, further intervention such as increasing the frequency of periodic follow-up after completion of phase I cardiac rehabilitation, providing education and counseling for physical activity, or continuing guidance on self- exercise with motivational coaching strategies and objective feedback on the training data will be necessary. This study has several limitations. First, it was a small, prospective, observational study without a control group, and most samples (85.4%) were men. For those reasons, all confounding factors that may affect the changes in the renal function were not sufficiently adjusted. We examined the control status of age, blood glucose level, lipid level, and blood pressure as background factors that may affect renal function. We analyzed these factors as covariates using GEE but did not confirm dietary intake status that includes management of salt and water intake because detailed evaluation after discharge of patients was difficult. Second, the verification of this study was limited to a follow-up period of 3 months. Whether the same results will be obtained in an extended observation period is unknown. Verifying whether the changes in renal function can be corrected by improving physical activity and physical function during the study is important, but this was not investigated in this study. For this reason, future research should verify the long-term effect of physical activity level on renal function among AMI patients in a larger cohort. # Conclusion The present study is the first to show the association between physical activity level and changes in renal function after the onset of AMI using an accelerometer and eGFRcys. High physical activity was suggested to suppress renal function decline in patients with AMI. Our findings support the importance of interventions to maintain high physical activity as a strategy for renal protection in AMI patients. Future research should verify the long-term effect of physical activity level on renal function among AMI patients. # Supporting information The authors thank Prof. O. Ito from Tohoku Medical and Pharmaceutical University for his valuable comments. We would also like to thank all patients whose participation made this investigation possible and all colleagues in Southern Tohoku General Hospital for their contribution to the medical care of the patients. [^1]: The authors have declared that no competing interests exist.

# Introduction Patient-reported outcomes (PRO) are important as endpoint in clinical trials and epidemiological studies. These outcomes comprise all self-reported measures by the patient regarding the patient’s health, the disease and its impact, or its treatment. They include health related quality of life, pain, patient satisfaction, psychological well-being, symptoms, treatment adherence/preference,… PRO have first gained importance as secondary endpoints because they can be helpful to evaluate the effects of treatment on patient’s life or to study the quality of life of patient along with the disease progression to adapt the patient’s care. They can also be used as primary endpoint, especially in chronic diseases such as cancer, to compare two standard treatments with comparable survival outcomes or to help decision making. The deleterious impact of each treatment on patient’s quality of life can also be evaluated. The singularity of PRO lies in the fact that the outcome, such as quality of life or wellness, cannot be directly observed. This particular outcome is defined as a latent variable. Generally, a questionnaire is the instrument that indirectly measures the latent variable and the responses of patients to items are further analyzed. Models from the Item Response Theory (IRT) link the probability of an answer to an item with item parameters and a latent variable. This theory has gained importance in Patient-Reported Outcomes area compared to the Classical Test Theory (CTT) where models are based on a score that often sums the responses to the items. IRT has shown advantages such as the management of missing data, the possibility to obtain an interval measure for the latent trait, the comparison of latent traits levels independently of the instrument, the management of possible floor and ceiling effects. With the development of patient-reported outcomes in clinical research, guidelines were edited for construction, validation and administration of questionnaires. However, the literature presents few references to the design stage. In particular, the sample size requirements when IRT is intended to be used for analysis of PRO seems to lack of theoretical work. When PRO are used as primary endpoint in a group comparison study, it is essential at the design stage to correctly determine the sample size to achieve the desired power for detecting a clinically meaningful difference in the future analysis. An inadequate sample size may lead to misleading results and incorrect conclusions. General recommendations on the sample size in the framework of education can be found. It should be highlighted that these recommendations are usually made without any theoretical justification. It is admitted that the sample size has to increase with the complexity of the model : a number of 50 individuals was proposed for the simplest model of IRT, the Rasch model, a sample size of 200 respondents for the two-parameter logistic model has been suggested and 500 examinees for the graded-response model. Consequently, publications on health outcomes assessments make generally only few comments on the sample size determination as no analytical formula for the sample size exists. It has been recently pointed out that the widely-used formula for the comparison of two normally distributed endpoints in two groups of patients was inadequate in the IRT setting. Indeed, the power achieved by the tests of group effects using IRT modeling in a simulation study was lower than the expected power using the formula for normally distributed endpoints. Subsequently, Hardouin et al have proposed a methodology to determine power related to sample size for PRO cross-sectional studies comparing two groups of patients in the framework of the Rasch model. The power determination depends on the difference between the expected means in the two groups (the group effect) and its standard error. The key point of the method is to estimate this standard error using the Cramer-Rao bound. This theoretical approach was first validated by simulation studies in some cases (small variance, appropriate questionnaire for the population under study) that may not reflect what is encountered in practice. Whether the method would perform as well in a large variety of situations often met in clinical and epidemiological studies remains unknown. As a matter of fact, the population of the study can have heterogeneous levels of the latent variable. Moreover, the PRO instrument might be more or less suitable for the population under study. Indeed, the items composing the instrument can be more or less relevant for the intended population of the study. For example, items from a disease-specific questionnaire (such as the QLQ-C30 evaluating the quality of life of cancer patients) can be too difficult in a newly-diagnosed population in the sense that items specific to the disease can almost never be encountered in a population where the disease was recently detected, potentially before most of the symptoms appear. The measures provided by the PRO might not be reliable for all patients and the power could therefore be impacted by the choice of the questionnaire. The purpose of this study was to validate the Cramer-Rao method for PRO cross- sectional studies comparing two groups of patients using the Rasch model. The impact of the variation of the variance of the latent variable (inter-patient heterogeneity regarding the latent variable) and of the distribution of the item parameters (appropriateness of the questionnaire for the population) on the proposed methodology has been studied by comparing the results of the Cramer-Rao method to the results of a simulation study. # Methods At the planning stage, the calculation of a sample size is usually based on a statistical test to detect a clinically meaningful effect at desired levels of type I and type II errors. In the case of the comparison of mean levels of PRO measures in two groups of patients, the widely-used formula for the comparison of two normally distributed endpoints may apply. The formula assumes that the two groups are independent and that the variance of the endpoint is common across the groups. The hypotheses for the two-sided test of comparison are defined as against, where and are the means of the endpoint in the first group and the second group respectively. The number of patients to be included in the first group is determined by specifying an expected difference in the means of the PRO measures and the common variance as well as the type I error and the desired power of the test.where is the number of patients in the second group and the percentile of the standard normal distribution. If this formula is adequate for manifest variables such as quality of life scores, it seems to incorrectly determine the sample size for latent variables as it doesn’t take into account the uncertainty due to the estimation of the latent variable. So, this formula is not adapted for studies intending to use IRT models for the analysis. ## Sample Size and Power Determinations in IRT ### The rasch model In IRT, the link between a latent variable, that is the non-directly observable variable that the PRO instrument intends to measure (quality of life for example), and item parameters is modeled. Amongst the large family of IRT models, the Rasch model, is largely used for dichotomous items in health sciences. It models the probability that a person answers a response to an item by a logistic model with two parameters, (i) the value of the latent variable of the person, and (ii) the item parameter associated with the item. For a questionnaire composed of dichotomous items answered by patients, the Rasch mixed model can be written as follows:where is a realization of the random variable ( for the most defavorable response, for the most favorable one). is also called the difficulty of item. As the value of increases, the item is more and more difficult which means that patients are less and less likely to answer positively to the item. For example, an item “Does your health allows you to run an hour?” will be more difficult than an item “Does your health allow you to dress yourself?” if the positive answer is defined as “yes”. is a realization of the random variable, generally assumed to have a gaussian distribution. In this case, the parameters of the Rasch model can be estimated by marginal maximum likelihood (MML). A constraint has to be adopted to ensure the identifiability of the model. The nullity of the mean of the latent variable is often used for this purpose. ### Power estimation using cramer-rao bound In the design of a cross-sectional study for the comparison of two groups of patients in IRT, we are interested in the evaluation of a group effect, defined as the difference between the means of the latent variable in the two groups. Let and be the expected sample size in the first group and the second group respectively. To identify the model presented above, the constraint of the nullity of the mean of the latent variable is adopted. The mean is the mean between and, each of them weighted by the sample sizes and. Consequently, Let be a random variable representing the latent variable with normal distributions and in the first and the second group respectively. The variance of the latent trait is assumed to be equal in the two groups. The mixed Rasch model including a covariate to estimate a group effect can be expressed as follows:with in the first group and in the second group in order to meet the constraint of identifiability. The sample size determination often relies on the Wald test to assess whether the group effect is significant. The following hypotheses are to be tested, against. To perform the test, an estimate of and its variance are required. The test statistic follows a normal distribution under. At the design stage, Hardouin et al. proposed to use Fisher’s information and the Cramer-Rao (CR) boundary property to obtain an analytical formula for the standard error of. This method takes into account the characteristics of the questionnaire by using the parameters of the items to estimate the variance of the group effect. It also incorporates the uncertainty related to the estimation of the latent trait in the IRT model. At the design stage, the item parameters are set to some planning expected values as well as, and. In addition, as the patient’s responses are not known, they should be determined. For each possible response patterns ( for binary response), the associated probability is computed for each group using the Rasch model, conditionally on the planned values of, and. The expected frequency of each response pattern in each group is then determined. The dataset created with the response patterns and their associated expected frequencies is analyzed using a mixed Rasch model including a group effect to estimate the variance of the group effect using CR and the power of the Wald test. The expected power of the test of the group effect based on the Cramer-Rao bound (CR), can be approximated by :with assumed to take a positive value, be the quantile of the standard normal distribution and evaluated using Cramer-Rao bound. The whole procedure has been implemented in the free Raschpower module accessible at <http://rasch-online.univ-nantes.fr>. This module determines the expected power of the test of the group effect based on the Cramer-Rao bound given the expected values of the sample size in each group, the group effect, the variance of the latent variable and the item parameters defined by the user. ## Simulation Study To validate the Cramer-Rao method, the power determined with this method was compared to the power obtained by a simulation study, used as a reference. ### Generation of data Responses to dichotomous items of two groups of patients were simulated using a mixed Rasch model where the latent variable has normal distributions and in the first and the second group respectively. To study the impact of the values of the item difficulties, the distribution of items could vary in two different ways according to the regularity of the spacing of the items and the gap between the mean of the latent variable and the mean of the items distribution. To obtain item difficulties that are quite regularly spaced, their values are set to the percentiles of a determined probability distribution. The normal distribution is used with the same mean and variance as the latent trait distribution. The questionnaire will therefore estimate the patients levels of quality of life with the same accuracy whatever the level of quality of life on the continuum of the latent trait as shown on (subfigure A). To obtain irregularly spaced item difficulties, an equiprobable mixture of two gaussian distributions was used. When the spacing is irregular, the estimates of the patients levels, of quality of life for example, will be more precise when difficulties are close to each other than when they are far apart from each other. We can see on (subfigure B) that the quality of life levels around −1 will be estimated more precisely than quality of life levels between −0.5 and 0.5. The case of irregular spacing of item difficulties is probably more encountered in practice than regular spacing. The distribution of the items could be centered on the same mean as the latent trait or a gap, between the means of the latent trait and the mean of the item difficulties could be simulated. A positive gap is illustrated in (subfigures C and D). The latent variable distribution and the items distribution are then no more overlaid. The most difficult items of the questionnaire (on the right of the distribution) will be too difficult for the population. Hence, a very small part of the patients will respond positively to these items while most of the patients will respond positively to the easiest items (on the left of the distribution) leading to a floor effect. Due to this floor effect, the estimates of the patients levels will be less accurate on the left of the latent trait distribution (for poor levels of quality of life for example). In practice, a floor effect can occur when a disease-specific population answers to a generic questionnaire. For example, patients with serious physical impairment won’t be likely to answer positively to physical functioning items such as the ability to walk a block, to run or to climb stairs (example of items from the physical functioning of the generic questionnaire SF-36). On the opposite, a negative gap will lead to a ceiling effect as the items will be too easy for the studied population. ### Parameters of the simulation study The following values of the parameters were used in the simulation study: - The number of individuals was equal in both groups and could take the value 50, 100, 200, 300 or 500. - The group effect was equal to 0, 0.2, 0.5 or 0.8. - The value of the variance of the latent trait could be 0.25, 1, 4 or 9. - The number of items was 5 or 10. - The item difficulties could come from a normal distribution (quite regularly spaced) or from an equiprobable mixture of and (irregularly spaced). The global mean of the latent variable was equal to 0. So, the distributions of items and latent traits were overlaid if the mean of item distribution was also equal to 0. The gap, was defined as 0 (overlaid distributions), 1, 2. As the gap becomes larger, the items distribution departs more and more from the latent traits distribution and floor effect could occur more frequently. In the case of a normal distribution with a null, the questionnaire is assumed to be appropriate for the population without a floor effect and the items are quite regularly spaced. The combination of all parameter values lead to 960 different cases. 1000 replications were simulated for each case. ### Evaluated criteria Each simulated dataset was analyzed with a mixed Rasch model including a covariate to estimate the group effect. A Wald test was then performed for assessing the significance of the group effect. For the simulations only, the type I error was estimated as the rate of rejection of the null hypothesis (null group effect) amongst datasets where the group effect was null. The confidence intervals of the type I error was computed as exact binomial proportion confidence intervals. The power of the test of group effect of the simulations, was estimated as the rate of significant tests amongst the simulations where the simulated value of was not null. This result was compared with the estimated power using CR, (eq. 2), computed with the Raschpower module of Stata. As the estimation of is based on the estimated value of the standard error of, a good estimation of the power requires a good estimation of this standard error. Hence, the estimated value of the variance of the group effect in the simulation, was compared with the estimated variance of the group effect using CR. # Results ## Estimation of the Variance of the Group Effect and show the estimated variance of group effect obtained either by simulations or using CR for all the values of parameters, for a group effect equals to 0 or 0.2 and 0.5 or 0.8, respectively. The estimations of the variance are close for both methods in general. As expected, the variance of the group effect decreases as and increase. Coherently, the variance of the group effect increases with the variance of the latent variable. It slightly increases with the value of the group effect. We note that the estimations of the variance for CR method are larger as compared to the simulation mostly when the gap is high and. The highest overestimated values of the variance for CR are observed for low values of the sample size and of the number of items, high values of the latent variable variance and a normal distribution of the items as compared to a mixture of normal distribution of items. ## Type I Error and Power of the Test of Group Effect For the simulations, the type I error is well maintained to the expected value of 5% in almost all scenarios (results not shown). The type I error fluctuates between 2.6% (for a mixture distribution of the item difficulties) and 6.8% (for a mixture distribution of the item difficulties). Amongst the 240 values of the type I error, only 9 confidence intervals at 95% of the estimated type I error don’t contain the expected value of 5%. None of the parameters seems to have an impact on the value of the type I error. and present the estimated values of the power obtained either by simulations or using CR for the values of all parameters for a questionnaire composed of 5 and 10 items, respectively. For simulations, the power was estimated by the rate of rejection of the null hypothesis amongst datasets where the group effect was not null. For all values of the simulation parameters, the estimated powers are close for each method (CR or simulations) when there is no gap. The difference between the powers obtained by simulation and using CR is around 0.003 in average and fluctuates between −0.034 (items normally distributed) and 0.059 (items normally distributed). As expected, the power increases as the sample size and the number of items increase and decreases as the variance of the latent trait increases. It also increases with the group effect. We observe an impact of the gap between means of the latent variable and item difficulties which is stronger when the gap is high. In these cases, the power obtained using CR is lower than the power of simulations. The loss of power is the highest when the variance ( = 4 or  = 9) and the group effect ( or) are high, the number of items is low and the distribution of the items is normal as compared to a mixture of normal distribution of items. The loss can exceed −20% in the worst cases. For example, when, and the distribution of items was normal, the estimated power is 83.4% for the simulations and 60.3% for CR. # Discussion The validity of the method to estimate the standard error of group effect and to determine the power of the test of group effect in IRT using Cramer-Rao bound was investigated for a large number of situations that may be often encountered in practice. The estimated variance of group effect and power obtained using Cramer-Rao were close to the estimations from the simulations when the distributions of the latent variable and the items were overlaid. As expected, the variance of group effect increased with the variance of the latent variable. This led to a decrease of the power of the test of group effect that does not differ for both methods (Cramer-Rao and simulations). The Cramer-Rao method seems to be still valid for high values of the variance of the latent variable. However, when the gap between means of the latent variable and item difficulties is high, we observed an inflated estimation of the variance of group effect and consequently a loss of power for CR compared to the simulations. The Cramer-Rao method seems to reach its limits for and high values of and. The impact of an underestimation of the power can have large consequences on the planned sample size. To achieve a power of 80% for a gap equal to when and, the Cramer-Rao method suggests to use patients per group whereas patients per group is a sufficient sample size to obtain a power of 83.4% according to the simulations. Hence, in this example, 200 patients in each group would have been unnecessarily included in the study to achieve a power of 80% using the Cramer-Rao method with a gap equals to. So, the choice of a questionnaire appropriate to the population at the design stage is an important issue. For example, the use of a disease- specific questionnaire in general population is not recommended as the population of the study will probably not encounter some of the symptoms strongly related to this disease. Thereby, some items evaluating the symptoms will have only few or no positive responses leading to a floor effect and an incorrect determination of the power with the Cramer-Rao method. We recommend taking time on the choice of the questionnaire before the study. To evaluate the suitability of a questionnaire, it seems important to first check that the items composing the questionnaire intended to be used are relevant for the population of study. An item is not considered as relevant if the population will answer mainly to one of its modality only and will lead to ceiling or floor effect. When choosing a questionnaire for a study, one has to take into account the characteristics of the population used for its former validation (type of the disease, seriousness of the pathology, …) in order to be suitable enough for the population to be studied. At the planning stage, the parameters of the distribution of the latent variable and the item parameters have to be fixed. To do so, it is easier to rely on a pilot study or on previous articles for example. Hence, it may be possible to evaluate if a gap between the mean of the latent variable and the mean of the items distribution is likely to occur. Despite all the precautions taken at the planning stage, a gap can be observed at the analysis stage. Unfortunately, the Cramer-Rao method would have underestimated the power in this case. Consequently, the number of subjects to be included in the study would have been overestimated which raise ethics and financial problems. Given the results, it does not seem reasonable to use the Cramer-Rao method for a gap equals or higher than. In fact, a gap equals to 2 standard deviations seems to already reflect a poorly suitable questionnaire, a generic questionnaire assessing health-related quality of life of a seriously ill population for example. However, the Cramer-Rao performs well in a large number of situations and can handle a moderate gap between the distributions of the latent variable and the items. We observed a slight impact between the quite regularly spaced items (normal distribution) and the irregularly spaced items (mixture of normal distributions) on the variance and the power when the gap was high. The normal distribution gave higher estimations of variance and so lower power than the mixture of normal distributions. This effect increased with the gap. It could be explained by the fact that, in the way the data were simulated in our study, the items coming from the mixture of distributions covers a wider part of the latent variable distribution as shown in. Furthermore, when the latent variable and items distributions are not overlaid, the easiest item coming from the normal distribution ( in (subfigure C) for example) is more on the right of the latent variable distribution than the easiest item coming from the mixture of distributions ( in (subfigure D)). Therefore, the floor effect, resulting from the gap, occurs at a lowest level of for the normal distribution than for the mixture of distributions. And so, the floor effect has more impact on the variance and power obtained using item parameters coming from the normal distribution. As this effect is linked with the simulation process, it can’t be interpreted as an impact of the regularity of the items on the performance of the Raschpower method. Beyond the impact of items and variance of the latent trait, the effect of the sample size, the number of items and the group effect were also studied. Their values were chosen to reflect what is frequently encountered in practice in health studies. However, some assumptions had to be made to perform the simulation study. Instead of the Rasch model, another IRT model for dichotomous items could be considered such as the 2-PLM or the OPLM. These models are more complex than the Rasch model in the sense that they include item discriminations in addition to item difficulties. The variance using Cramer-Rao could probably be estimated with the same efficiency by adapting the formula and fixing the item discrimination to known values as made for the item difficulties. The estimation of the variance and the determination of the power are based on the expected planned values that are fixed. This is usual at the design stage but it can turn out to be problematic if no previous studies can provide some information on the values of the parameters. If the planning values are far from the estimated values in the study at the analysis stage, the variance could be incorrectly estimated and the power for a determined sample size could then not be achieved. It seems important to further study the impact of misspecifications in the choice of the planning values on the performance of the Cramer-Rao method. The robustness of this method when some of the assumptions on the model are violated should also be evaluated to identify settings where the method should or should not be used. For now, the main limitation of the Cramer-Rao method is that the variance can only be estimated in the frame of Patient-Reported Outcomes evaluated with dichotomous items in a cross-sectional setting. Two major developments seem to be necessary to make this method applicable in almost all studies in health sciences. First, the method should be able to deal with polytomous items. The estimation of the variance can be based on the partial-credit model or the rating-scale model, which are extensions of the Rasch model for this type of items. The introduction of such models will lead to a more complex procedure of estimation as the number of parameters will increase with the number of modalities of the items. Second, the study of the evolution of a criteria is often of interest in health sciences. Patients’ evolution of PRO through time are often evaluated in longitudinal studies. The validity of the Cramer-Rao method in this context has to be studied as the correlated measures of patients bring into play a more complex model than in cross-sectional studies. The estimated variance of group effect and power obtained using Cramer-Rao were close to the estimations from the simulations in most cases. These results show that the variance using Cramer-Rao bound correctly estimates the variance of the group effect. Hence, the Cramer-Rao method can be used to determine the power of the test of group effect at design stage for two-group comparison studies including patient-reported outcomes for many situations in health sciences. The important recommendation is to choose the most appropriate questionnaire for the population. Otherwise, sample size might be misspecified by this methodological approach. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: VS JBH MB. Performed the experiments: MB. Analyzed the data: VS JBH MB. Wrote the paper: WV JBH BF FG MB.

# Introduction Since matrix metalloproteinases (MMPs), a family of calcium and zinc-dependent extracellular matrix (ECM) degrading enzymes, are expressed in physiological and pathophysiological processes, and process virtually any component of the ECM, it has been suggested that MMPs may be important in inflammatory conditions such as rheumatoid arthritis and periodontitis where they participate in inflammation. There are no basal levels of MMP-3 mRNA and protein in cells, and MMP-3 synthesis is tightly controlled *in vivo*. Significantly elevated levels of MMP-1, MMP-2 and MMP-3 have been detected in acute and chronic rheumatoid arthritis, inflamed pulps and in periapical lesions when compared with healthy tissues,. While it is intuitive that dental pulp destruction may be a function of MMPs, our previous study reported that MMP-3 actually accelerates wound healing following dental pulp injury. This observation indicates that MMP-3 may be involved in both ECM degradation and the subsequent morphogenesis, wound repair, and angiogenesis in the inflamed tissue. The regulation of metalloproteinases is very complex. Regulation occurs at different levels and on a broader scale, three levels of endogenous control exist: transcriptional regulation, zymogen activation and regulation on the level of enzymatic activity by different endogenous regulators. A wide variety of cytokines, growth factors and oncogene products stimulate MMP expression. Tumor necrosis factor (TNF)-α and interleukin (IL)-1 are regularly involved in metalloproteinase gene induction. Although we previously reported that a proinflammatory cytokine mixture (IL-1β, TNF-α, and IFN-γ) could induce MMP-3 activity in odontoblast-like cells derived from mouse embryonic stem (ES) cells, the identification of the principal cytokine responsible for this MMP-3 stimulus remains unresolved. As a trophic factor, the multifunctional cytokine IL-1 plays an important role in the proliferation of cells at the site of tissue injury, ; although the signals that control cells from proliferation during wound healing are unclear. In general, a relatively low level of IL-1 can induce cell proliferation; however, high levels of the cytokine cause apoptotic cell death. Given that IL-1β has been detected in inflamed dental pulps and was associated with periapical disease, this cytokine was believed to be essential in the pathogenesis of pulpitis. In particular, treatment with IL-1β resulted in the potent induction of *MMP-3* expression in dental pulp, which contains large numbers of odontoblasts. Taken together, these studies suggest that MMP-3 induced by the proinflammatory cytokine IL-1β contributes to the pathophysiology of inflamed dental pulp. In particular, the dental pulp tissue consists predominantly of odontoblasts, with small populations of fibroblasts, blood vessels and neurons, therefore, odontoblasts may represent a new target for therapeutic strategies. Due to the challenges associated with obtaining sufficient amounts of purified odontoblast cells, no study has focused on odontoblast cells following the induction of inflammation. The heterogeneous nature of cells in the dental pulp obfuscates direct investigation of MMP-3 effects in whole dental pulp. Moreover, while the development of our basic knowledge with regard to stem cell differentiation is highly valuable, the use of human ES cells is ethically controversial and treatments employing these cells are unlikely to be realized in the near future. Consequently, we undertook our experiments using purified odontoblast-like cells derived from induced pluripotent stem (iPS) cells and ES cells, which are excellent models in which to examine the mechanism of wound healing *in vitro*. We have addressed several points described above using odontoblast-like cells derived from mouse iPS cells and ES cells.Here, we focus on the relationship between IL-1β-induced MMP-3 accumulation and the responses of odontoblast-like cells derived from iPS cells and ES cells *in vitro*. We used siRNA directed against *MMP-3* transcripts to examine whether IL-1β-induced changes in cell proliferation and apoptosis of odontoblast-like cells derived from iPS cells is associated with an increase in the expression and activity of MMP-3. # Materials and Methods ## Materials Mouse recombinant IL-1β was obtained from PeproTech (Rocky Hill, NJ, USA). Recombinant human MMP-3 was obtained from Chemicon (Temecula, CA, USA). Exocytotic inhibitor (Exo) 1 and 2-(4-Fluorobenzoylamino) methylbenzoate, an inhibitor of protein trafficking emanating from the ER, which acts by inducing the rapid collapse of the Golgi, were obtained from Sigma-Aldrich (St. Louis, MO, USA). ## Cell cultures The mouse iPS cell line iPS-MEF-Ng-20D-17 was a gift from Prof. Yamanaka (Kyoto, Japan) and was maintained as described previously. An E14Tg2a ES cell was a kind gift from Dr. Randall H. Kramer (UCSF, San Francisco, CA, USA) and maintained as described previously. Moreover, B6G-2 ES cells were from the Riken cell bank (Ibaraki, Japan) and were maintained as described previously. B6G-2 cells require feeders, whereas E14Tg2a cells do not require feeders, thus both cells were used for comparison. Rat odontoblast-like cells (KN-3 ; kindly provided by Dr. Chiaki Kitamura, Kyushu Dental College, Kitakyushu, Japan) were maintained as described previously and used as an authentic control. Purified odontoblast- like cells derived from ES cells were prepared as reported previously. Purified odontoblast-like cells derived from iPS cells were also prepared as reported. The monoclonal anti-α2 integrin antibody is known to potently suppress the expression of odontoblastic markers in these cultured systems. Thus, we could confirm that the expression of α2 integrin in ES cells triggered their differentiation into odontoblast-like cells. The proportion of α2 integrin- positive cells in the total differentiated odontoblast-like cell population is a measure of the purity of the B6G-2- and E14Tg2a-derived odontoblast-like cells, and was estimated by FACS analysis to be 98.63±0.74% (iPS-derived odontoblast- like cells; *n* = 3), 98.53±0.88% (B6G-2-derived odontoblast-like cells; *n* = 3), or 98.79±0.43% (E14Tg2a-derived odontoblast-like cells; *n* = 3). The resultant differentiated cells were shown to have odontoblast-like physiological characteristics (e.g., calcification activity and alkaline phosphatase activation) up to day 21 of the culture. ## Determination of apoptotic cell death by ELISA Cellular DNA fragmentation was assessed by detection of BrdU-labeled DNA fragments in the cytoplasm of cell lysates using solid-phase-immobilized anti- DNA monoclonal antibodies and anti-BrdU monoclonal antibodies labeled with peroxidase (Cellular DNA fragmentation ELISA; Roche Applied Science, Mannheim, Germany), according to the manufacturer's instructions and previously published studies. ## Cell proliferation assay Cell proliferation was evaluated using the BrdU-cell proliferation ELISA (Roche Applied Science, Mannheim, Germany) as described previously. The cells were seeded into 96-well tissue culture plates at a density of 1×10⁵ cells/cm². ## Semi-quantitative reverse transcription polymerase chain reaction (RT-PCR) analysis The cells were seeded into 6-well tissue culture plates at a density of 1×10⁵ cells/cm². The cells were cultured for 2 h with or without IL-1β, and total RNA was isolated from the cultured cells using an RNeasy Mini Kit (Qiagen, Inc., Valencia, CA, USA). The amount of RNA was equalized using a glyceraldehyde 3-phosphate dehydrogenase (GAPDH) competitive PCR kit (Takara Shuzo Co., Shiga, Japan). The mRNA was converted to complementary DNA (cDNA) with the GeneAmp RNA PCR kit (Perkin Elmer, Branchburg, NJ, USA). The primer sequences and sizes, and accession numbers of the amplicons are as follows: rat MMP-3 (sense 5′-cctgagaccttaccaatgtgtagc-3′, antisense 5′-caacatggatgctgcatatgaagtt-3′; 189-bp amplicon, NM_133523), mouse MMP-3 (sense 5′-aggatttcccaggaagatagctgag-3′, antisense 5′-aattccaacagcgaagatccact-3′; 114-bp amplicon, NM_010809), mouse TIMP-1 (sense 5′-cagcaaagagctttctcaaagacct-3′, antisense 5′-tagataaacagggaaacactgtgca-3′; 70-bp amplicon, NM_011593), mouse TIMP-2 (sense 5′-ggacctgacaaagacttcgagttta-3′, antisense 5′-ccatctccttctgcctttcctg-3′; 119-bp amplicon, NM_011594), rat GAPDH (sense 5′-gctctctgctcctccctgttc-3′, antisense 5′-cgtccgatacggccaaatcc-3′; 113 bp amplicon, NM_017008) and mouse GAPDH (sense 5′-aatggtgaaggtcggtgtgaac-3′, antisense 5′-cgtgagtggagtcggaac-3′; 155-bp amplicon, NM_008084). The primer mixture, a total volume of 25 µl, contained 50 µM deoxynucleoside triphosphates, 0.5 units of Tag DNA polymerase, 1.5 mM MgCl₂ and 1 µM of forward and reverse primers in a 10× PCR reaction buffer (200 mM Tris-HCl, 500 mM KCl, pH 8.3). The PCR reaction within the exponential phase of the amplification curve was performed for 25 cycles for MMP-3, TIMP-1, TIMP-2 and GAPDH under the following conditions: initial denaturation at 94°C for 2 min, denaturation at 94°C for 30 s, annealing at 64°C for rat MMP-3, at 67°C for mouse MMP-3, at 66°C for mouse TIMP-1, at 66°C for mouse TIMP-2, at 61°C for rat GAPDH and at 65°C for mouse GAPDH for 30 s, and an extension period at 72°C for 1 min. The PCR products were loaded in a 1.5% agarose gel, visualized with ethidium bromide and photographed. The intensities of the PCR products were quantified using a scanner and digital image analysis software (Multi Gauge-Ver3.X, Fujifilm, Tokyo, Japan). ## Western blot analysis MMP-3 protein levels in the cell lysate were determined by western blot analysis. Cells were cultured for 6 h with or without IL-1β, lysed and the protein lysate separated on SDS-polyacrylamide gels (12%) in preparation for western blot analysis using anti-MMP-3, anti-TIMP-1, anti-TIMP-2 and b-tubulin polyclonal antibodies (sc-6839, sc-5538, sc-6835 and sc-9935, respectively; Santa Cruz Biotechnology, Inc., Santa Cruz, CA, USA). The anti-MMP-3 antibody showed no significant cross-reactivity with other MMPs (data not shown). Visualization and quantification of blotted protein bands were performed with the Multi Gauge-Ver3.X software (Fujifilm). ## Measurement of MMP-3 activity The protocol for measuring MMP-3 activity was described previously and has been incorporated into a commercially available MMP-3 activity assay kit (SensoLyte™ 520 MMP-3 assay kit; AnaSpec, San Jose, CA, USA). Prior to detection, MMP-3 was immunoprecipitated from the culture medium using a goat anti-MMP-3 antibody (sc-6839, Santa Cruz Biotechnology, Inc.) and protein A/G–agarose for 6 h at 4°C. After centrifugation, the agarose pellets were suspended in a MMP-3 assay buffer (containing the MMP-3 substrate as 5-Carboxyfluorescein (FAM)/Arg-Pro- Lys-Pro-Val-Glu-Nva-Trp-Arg-Lys-QXL™ 520-NH2 fluorescence resonance energy transfer (FRET), peptide) supplied in the assay kit, and the activity of MMP-3 was determined according to the manufacturer's instructions. ## Silencing of the MMP-3 gene by siRNA transfection The anti-MMP-3 siRNAs for gene silencing were acquired commercially (sc-37265 and sc-61874, Santa Cruz Biotechnology, Inc.) and transfected into cultured cells using a siRNA reagent system (Santa Cruz Biotechnology, Inc.) according to the manufacturer's protocol. An anti-GAPDH siRNA and a siRNA with no known homogeny for any vertebrate sequence (Thermo Scientific, Lafayette, CO, USA) were used as positive and negative controls, respectively. ## Statistical analysis Data are presented in bar graphs are the means ± standard deviations (SD) of 4–6 independent experiments. Statistical significance was assessed using the Mann- Whitney U-test. *P*\<0.05 was considered as statistically significant. # Results ## IL-1b induces the expression of MMP-3 mRNA and protein, and MMP-3 activity in odontoblast-like cells The odontoblast-like cells derived from iPS cells, ES cells and KN-3 were cultured in the presence of three concentrations of IL-1β (0, 0.25, 2.5 and 25 ng/mL) and MMP-3 induction was assessed using RT-PCR and western blot analysis. We found that 0.25 and 2.5 ng/mL IL-1β induced *MMP-3* mRNA and protein expression, whereas 25 ng/mL IL-1β did not affect MMP-3 levels. MMP-3 activity is precisely regulated at the level of transcription by the activation of their precursor zymogens, and by the action of endogenous inhibitors, namely TIMPs. Although it in known that TIMP-2 is inducible by cytokines, we confirmed that both TIMP-1 and TIMP-2 mRNA and protein levels were constitutively expressed in all experimental conditions. To normalize the increase in MMP activity, it is necessary to describe the activity in the context of TIMP protein expression. As shown in, TIMP-1 and TIMP-2 were present continuously at stable levels under all experimental conditions. To assess the intracellular MMP-3 peptidase activity induced by IL-1β treatment, we assayed MMP-3 activity in immunoprecipitates prepared from cell culture medium following cell treatment with various concentrations of extracellular IL-1β (0, 0.25, 2.5 and 25 ng/mL). We observed significant increases in MMP-3 activity in the culture medium from cells treated with 0.25 and 2.5 ng/mL of cytokine, but not in that from cells treated with 25 ng/mL of IL-1β (*P*\<0.01;), as compared with control cultures. The level of mRNA is somewhat irrelevant because it does not give any information about how the MMP-3 is released once it is made, i.e., it could be released exocytically or by cell death. When we performed a simple experiment in the presence and absence of an exocytosis inhibitor, such as Exo1, to demonstrate that inhibiting exocytosis precludes the appearance of MMP-3 in the conditioned medium, we demonstrate that Exo1 (3 µM) could attenuate the appearance of MMP-3 in the conditioned medium. Therefore, we confirmed that IL-1β-induced MMP-3 is released exocytically from these cells. ## Effect of exogenous IL-1β on DNA fragmentation and cell proliferation in odontoblast-like cells We next sought to assess the effect of IL-1β on MMP-3-mediated apoptosis and cell proliferation. At 25 ng/mL, IL-1β was found to induce a significant (\>400%) increase in apoptosis (*P*\<0.01;) and a decrease in proliferation. The lower concentrations of exogenous IL-1β (0.25 and 2.5 ng/mL) were found to increase cell proliferation of odontoblast-like cells derived from iPS cells, ES cells and KN-3 (*P*\<0.05;) and apoptotic cell death was not detected. Since apoptosis and proliferation rates appear to be inversely correlated, the increase in apoptosis induced by the IL-1β at a concentration of 25 ng/mL may cause a marked decrease in proliferation observed in these odontoblast-like cells under these conditions. ## Effect of MMP-3 siRNA on IL-1b-induced MMP-3 expression in odontoblast-like cells We next employed siRNA directed against MMP-3 to determine if the effects observed with IL-1β stimulation were mediated by MMP-3 in each cell type. Cells were transfected with MMP-3 siRNA or a control siRNA and then stimulated with IL-1β as described above. *MMP-3* was expressed in cells transfected with control siRNA but not in cells transfected with the MMP-3 siRNA. We observed no change in the expression of the internal control (*GAPDH*) after transfection, and the negative control siRNA confirmed that the effect on the cells was specific to MMP-3. Moreover, western blot analysis confirmed the efficient silencing of the MMP-3 gene at the protein level in the presence of IL-1β stimulation. Transfection with siRNA had no effect on the level of expression of β-tubulin, which was used as a control to ensure equal loading. ## Effect of MMP-3 siRNA on IL-1β-induced MMP-3 activity in odontoblast-like cells Next, we demonstrated that transfection of MMP-3 siRNA also efficiently down- regulated MMP-3 activity (*P*\<0.01;). The control siRNA had no effect on the changes in *MMP-3* mRNA, MMP-3 protein levels, or MMP-3 activity induced by IL-1β (0.25 and 2.5 ng/mL;). ## Effect of MMP-3 siRNA on DNA fragmentation and cell proliferation in odontoblast-like cells Under the same culture conditions described above, we next tested the effect of MMP-3 siRNA on IL-1β–induced changes in apoptosis and cell proliferation. Silencing the MMP-3 gene considerably increased the number of apoptotic cells compared with the non-transfected control group following treatment with the IL-1β. The degree of apoptotic cell death in MMP-3 siRNA-transfected cells treated with the cytokine was 3-fold higher than in the non-transfected control group (*P*\<0.01;). There was no such increase in apoptosis in cells treated with the control siRNA. The control siRNA did not attenuate the cell proliferation induced by IL-1β (0.25 and 2.5 ng/mL). The number of cells with proliferative potential was 60% less in MMP-3 siRNA-transfected cells than in non-transfected cells (*P*\<0.05;). ## Effect of exogenous MMP-3 on cell proliferation in odontoblast-like cells We tested whether exogenous MMP-3 could enhance proliferation in odontoblast- like cells derived from iPS cells. Interestingly, the addition of MMP-3 (10, 30, or 50 ng/mL) increased cell proliferation (*P*\<0.05;) in a dose-dependent manner, but caused a dramatic decrease in cell proliferation at the highest concentration tested (100 ng/mL). Exogenous MMP-3 also induced cell proliferation in odontoblast-like cells from ES cells and rat KN-3. ## Effect of exogenous MMP-3 on IL-1 β induced DNA fragmentation in odontoblast-like cells Finally, we next tested whether exogenous MMP-3 could rescue the apoptotic effects seen with MMP-3 siRNA. Our results confirmed that exogenous MMP-3 (5, 10, 20 and 30 ng/ml) could rescue the cells from undergoing IL-1β-induced apoptosis (0.25 ng/ml) in the presence of MMP-3 siRNA (*P*\<0.01;). # Discussion Currently, the effects of suppression of MMP-3 activity on the apoptosis of odontoblast-like cells derived from iPS cells has not been reported previously, because purified odontoblast-like cells were not available. In our previous report, a cytokine mixture induced MMP-3 regulated cell proliferation and suppressed apoptosis in mouse ES cells derived odontoblast-like cells. IL-1β also induces MMP-3-regulated cell proliferation and suppresses apoptosis in rat dental pulp cells. However, as dental pulp cells consist of fibroblasts, odontoblast cells, vessels and neuronal cells in the tissue, it was unclear which cells were mainly regulated by IL-1β-induced MMP-3. In the current study, IL-1β was found to activate MMP-3 in odontoblast-like cell, and this observation represents a novel physiological function of MMP-3. Our data highlights four main points. First, this is the first report that involved the use of siRNA that targets *MMP-3* to elucidate the mechanism on IL-1β-induced proliferation of odontoblast-like cells derived from iPS and ES cells. Given the challenges of obtaining sufficient amounts of purified odontoblast-like cells, this is the first study to investigate odontoblast-like cells after induction with a proinflammatory cytokine. As shown in the method described above, the purity of these cells was ∼98% and the cells display the physiological functions of odontoblast-like cells. It is likely that the characterization of the cells derived from iPS cells is different from the one derived from ES cells. Since odontoblast-like cells derived from iPS cell showed similar responses as those observed for odontoblast-like cells derived from ES cells, we demonstrated a good correlation and the availability of two-cell lines to produce odontoblast-like cells. Secondly, since we confirmed that a relatively low amount of exogenous MMP-3 (30 and 50 ng/mL) increases cell proliferation and rescues cells from undergoing apoptosis (MMP-3: 5–30 ng/mL;), it is possible that minor accumulation of MMP-3 stimulates the regulation of cell proliferation within a microenvironmental site (e.g., at sites of inflammation). Thirdly, since we demonstrated that a cytokine mixture induced MMP-3 also regulated cell proliferation of odontoblast-like cells from ES cells, the effects of the cytokine mixture on the odontoblast-like cells may be primarily due to the presence of IL-1β. Finally, while relatively high amounts of IL-1β caused apoptotic cell death with no induction of MMP-3, low amounts of IL-1β could produce MMP-3 in the cells tested. Although IL-1-induce MMPs are involved in the breakdown of ECM in disease processes, such as arthritis and tumor metastasis,, there is no MMP-3 mRNA and MMP-3 under normal physiological conditions because of the absence of the required cytokines to stimulate production of this protein, as shown in. Taken together, the physiological function of IL-1β-induced MMP-3 might play a role in anti-apoptotic activity, but not a destructive role in cells during the early phase of inflammation. It remains to be established how IL-1β-induced MMP-3 activity regulates the anti-apoptotic effect of odontoblast-like cells. We hypothesize that, with so much apoptosis occurring as shown in and, there may also be a limited amount of uncontrolled cell death that leads to the release of MMP-3 into the conditioned medium. To test this concept we measured MMP-3 levels in the presence and absence of an exocytosis inhibitor, Exo1, to demonstrate that shutting down exocytosis suppressed the appearance of MMP-3 in the conditioned medium. Taken together, we believe that the presence of MMP-3 in the culture corresponds to exocytosis of the enzyme, and not to passive output as a consequence of cell death. It is unclear how cytokine-induced MMP-3 regulates odontoblast-like cells proliferation and what molecular pathways upregulate MMP-3 following exposure to IL-1β. Recent reports demonstrated that IL-1β-induced MMP-3 is associated with the secreted glycoprotein Wnt signal pathway. Blockage of JNK signaling impairs the Wnt-5A-induced up-regulation of MMP-3. Thus, Wnt-5A may be associated with cartilage destruction by promoting the expression of MMP-3. MMP-3 also induces hyperplastic mammary epithelial growth and regulates Wnt signaling by antagonizing Wnt-5B function. The nature of the pathway by which MMP-3 induces cell proliferation, however, remains to be elucidated. Although MMP activity is but one factor that regulates the destruction of tissue, our previous study demonstrated that MMP-3 accelerates wound healing following dental pulp injury. We also confirmed that the exogenous MMP-3 could induce cell proliferation in odontoblast-like cells. Taken together with our previous report, the current evidence suggests that lower concentrations of IL-1β can induce MMP-3-drived increases in odontoblast-like cell proliferation, whereas higher concentrations inhibit cell proliferation and promote apoptosis. We thank Dr. Randall H. Kramer for the donation of experimental reagents and for helpful discussions. [^1]: The authors have declared that they have no competing interests with regard to the publication of this paper. [^2]: Conceived and designed the experiments: NO MM. Performed the experiments: TH NO HY RK KN AK. Analyzed the data: TH NO HY RK AK. Contributed reagents/materials/analysis tools: NO MM. Wrote the paper: TH NO MM HN.

# Introduction Disparities in SARS-CoV-2 infection, severity, and mortality have been documented between racial and ethnic minority groups and non-Hispanic Whites in the United States since March 2020. In California, Black individuals had higher odds of hospitalization and mortality due to COVID than White individuals. Latinx individuals in California were found to be over represented in COVID-19 case counts but underrepresented in testing and vaccine access. However, our understanding of racial/ethnic differences in COVID-19 incidence, severity, and mortality in California is restricted by available data that uses a limited number of racial/ethnic signifiers. As a result very little is known about how COVID-19 has affected other minoritized groups, including for example, Arab Americans. Arab Americans are individuals who have a cultural, ethnic, or linguistic origin in the Middle East and North Africa. It is estimated that there are over 3.5 million Arab Americans residing in the United States with the largest population residing in California. Arab Americans are not easily identifiable from standard surveys and electronic health records due to their racial categorization as White according to the federal government. Much of the research on Arab American health has focused on ethnic enclaves, like Dearborn Michigan, where large numbers of Arab Americans live in close geographic proximity. There are a number of reasons to believe that Arab Americans may be at increased risk for SARS- CoV-2 acquisition, severity and mortality including large household sizes, occupation type, and known lack of engagement with preventive services. The city of El Cajon in the east side San Diego county is a known Arab ethnic enclave. Ethnic enclaves are concentrated populations of one ethnic minority group that are often created through chain migration. While the majority of the original immigrants to the El Cajon community were from Iraq (mostly Chaldeans), many recent immigrants are from Syria, Egypt, and Palestine. El Cajon is one of three cities, with La Mesa and Santee, that comprise what is referred to as East County, San Diego. Between 50–60,000 residents are believed to be of Arab descent in East County. Most of these residents are recent immigrants who came to the United States as refugees or asylum seekers. One of the primary healthcare centers in the area that caters to the El Cajon community is Sharp Grossmont Hospital in La Mesa. Sharp Grossmont hospital offers a unique opportunity to assess the potential impact of COVID-19 on a dense Arab American community. The aim of this study was to quantify whether Arab American patients in the El Cajon region of California were experiencing higher levels of SARS-CoV-2 infection, COVID-19 severity and in-hospital mortality when compared to non- Hispanic White, Black, Hispanic and Asian patients at Sharp Grossmont Hospital. This analysis represents the first quantification of disparities in COVID-19 testing, admissions, and mortality among Arab Americans nationally. # Methods ## Study design A retrospective study was conducted using Sharp Grossmont Hospital’s electronic medical record. Patient records were included in the study if: patients were 18 years of age or older, were tested for SARS-CoV-2, were admitted for COVID-19 infection, or had COVID-19 listed as a cause of death between March 1, 2020 and January 31, 2021. This research was approved by the Sharp HealthCare Institutional Review Board. De-identified data were analyzed and therefore consent was not obtained. It was necessary for the Sharp HealthCare Clinical Analytics team to have initial access to the last name of the patients in order to determine potential inclusion in the Arab American group. Subsequent to coding patients (using numerical values) into the respective racial/ethnic groups, the last names were deleted to produce a fully de-identified data set. The research members employed by Sharp HealthCare’s Clinical Analytics team accessed the electronic medical records during the time period of December 2020 through March 2021. ## Primary exposures and outcomes The primary exposure of interest was race and ethnicity. Arab American ethnicity was determined in one of three ways: if place of birth was one of 22 Arab League countries, preferred language was Arabic, or if the last name matched a surname list that was applied to the data extracted from the medical record. Race and ethnicity for non-Arab American patients was determined from the medical record and was categorized as non-Hispanic White (NHW), Hispanic, non-Hispanic Black (NHB), non-Hispanic Asian (NHA), American Indian, Native Hawaiian/Pacific Islander, Other, or Unknown. Due to small sample sizes American Indian, Native Hawaiian/Pacific Islander, Other, and Unknown individuals were categorized together for all analyses. The primary outcomes of interest were a positive COVID-19 test result, admission to the hospital due to COVID-19, and in-hospital COVID-19 related mortality. COVID-19 test results were noted in the medical record along with the date of the test. Patients were considered admitted to the hospital due to COVID-19 if they had a positive test result that occurred at the same time or prior to their admission date and were noted as receiving inpatient care. In-hospital mortality was noted in the medical record. Other data collected from the electronic medical record included sex at birth, age, visit type (inpatient, outpatient/emergency, observation), smoking status (ever smoker, never smoker), marital status (married, single/widowed/separated/divorced), insurance type (private, Medi-Cal, other public insurance, other insurance), date of SARS-CoV-2 test, date of admission, date of death, diagnosis with co-morbidities associated with the Charlson Comorbidity Index (CCI, e.g. obesity, chronic pulmonary disease, diabetes with and without complications). Zip code was extracted and compared to data from 2019 American Community Survey to determine whether the individual lived in a zip code with \>12% family poverty prevalence. ## Analysis Summary statistics were calculated for all outcomes and covariates and were stratified by race and ethnicity group for all eligible individuals. Statistics were compared using chi-squared tests across all groups. Additional summary statistics and comparisons were made among those individuals who had tested positive for SARS-CoV-2 stratified by race and ethnic group. Logistic regression models were run with testing positive for SARS-CoV-2, admission due to COVID-19, and in-hospital COVID-19 related mortality as the primary outcomes. The primary exposure variables in these analyses compared Arab American patients to NHW, Hispanic, and NHB patients. Models were adjusted for sex at birth, age group, living in a high poverty area, CCI, marital status, insurance status, months since March 2020, and obesity. All analyses were run using SAS software version 9.4 (SAS Institute Inc., Cary, NC, USA). # Results ## Overall sample A total of 25,216 individuals met inclusion criteria between March 1, 2020 and January 31, 2021. The largest racial/ethnic group were NHW (N = 10188, 40.4%) patients, followed by Hispanic (N = 6164, 24.4%), Arab American (N = 2744, 10.9%), NHB (N = 1902, 7.5%), and NHA (N = 641, 2.5%) patients. Individuals who identified as American Indian, Pacific Islander, Other, or had unknown race or ethnicity accounted for 14.2% of the sample (N = 3577). Differences in sociodemographic factors were observed across the race and ethnicity groups. Most notably, the majority of Arab American patients (74.9%) were primarily insured through Medi-Cal (California’s Medicaid option). Living in a high poverty area was most common for Arab American (58.6%) followed by Hispanic (37.7%) patients. Overall, fewer Arab American patients had pre- existing health conditions than the other groups including chronic pulmonary disease, hypertension, and a lower CCI. ## COVID-19+ sample A total of 5,513 individuals had a positive SARS-CoV-2 test between March 1, 2020 and January 31, 2021. Hispanic (N = 2041, 37.0%), NHB (N = 360, 6.5%), and Arab American (N = 1091, 19.8%) individuals were disproportionately represented in the COVID-19+ sample when compared to NHW (N = 1192, 21.6%) and NHA (N = 111, 2.0%) individuals. Arab American patients who tested positive were most commonly in the 30–49 age category (37.7%) which differed from the age distributions of positive patients in the other race and ethnic groups. A smaller proportion of Arab American patients were admitted after a positive test (15.5%) when compared to NHW (42.0%), Hispanic (26.9%), NHB (26.4%), and NHA (46.0%) patients. A smaller proportion of Arab American patients (2.4%) experienced in-hospital mortality when compared to other ethnic groups, most notably NHA (16.2%) patients. Presenting symptoms also differed between racial and ethnic groups who tested positive. ## Models In both unadjusted and adjusted models, Arab American individuals had higher odds of testing positive for COVID-19 between March 2020 and January 2021 than NHW and NHB patients at Sharp Grossmont. Arab American patients had lower odds of being admitted after a positive COVID-19 test when compared to NHW (AOR: 0.54, 95% CI: 0.38, 0.78) and Hispanic (AOR: 0.47, 95% CI: 0.36, 0.63) patients in adjusted analyses. In adjusted analyses, Arab American patients had lower odds of experiencing an in-hospital mortality than Hispanic patients (AOR: 0.43, 95% CI: 0.28, 0.65). # Discussion Using electronic medical records from a large hospital in Southern California, differences in COVID-19 testing, admission, and in-hospital mortality were documented for a range of ethnic groups including Arab American patients. Specifically, we find that Arab Americans had greater odds of testing positive for COVID-19 but lower odds of being admitted and dying in the hospital than other racial and ethnic groups. In the first analysis of its kind, we show that there are distinct patterns for COVID infection, severity, and mortality for this largely invisible minoritized group in Sothern California. These findings suggest that the Arab American community may have protective and resiliency factors that may support their recovery from COVID-19 that other communities and groups do not have. Some of the most salient predictors for poor COVID-19 outcomes are age and the presence of co-morbidities. Arab Americans who tested positive for COVID-19 in our sample were younger and had lower prevalence of comorbidities than NHW, Hispanic, and NHA patients. Yet, even after adjustment by age and CCI, Arab Americans still had reduced odds of being admitted with COVID-19 than NHW and Hispanic patients. This suggests that the population of Arab Americans in this sample was healthier than other racial and ethnic minority groups in the area. El Cajon is mostly comprised of recent immigrants to the United States, and it is possible that this represents a healthy migrant effect in this group, even if other work has not borne out the healthy migrant effect among Arab Americans. Because the majority of Arab Americans in this study lived in the Arab ethnic enclave of El Cajon, it is also possible that these patterns represent health of this group in an ethnic enclave. Evidence regarding the overall effects of living in ethnic enclaves on health has been mixed. Ethnic enclaves have been found to be beneficial to first generation immigrants due to increased access to ethnic goods, social capital, and community services but detrimental for later generations due to low levels of resources for employment and concentrated poverty in some areas. While Arab Americans in our study had higher odds of testing positive for COVID-19, and had many social determinants of health that increased their risk for COVID-19 acquisition (large household size, low socioeconomic status), there seemed to be protection from increased severity (measured through admission) and in-hospital mortality for this population. This may suggest that while ethnic enclaves do not reduce risk of acquisition, there may be protection against serious disease and infection. Additionally, because more Arab Americans were married, there is the potential for supportive structures at home for Arab Americans dealing with infection, which may contribute to reduced severity. There is also the potential for bias against admission for this population whether due to differential practices by medical staff or due to resistance from patients. It is important to contextualize our study results with that of other minoritized populations’ experiences with COVID in California and across the US. Data at the national level shows that Black Americans have higher odds of both hospitalization and mortality due to COVID-19 than White Americans. While these patterns are likely to vary within various geographic contexts, they have been replicated in California within a large health system. The odds of admission and mortality for Arab Americans did not differ from those for Black Americans at the same hospital suggesting the need for dedicated interventions for both minoritized groups. While differences in COVID-19 outcomes between minoritized groups and White Americans have been attributed to differences in socioeconomic status, living conditions, occupation, comorbidities, and access to healthcare, the impact of structural and systemic racism underlies many of these differences in opportunity and lived experiences. The results of this analysis should be seen within the limitations of the methods. First, Arab Americans in this analysis were primarily identified using an Arab surname algorithm which may mean that some Arab Americans were included in the comparison groups and that others who are not Arab Americans may be categorized with this ethnic group. This surname algorithm has been used and validated in other studies in populations specific to California, but remains an imperfect approach to classifying Arab Americans. Second, patterns in disease incidence and severity are often influenced by acculturation and immigration variables. Due to our reliance on electronic medical records, these variables were not available for adjustment and inclusion in our analysis plan. Third, this analysis required that Arab Americans showed up to the hospital for a COVID-19 test and may not be representative of the whole Arab American population in El Cajon, specifically those who were uninsured or health illiterate. Our results can only be generalized to the population with health insurance who was comfortable attaining healthcare from this particular hospital setting. Finally, without contextualizing our results with multiple data points, biological specimens or qualitative information on patient experiences and provider practices, we cannot distinguish between the potential mechanisms explaining the disparity in admissions for this population. In the first study of its kind, this analysis examined disparities in COVID-19 acquisition, severity, and mortality among Arab Americans as compared to other racial and ethnic minority groups in a hospital in Southern California. This work adds nuance to existing studies examining health inequities in COVID-19 incidence, severity, and mortality in California. The study shows that dedicated interventions need to be put in place to reduce COVID-19 acquisition among Arab Americans. Future work should explore the protective factors that reduce severity and mortality in this population. Without a dedicated ethnic identifier, the health disparities facing Arab Americans will continue to go undocumented. Providers and hospitals in areas with large Arab American populations would benefit from collecting data explicitly on this population to better serve the community and its needs. The authors thank Jennifer Sheppard, MBA and James Curtis, Sharp HealthCare Clinical Analysts, for their contributions in extracting and validating the data sets used for analysis. [^1]: NNA and SG declare that they have no conflicts of interest and have received no funding for this study. KLG is an employee at Sharp HealthCare Center for Research and received no funding for this study. RA is a member of the medical staff at Sharp Grossmont Hospital and received no funding for this study. This does not alter our adherence to PLOS ONE policies on sharing data and materials

# Introduction Autistic people are thought to account for at least 1% of the global population. Individuals with a diagnosis of autism have differences in social communication and are more likely to have intense interests. People with autism belong within a spectrum of neurodiversity that is important for society and evolution. For the purpose of this systematic review we have followed the established international diagnostic criteria and the corresponding nomenclature. From herein we will use the associated terminology, autism spectrum disorder (ASD), although we acknowledge that different perspectives exist regarding language and terminology preferences. The genomic landscape of ASD is complex, however a strong genetic aetiology is recognised with twin studies estimating heritability between 70–90%. Access to broad genomic testing is reshaping our understanding of ASD, which appears to encompass a collection of broad, heterogenous and variable conditions with overlapping neurobehavioral phenotypes. These may be considered on one hand as complex or syndromic when ASD symptomatology features alongside intellectual disability, facial dysmorphism or congenital malformations. On the other hand, non-syndromic ASD symptomatology may comprise a broader understanding of neurodiversity. High impact genetic variants are reported to occur in around 15% of individuals with ASD, which are predominantly caused by nuclear sequence-level and structural variants, or less commonly mitochondrial variants. It is important to recognise the variable contribution genetic variants have made towards ASD symptomatology, which frequently demonstrate incomplete penetrance and variable expressivity. Proposed explanations for the high heritability, but low monogenic diagnostic findings in ASD include oligogenic and polygenic models of aetiology. Other proposed genetic aetiologies include the imprinted brain theory where there is a paternal bias in the expression of imprinted genes and epigenetic contribution. Given that most nucleotides in the human genome are outside of open reading frames of protein coding genes, yet around 75% of the genome are transcribed, this draws our attention inexorably to non-coding RNA transcripts that comprise functional molecules that may play an important role in gene expression and gene-environment interactions in ASD. A good starting point is a synthesis of the ncRNA gene expression literature to delineate further promising avenues of enquiry for ASD research. ## Classification of non-coding RNA ncRNA are described in detail elsewhere. They have historically been categorised by size, where long non-coding RNA (lncRNA) are 200 or more nucleotides and short ncRNA are less than 200 nucleotides in length. The terms “short” or “small” however, are being used less to describe ncRNA, and do not feature in the current approved nomenclature. Many ncRNA molecules regulate gene expression via RNA interference, epigenetic modification and inhibition of translation related mechanisms. Secreted extracellular circulating ncRNA are, in many cases, highly stable and detectable in multiple biological fluids such as blood, saliva and urine. There is great interest in developing ncRNA expression assays translatable into a clinical setting that may be capable of supporting ASD diagnostics and providing phenotypic or prognostic information to enhance ASD care. The HUGO Gene Nomenclature Committee (HGNC) have worked with specialist advisors to define the accepted nomenclature for ncRNA. HGNC define nine major classes of ncRNA annotated in the human genome. In this systematic review, we are interested in the shorter classes ncRNA and their relative gene expression in ASD. We are not considering genomic variation within ncRNA genes, or the expression of larger ncRNA such as ribosomal RNA or long non-coding RNA (lncRNA). The shorter ncRNA HGNC approved gene classes included in this systematic review are: microRNA (miRNA), piwi-interacting-RNA (piRNA), small NF90 (ILF3) associated RNA (snaR), small nuclear RNA (snRNA), small nucleolar RNA (snoRNA), transfer RNA (tRNA), vault RNA (vtRNA), and Y RNA. They have been summarised in. Whilst we acknowledge that HGNC approval is only in place for piRNA gene clusters, given the likely expansion to include individual piRNA genes in the future and given that annotation exists elsewhere, they have also been included. For simplicity, we will collectively refer to the shorter ncRNA classes included in this systematic review as ncRNA from herein. ## Rationale for systematic review A systematic review is warranted for a few key reasons. Firstly, much of the early research examining ncRNA expression profiles in association with ASD examines miRNA alone. We may be missing other important classes of ncRNA. To our knowledge there has been no systematic review exploring gene expression of other ncRNA. Secondly, a large proportion of early research in this field is from post-mortem samples from brain tissue. These are important for discovery but may lack clinical translatability. To realise the potential of ASD ncRNA gene expression assays for biomarker use, we require an appreciation of the combined expression data from living patients with ASD from clinically available samples. To date there have been some narrative, discursive, selective or scoping reviews and just one recent systematic review that only examines miRNA expression associated with ASD that is missing some studies. Finally, in view of the recent international nomenclature describing ncRNA with HGNC approved human genome annotation, we are keen to collate and present up to date and standardised ncRNA gene expression data associated with ASD. We acknowledge that there may be a paucity of evidence for classes of ncRNA other than miRNA, but demonstrating and delineating this clearly by systematic review is important to help shape future research directions. # Methods PROSPERO registration number: CRD42020208233. ## Study eligibility criteria The inclusion criteria were as follows: <u>P</u>opulation: Human subjects with a diagnosis of ASD compared with controls without ASD. <u>E</u>xposure: ncRNA gene expression profiles from biosamples measuring HGNC approved ncRNA genes or piRNA genes listed in piRBase v3.0. <u>O</u>utcome(s): Expression profile of any of the following ncRNA genes: miRNA, piRNA, snaR, snRNA, snoRNA, tRNA, vtRNA, and Y RNA; using validated scientific methodologies. <u>S</u>tudies: Peer reviewed publications, conference abstracts or dissertations. The exclusion criteria were as follows: studies not published in English, duplicated data, non-human studies, review articles, hypothesis papers, narrative reviews, fact sheets and letters to the editor that did not present unique or new data, unpublished materials and studies published before 2000. ## Search strategy The following electronic databases were searched: Cochrane, EMBASE, Science Direct, Medline, PubMed, Scopus, Web of Science, PsychInfo, ERIC, AMED, and CINAHL. We searched databases from January 2000 to May 2022. Medical Subjective Heading (MeSH) search terms were used for autism spectrum conditions including ‘autism’, ‘autistic’, autism spectrum disorder, ‘ASD’, autism spectrum condition (ASC), ‘Asperger’, ‘pervasive developmental disorder’ and ‘PDD’ in all combinations with the terms ‘short non coding RNA’, ‘non-coding RNA’, ‘RNA’, ‘miRNA’, ‘miRNA’, ‘piwi interacting RNA’, ‘piRNA’, ‘ribosomal RNA’, ‘rRNA’, ‘small NF90 associated RNA’, ‘small NF90 (ILF3) associated RNA’, ‘snaRs’, ‘small nuclear RNA’, ‘snRNA’, ‘small nucleolar RNA’, ‘snoRNA’, ‘transfer RNA’, ‘tRNA’, ‘vault RNA’, and ‘Y RNA’. The references cited in identified publications were also searched to locate additional studies. Data related to ncRNA expression profiles was extracted where available, including information related to normalisation strategies, ncRNA gene expression fold change, P values and confidence intervals. Given the varied nomenclature used for ncRNA, gene names will be recorded together with HGNC codes, accession IDs from miRBase database v22.1 ([mirbase.org](http://mirbase.org)) or piRBase database v3.0 (bigdata.ibp.ac.cn/piRBase). ## Procedure Two reviewers independently screened the titles and abstracts to identify all eligible studies identified by the searches. Any discrepancies were adjudicated by a third reviewer. The reference lists of selected articles were used to identify additional papers for screening. The included studies were assessed using the Preferred Reporting Items for Systematic Review and Meta-Analysis (PRISMA) guidelines. Data extraction took place and was recorded in a dedicated data extraction form generated using Microsoft Excel for further evaluation of study quality and data synthesis including functional enrichment analysis of the significant differentially expression miRNA genes. Raw data was retrieved from published papers, supplementary materials or by contacting the corresponding authors. ## Data synthesis and quality assessment We planned to perform meta-analysis of ncRNA gene expression using the statistical techniques employed by Zhu and Leung, including a random effects model to examine differentially expressed ncRNA genes in ASD compared with controls. We expected between study heterogeneity and subgroup analysis were to be used to explore possible sources, including source of patients, source of control (such as healthy control or disease control), participant ethnicity, ncRNA profile (single ncRNA and multiple ncRNA) and sample specimen (blood, saliva, urine, cultured lymphoblastoid cells, fibroblast cells, neural tissues, and others); living or post-mortem. We planned to analyse the statistical heterogeneity of the meta-analysis by x-squared (x²)-based Q statistic test when I² (I-squared or I2) exceeded 50% or P \< 0.1. Receiver-operating characteristics (ROC) curves were planned to be generated with sensitivity, specificity and positive predictive values based on known assessments of participants with ASD or without ASD. The area under the curve (AUC) was planned to be calculated both overall and for any subgroup analysis. Statistical tests were intended to be two-sided, with P \< 0.05 considered statistically significant. Functional enrichment analysis of statistically significant differentially expressed miRNA genes as determined by meta-analysis would be performed using DIANA-miRPath v3.0 and executed using the online DIANA- microT-CDS web-server algorithm to examine Gene Ontology (GO) with ‘categories union’. P-value and microT thresholds would be set at \< 0.05 and 0.8, respectively with False Discovery Rate (FDR) correction applied. Targeted pathways and significance clusters will be generated and a related heatmap constructed. We planned an assessment of publication bias using Egger’s graphical test to construct a funnel plot of all studies included in the meta-analysis and explore the symmetry of the study distribution on the plot. Begg and Mazumdar’s Rank Correlation test would be used to correlate the ranks of effect sizes and the ranks of their variances and Orwin’s Fail-Safe N test would determine the presence of missing studies that may skew the regression line in the funnel plot, with Duval and Tweedie’s Trim and Fill method being used for imputation of the missing studies. The methodological quality of all included studies was assessed by two reviewers independently using a quality assessment template based on Quality Assessment of Diagnostic Accuracy Studies (QUADAS-2). # Results ## Studies identified for selection The systematic review search strategy yielded 5250 publications, with 1221 being duplications. The titles and abstracts of 4029 papers were screened and 168 papers were assessed in full for eligibility. 48 studies were identified for inclusion in the systematic review for data extraction. This process is outlined along with reasons for exclusion in the PRISMA flow chart. ## Summary of eligible studies This systematic review has brought together the findings of 48 studies involving over 1800 individuals with ASD compared with over 1400 controls. The year of publication ranged from 2008 to 2021. ASD ncRNA gene expression studies have been conducted in numerous countries across the world, including Brazil, Bulgaria, China, Egypt, Iran, Italy, Japan, United Kingdom, and United States of America (USA). The most prolific country for publication was the USA with 12 studies. Considering all included studies, the diagnosis of ASD of study participants in 16 studies reported the use of both a validated assessment tool and Diagnostic and Statistical Manual of Mental Disorders (DSM-5) criteria. There were 14 studies that only reported the use of a validated assessment tool, the most common being the Autism Diagnostic Interview-Revised (ADI-R) and the Autism Diagnostic Observation Schedule (ADOS). Eight studies solely used The World Health Organisation (WHO) or DSM diagnostic criteria without a validated assessment tool and 10 studies did not state the method of ASD diagnosis. The vast majority of studies (N = 46) examined miRNA gene expression; 41 studies did so exclusively and 7 studies examined other classes of ncRNA, of which 5 studies also measured miRNA gene expression (including a genome wide study ncRNA expression study encompassing miRNA genomic loci). Fourteen studies used pre- selected candidate-driven ncRNA expression approaches, for example where specific miRNAs had been investigated, in contrast to 34 studies that investigated unselected or larger populations of ncRNA genes including those examined using genome wide approaches. Many of these studies went on to examine (‘validate’) a selected population of miRNA genes identified by an initial unselected approach such as microarray or from RNA-seq. Thirty-three studies reported ncRNA expression findings using tissue samples and laboratory methodologies that could feasibly be implemented into clinical practice (i.e., those from living individuals, with routine sampling methodology of easily obtainable tissue such as blood or saliva and routine laboratory work). These studies had a male to female ratio of participants of 3.5 to 1. There were 15 studies that exclusively reported findings from studies with less or unfeasible clinical implementation possibilities (i.e., when samples derived from post- mortem brain tissue or studies from living individuals requiring specialist sampling procedures such as biopsies, or those with complex or time-consuming laboratory work such as cell culturing). These studies had a male to female ratio of participants of 4.8 to 1. There were two studies that examined ncRNA expression from both clinically feasible and unfeasible samples. ## Characteristics of eligible studies provides a summary of 33 studies describing methods feasible for clinical implementation. Of these, 29 reported ncRNA gene expression from peripheral blood and 4 reported from saliva samples. We found no studies exploring ncRNA gene expression from other bodily fluids such urine or sweat. summarises the studies with less or unfeasible clinical implementation. From these 17 studies, 10 were from post-mortem brain tissue samples, five were from cultured lymphoblastoid cell lines, one was from reprogrammed induced pluripotent stem cell-derived neurons, and a further study reporting both olfactory mucosal stem cells and primary skin fibroblasts. Two of these studies examined ncRNA gene expression from both clinically feasible and unfeasible samples, and therefore feature in both Tables and. provides an overview of the individual ncRNA genes (all of which are miRNA genes) that have been reported to have increased or decreased expression in ASD cohorts in more than one study. The individual miRNA genes are listed with the direction of expression change presented by broad tissue sample types: blood, saliva, cultured lymphoblastoid cells (unless otherwise specified) and post-mortem brain samples. The seven studies examining differential expression of ncRNA genes other than miRNA have been presented in a separate table. ## Non-coding RNA with differential expression in ASD The systematic review revealed 64 miRNA genes with differential expression in more than one study. Twenty-nine of these miRNA genes had differential expression in opposing directions. Four miRNA genes had differential expression in the same direction in the same tissue type in at least 3 separate studies. These were in bloods samples for *miR-106b-5p* and *miR-328-3p*, which had increased and decreased expression, respectively. The other miRNA gene was *miR-155-5p* which had increased expression in post-mortem brain samples. Finally, *miR-146a-5p* had consistent, increased differential expression across several different tissue types as reported in four studies. These were from saliva, primary skin fibroblasts, lymphoblastoid cell lines, olfactory mucosal cells and post-mortem brain samples from the pre-fontal cortex and temporal lobe, respectively. Seven research studies examined ncRNA gene expression, other than miRNA, in association with ASD. From these studies, differential expression was reported in individual genes from ncRNA classes including: snoRNA, snRNA, piRNA and Y RNA genes. Differential expression of one or more snoRNA genes was reported by six studies, but no individual snoRNA gene or other individual ncRNA genes (excluding miRNA genes) had differential expression reported in more than one study. ## Data synthesis and meta-analysis Functional enrichment analysis using DIANA-miRPath v3.0 online interface was performed with interrogation of the four key miRNA genes identified in this review (*miR-106b-5p*, *miR-328-3p*, *miR-146a-5p* and *miR-155-5p*) versus Gene Ontology (GO) categories. Clustering with the highest enrichment significance levels were seen in ‘ion binding’ and ‘organelle function’ GO categories, which can be visualised within the heatmap generated. For the planned meta-analysis, we extracted all available ncRNA expression data from each study. Data for most studies was incomplete for meta-analysis, therefore we contacted corresponding authors, but were only able to obtain raw datasets from a small number of studies. Considering all included studies, a range of data elements were used to capture ncRNA gene differential expression. Only 17 papers included both fold change and associated statistical findings and often the latter was not corrected for multiple testing. The different data types, levels of data processing and in many instances inaccessible data, made meta-analysis unsuitable. Many papers only reported p-values for miRNAs that were found to be differentially expressed (i.e. did not report those with non-significant expression). Although methods to combine p-values have been proposed, we found that the use of different statistical tests, different hypotheses (one-sided or two-sided) and any adjustment for multiple comparisons often being unknown, made this inappropriate. Originally plans were made for a series of statistical and publication bias analytical assessments as part of a meta-analysis, but these were not possible. We added a field into the quality assessment related to statistical analysis given the complexities of analysing complex data sets and multiple testing in the included studies. The quality assessment using adapted QUADAS-2 is shown in. # Discussion We consider here our findings from 46 studies that examined miRNA gene expression and 7 studies examining other classes of ncRNA. ## Differential expression of miRNA genes in ASD Several miRNA genes have been reported to have differential expression in two or more studies. Whilst this initially appears promising, many of these are in opposing directions. This may relate to tissue specificity of miRNA gene expression, type I errors related to high numbers of miRNA genes tested and/or statistical tests being performed or reflect the heterogenous nature of ASD aetiology or the study populations examined. Further issues around study quality and bias are considered later in the discussion. Only four miRNAs had differential expression in the same direction and tissue type in at least three studies: *miR-106b-5p*, *miR-146a-5p*, *miR-155-5p* and *miR-328-3p*. Intriguingly, in addition to the studies in our systematic review, a further single case study examining genome-wide differential miRNA gene expression from the post-mortem prefrontal cortex of a single deceased individual with ASD compared with a non-ASD sibling control without ASD (i.e. ASD of N = 1) also found *miR-106b-5p* and *miR-146a-5p* were in their top six differentially expressed miRNA genes. It is instructive to consider the 4 notable miRNA genes identified in our systematic review in more detail, although caution should be exercised given the possibility of selective research and/or reporting and high levels of potential bias found from our quality assessments. ## Four notable miRNA genes with differential expression in ASD ### miR-106b-5p It has previously been reported that *miR-106b-5p* has altered expression in schizophrenia. The finding that ASD and childhood onset schizophrenia both share altered expression is under research scrutiny, although both these groups also have high associated rates of pathogenic copy number variants and brain trauma and there is a long history of some diagnostic overlap. *miR-106b-5p* has a wide influence on various biological processes including cancer and in isolation is unlikely to demonstrate disease specificity. ### miR-146a-5 *miR-146a-5p* was found to have uniformly increased expression in our systematic review across a wide range of tissue types including saliva, primary skin fibroblasts, lymphoblastoid cell line, olfactory mucosal cells and post-mortem brain samples from the prefrontal cortex and temporal lobe, respectively. One of these studies examined tissue and disease specificity of *miR-146a-5p* (with three other miRNA genes) and found no differential expression in peripheral blood mononuclear cells (PBMC) from a group of ASD patients compared to controls. *miR-146a-5p* has also been implicated in a number of biological processes including regulation of the development of viral infections and cancer tumour suppression, for example in the inhibition of both EGFR and NF-kB signalling and reduction of the metastatic potential of cancers. ### miR-155-5p *miR-155-5p* showed a degree of uniformity in our systematic review, with increased expression in the amygdala, prefrontal cortex and temporal cortex regions in three post-mortem studies but with no significant differential expression found in dorsolateral prefrontal cortex. *miR-155-5p* has been implicated in inflammatory processes and the modulation of cancer. *miR-155-5p* expression appears to be involved in impaired development of dendritic cells, B cells and T cells and is important for immune response. Moreover, it was one of several differentially expressed miRNA genes associated with a basket of neurodegenerative diseases, including idiopathic Parkinson’s disease, where *miR-155-5p* has been reported to have increased expression. ### miR-328-3p In our systematic review, *miR-328-3p* was found to have decreased expression in peripheral blood samples in three studies examining serum and plasma, respectively but a further study reported increased expression in peripheral blood. *miR-328-3p* has been thought to have a role in cancer, whereby suppression is believed to impair stem cell function, a mechanism hypothesised to prevent ovarian cancer metastasis. ### Functional enrichment analysis The output of functional enrichment analysis by DIANA-miRPath v3.0 with the four key miRNA genes identified in this systematic review (*miR-106b-5p*, *miR-146a-5p*, *miR-155-5p* and *miR-328-3p*) versus gene ontology categories identified the most significant levels of enrichment in ‘ion binding’ and ‘organelle function’ GO categories. Ion binding is an interesting finding, given the theories of channelopathy dysregulation in the pathogenesis of ASD. However, there are well articulated concerns related to the cautious interpretation of functional enrichment and pathway analysis of miRNA that have been raised within the miRNA research community. For example, there have been suggestions that the results from standard analyses are biased by over-represented terms and may suffer from ascertainment bias for the most studied molecular pathways and be limited by selective coverage of annotated genes within a gene set. Some solutions to these challenges have been proposed but are beyond the scope of this review. ## Other ncRNA with differential expression in ASD Whilst the majority of papers identified in this systematic review examined miRNA gene expression, other ncRNA genes with differential expression were reported in seven papers including differential expression of snoRNA, snRNA, piRNA and Y RNA genes. One of these studies Salloum-Asfar and colleagues (2021) was the first to report stable expression of piRNA, snoRNA, Y RNA and tRNA genes in plasma, a helpful attribute for further research. Two of the seven papers were published by the same research group, and described overlapping ASD ncRNA ‘diagnostic signatures’ that derived from re-annotation and analysis of expression data from an external dataset with validation using recruited participants. Together these two studies described nine snRNA genes, one snoRNA gene and one Y RNA pseudogene in overlapping ncRNA expression diagnostic models measured in blood. Unfortunately the corresponding raw data, strength and direction of expression change, and how each ncRNA gene contributed to their ‘signature formula’ models were not clearly reported. The small number of ncRNA gene expression studies in cohorts of individuals with a diagnosis of ASD is in itself an important finding to report, to help shape future research directions, given their cellular mechanisms and theoretical links with ASD. Each ncRNA class with reports of differential expression in ASD found in our review, have been discussed further, in turn. ### Small nucleolar RNA Six studies examined differential expression in snoRNA genes in ASD. snoRNA can be divided into three major classes: C/D box snoRNAs (SNORDs), H/ACA box snoRNAs (SNORAs) and small Cajal body‐specific RNAs (scaRNAs). snoRNAs accumulate in the nucleoli of the cell and have roles in post-transcriptional modification and maturation of ribosomal RNA and snRNA and roles in mRNA processing and splicing. There is interest in snoRNA splicing disruption affecting neuronal development and function. snoRNA have been associated with a range of human diseases including ASD, and are gathering interest \[, –\]. Differentially methylated genomic regions of paternal sperm samples have been associated with ASD-related phenotype at 12 months of age. The paternal sperm genomic loci region exhibiting differential methylation in this study contains fifteen snoRNA genes within the *SNORD-115* cluster, which lies within the Prader-Willi syndrome critical region on chromosome 15. Prader-Willi syndrome is an imprinting condition that can manifest with a neurobehavioral phenotype with aspects of ASD symptomatology. ### Small nuclear RNA Most snRNA are involved in the major and minor spliceosome complex to splice the introns from pre-messenger RNA. snRNA and the related core spliceosomal U-snRNP complexes are associated with numerous diseases including those with neurological manifestations such as spinal muscular atrophy (SMA), amyotrophic lateral sclerosis and Burn‐McKeown syndrome. Some authors have proposed an association of with snRNA with ASD, including Zhou and colleagues (2019), identified in this review, who report an ASD-ncRNA ‘diagnostic signature’ in blood comprising entirely of snRNA genes. ### Piwi-interacting RNA piRNA are frequently considered with miRNA, given their comparable size, and overlapping molecular functions. In contrast to miRNA, piRNA are predominantly expressed in germline cells and function to silence transposable elements and regulate gene expression through RNA cleavage and methylation mechanisms. The role of piRNA is increasingly being described in somatic cells, such as in the nervous system and they have been implicated in neurodevelopmental and neurodegenerative disorders. Rett syndrome is an X-linked dominant neurodevelopmental condition affecting females caused by pathogenic variants in the *MECP2* gene. Rett Syndrome is characterised by developmental regression following a period of apparently normal development, an ASD neurobehavioural phenotype and repetitive hand movements. The *MECP2* gene is responsible for binding to methylated genomic DNA and has epigenetic functions required for neuronal development. Interestingly, *MECP2* knockout mice have increased piRNA expression profiles in the cerebellum. *MECP2* also has roles related to miRNA biogenesis, miRNA binding and lncRNA interactions. ### Y RNA One study identified in this systematic review reported five Y RNA genes (*RNY4P36*, *RNY4P6*, *RNY4*, *RNY4P25* and *RNY4P18*) with decreased plasma expression in ASD compared with controls. The same study reported four other Y RNA genes with differential expression associated with ‘more symptoms’ of ASD, with increased expression of *RNY4P29* and decreased expression *of RNY3P1*, *RNY3 and RNY4P28*, respectively. Whilst there were no other studies reporting Y RNAs, Cheng and colleagues (2020) included a single Y RNA pseudogene known as *RNY1P11* within their ASD ncRNA diagnostic signature in blood, but had no HGNC approved Y RNA genes within their model. Y RNAs were first discovered in the serum of people with systemic lupus erythematosus (SLE), a multisystemic autoimmune condition that can involve the brain. Y RNA have cellular roles related to DNA replication, RNA stability and cellular stress responses. ### Other classes of ncRNA lacking ASD differential expression evidence Whilst no differential expression findings were forthcoming from this systematic review in relation to vtRNA, tRNA and snaR, we have highlighted some interesting literature relevant to ASD, worthy of further discussion. ### Vault RNA vtRNA plays a role in neuronal synapse formation and so are of interest in ASD given postulated aetiologies such as altered neurone development including synapse formation. vtRNA bind to and activate a mitogen-activated protein kinase (MEK) to amplify the RAS-MAPK signalling pathway. There is emerging evidence associating RASopathies (a group of inherited disorders caused by pathogenic variants of genes encoding regulatory proteins within the RAS-MAPK signalling pathway) with an increased prevalence of ASD. One such RASopathy is Legius syndrome, which interestingly has a murine model where the ASD-like neurobehavioral phenotype is ameliorated by MEK inhibitors. Further work related to vtRNA expression in ASD could complement this research to support the possible clinical translation of ASD-related MEK inhibitor drug therapy. ### Transfer RNA tRNA genes are encoded for by both nuclear and mitochondrial genomes. The mitochondrial genome has been proposed as a genetic modifier for ASD and theories related to mitochondrial dysfunction in ASD have been hypothesised. The mitochondrial genome encodes 22 transfer RNA genes and harbours the majority of pathogenic variants that result in broad and disparate disorders. One report demonstrated a mitochondrial tRNA variant within a single family that was attributed as causative for a heterogeneous group of neurological disorders where ASD was a feature. ### Small NF90 (ILF3) associated RNA snaR gene expression may also be worthy of further examination in ASD, given their abundant expression within the testis and discrete regions of the brain. Evidence from meta-analysis reports that advanced paternal age as a risk factor for ASD, which may be related to increased rates of genomic and epigenomic abnormalities within the germline cells. It is also interesting that polymorphisms of *SNAR-I* (one of twenty snaR genes), is associated with increased lateral ventricle volume, which is one of two neuroimaging distinguishing features (alongside increased Pallidum volume) found in a large ASD cohort that underwent high-resolution structural brain scans. ## Limitations and quality assessments of studies ### Quality of data and reporting There are several limitations that need to be taken seriously both in interpreting the results from this systematic review and in planning for future research. The exact number of ASD participants from all included studies was difficult to ascertain as certain studies were not explicit in descriptions of study populations, and there were occasions where it was difficult to exclude some study population overlap. The use of external datasets and biobank sample resources also made this challenging, with some instances where the same Gene Expression Omnibus (GEO) dataset was used (for example GSE18123 in three studies). Two of these studies were from one research group that also appeared to use the same internal datasets in both of their studies, but this was not readily apparent in their described methodologies. Most studies use small sample sizes and several studies do not report how the diagnosis of ASD was established (e.g., whether they used validated measures). We identified studies that included participants with ASD present alongside confounding phenotypes for example, individuals with ‘high-functioning’ ASD, those who recruited both ASD and control participants from an allergy/immunology clinic, and individuals with high levels of consanguinity, epilepsy and dysmorphism, that may influence miRNA expression. Participants were frequently recruited from convenience samples or clinic populations and many studies had a limited description of control groups with few or no assessments to characterise phenotype variations. These factors are further challenged by the heterogeneity of ASD and the use of small sample sizes. We also recognised a large variation in the methods used to determine ncRNA gene expression and many studies omit important methodological details related to these. ### Meta-analysis and data synthesis Statistical methodological quality in the studies are highly variable with many instances of small sample sizes and studies using inappropriate statistical tests. It is unclear in some studies whether correction for multiple testing has been applied and, where stated, different methods have been used such as Bonferroni or Benjamini-Hochberg correction. For meta-analysis, we considered methods to combine p-values, such as Stouffer’s method that is generally preferred when different weights are attributed to the p-values being combined. However, it is not clear how the direction of differential expression (often presented as fold change) should be incorporated. Some authors recommend the removal of genes with conflicting differential expression, so that only the genes with the same fold change are combined and others suggest that one-sided p-values can be used to take the direction of fold-change into account. When not specified, the p-values given are presumably two-sided but one-sided p-values are sometimes reported. We also observed different statistical tests, including t-tests, Mann-Whitney U-tests and Tukey’s multiple comparison tests to provide the p-values. These were often reported as simple inequalities rather than precise values, making it unlikely that useful information could be extracted from their combination. High degrees of heterogeneity were apparent across studies with respect to participants, sample types and expression assays. It is well recognised that different cell types have tissue specific ‘miRNomes’ and comparing this ncRNA expression data therefore might not be appropriate. Despite contacting several authors, we were not able to obtain full data sets in several cases. In summary, our planned strategy for meta-analysis and integration of the findings from different studies was not possible because of the large variation in data presentation, availability, statistical analysis used and many instances of poor reporting. ### Factors affecting ncRNA gene expression The field of ncRNA gene expression studies is littered with challenges in the interpretation of findings. Disease or developmental states may not be the only factors altering ncRNA expression. Exercise, sleep, nutritional intake and infection are just some factors that may impact ncRNA expression. Interestingly, sleep, nutrition, bowel habit and exercise may be markedly different in people with ASD compared to neurotypical people, raising the prospect that ncRNA differential expression findings may be as a result of ASD and its patterns, lifestyles and associations rather than (or as well as) aetiological. This is currently unclear and so research methodologies should attempt to examine and control for this where possible. There are numerous ways that ncRNAs are deployed in biological processes. As in multifactorial models of ASD aetiology, the role of ncRNAs may also be multifaceted and interactive. We also know that sample collection, RNA extraction, purification, storage, handling, and testing conditions can greatly impact ncRNA expression. For example, the use of an EDTA anticoagulant appears to influence specific miRNA expression, particularly after longer EDTA exposure times. In our systematic review, EDTA blood tubes were used in several studies with only a few studies using PAXgene blood RNA tubes and many studies omitted details about blood sample collection, including anticoagulant exposure timings. Quantity and quality of centrifuging in blood has also been shown to alter the proportion of intra and extracellular components that may demonstrate different miRNA expression properties. Challenges related to ncRNA data normalisation approaches also support the need for standardisation. Caution is also required for the interpretation of post-mortem samples. In life, hypoxia is known to change miRNA function and expression and so it is not surprising that post-mortem miRNAs are altered through the process of death with degradation happening in different ways at different rates. Post-mortem ncRNA gene expression studies therefore need to include supplementary tests to explore degradation to aid interpretation. In summary, the process of measuring ncRNA gene expression requires quality control and clear detailed reporting to allow comparison between studies for meaningful interpretation. ### Differential gene expression in opposing directions Our review findings of studies reporting miRNA genes with differential expression in opposite directions needs further consideration. Another systematic review in type two diabetes mellitus reported that two thirds of differentially expressed miRNA genes were found in opposite directions. Whilst this may suggest poor methodologies or reporting bias we should be cautious about how we interpret this. Some miRNA genes appear to have greater tissue specificity than others. In the context of cancer, opposing directions of miRNA differential gene expression in *miR-125b* is thought to represent oncogenic characteristics when expression is increased and loss of tumour suppressive functions when expression is decreased. Differential expression in opposing directions of individual miRNA genes was observed in this systematic review on a population level, but also on an individual level. There is evidence that direction of miRNA (and other ncRNA) differential expression may change with age or over time and may respond to environmental exposures such as smoking and alcohol. Whilst numerous miRNA genes have been associated with neurodevelopmental or neurodegenerative diseases there is still much work to be done to understand whether miRNA differential expression may play a role in aetiology or to the numerous other factors described above including a response to the condition itself. ### Expression assays for ncRNA Various technologies for measuring ncRNA expression levels have been used in the studies, each with different strengths and limitations. Quantitative polymerase chain reaction assays (qPCR) are based on the amplification of target ncRNA genes of known sequence. Although qPCR assays are known for their high sensitivity and specificity, the sensitivity does depend on the target abundance and the efficiency of the amplification. If there are closely related sequences to the target sequence, there is a risk of false amplification. The many different protocols, reagents, and analysis methods and lack of technical information led to recommendations for qPCR assay design and data reporting, or “minimum information for the publication of qPCR experiments” (MIQE). qPCR assays can be expensive as each target requires specific primers and probes and they are commonly used to validate gene expression changes identified by other methods, such as microarrays or Next-Generation Sequencing (NGS). Microarrays are cost-effective and have been widely used in ncRNA gene expression research. However, they may not be sensitive enough to detect expression of low-abundance ncRNA genes and can suffer from dynamic range issues which affect the quantification of highly abundant transcripts. Microarray results can also be influenced by probe design bias, as the performance of the probes may vary depending on their sequence. Differences in hybridisation as well as normalisation issues mean that RNA sequencing is sometimes preferred. NGS has revolutionised ncRNA research by allowing comprehensive profiling of ncRNAs. However, biases in library preparation methods, including at ligation, reverse transcription, and amplification steps, and sequencing errors, can all affect the accuracy of ncRNA identification and quantification. Furthermore, NGS generates huge amounts of data, requiring advanced bioinformatics tools and computational resources for data analysis. ### Implications for clinical practice At the current time there are no implications for clinical practice that we could reliably draw from these results, with limited evidence to support ncRNA gene expression as biomarkers for ASD. The ncRNA genes with differential expression identified in this systematic review have all been implicated in several other diseases and biological processes and there is limited or no reporting of any high sensitivity and/or specificity scores or validation studies. There are also limited descriptions of phenotypes in the ASD groups. There is, however, enough promise to suggest that continuing to research in this field has potential to improve our understanding of mechanisms associated with neurodevelopmental differences such as ASD. ### Implications for research By contrast there are many implications for research to consider. The finding that there is limited research examining gene expression in classes of ncRNA other than miRNA is important to report. This shines a light on the omission in the research literature. Given that miRNA gene silencing occurs in many tissue types including in the developing brain, it is intriguing that four proteins critical for miRNA biogenesis are encoded by genes associated with Mendelian disorders where ASD and overlapping neurobehavioral phenotypes are highly prevalent: *DRCG8* (included within the deleted region in chromosome 22q11.2), *MECP2* (Rett syndrome), *FOXG1* and *FMR1* (Fragile X). As key regulators of gene expression, miRNA may have a role in modifying genetic variants demonstrating incomplete penetrance and variable expressivity. This theory is interesting, considering the multiple examples of recurrent pathogenic CNVs associated with variable ASD risk. Some standardisation is required to overcome the large variability in quality and reporting of ncRNA gene expression in ASD. Improved methodologies and reporting would greatly benefit the research endeavour. Alongside MIQE mentioned above, we recommend researchers work to the FAIR Guiding Principles for scientific data management and stewardship (2016) to improve the findability, accessibility, interoperability, and reusability of ncRNA expression data in ASD and other ncRNA expression studies. This would provide the standardisation and authentication necessary for data to be reusable. Feature level extraction output (FLEO) files have been recommended as published gene lists (PGL data) and gene expression data matrices (GEDMs) have been deemed unsuitable for meta- analysis due to their dependence on the pre-processing used. Sharing research data between research groups comes with challenges and public sharing of raw data in biomedical microarray studies appears to be more likely for studies published in high impact journals and when lead authors are more experienced researchers. The majority of journals and funders now have data sharing policies. National and international data protection laws restrict data sharing by genomic researchers but a number of initiatives have been developed to promote successful data sharing including those hosted by the European Molecular Biology Laboratory’s European Bioinformatics Institute, the International Cancer Genome Consortium’s project, the Pan-Cancer Analysis of Whole Genomes (PCAWG) and the Human Cell Atlas. The researchers involved in setting up PCAWG have called for an international code of conduct to overcome issues with data protection and provide guidelines for researchers. # Conclusion The search for discrete genetic, immunological, metabolic, neurological/neurophysiological and behavioural associations with ASD continues. Differential expression of ncRNA genes have shown much promise in various conditions and may be playing a role in the multifactorial aetiology of ASD. At present, no clear conclusions can be drawn from this systematic review for implementation into clinical practice. The key recommendations from our study are to improve research methodologies, reporting and data sharing in this field and to fund and deliver larger studies with more power that will increase the likelihood of being able to answer important questions. # Supporting information We thank David Brown from the Academic Liaison Information Services Team at the University of York for helpful guidance related to the library databases and the search strategy. [^1]: I have read the journal’s policy and the authors of this manuscript have the following competing interests: S G-J is the Project leader of the miRBase project at The University of Manchester. This does not alter our adherence to PLOS ONE policies on sharing data and materials.

# Introduction The effect of weekend admission on increased hospital mortality is well described, but with conflicting consistency across hospitals and specialties. In-hospital factors such as lower overnight staffing, less experienced physicians, fewer ward rounds, delayed assessment and management–and pre- hospital factors such as sicker patient populations or higher acuity of illness, are all posited as potential contributors toward the phenomenon of higher mortality for those admitted to hospital over the weekend (henceforth referred to as the ‘weekend effect’). Because the incidence of critical illness does not change over the weekend, and patients admitted to intensive care (ICU) require timely, intensive, 24-hour care, ICUs have increased levels of weekend staffing and experience compared to the rest of the hospital, in an effort to minimize the weekend mortality discrepancy. Despite this, there is a demonstrably higher mortality risk for patients admitted to ICU on a weekend. A thorough investigation of the weekend effect for stroke patients admitted to ICU, assessing not only immediate mortality, but longer-term mortality and morbidity outcomes, is yet to be described. While the weekend effect for stroke patients admitted to a hospital ward has been shown to be minimal, pre-hospital factors influencing mortality (like stroke severity) may uniquely apply to the population of patients admitted to an ICU rather than a ward, over the weekend. Stroke patients admitted to an ICU have likely suffered more severe strokes than those admitted to a ward, and are likely at greater risk of sudden deterioration; requiring particularly close monitoring for potentially immediate intervention. Because of this, weekend staffing and experience differentials in the ICU may affect outcomes of critically unwell stroke patients (where timely assessment and management is particularly important) differently to how it affects stroke patients admitted to a general ward. Finally, extending outcome measurements to determine level of disability at hospital discharge, and long-term mortality, may outline whether the implications of weekend admission to an ICU for stroke patients extend beyond acute hospital outcomes. Our study describes short and long-term morbidity and mortality outcomes of an adult population suffering stroke, admitted to an academic center ICU in Boston, Massachusetts, USA, with the intention of eliciting so-far undescribed short and long-term admission outcomes for a critically unwell patient population, admitted on a weekend. # Methods This was a hospital-wide, retrospective cohort study using the Medical Information Mart for Intensive Care III (MIMIC-III) database from the Beth Israel Deaconess Medical Centre (BIDMC). MIMIC-III is a large, publicly- available database of de-identified health-related data, collected prospectively for over 40,000 patients who were admitted to the BIDMC ICU. The database includes basic demographic information, admission details (laboratory findings and investigations, medications, interventions and outcomes) and encompasses a diverse, large population of critically ill patients in the United States. The data in this study was limited to patients aged 16–89 years of age. All data is de-identified in accordance with the Health Insurance Portability and Accountability Act (HIPAA) standards, including removal of names, phone numbers, and addresses. Date-shifting was utilised to preserve de-identification, however time of the day and day of the week are preserved in the MIMIC III database. The institutional review boards of the Massachusetts Institute of Technology (No. 0403000206) and BIDMC (2001-P-001699/14) both approved the use of the database for research. Stroke was defined using ICD-9 codes recorded in the MIMIC III database. We defined stroke with ICU admission diagnosis classified as ICD-9 code 430 (subarachnoid haemorrhage), 431 (intracerebral haemorrhage), 432 (other and unspecified intracranial haemorrhage), 433 (occlusion and stenosis of precerebral arteries), or 434 (occlusion of cerebral arteries). Stroke was further subclassified into either haemorrhagic (ICD 430, 432, 432) or ischemic (ICD 433, 434) stroke. Weekend ICU admission was defined as any admission during the 48-hour period between midnight at the beginning of Saturday, and midnight at the end of Sunday. The MIMIC III database records acute severity of illness using the Oxford Acute Severity of Illness Score (OASIS); a simplified illness severity score developed using machine learning. It has a discrimination, calibration and prediction of equivalent accuracy to more complex models like APACHE IV, and is used in our analysis as an indicator of illness severity. Only the first index admission to ICU for each patient was included in the analysis, to avoid multiple analyses for patients re-admitted to the intensive care for a presentation associated with their initial cerebrovascular accident. Demographic and outcome data is presented as either counts and percentages (for categorical data), or means and standard deviations (for continuous data). Ethnicity is divided into four primary categories (“Other” inclusive of Hispanic, Latino, Asian and other patients) to preserve statistical power, insurance status as either government (Medicare, Medicaid) or non-government, and ICU admission type as either emergency (urgent or emergency admission to ICU) or elective. The ICU and hospital length of stay (LOS) was calculated only for those who survived the index ICU or hospital admission. The discharge home and 6-month mortality outcomes were calculated only for those who survived the index hospital admission, in order to represent the difference in the level of disability at the end of admission between groups, and the longer-term mortality of those who survived initial hospital admission, respectively. Differences in demographics and outcome data were calculated using Student’s t-test, Chi-squared test, unadjusted univariable logistic regression and adjusted multivariable logistic regression for significant confounding variables (adjusted for age, gender, and admission type). A p-value of less than 0.05 was considered statistically significant, and odds ratios are presented with 95% confidence intervals. Missing data was present for \<0.1% of all patient information, and any patient with missing data was excluded from that particular analysis. All analyses were performed using Python 3.6.9. This project is an output of Hack Aotearoa 2020, a health datathon organized by the University of Auckland and MIT Critical Data. This research received no specific grant from any funding agency in the public, commercial or not-for- profit sector. There are no competing interests to declare by any of the authors. # Results A total of 3,729 patients aged 16–89 years old were admitted to the Beth Israel Deaconess Medical Centre (BIDMC) ICU between 2001 and 2013 for an acute cerebrovascular event (stroke). Missing data was present for \<0.1% of participants during this period, and patients with missing data were excluded from any analysis for that domain. ## Baseline characteristics Overall, 23% of BIDMC ICU stroke admissions occurred over the weekend, and 77% occurred during the week. Stroke patients admitted to ICU over the weekend were more likely to have suffered haemorrhagic strokes vs ischemic strokes (60.6% vs 39.4%), with more similar distribution of haemorrhagic vs ischemic stroke for those admitted during the week (47.9% vs 52.1%). Those admitted on the weekend were younger (65.4 vs. 67.2 years old), and more likely to be male (55.7% vs. 53.6%) and unmarried (48.6% vs. 47.7%). Ethnic subcategory representation was comparable between weekend and weekday ICU admission, and those admitted to the ICU on the weekend were less likely to be publicly insured (61.5% vs. 65.4%) than those admitted on a weekday. Those admitted on the weekend were more likely to be emergency admissions, compared to those admitted during the week (8.7% vs. 90.2%). The OASIS severity of illness score was similar for patients admitted to ICU on the weekend and the weekday (32.5 vs. 32), and the lowest day-one Glasgow Coma Scale (GCS) score was also similar between both groups (12.6 vs. 12.9). ## Intensive care outcomes While ICU length of stay appeared similar for stroke patients admitted on the weekend compared to those admitted on a weekday (5.6 vs. 5.3 days), results were non-significant. The ICU mortality rates were significantly higher for stroke patients admitted on the weekend compared to a weekday (18.8% vs. 14.9%, p = 0.006). Unadjusted odds of ICU mortality were significantly higher for stroke patients admitted to intensive care over the weekend (OR 1.32, CI 1.08–1.61, p = 0.006). When adjusting for age, gender, and ICU admission type, odds remained significantly higher, however when adjusting for type of stroke, the higher likelihood of ICU mortality for those admitted over the weekend became non- significant (OR 1.17, CI 0.95–1.44, p = 0.13). ## Hospital outcomes While hospital length of stay appeared similar for stroke patients admitted on the weekend compared to those admitted on a weekday (11.2 days vs. 11.3 days), results were non-significant. The hospital mortality rate was significantly higher for stroke patients admitted to ICU over the weekend (26.6% vs. 20%, p\<0.001). Unadjusted odds of hospital mortality showed those admitted on the weekend had 1.45-times the odds of death compared to those admitted during the week (CI 1.22–1.73, p\<0.001). When adjusted for age, gender, ICU admission type and type of stroke, the odds of in-hospital death for those admitted to ICU over the weekend remained significantly higher (OR 1.31, CI 1.09–1.58, p = 0.004). ## Discharge outcomes The likelihood of being discharged to home for those admitted to ICU on the weekend compared to those admitted on a weekday was similar (37.1% vs. 40.1%, p = 0.10). Unadjusted odds of being discharged home for those admitted on the weekend compared to a weekday was similarly non-significant (OR 0.86, CI 0.72–1.03, p = 0.10), and remained non-significant when adjusting for age, gender, admission type, and type of stroke (OR 0.88, CI 0.72–1.06, p = 0.18). ## Long-term mortality outcomes Stroke patients admitted to the ICU on the weekend appeared to have no significantly higher risk of 6-month mortality (14.3% vs. 12.7%, p = 0.30). Unadjusted odds of 6-month mortality for those admitted on a weekend compared to a weekday were non-significant (OR 1.14, CI 0.89–1.48, p = 0.30). When adjusted for age, gender, admission type and type of stroke, the odds of longer-term mortality for those admitted on the weekend compared to a weekday remained non- significant (OR 1.12, CI 0.86–1.46, p = 0.41). # Discussion The major findings of this study are an initially higher ICU mortality for stroke patients admitted over the weekend that becomes non-significant when adjusting for type of stroke, and a higher in-hospital mortality thereafter which remains significantly higher when adjusting for illness severity and type of stroke on admission. There was no significant difference in length of ICU and hospital admission between groups in this study involving more than 3,700 patients. While differences in the level of disability at discharge (denoted as those well enough to be discharged back home after their hospital admission) and longer-term outcomes (denoted as 6-month mortality of the hospital survivors) were non-significant, patients admitted on the weekend did appear to fare more poorly in both measures. This suggests the need for a larger study of the longer-term implications of weekend intensive care admission after a severe stroke as the effect size may not be big enough to detect given the sample size. Although the weekend effect for patients admitted to intensive care has been previously reported as non-significant in older studies, a recent systematic- review and meta-analysis by Cavallazzi and colleagues concluded that patients admitted to intensive care over the weekend were at significantly higher risk of both ICU and in-hospital mortality across multiple diagnoses; partially consistent with our findings. This was posited to be due to in-hospital factors; lower levels of staffing and intensity of care over the weekend. However, this analysis studied a general ICU population, rather than a sub-population for whom timely assessment and expert management may be particularly important, and the implications of a weekend admission more exaggerated. More recently, a systematic-review and meta-analysis by Galloway and colleagues found that admission to ICU over a weekend was also associated with a significant increased odds of death, also partially consistent with our findings; attributing in- hospital factors (an absence of on-site intensivists over the weekend compared to a weekday) toward the mortality differential. Of note, while on-site specialists are present in the BIDMC ICU during both weekdays and weekends, a significant proportion of specialists and nurses in the ICU over the weekend usually work in different departments (i.e. ambulatory) during the week; assisting weekend ICU staffing only where needed. There are also fewer junior medical staff working in the ICU over the weekend. In addition to the in-hospital factors that may influence weekend mortality, pre-hospital factors like severity of illness have also been posited as influential. A 2019 systematic-review and meta-analysis by Chen and colleagues concluded that the significant increase in in-hospital weekend mortality was in part because patients admitted over the weekend were more severely ill. In the present study, pre-hospital illness severity measures (OASIS, GCS) were similar between weekend and weekday admissions and did not influence mortality outcomes. However, those admitted over the weekend were considerably more likely to have suffered haemorrhagic strokes; considered more severe than ischemic strokes and with previously reported higher rates of in-hospital mortality for those admitted over a weekend. Haemorrhagic strokes account for the vast minority of strokes, and the reason they were so disproportionately over-represented on weekend compared to weekday ICU admission in our cohort cannot be concluded by the present study. But when adjusting for type of stroke, higher odds of in-ICU mortality for those admitted over the weekend became non-significant; consistent with the notion that pre-hospital factors may play a particularly important role in the weekend mortality effect for those in ICU; in contrast to the suggested in-hospital contributors by Cavallazzi and colleagues and Galloway and colleagues (above). Importantly, our study demonstrated higher in-hospital mortality for those admitted to ICU over the weekend even when accounting for pre-hospital factors; contrasting recent stroke-specific findings: Inoue and colleagues’ analysis of the weekend effect for stroke patients found no significant increased risk of in-hospital mortality for those admitted to a stroke-ICU over the weekend, after adjusting for severity of illness. They studied ischemic stroke only, in an exclusively Japanese population, and with alternative acute illness severity and comorbidity measures (the modified Rankin Scale); which may in part explain the different findings of their study. A less recent 2010 study by Hoh and colleagues similarly described no significantly different adjusted in-hospital mortality for stroke patients admitted on the weekend (similarly contrasting our findings). However, while their study size was significantly larger, they did not include or adjust for any measures of severity of illness or co-morbidity status (which may have considerably altered their findings), and similarly only studied an ischemic-stroke population. Neither study examined the implications of weekend admission on longer-term mortality patterns. The effect of weekend admission to ICU for patients suffering severe stroke on in-hospital mortality appears significant for our study population, notably persisting even when adjusting for type of stroke (in contrast to previous studies). This outlines the potential effect of in-hospital factors on the ‘weekend mortality effect’, as alluded to by Zampieri and colleagues. Further, the non-significant higher odds of in-ICU mortality for those admitted over the weekend demonstrates the potential influence of pre-hospital factors on ICU mortality for weekend admissions. The present findings underline the particular need for urgent review and management of patients who suffer severe stroke of any kind, where delay may substantially affect outcomes more-so than for other patient sub-populations. Lower numbers of on-site senior staff, lower nurse-to- patient ratios, fewer standardized protocols, and subsequent delays to expert management, may have more serious ramifications for severe stroke patients, underlying a sub-population of critically ill patients who need particular care when admitted over the weekend. Our findings are relevant for all healthcare professionals working both in hospital, or on-call from home, responsible for such an at-risk patient population. ## Strengths and limitations This is the first study to analyse both the short-term and longer-term morbidity and mortality outcomes for stroke patients admitted to an intensive care unit over a weekend. Our database is large, inclusive of important information like severity of illness, and includes high-resolution patient information from a single center academic institution. Our analysis team comprised of clinicians, data scientists and statisticians, and the database used is publicly-available. This is congruent with an effort to move toward a more reproducible multidisciplinary, collaborative style of research. As with all retrospective database research, there are limitations to this study. Information bias exists; stroke-specific markers of disability (i.e. the NIH Stroke Scale) were not recorded, so surrogate markers (discharged to home) used instead. And despite being a large dataset, this was only a single-center study, which limits generalizability # Conclusion In this large study of a diverse US population, the effect of weekend ICU admission for severe stroke patients appears to be significant for in-hospital mortality, irrespective of severity of illness or type of stroke. There were no significant differences in adjusted ICU mortality, or in ICU or hospital length of stay for those admitted on a weekend. While longer-term morbidity and mortality findings were non-significant, patients admitted over the weekend did appear to fare more poorly than those admitted on a weekday, which may warrant further exploration. Our findings are relevant to healthcare professionals working either in hospital or on-call from home, responsible for caring for such an at-risk patient population. Other declarations: This project is an output of Hack Aotearoa 2020, a health datathon organized by the University of Auckland and MIT Critical Data. [^1]: The authors have declared that no competing interests exist.

# Introduction Dioxins, which refer to a family of structurally and chemically related polychlororinated dibenzo-*p*-dioxins (PCDDs) and dibenzofurans (PCDFs) are lipophilic chemicals resistant to degradation and categorized as persistent organic pollutants, with 2,3,7,8- tetrachlorodibenzo-p-dioxin (TCDD) being the most toxic dioxin. Dioxins are suspected of interfering with the endocrine systems of humans and wildlife – causing a broad spectrum of adverse effects including developmental and reproductive toxicity in the offspring of laboratory animals, and perhaps in humans. These disorders are of very high concern, because they occur at much lower doses than those causing wasting syndrome or carcinogenesis. Moreover, dioxins tend to accumulate in the food chain, essentially fatty food including breast milk, and may also cross the placental barrier. These data emphasize that foetuses and neonates are vulnerable populations. For example, it was recently demonstrated that breast-fed but not formula-fed sons from mothers exposed to dioxin after the accident in Seveso had permanently reduced sperm quality. In utero exposure to TCDD also impaired prostate development in many mammals including rats and mice. However, data on testicular development are debated. In rat females, it has also been shown that TCDD exposure in utero is associated with malformations of external genitalia, reduced fertility, and disruption of estrus cycles and inhibition of ovulation, . Most of the toxic effects of TCDD are mediated through the binding and activation of the aryl hydrocarbon receptor (AHR) and subsequent alteration of target gene expression. The classic battery of dioxin-responsive genes exhibit dioxin response elements in their promoter moiety and includes phase I and phase II enzymes of the detoxification machinery, such as the cytochrome P450 (Cyps) 1a1 and 1b1. However, a recent study pointed out that 65% of the gene expression responses elicited by TCDD do not involve direct AhR binding to a Xenobiotic Response Element (XRE). It suggested that expression of genes lacking a XRE element reflected indirect AhR-mediated signalling, and that such changes in expression could be attributed to latent secondary effects. It could as well suggest that non-consensus XRE element may confer TCDD inducibility as shown recently with PAI-1. These two mechanisms may well explain the pleiotropic nature of TCDD toxicity illustrated with the large panel of target genes identified that include for example genes involved in cellular growth, differentiation and inflammation. Global gene expression technology provides a comprehensive strategy whereby critical AHR-regulated genes apart from the classic battery of genes can be identified, and has been used to elucidate target pathways involved in the aetiology of TCDD and related compounds toxicity in liver and kidney. Using this strategy, we focused our study on developing gonads, testes and ovaries, in rats. The age of 5 days was chosen for males because this time period coincides with germ cell reentry into mitosis, the set-up of the spermatogonial program including stem cell self-renewal and the maturation of the somatic cell lineages, i.e. the Sertoli and Leydig cells. Regarding females, we concentrated on prepubertal period and specifically infantile period (from 7 to 20 days of age) when the ovary as well as the pituitary displayed an intense endocrine activity with high levels of estradiol and gonadotropins. Real-time PCR was further used not only to validate microarray data generated but also to gain more insight into the kinetics of regulation of the identified genes along with crucial developmental periods. In parallel, the expression level of the classical battery of detoxifying genes was surveyed. Overall, our results evidenced that both sex gonads responded differently to TCDD exposure with respect to enzymes of the detoxifying machinery. We illustrated that inflammatory pathway is one pathway activated by TCDD in gonads. We identified several new genes targeted by TCDD including Fgf13 in testis and one gene, Ptgds2/Hpgds regulated in the two sex gonads. # Materials and Methods ## Experimental Design Time pregnant Sprague-Dawley females of embryonic day 12 were purchased from Janvier’s Breeding (Le Genest, France). They were housed individually in plastic cages with food (Altromin 1310; Genestil, Royaucourt, France) and water provided ad libitum at 23°C and a 12∶12 photoperiod. Animals were randomly assigned to treatment groups. Dams were allowed 3-day acclimatization and were given one oral dose of 2,3,7,8-TCDD (ref ED-901-C) (LGC Promochem, Molsheim, France) 200 ng/kg body weight (bw) in sesame oil on embryonic day 15. Control animals received sesame oil. Pups were sacrificed by cervical dislocation under CO₂ anesthesia at various ages, at 3, 6, 10, 12, 14 and 25 days for females, and at 5, 28, 40, 67, 145 days for males. Ovaries and testes were dissected and snap-frozen. Pituitaries were collected from female pups of 3, 6, 12 and 14 days of age. In addition, livers were collected at 5 and 28 days for males and 3 and 6 days for females. All organs were kept at −80°C before use. Throughout the manuscript, treated organs refer to organs collected from animals born from TCDD-exposed dams. Control organs refer to organs collected from animals born from sesame oil exposed dams. Animals were housed and maintained according to published European communities’ guidelines (86/609/CEE) and all the performed experiments on animals were approved by the experimental animal committee of the Paris Rive Gauche site (Agreement A75-13-17, Centre National de la Recherche Scientifique, Paris 7 University, Paris, France). A detailed protocol of TCDD exposure and follow-up of the animals has been published. ## RNA Preparation and Microarray Analysis Changes in global gene expression induced by TCDD were analyzed in 5-day old testes and in 14-day old ovaries. In both cases, 3 rats treated in utero by TCDD were compared to 3 rats treated with sesame-oil vehicle. RNA extraction was performed with the RNeasy Mini RNA extraction kit (Qiagen, Courtaboeuf France). Microarray analysis has been performed in the genomic and microgenomic core facility profileXpert (Bron, France) as described previously, using a high- density oligonucleotide array (GeneChip Rat Genome 230 2.0 array, Affymetrix, Santa Clara, CA, USA), and following Affymetrix protocol (<http://www.affymetrix.com>). The arrays were read with a confocal laser (Genechip scanner 3000, Affymetrix). Then CEL files were generated using the Affymetrix GeneChip Command Console (AGCC) software 3.0. The obtained data were normalized with Affymetrix Expression Console software using MAS5 statistical algorithm. Normalized data were compared and filtered using Partek Genomic Suite software 6.5 (Partek Inc., St. Louis, MO, US). Microarray data are available in the GEO database under the number GSE32890. ## Analysis of microarray data A principal component analysis (PCA) of the complete list of genes present in the chip and two sample t-test were performed between TCDD samples and control samples for ovaries and for testes. Only genes showing a variation of at least 1.5-fold and a p-value less than 0.05 were retained. Then, a gene was considered as differentially expressed between groups only if the detected signal was above the background for at least one of the compared groups. David functional annotation clustering (<http://david.abcc.ncifcrf.gov/conversion.jsp>) was performed to identify enrichment in biological functions and pathways. Promoter sequences (650pb) of the regulated genes were extracted from Rat RGSC 3.4 assembly using BioMart, and firstly analyzed for pattern matching using Common TFs from the Genomatix software package (Genomatix software suite v2.5, München, Germany). Briefly, the sequences were scanned for matches to Transcription Factor binding sites and only matrices displaying an enrichment p-value \<0.05 were considered enriched in the promoters of interest compared to the rest of the vertebrate promoter sequences. ## Reverse Transcription and Real-time RT-PCR Reverse transcription was carried out with 250 ng of total RNA recovered from testes (of 5, 28, 40, 67, 145 days of age), ovaries (of 3, 6, 10, 12, 14, 25 days of age), pituitaries (of females aged of 3, 6, 12, 14 days of age) or livers (of 5 and 28 days for males and 3 and 6 days for females), 400 ng oligo- dT primers (Qiagen) and Superscript II reverse transcriptase (Invitrogen, France). Real-time RT-PCR (QRT-PCR) was performed in a LightCycler 480 instrument (Roche Diagnostics, Mannheim, Germany) using the LightCycler 480 SYBR Green I Master mix according to the manufacturer’s protocol. Primers used are listed in. They were tested before use for specificity and efficacy. Amplification conditions were as following: 10 min at 95°C followed by 40 cycles of denaturation (10 sec at 95°C), annealing (10 sec at 60°C), and extension (10 sec at 72°C) with single acquisition of fluorescence at the end of each extension step. The specificity of PCR products was confirmed by analysis of the melting curve and agarose gel electrophoresis. All samples were run in quadriplicate reactions (duplicate of two dilutions). Quantification of gene expression was performed using the Relative Quantification Software (Roche Diagnostics), and data were expressed as a ratio of target gene to the reference gene hypoxanthine phosphoribosyltransferase, HPRT. Statistical analyses were done using Statview 5.0 software package (SAS Institute Inc. Cary, NC 27513). Comparisons between treatments were made by one-way analysis of variance (ANOVA) followed by the post hoc Fisher PLSD test for multiple comparisons. A *p* value of less than 0.05 was considered significant. # Results ## Reproductive Parameters of the F1 Offspring A follow-up of the female offspring including body weight, fertility assessment and measurement of mRNA levels of some key genes involved in the endocrine function of the ovary during prepubertal period is provided. Except for body weight, no significant differences were obtained. Regarding the male offspring exposed in utero to TCDD, testicular and epididymal weight did not change. Testis histology was grossly normal so were testosterone levels throughout development. A precise description was provided previously. ## Global Analysis of Testes and Ovaries Transcriptomic Data The total amount of genes expressed in ovaries and testes was approximately of 65% of the rat genome, indicating that an average of 20,000 genes is expressed in the gonads. PCA analysis did not allow segregating genes between treated- testes and control testes, or between treated-ovaries and control ovaries. It indicated that the sample to sample variation within the group (male or female) was bigger than the variation due to the treatment (not shown). A two sample t-test was next performed and a list of 113 and 56 differentially expressed genes showing a variation of 1.5-fold and a p-value less than 0.05 came out in ovary and testis, respectively. Of the 113 genes responding to TCDD in ovaries, 38 were up-regulated and 75 down-regulated. In testes, 27 were up-regulated and 29 down-regulated. About, half of them are not fully characterized and are identified by their Affymetrix probeset ID. Some of them are transcribed locus or expressed sequence tags (ESTs). A sub-list of characterized genes differentially expressed in response to TCDD, in testes and in ovaries is given in and, respectively. We also identified a total of 7 genes common to testes and ovaries if cross-comparing the list of regulated genes that exhibited at least a 1.5 fold change over its respective control. There were Ahr repressor (Ahrr), prostaglandin D2 synthase 2, hematopoietic (Ptgds2/Hpgds), the chemokines Rantes/Ccl5 and Pf4/Cxcl4, and 3 uncharacterized genes which were not further studied. David functional annotation clustering indicated one cluster with enrichment score of 1.61 and a p-value of 4.9.10⁻³ using the total list of testis genes. It corresponded to immune and defense response with 6 transcripts including Ccl5 and Pf4. No cluster came out if only the list of the up or down- regulated genes was considered. Using the list of the up-regulated genes in the ovary, 4 clusters came out with enrichment score comprised between 1.56 and 2.43 and corresponding to immune response, positive regulation of response to stimulus, chemotaxis and chemokine activity, and response to steroid hormone stimulus. In addition, KEGG pathway pointed to the chemokine signaling pathway associating Ccl5, Ccl6 and Pf4. A single cluster with an enrichment score of 1.4 and corresponding to behavior came out when uploading the list of down-regulated genes. ## Expression of Genes Belonging to the Classic Battery of TCDD Target Genes To extend microarray data, we first focused on Ahrr strongly induced in both gonads, and other genes belonging to the TCDD-inducible Ahr gene battery. The TCDD battery includes in addition to Ahrr which acts as a dominant negative factor to repress Ahr induced signalling pathways, four phase I xenobiotic metabolizing enzymes, i.e., Cyp1b1, Cyp1a1, Cyp1a2, and Cyp2s1, and four phase II xenobiotic metabolizing enzymes, i.e., NADP(H) quinone oxidoreductase (Nqo1), glutathione transferase a1 (Gsta1), cytosolic aldehyde dehydrogenase-3 (Aldh-3a1), and UDP glucuronosyltransferase 1a6 (Ugt1a6). Three genes, Cyp1a1, Cyp1a2 and Aldh-3a1 were not found at detectable levels in the gonads. Two genes appeared to be constitutively expressed (Cyp2s1 and Gsta1) or weakly enhanced (Nqo1 and Ugt1a6) after TCDD treatment. Cyp1b1 was significantly up-regulated in ovaries (FC 1.63) but not in testes (FC 1.22 with a *p*-value of 0.36). Finally, gene coding Ahrr was the most up-regulated gene in both gonads of treated animals (FC 4.2 in ovaries and 3.34 in testes). To validate and extend the microarray results, we analyzed the expression of Ahrr and Cyp1b1 as well as Cyp1a1 and Nqo1 using real-time PCR. This was done not only in ovaries at 14 days and in testes at 5 days but also during a period of time extending from 3–25 days in females and 5–145 days in males to cover critical developmental steps in both gonads. Real-time PCR confirmed the findings of the microarray in gonads. For example in testes, Ahrr was significantly enhanced at 5 days of age although the fold-change was lower than the one observed using microarray. In addition, none of the metabolizing enzymes had their gene expression levels altered after TCDD treatment. Noticeably, Cyp1a1 gene was not expressed in testis, either control or TCDD-treated (not shown). Cyp1b1 and Nqo1 expression levels increased as a function of age from 5 to 145 days. In contrast, neonate ovaries recovered from the same litters were strongly responsive to TCDD exposure. Expression levels of Ahrr were maximally up-regulated at 3 and 6 days. A significant enhancement was still observed until 14 days, consistent with the microarray data. No effect was observed in ovaries of 25 days. Of note, Ahrr gene expression levels were almost undetectable in control ovaries. Cyp1a1, the hallmark of TCDD-inducible Ahr gene, although not present in control ovaries was highly induced in TCDD-treated ovaries of 3 to 10 days, peaking at 10 days. No expression was detected at 14 days consistent with the microarray data. Gene expression levels of Cyp1b1 and Nqo1 were at detectable levels in control ovaries and roughly constant from 3 to 25 days of age. Treated ovaries had enhanced levels of Cyp1b1 and Nqo1 from 3 to 14 days, also in agreement with the microarray data. No differences were observed at 25 days. We also studied an additional endocrine organ, the pituitary, and the liver as the primary detoxifying organ. Data synthesized on, indicate a strong and significant enhancement of Ahrr, Cyp1a1, Cyp1b1 and Nqo1 at 6 days of age in pituitary of TCDD-treated animals. Although significant, induction was less pronounced at 14 days of age. As expected, RNA from both male (5 days old) and female (6 days of age) liver showed increased Ahrr and Nqo1 expression levels (data not shown) in treated animals consistent with the induction in Cyp1a1 mRNA levels reported previously. ## Expression of Chemokines in Response to TCDD in Gonads but also in Pituitary and Liver Consistently with our previous data, we observed down-regulation of Ccl5 and upregulation of Pf4/Cxcl4 in the testes of 5-day old rats of TCDD-treated dams. In females, microarray studies indicated that both Ccl5 and Pf4/Cxcl4 were significantly up regulated 2.5- and 1.57-fold, respectively. In addition, KEGG pathway pointed to the chemokine signaling pathway associating Ccl5, Ccl6 and Pf4/Cxcl4, in ovary and one enriched cluster comprised of Ccl5 and Pf4/Cxcl4 in testis. To define if chemokines could be targeted by TCDD in other organs, real- time PCR was performed with RNA from pituitary and liver. We found that pituitaries recovered from 14-day-old female rats from TCDD-treated dams had increased Ccl5 (3.26 fold-change, n = 4; p\<0.05) and Pf4/Cxcl4 (1.81 fold- change, n = 4; p\<0.05) versus time-matched controls. Ccl5 but not Pf4/Cxcl4 was also up-regulated (2.12 fold-change, n = 4; p\<0.05) in the liver recovered from 28-day old male rats (data not shown). ## TCDD-regulated Genes in Testis and Comparison with Two Endocrine Organs (Ovary and Pituitary) and Liver Microarray data were further exploited by real-time PCR to analyse the testis response to TCDD exposure. Genes selected included Art2b, Gzmf/Nrkp7, and Fgf13 because they were among the most regulated genes in testis, and Ptgds2/Hpgds because it was regulated in both sex gonads. We observed up-regulation of Gzmf/Nrkp7, Art2b, Ptgds2/Hpgds and down-regulation of Fgf13 (p\<0.05) at 5 but not at 28 or 67 days. In ovaries, Ptgds2/Hpgds was present at all ages investigated from 3 to 25 days. Enhanced expression levels of Ptgds2/Hpgds were observed in the ovaries of rats from TCDD-treated dams at 6, 12 and 14 days but not at 3 and 25 days of age. Fgf13 was present in neonate ovaries (3 and 6 days of age) but its expression levels did not change in treated ovaries. Art2b and Gzmf/Nrkp7 were neither expressed in the ovary at 6, 10 and 14 days of age, nor induced in the ovaries of rats from TCDD-treated dams at these age-time points (not shown). In pituitaries, Gzmf/Nrkp7 and Art2b were present at 3 and 12 days of age and organs from treated animals exhibited levels higher than 2-fold over control levels. Fgf13 was also present in pituitaries collected from 3 and 12 days old animals, and its expression levels were significantly decreased (a 20% decrease) in treated organs. Ptgds2/Hpgds was expressed in pituitary but not TCDD- regulated. Finally, real-time PCR using liver from 3 and 6 day-old animals revealed transcripts for Gzmf/Nrkp7, Art2b and Ptgds2/Hpgds but not Fgf13, with TCDD up-regulation of Gzmf/Nrkp7 and Art2b (p\<0.05) at 3 but not at 6 days of age. Ptgds2/Hpgds was not targeted. recapitulates data. ## Identification of DNA Matrices, Potential Sites for Transcription Factors To find out whether certain transcriptional factor binding site(s) were enriched in the TCDD-regulated genes identified in this study, we selected 650bp promoter of each gene, and searched for DNA matrices families using Genomatix promoter analysis. A total of 25 promoter sequences for testes regulated genes and 54 promoter sequences for ovaries regulated genes could be processed. We observed that the DNA matrices family ETSF was consistently identified in the promoters of all 25 TCDD regulated genes in testes and the 54 regulated in ovaries. We also identified 10 other overrepresented matrices families’ specific to either testes or ovaries which may account for the gender differences that we detected in this study. # Discussion The objective of our project was to determine the impact of an in utero exposure to 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD) on reproductive function of male and female offspring in the rat with a special emphasis on the immature period, in conditions in which reproductive parameters are grossly normal both in males and in females (this study). To this end, animals were exposed during gestation to low doses to avoid collateral toxic effects. We also considered the TCDD half-life of 3 weeks in rodents, and conclusions published indicating that the majority of occurrences of TCDD in offspring of dosed dams arise from lactational transfer of TCDD. In addition, given the unexpected microarray results highlighting gender difference responses; we developed a real time PCR approach on gonads at various developmental periods of interest. On the one hand, this approach permitted to extend our study on male and female gonad development following an in utero exposure to TCDD. On the other hand, this approach allowed comparing common stages in both sexes (specifically, 5 and 28 days in testes, and 6 and 25 days in ovaries). The microarray technology is a powerful method allowing full gene analyses of different samples. It is especially fruitful when scarce differences are expected. In the present study, less than 1% of the expressed genes in gonads were found to be altered following embryonic TCDD exposure. Specifically, we identified a total of 113 genes in ovaries and 56 genes in testes differentially regulated by the treatment over the 20,000 genes found expressed in the ovary and testis, respectively. The low number of regulated genes found in testis versus the ovary probably suggested that TCDD may have less deleterious effects in testes than in ovaries. This could result from the lack of response of the TCDD battery of detoxifying genes in testis by comparison with the neonate ovary. Indeed, TCDD only induced gene expression of Ahrr in neonate testes while in ovaries, Cyp1a1, Cyp1b1, and Nqo1 exhibited a very high up-regulation in addition to Ahrr. Ahrr functions as a naturally occurring dominant-negative factor. Hence, Ahrr is an important determinant of tissue specific responsiveness to TCDD, and strong evidences have been reported on an inverse relationship between Ahrr expression and sensitivity to induction of xenobiotic- metabolizing enzymes caused by TCDD. This is coherent with elevated constitutive levels of Ahrr in testis. The data presented herein i.e., high expression of Ahrr and no detectable Cyp1a1, Cyp1b1 and Nqo1 induction in testis are in keeping with these observations. Nonetheless, the presence of Ahrr and its up-regulation in TCDD treated testes indicate that the Ahr canonical pathway is active in testis even though the role played by Ahr in testis is misunderstood. Impairment of the urogenital sinus is the major phenotype in Ahr^−/− male mice. Testis alteration was also reported in aged Ahr^−/− mice with reduced testosterone production and sperm numbers. It is of interest that enzymes of the TCDD battery of inducible genes expressed in testis including Cyp1b1 and Nqo1 had their expression levels increasing as a function of time, from infancy to adulthood. It may suggest that these enzymes, which are confined to Leydig cells in testis , exert a physiological role during development in relation with the endocrine status of the animal. Our data also illustrated that in addition to liver, which is a primary detoxification organ, endocrine organs such as ovary and pituitary displayed an up-regulation of the expression of genes coding enzymes of the detoxifying machinery. For example, in both ovary and pituitary there was an induction of Cyp1a1 which is the hallmark of TCDD exposure. Nonetheless, Cyp1a1 induction was no longer observed at 12 days of age in the ovary while still detected at 14 days in pituitaries (and 25 days, not shown) and 28 days in livers, illustrating organ specificity in the timing of the response. Indeed, the other detoxifying genes studied in ovary, i.e., Ahrr, Cyp1b1 and Nqo1 were still enhanced at 12 and 14 days of age. Together, these data are in keeping with previous studies demonstrating enhanced Ahrr and Cyp1a1 in the pituitaries of male rats exposed to an acute-dose of TCDD at the adult age. Regarding the ovary, evidences have been brought indicating enhanced Cyp1a1 in response to dioxin. However, to our knowledge, this is the first report illustrating the TCDD-induction of Ahrr in ovary. Present data extend our previous study reporting alteration of chemokines in the testes of TCDD-treated rats, with up-regulation of Cxcl4 and down-regulation of Ccl5, and other studies pointing out that various chemokines were targeted by TCDD exposure including Ccl5 in a model of endometriosis, Ccl1, Ccl2. In this study, the chemokine pathway associating Ccl5, Pf4/Cxcl4 and Ccl6 was induced in TCDD-treated ovaries. Therefore, in addition to growth factors and cytokines, chemokines might be added to the list of the TCDD-targeted genes, extending the notion that TCDD interacts with the inflammatory pathways. Our results also evidenced a sex specific response of gonads to the TCDD exposure with Fgf13, Art2b, and Gzmf regulated in testes but not in ovaries, and, a gonad gene expression signature with Ptgds2/Hpgds. Indeed, this study demonstrated that Ptgds2/Hpgds, although expressed in various tissues, was only regulated by TCDD in the gonads. Interestingly, none of these genes, and Ccl5 and Pf4/Cxcl4 exhibited-XRE elements in the 2,0000 bp upstream of the transcription sites, in rats (not shown). This situation contrasted with the mouse orthologs of Art2b, Fgf13, Ptgds2/Hpgds and Ccl5 having XRE elements in the 2,000 bp upstream of the transcription sites (<http://drgap.nies.go.jp/pub/page/element>). Nonetheless, these genes may be direct targets if considering that only 1/3 of the gene expression responses elicited by TCDD involves direct AhR binding to a XRE. It may also indicate that the TCDD-induced enhancement of these genes is secondary to a primary event, which remains to be defined. It may well be the case for Ptgds2/Hpgds because its induction in ovary was not immediate and could not be detected at 3 days of age, an age at which all the detoxifying genes have been shown to be deregulated. We cannot ascertain that this situation is female-specific because we did not recover males younger than 5 days of age. Ptgds2/Hpgds is a cytosolic protein responsible for the biosynthesis of prostaglandin D2 (PGD2) in immune and inflammatory cells, being widely distributed in antigen presenting cells. Interestingly, very recent data have brought new insight into involvement of Ptgds2/Hpgds in differentiation and function of gonads. Ptgds2/Hpgds is expressed in the early embryonic gonad in both sexes and participates to the initial nuclear translocation of the Sox9 protein triggering Sertoli cell differentiation in males. It is also expressed in the adult ovary where it indirectly participates to the regulation of progesterone secretion. Therefore, more studies need to be done regarding to Ptgds2/Hpgds because of the stimulatory effect of PGD2 on steroidogenesis regulation. The 3 other genes, Fgf13, and Gzmf and Art2b, were among the most differentially expressed genes in testis, in response to TCDD. Fgf13 (also called Fgf homologous factor 2) was first shown to be expressed in human fetal and adult brain and in adult kidney. Fibroblast Growth Factors (FGFs) form a large family of conserved signaling proteins with essential developmental functions in organ patterning and morphogenesis. Fgf13 belongs to the intracrine FGF family indicating that it is not secreted and that it functions in an FGF receptor- independent manner. In fetal mice, Fgf13 expression was detected within the mesonephros of both sexes at 12.5 embryonic day, and restricted to testis by embryonic day 13.5. In the present study, Fgf13 showed high inhibition in testis microarray and data were validated by real time PCR with gene expression levels halved in testis from TCDD-treated rats. Interestingly, its expression was not regulated in ovaries while down- regulated in pituitary; it may suggest that Fgf13 is under the regulation of transcription factors commonly expressed in testis and pituitary. There is little information regarding Gzmf and Art2b and their possible regulation by TCDD. Granzyme genes are serine proteinases previously shown to be implicated in tissue remodelling in the placenta and in the testis. Art2b is an ADP-ribosyl transferase gene, and 5 Arts have been described in the mammalian genome. ADP ribosylation is a reversible post-translational modification that can be used as a mechanism to regulate endogenous functions. The full significance of the alteration of these genes in the testis and pituitary, but not in the ovary recovered from siblings’ warrants further investigation. Interestingly, through cross-comparison between the lists of DNA matrices present in the proximal promoters of 25 genes in testes and 54 genes in ovaries, all regulated by TCDD, we extracted potential families of transcription factors allowing regulation by TCDD in testis and/or ovary. However, the importance of these findings remains to be determined. Overall, our results evidenced that male and female gonads responded differently to TCDD exposure showing, for example, the induction of the canonic battery of genes coding enzymes of the detoxifying machinery in ovaries but not in testes. We illustrated that inflammatory pathway is one of the pathways targeted by TCDD in gonads. Finally, we identified several new genes targeted by TCDD, including Fgf13 in testis and one gene, Ptgds2/Hpgds targeted in testis and ovary. # Supporting Information We thank the profileXpert transcriptomic platform (Bron, France), and Séverine Croze and Nicolas Nazaret for helpful assistance in microarray experiments. [^1]: Conceived and designed the experiments: BLMB SM. Performed the experiments: SM DR MI RW CD EM. Analyzed the data: SM BLMB. Wrote the paper: BLMB SM HV JCT. [^2]: Current address: MRC Centre for Reproductive Health, University of Edinburgh, The Queens Medical Research Institute, Edinburgh, United Kingdom [^3]: The authors declare that DR received funding from Shering-Plough. There are no patents, products in development or marketed products. This does not alter the authors' adherence to all the PLoS ONE policies on sharing data and materials.

# 1 Introduction According to the World Health Organization (WHO), 80% of people with disabilities live in low- and middle-income countries (LMICs). Lower-limb amputees in LMICs, which number is expected to increase due to the increasing diabetes prevalence globally, face two challenges: the limited access to rehabilitation services and the low availability of prosthetic components which are appropriate to the context of use. Particularly, prosthetic feet (further described as "feet" for ease of reading) with advanced biomechanical features (i.e., inversion/eversion, plantar/dorsal flexion, energy storage and release), also known as "dynamic feet", are usually made of expensive materials such as autoclaved carbon fiber prepregs that make them out of reach to a majority of people living in LMICs. Therefore, despite being needed for patients with higher mobility potentials, there is a lack of affordable dynamic feet suited for low- income countries. Some commercial feet (e.g., Blatchford SuperSACH foot, Jaipur foot) try to address this need but have limited functionality and do not allow for activity level 3 (ambulation with variable cadence, ability to traverse most environmental barriers) which is expected from dynamic feet. Activity levels, also known as K-levels, range from 0 (no ambulation ability) to 4 (exceeds basic ambulation skills) in the US Medicare classification. The Niagara Foot also targets the same audience and is suitable for level 3 users. Unfortunately, this foot only fits a limited range of patients due to its high profile (higher than a SACH foot, not suited to patients with longer stumps) and limited sizes available for both the keel and the cosmesis. Recent studies proposed innovative solutions, but the design process rarely included optimizing the foot production costs at each step of the conception, nor did they include a functional evaluation. As a major actor in the rehabilitation of lower-limb amputees in conflict- affected areas and LMICs, the International Committee of the Red Cross (ICRC) aims to provide affordable prostheses to the highest number of beneficiaries. Hence, through its Physical Rehabilitation Programme (PRP), the ICRC developed the Polypropylene (PP) Technology, a well-adapted and reliable solution for the specific context of low-resource settings. Over the past decades, the ICRC has fitted lower-limb amputees with CR-SACH feet (CR Equipements SA, CH), a cost- effective, solid-ankle cushion-heel foot suited for mobility levels 1 and 2 (K1–K2) as it features limited biomechanical properties. According to the ICRC, no K3 feet on the market currently meet their needs in terms of price, profile height, durability, and appearance. Therefore, this study aimed to design, manufacture, and test a dynamic foot according to the following specifications: (1) The total manufacturing cost should be maximum 100 USD. (2) The foot must be compatible with the PP Technology (i.e., can be bolted to the prosthetic leg through the foot arch). (3) The targeted performance should meet the Medicare level K3, which corresponds to WHO’s International Classification of Functioning, Disability and Health code d4602, and take into account users weight range. (4) The mechanical performance should comply with the standard ISO 10328:2016 and relevant P levels (the prototype foot targeted maximum 100kg users, corresponding to testing level P5). (5) The foot must comprise a fully-encapsulated keel in a real-looking non- removable cosmetic shell with a split toe to ensure durability (i.e., the cosmetic shell acts both as a protection against environmental and wear factors and to decrease the stigma of disability); with external dimensions, especially the profile height, of the foot being similar to the ones of the CR-SACH to fit the maximum users possible (stump length). # 2 Materials and methods ## 2.1 Project framework The project consisted of two main phases: (1) understand the needs and set the objectives, and (2) design a foot according to the requirements set in (1). Each phase was divided into inter-related sub-tasks, which were, in some cases, repeated several times within the course of the study. Based on ICRC’s needs, an analysis of the value chain and the future foot’s life cycle was carried out using EssentialTech’s (EPFL) methodological approach. This method clearly defined the objectives of the project and identified the resources required. Hence, a team of three EPFL laboratories was set up to face the challenges related to biomechanics and gait analysis (LMAM), mechanical design and simulation (LMAF), and materials science, composites processing and technical cost analysis (LPAC). During the project’s course, this approach, proposed by EssentialTech, continually adjusted the research efforts to the needs of the beneficiaries, and other stakeholders, in close collaboration with the ICRC physical rehabilitation team. To gain insight into the mechanical response of existing feet, we carried out a set of in-lab gait analysis tests and mechanical characterizations. Four feet were selected based on their Medicare classification (K-level) and their range of applications. These were: P_SACH (CR-SACH, CR Equipements SA, CH) a solid ankle cushion heel foot, not dynamic, activity level K1-K2, low profile, polymer keel encapsulated in foam and widely used for humanitarian applications; P_{K3_C} (Model 2, Niagara, CA) a dynamic foot, activity level K2-K3, high profile, polymer keel in cosmesis and designed for humanitarian purposes; P_{K3_E} (1D35, Ottobock, DE) a dynamic foot, activity level K2-K3, low profile, short glass fiber-reinforced composite keel encapsulated in a foam; P_K4 (Vari-Flex, Össur, IS) a dynamic foot, activity level K3-K4, high profile, long carbon fiber-reinforced composite keel in a cosmesis. The results from this characterization process were used to set the mechanical requirements of the new foot. After the objectives were defined during the first phase, the design of the prototype foot (P_PRO) was started. The design of P_PRO, based on Finite Element (FE) simulation, used the feedback from the in-lab biomechanical (i.e., gait analysis, load vs. pitch angle) and mechanical (i.e., reaction load & moment vs. deflection) evaluations to improve the model. These simulations also influenced the materials selection process, which identified candidate materials based on their cost and mechanical properties. Several design iterations were required to obtain a satisfying version of P_PRO that would meet all requirements and objectives. This final version was tested in Vietnam with actual prosthetic users and provided valuable feedback for future improvements. ## 2.2 In-lab characterization of commercial prostheses ### 2.2.1 Protocol In total, 13 non-amputee volunteers (age: 28 ± 8 years; weight: 74.6 ± 7.6 kg; size: 174.8 ± 6.1 cm) participated in the study. Each participant wore a pair of 2.4kg orthotic boots (Body Armor Walker 2, DARCO International, USA), similar to those previously validated for gait analysis. The participants performed a protocol composed of the following tasks: (1) alignment of the prosthetic system, (2) habituation period (minimum 15min), (3) sensor calibration, (4) level walking, (5) stairs climbing and (6) ramp climbing both frontally and laterally. These tasks were repeated for each foot (i.e., P_SACH, P_{K3_C}, P_{K3_E}, and P_K4) with the feet selected in a randomized order for each participant. A certified orthotist/prosthetist performed the alignment of the prosthetic system, and the order of the feet was randomized for each participant. Throughout the course of the project, three iterations of the biomechanical tests were performed: during the first iteration, the four commercial feet were involved, while in the second and third tests, different versions of P_PRO were evaluated. The protocol has been approved by EPFL’s ethical committee (HREC 001–2017) and was designed in agreement with the Declaration of Helsinki. Written informed consent was obtained from all the participants prior to the measurements. ### 2.2.2 Measurement systems The kinematics and kinetics of the foot were recorded using several measurement systems. A ground-integrated force plate (Kistler, CH) measured the three- dimensional ground reaction forces (GRF) at 200 Hz during level walking. A stereo-photogrammetric motion tracking system with 11 cameras (Vicon, UK) recorded at 200 Hz the position of 20 reflective markers affixed on the prostheses. One inertial measurement unit (Physilog 4, GaitUp, CH) was placed on the dorsum of each foot and set to measure the acceleration and angular velocity at 200 Hz. The data from these units were used to empirically confirm that the previously developed algorithms for sound leg gait analysis were functioning on the prosthetic foot and could be used for field tests. Pressure insoles (Pedar, Novel, DE) recorded the plantar pressure at the interface between the prosthetic foot and the sole of the shoe at 100 Hz. All measurement systems were electronically synchronized using a 5V trigger impulse. ### 2.2.3 Gait analysis Standing still posture measurements were used to define the shank, fore and rear foot frames with the vertical axis perpendicular to the ground surface and the longitudinal axis parallel to the line connecting the m_mid,fore and m_mid,_rear markers. We defined the pitch angle (*α*_*pitch*) as the sagittal plane angle between the longitudinal axis of the rear-foot segment and the ground. We also defined the flexion angle (*α*_*flexion*) as the sagittal plane angle between the fore and rear foot frames. The stance phases during gait were detected based on the vertical GRF (vGRF) obtained from the force plate. For each foot, we computed the mean values of the vGRF, *α*_*pitch*, and *α*_*flexion* for all the steps of each trial, and then calculated the mean ± STD over all subjects. However, we used the median, and the interquartile range (IQR) as inter-steps statistics (i.e., calculated over all the steps regardless of the subject) as histograms and the quantile-quantile plots (Q-Q plots) suggested a non-Gaussian distribution. To further characterize the feet, their behavior during the stance phase was assessed at 30%, 50%, and 100% of body weight (BW) during the loading and pushing phase, respectively. For each foot, the values of the average angles (*α*_*pitch* and *α*_*flexion**)* were estimated at the three *Loading*, and the three *Pushing* times over all the steps and median ± IQR was then calculated over all subjects. Since we considered P_K4 as the foot with the highest performance, a pairwise distribution difference between P_K4 and the other feet was calculated using the non-parametric Kruskal-Wallis test. ## 2.3 Design, materials, and simulation ### 2.3.1 Concept design We divided the design of the keel into three elements: a blade, an ankle, and a filling foam ((iv), (iii), (ii)). We then encapsulated these elements into the CR-SACH cosmetic shell ((i)). The foot is connected to the rest of the prosthetic leg with a removable M10 bolt (; (vii)) passing through a 20 mm hole at the foot arch. The design of the keel was based on three main criteria: cost, structural integrity (ISO 10328:2016), and the targeted performance (d4602) for an 80 ± 10kg patient using a 26 cm size foot. Also, the heel was designed 10 mm higher than the forefoot to fit the cosmetic shape. This design took into account the strength limits and stiffness of each part and was optimized to exhibit the best deflection vs. pitch angle response based on reference measurements obtained with commercial dynamic prostheses. Since the ankle must allow high elastic deformation during the forefoot loading case, we opted for a “C” shape design to maximize the bending length. We added a security system to create a mechanical contact between the upper and lower part of the ankle to avoid an overload failure. The blade was set to be mechanically constrained during the gait cycle; it works as a beam on the heel and the forefoot. The strength and the elastic behavior of the blade were optimized following several set targets: minimal use of carbon fiber reinforced material to limit cost (the target cost of this part was 35 USD), geometrical constraints due to the presence of the split toe, flexural fatigue resistance of 2 million cycles at 750 MPa stress conditions, and a flexural modulus in the range defined by the observation of commercial feet, to 100 GPa. Mainly, three parameters were explored: the general blade shape, thickness, and carbon fibers orientation in the layup (as well as the type of carbon fiber with good price/intermediate modulus and high strength). Both the C-shaped ankle and the blade feature a slit in the forefoot to allow eversion and inversion up to 15°. Furthermore, the foam had three functions: (1) filling the volume inside the foot to stabilize the outer cosmetic shell, (2) preventing the penetration of liquids or dirt while being light, and (3), when the foam reaches its compaction point, it also limits the deformation between the blade and the ankle. Hence, changing the foam density allowed to adjust the overall foot rigidity based on the user’s target weight. Finally, we further reduced the costs by decreasing the number of items to be produced; the parts were symmetric and could therefore, be encapsulated in either left or right cosmetic shells. To respond to a various range of patient characteristics, the parts have been designed in 5 different sizes 22, 24, 25, 26, and 28. ### 2.3.2 Materials and process selection methodology Different materials were screened for the three elements of the keel, namely the blade, ankle, and foam parts, based on the mechanical property requirements and possible processing methods guided by the cost and design constraints. A manufacturing cost modeling was performed based on the activity-based technical cost model described by Wakeman et al. and previously implemented in industrial composite manufacturing projects at EPFL. This methodology takes into consideration all material costs, energy costs, labor costs, machine depreciations and various practical aspects get to a realistic cost assessment. Additional details are provided in the supplementary information on the cost analysis and various materials tested in this research, scalability for different feet sizes and stiffness requirements were taken into account for the materials and processes selection, together with the targeted volume of production of the parts (in the range of 5000 parts per year per size). Specimens (100 mm long, 10 mm wide, and thickness varying from 4–4.2 mm) of the ankle and blade materials were prepared and tested for fatigue performance using a mechanical testing machine (ElectroForce 3400 from BOSE, USA). Three-point bending loading mode was used with a span to length ratio of 16:1 and at 10 Hz frequency for the ankle materials and 5 Hz for the ankle and blade materials. Compression tests at 2 Hz and 70% deformation were carried out on cylindrical filling foam specimens (13 mm diameter and 10 mm height) to screen suitable candidates for the foam part and to provide the constitutive laws for FE analysis. The cosmetic shell was made of thermoplastic polyurethane (TPU) based foam formulations similar to and compatible with the filling foam material. ### 2.3.3 Mechanical simulation Non-linear FE analysis was carried out, in which all parts were simulated with different material properties with their relevant constitutive laws. Materials properties were characterized using quasi-static FE models reproducing compression and three-point bending tests to adjust material parameters until the experimental and numerical results matched. The cosmetic shell material was simulated using a polynomial hyperelastic model, which fitted the uniaxial test data carried out in the laboratory. The foam part was simulated using Ogden’s hyper foam potential where the model coefficient were identified from experimental compression test data. Continuous fiber composites (C.F.C) used for the blade were simulated using properties of one of the promising candidate materials, Gurit SE70 carbon-epoxy composite. Long Fiber Thermoplastic (LFT) materials used for the ankle were simulated using an isotropic elastic modulus of E = 10.5GPa. The foot stiffness and strength characterizations were performed using a quasi- static FE model reproducing a compression test on the entire foot assembly to evaluate the vertical load-deflection characteristics and ensure the integrity of the foot for different loading cases including ultimate (1) and fatigue (2) loads required to comply to the standard ISO 10328:2016. The ankle, the foam, and the cosmetic shell were meshed using quadratic solid tetrahedron elements. The blade, represented by a composite shell with a variable thickness, was meshed using a quadratic shell element. The final FE characteristic element size (\~5mm) was determined using a mesh convergence study on the load-displacement response. The FE assembly was first oriented to model the desired loading angle and then clamped through a reference point, which was coupled to the upper horizontal surface of the prosthesis interface. The vertical displacement of a rigid horizontal plate representing the ground was simulated as in the experimental setup. Frictionless hard contact was used to model the foot-ground interaction. The stiffness of the prototype was optimized manually with incremental modification of different variables in the CAD & FEA model, to reach a similar level of deflection and reaction moment (top of the ankle part) compared to the selected reference commercial feet for different stance phases (pitch angles) and load levels corresponding to gait loads (as depicted). The optimization variables were the key ankle geometry dimensions such as: the skins and the ribs thicknesses; the foam type, the composite layup, and the blade’s thickness distribution. The design constraints were the manufacturing limitations, such as, available foam, composite material grades and injection molding requirements. The position of the injection nozzle and the flow cross section surface have been optimized with the manufacturing partner in a way to avoid pressure drops before the end of the injection (molding flow simulation). Finally the design satisfied the material strength criteria for both fatigue (2M cycles) and static overloading. ## 2.4 Prototype manufacturing The prototype was produced in a four steps process which involved: (1) injection molding of the ankle part, (2) over-molding of the filling foam onto the ankle within a mold, (3) production of the blade by autoclave processing, followed by CNC machining and assembly to the ankle by fasteners, and (4) cosmetic shell molding over the keel-blade-foam assembly for complete encapsulation. The ankle, foam, and cosmetic shell injection molding were performed with an injection molding machine (Arburg Golden edition 470C, Germany) within an industrial context, based on the process recommendations from the material suppliers and the materials characterization completed in the EPFL laboratories. The composite blade prototypes were manufactured at EPFL, reproducing industrial conditions. ## 2.5 Mechanical compression tests Feet stiffness and hysteresis were quantified and evaluated for reproducibility. Tests were performed using an axial-torsion servo-hydraulic mechanical testing system (MTS 809, USA) equipped with a 10kN load-cell, at a constant load rate of 800N/min. The vertical displacement and forces data were registered at 0.5Hz. Five cycles of load (800 N) and unload (10 N) were set using a sinusoidal signal. The feet were neutrally aligned on the frontal plane and tilted at different pitch angles: -30°, -20°, -10°, -5°, 0°, +5°, +10°, +15°, +20°, with a mechanical system (V). We considered a negative pitch angle for the forefoot loading case and a positive pitch angle during heel loading. To ensure a purely vertical force recording, hence without shear forces at the plantar surface, we placed a low friction linear bearing guide (, II) with a Teflon tape under the foot. We tested P_SACH, P_{K3_E}, P_K4, and P_PRO in such a setting and compared their stiffness. These recordings were then compared with the nominal vertical ground reaction forces (NvGRF) observed during the in-lab gait analysis. Four strain gauges were placed with 2 Wheatstone bridges (IV) to measure My and Mx moments. The second series of compression tests were carried out on the prototype to meet the existing standard ISO 10328:2016, with P5 loading cases (user weight up to 100kg). Ultimate strength tests were performed at -20° and +15° consecutively until 4480N, with a constant load rate of 200N/sec and a flat step of 30 sec at the load. Cyclic fatigue tests were performed with a brand-new prototype on a 3400 electro-force BOSE fatigue machine (BOSE, USA). Two million load cycles were applied successively on the forefoot (-20° pitch angle) and heel (+15°) with a loading range of \[50N – 1330N\] and a frequency of 2Hz. As required by the standard, we performed a final compression test at 2240N at both angles with a constant load rate of 200N/sec and a flat step of 30 sec at the maximum load to verify the remaining strength of the prototype foot after fatigue. ## 2.6 Field testing ### 2.6.1 Protocol The field test aimed to assess the real-life usability and performance of the new foot with actual prosthetic users and gain insight for further improvements. We recruited 11 healthy transtibial amputees (weight: 72 ± 11 kg, size: 170 ± 2 cm, amputation side: 6L and 5R, foot length: 25.5 ± 0.5) without other disabilities. The measurements took place at the Vietnamese Training Centre for Orthopaedic Technologists (VIETCOT, Hanoi, Vietnam), where participants were fitted by two certified orthotist/prosthetist with new transtibial prostheses (comprising a CR-SACH foot) two weeks before the data collection. Two feet, P_SACH and P_PRO, were tested in a randomized order. The real-life usability protocol included: (1) checking of the complete prosthesis, (2) habituation (\~ 30min), (3) 6 Minute Walk Test (6MWT) and (4) 48 hours of daily-locomotion monitoring using shoe-worn inertial sensors (Physilog5, GaitUp, CH). In addition, we used a foot assessment questionnaire adapted from the Prosthesis Evaluation Questionnaire (PEQ) to evaluate comfort of the users based on (1) stability while standing, (2) stability while walking, (3) powerful push- off, (4) feeling of firm contact with the ground, (5) weight of the prosthetic foot, (6) comfort of the prosthetic foot and (7) overall score of the prosthesis. This protocol was accepted by the local ethical committee at the University of Labour and Social Affairs (ULSA, Vietnam) and written informed consent was obtained from all the participants prior to the measurements. ### 2.6.2 Data analysis The detection of the locomotion periods and their classification for the daily measurement trials were completed using previously validated algorithms. From these algorithms, the number of walking bouts per day, and the mean duration and cadence of each bout was extracted as indicators of the daily physical activity of the patients. The results were grouped by the foot type and the inter-trial mean, median, minimum, and maximum calculated for each of the aforementioned parameters. The questionnaire was composed of 14 explicit and 2 general questions. The mark of each question was reported relative to the length of the linear analog scale with a 100% score indicating a perfect evaluation. The inter-subject mean score of each question was calculated for each prosthesis, and results reported only for the questions where we observed a difference greater or equal to 2% (threshold empirically set based on the width of the pencil marks and the width of the scale). # 3 Results ## 3.1 Biomechanical specification of PPRO versus commercial feet In total, 840 steps (respectively: P_SACH: 205, P_{K3_C}: 193, P_{K3_E}: 195, P_K4: 192 and P_PRO: 55) of level walking have been analyzed with more than 10 steps recorded per participant and per foot. The average (min, max) number of steps per participant was 71 (51, 83) steps. This number varied between subjects as invalid steps (e.g., foot partially out of the force recording area, occlusion of the markers) were removed from the data set. The evaluation of the final foot prototype was performed during the third and final test, with only 7 of the 13 initial participants hence the lower number of steps recorded. The normalized mean ± STD of vGRF as a function of the pitch angle for each foot was compared and illustrated in. The inter-steps median and IQR of the six loading times of the stance phase and the corresponding *α*_*pitch* and *α*_*flexion* are shown in with the p-values indicating a difference between each foot and P_K4 (highest profile). ## 3.2 Prototype design, materials and process selection ### 3.2.1 Prototype design presents the final design of a prototype foot, as well as the assembly showing the inner elements before the final over-injection step. After several iterations, the keel was successfully encapsulated in the cosmetic shell. The C-shaped front of the ankle ensures the integrity of the part in case of overload. This area has a gap of approximately 10mm, which closes when a high load is applied, limiting any structural failure in the ankle part in case of overloading. ### 3.2.2 Materials and process selection After the cost modeling of each material and process selection step, the material choice, cost index, and preferred processes that satisfied the requirements are presented in. For the blade part, it appeared that carbon continuous thermoset composites, which are widely used in high-performance prosthetic feet, are necessary for the required stiffness and elastic energy restitution within the geometric thickness constraints. Thermoplastic-based materials and processes were considered to limit the cost associated with the complex shape of the ankle. A similar shape produced by traditional thermoset composite processes did not fit our target costs. Additionally thermoplastic processes such as injection molding provided full automation and low cycle times leading to cost savings. Carbon fiber reinforced Polyamide LFTs were selected due to their sufficiently high mechanical properties and reasonable cost. Injectable TPU compounds with foaming agents were selected for the foam part to preserve compatibility with the existing cosmetic shell material and achieve good adhesion. The selected keel materials were successfully qualified by fatigue testing to ensure durability as verified with the results presented in. All the materials passed the required fatigue resistance without showing any signs of failure and thus, were selected for the final prototype manufacturing. ## 3.3 Prototype testing For each foot, the load-deflection response was measured at different pitch angles to extract its non-linear stiffness characteristics starting from a vertical preload of 70N, as shown in. The vertical deflection was recorded during mechanical testing for each pitch angle at the corresponding level of GRF measured during gait testing, where 100% GRF was defined as 800N. The displacement and pitch angle chart summarizes the behavior of the feet during a complete stance phase, in other words it represents the vertical displacement observed during mechanical testing in function of the pitch angle at the corresponding force measured during the gait testing. As expected, all feet presented a higher rigidity (and lower displacements) at the heel compared to the forefoot, and most of the feet showed highly non-linear stiffness characteristics. P_PRO, P_{K3_C}, P_{K3_E}, and P_K4 exhibited higher forefoot flexibility than P_SACH with maximal displacements in the range of 9 to 10mm. Good agreement, within 1.5 mm deviation, between the FE model (P_promodel) predictions and P_PRO was observed for the heel loading case, while the P_PRO was slightly stiffer than expected in forefoot load cases. Overall, P_PRO showed a moderately stiffer behavior than P_K4, its response sitting between P_K4 and P_K3, but was twice more compliant in the forefoot compared to P_SACH. The P_SACH measured moment (My) shows a peak at the swing phase (around 2deg), while the moment peak of the other feet appears later, around -10deg with lower values; P_K4 having the lowest one. With the measured moment and force, the center of pressure has been calculated and represented. As expected, the center of pressure on the X-axis of P_SACH shifts sooner than others at the forefoot. This behavior is mainly explained by the geometry and the stiffness of the P_SACH keel. Tests according to ISO 10328:2016 P5 led to permanent deformations of less than 5mm at the forefoot and less than 2mm at the heel. No evidence of failure was detected. For fatigue tests, P_PRO was loaded successfully with a sinusoidal force between 114N and 1320N for 2M cycles at the forefoot and the heel successively, without any signs of damage. The minimum test loads slightly differed from the standard recommendations because of the machine’s stroke limitations. After fatigue testing, a final monotonic load test at 2240N at +15 and -20° was performed with no evidence of failure. The classification of the developed foot was validated by measuring the energy restitution as defined in. ## 3.4 Field testing Out of the 44 days of activity monitoring, 14 days were removed from the data set because of involuntary false manipulation by the participants, sensor errors, and extreme weather (i.e., heavy rainfalls during monsoon season affected the daily activity of two participants). A total of 11142 walking bouts have been detected with 10937 bouts shorter than 30 sec, 192 bouts lasting between 30 sec and 2 min, and 13 bouts longer than 2 min. In, the results for P_SACH and P_PRO are compared, and the inter-trials statistics of the daily-locomotion features and the average score of the evaluation questionnaire are shown. Cadence statistics were computed over the 95% interval in order to remove outliers’ errors. Finally, no statistically significant differences were observed in the distance traveled during the 6MWT and in the Q1-Q7 answers of the questionnaire. # 4 Discussion The functional prototype designed and tested in this study was conceived in a relatively short time frame (1.5 years) and, although further improvements might be required, the foot displayed promising results such as mechanical performance and user acceptance. Thanks to its modular design offered by the tuning of its foam properties (i.e., density and compaction point), and the tuning of the blade thickness, the foot proposed in this study has been adapted to five different size (22, 24, 25, 26, and 28) and multiple weight classes. Moreover, the chosen materials and related processing routes allow for a mechanical performance and energy storage and release that can satisfy the K3 level requirements while reaching the targeted low manufacturing costs. Every stakeholder (i.e., ICRC and academics) shared valuable inputs and expertise to build a collective knowledge encompassing different research fields, the key findings of this interdisciplinary approach are discussed here. We focused the output of the in-lab gait analysis on metrics directly usable in the design process (i.e., conceptualization and testing); hence the relations between the ground reaction forces and the foot orientation were extensively investigated. We identified the mechanical behavior of the P_K4 foot as the performance target based on the reported K level and the perceived comfort of the participants during the first in-lab tests. Although P_K4 is not adapted for humanitarian applications (e.g., too expensive, keel not fully encapsulated in a cosmetic shell), its mechanical efficiency in executing specific tasks became the objective targeted for the P_PRO. The gait analysis results show significant differences (p \<.01) in *α*_*pitch* during the loading phase for P_SACH and P_{K3_C} compared to the targeted P_K4 foot. In both cases, the results show that a higher vGRF was applied on the feet at positive foot strike angles; hence participants possibly felt more stable at high pitch angle (i.e., \> 10°) with P_K4 and P_{K3_E} than with P_SACH and P_{K3_C}. Interestingly, these differences occurred even though the feet were confined within a shoe with a 10mm rear-foot cushioning, which decreased the influence of the rear-foot stiffness. Moreover, during the pushing phase, the vGRF recorded for P_K4 decreased almost linearly with *α_pitch*. This mechanical feature was only observed for P_K4 and could explain the feeling of “fluidity” perceived by the participants during the pushing phase. In other word this could be interpreted as a progressive feeling during the pushing phase, while a nonlinear decrease of vGRF could lead to a feeling of control lost before the end of the gait. In addition, the plantar pressure distribution confirmed that the amount of pressure on the forefoot region was higher and almost equally distributed for P_K4 and P_{K3_E} than for P_SACH and P_{K3_C}. This corroborates with the findings which infer that the mechanical properties of P_K4 and P_{K3_e} allow for a higher and more comfortable vGRF generation to drive the body forward. Although a habituation period took place after the alignment of each foot, the orthotic boots used in this study indisputably increased gait variability. However, as the order of the feet was randomized, we assumed the learning bias to be similar for each foot. Also, the additional height and weight of the orthotic boots imply that higher torques were applied to the feet, which might have influenced the mechanical response of the system. Nonetheless, based on our observations and in conformity with the K-level of the feet, the conception of P_PRO aimed for mechanical properties bound by the observations made on the P_SACH and P_K4 feet, acting as lower and upper-performance limits, respectively. The final in-lab gait analysis tests revealed that, during the loading phase, the participants applied significantly (p \<.01) more vGRF at high *α*_*pitch* compared to the other feet and P_PRO exhibited a gait pattern close to P_{K3_E} in the pushing phase. Moreover, the mechanical compression tests and simulations were restricted to the angles observed during the in-lab gait analysis and to the angle set in the standards ISO10328:2016. We combined the results from the compression tests with the vGRF observed in-lab and showed that P_SACH was more rigid in the forefoot region compared to the other feet; a lack of vertical deformation could create instability and limit the amplitude of the pitch angle during the stance phase. From 0 to -20°, P_K4 showed the most linear vertical deformation, which corroborates with the results obtained with the participants in-lab. The shape of the keel had a significant impact on this behavior; P_K4 has a higher and longer profile, with a longer bending length. Compared to the high profile P_K4, the current design of P_PRO was restricted in terms of flexibility by the 20 mm hole at the heel (insertion of the M10 bolt) and, at the forefoot, by the low-profile and split toe requirements which limit the active bending length of the blade. A careful optimization of the composite layup and thickness distribution was required to achieve both a sufficient blade flexibility while fulfilling the strength and fatigue requirements. To partially alleviate the risk of overloading the blade, we used a foam with a specific compaction point on the heel and forefoot, to bypass the load transferred through the blade in order to fulfill the overload tests required by standards. The high load observed during the heel contact phase for P_PRO led to a moderate vertical deformation of the heel because of a higher stiffness of the heel compared to the commercial feet. This difference was not perceived as we used a cushioned heel shoe during in-lab gait tests. Indeed, shoes increase energy dissipation and reduce the incidence of injury to residual limb soft tissue. To complete this rigidity analysis, we evaluated the bending moment at the ankle to reconstruct the position of the center of pressure as a function of the pitch angle during stance. It was observed that P_SACH shows an abrupt jump of the contact point position from the heel to the toe during the rolling forward of the foot while all the other feet including P_PRO exhibit a progressive motion of the center of pressure during stance. The hysteresis during the compression test was measured to evaluate the overall energy restitution of P_PRO (averaged on all angles and on 5 loading- unloading cycles) which was 82% when brand-new and 90% after fatigue (indicating a required run-in time of the encapsulation foam for full performance), compared to 87% for P_{K3_E} and 93% for P_K4. Finally, the eversion/inversion capacity was verified by evaluating the load required to ensure the contact on both sides of the foot at 0° pitch when contacting a 15° side slope. A closure force of less than 40% vGRF was found for P_PRO which is comparable to that of reference prostheses such as P_K4 (approximately 300N). The pragmatic approach using cost and performance as coupled guidelines led to find a compromise and step away from a full carbon epoxy composite design. As shown in, when mechanical properties are compared, long fiber carbon-epoxy composites are an obvious choice for a dynamic-foot but are beyond the cost target placed in our development. Instead, injection-molded long fiber thermoplastic materials used in the automotive industry were investigated and showed adequate stiffness and fatigue performance while strongly decreasing the cost for the targeted production volume. This combination of design, materials and manufacturing choices for each part in the prosthetic foot allows to reach the target performance and manufacturing cost of less than a 100 USD. Finally, the results from the field study revealed that the participants walked more (+70%) and for more extended periods (+90%) with P_PRO than with the P_SACH foot, but no effect was observed on the walking cadence. According to the evaluation questionnaire, P_PRO received a better average evaluation for six of the seven most significant questions. Only the mean score of the perceived walking stability was lower (-6%) for the prototype foot. A potential explanation was found in the users’ general comments; they claimed that the transition of body weight on the forefoot was difficult, and in some extreme cases, a slight backward torque was observed. This rigid behavior might originate from the fact that the participants were lighter and shorter than announced while planning the measurements; hence some participants were at the limit of the prosthesis’s range of use. Overall, the participants were satisfied with P_PRO (overall reported score of 81%). Although the questionnaire was inspired by the PEQ and not the actual validated PEQ, which we agree is not ideal and constitutes a limitation of this study, these are promising preliminary results, especially considering that the participants were equipped with the CR-SACH foot two weeks before the measurements, thus more accustomed to it. Moreover, the lack of statistically significant difference in the 6MWT and questionnaire results is in accordance with previous studies comparing prostheses of different K-levels. Hence, further field research and over a broader range of humanitarian environments should be carried out. # 5 Conclusion A multi-partner collaboration framework, including an academic program which focuses specifically on technologies and product development for low-resource settings, was established for this study. Through an interdisciplinary and functional approach, we developed a prosthetic foot that met the needs of ICRC’s and the particular context of use in conflict-affected areas and LMICs. We first characterized the mechanical behavior of several commercial feet and set our conception objectives accordingly. We showed that the pushing phase played a key role in K3-K4 feet compared to K1 ones. We then designed a prototype based on these gait analysis results, and the requirements set by the ISO 10328:2016 structural testing standard. This process required multiple design iterations to reach a functional foot design that was tested under different conditions. Moreover, we performed a rigorous cost analysis of materials and processing routes assessment to ensure low manufacturing costs for the considered production volume; hence, a careful screening of a multitude of materials was performed, including some that are generally not considered for such an application. Finally, tests carried out in the real-world showed a good performance of the prototype and paved the way for a long-term evaluation in the field. Despite the promising results obtained some limitations must be considered, which will influence the next steps of the project. First, the number of patients who took part in the field tests: the authors consider that 11 patients is too few to really claim clinical evidence at this stage; further testing at a larger scale has to be performed to strengthen the first findings. Second, even if we mechanically tested the prototype with conditions as close as possible to the ISO standard, complete testing by an independent certified laboratory is required to ensure the developed foot complies with it. The technical content of the project has been transferred to the ICRC, and the manufacturing processes should be industrialized in order to reach mass production and supply humanitarian actors and others active in LMICs. # Supporting information Special thanks to the following people who brought expertise, supported, or directly took part in the project: Helena Texido Pedarros, Guillaume Martini, Pascal Morel, Pascal Hundt, Dikolela Kalubi, Marc Zlot, Pierre Gauthier, Miguel Fernandes, Jean-Daniel Broillet, Thierry Barras, Denis Müller, Robert Perrin, Elise Carron, Fabian Meylan, Wolfram Hang, Corentin Legris, Mr. Thanh, Mrs. Ha, all the staff at VIETCOT, patients at VIETCOT, ATMX workshop (EPFL), ATME workshop (EPFL), EMS-Chemie Holding AG, Polyone Corp., BASF SE, JR Purtec Gmbh, Gurit AG, CMP Gmbh, Fondation Philanthropia, Philip Morgan, Phalla Keo. 10.1371/journal.pone.0266656.r001 Decision Letter 0 Eshraghi Arezoo Academic Editor 2022 Arezoo Eshraghi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 6 Oct 2020 PONE-D-20-28531 A user-centric approach towards the design, development, and test of an affordable dynamic prosthetic foot PLOS ONE Dear Dr. Falbriard, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. As included in the reviewers' comments the manuscript lacks clarity and justification as well as important details about the methodology, results and discussion of the results. Please submit your revised manuscript by Nov 20 2020 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols> We look forward to receiving your revised manuscript. Kind regards, Arezoo Eshraghi, Ph.D. Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2.Please provide additional details regarding participant consent. In the ethics statement in the Methods and online submission information, please ensure that you have specified (1) whether consent was informed and (2) what type you obtained (for instance, written or verbal, and if verbal, how it was documented and witnessed). If your study included minors, state whether you obtained consent from parents or guardians. If the need for consent was waived by the ethics committee, please include this information. If you are reporting a retrospective study of medical records or archived samples, please ensure that you have discussed whether all data were fully anonymized before you accessed them and/or whether the IRB or ethics committee waived the requirement for informed consent. If patients provided informed written consent to have data from their medical records used in research, please include this information. 3\. Please ensure that you refer to Figure 7 in your text as, if accepted, production will need this reference to link the reader to the figure. 4\. Please note that in order to use the direct billing option the corresponding author must be affiliated with the chosen institute. Please either amend your manuscript or remove this option (via Edit Submission). \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: No Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This paper and work aim to address an important gap in prosthetic practice in LMICs by designing a new dynamic foot. The work involves a number of design and testing stages, as would be typically needed for the development of prosthetic componentry. The design process is in itself complex, and not all facets should be considered as research. While I can appreciate the work that has gone into developing the foot, the novelty is not entirely clear. Moreover, the paper appears to lack adequate rigour and details in many facets of the work as detailed below. Abstract: While it may be true that the prototype foot “showed a better performance and acceptance by users compared to a SACH foot”, did it perform adequately well to be deemed a dynamic foot as was the goal? The paper needs to clearly present this here, and in the discussion section. Introduction: The introduction does not adequately address the state of the science. What are the benefits of dynamic feet compared to SACH, and why is this critical for LMICs? Reasons other than affordable technology such as a lack of prosthetists should be acknowledged as a part of the limited access to prostheses in LMICs. There exist highly affordable feet such as the Niagara foot, and this background into existing solutions is not adequately acknowledged. In fact the Niagara foot is priced well below \$100. Information about the manufacturing costs of feet needs to be dealt with and presented. What are the clinical and user issues with SACH feet? These limitations need to be presented. Methods: The authors define and use data relating to pitch angles and GRFs in the assessment and design of the prototype foot. Please provide some justification and citations to support that these metrics are in fact important aspects of dynamic foot design. Relating to this, the authors need to very clearly define and if applicable quantify what constitutes the differences between different foot classifications (K1 vs K2 vs K3). Please more clearly define what distinguishes a K3 foot from a K2 and K1 foot in terms of the functional or mechanical properties. The FEA lacks detail about constraints, loading, meshing, convergences etc. The optimization on line 225+ is vaguely described, and information is needed about what was actually done. Line 292 – what is the 2% threshold based on? Results: The removal of 14 of 44 days is a significant amount (1/3 of data) and this could bias the results. How did the authors know that data were manipulated? What is the basis for removing data related to extreme weather? Monsoons are a normal occurrence in Vietnam and should therefore not be excluded. More so, did the monsoon days effect one foot more than the other? Please provide some justification that bouts of walking are a good measure of prosthetic performance. The differences in bouts could be attributed to other factors. People in places such as Vietnam use walking as a primary means of mobility. They need mobility to pursue ADLs and other necessary tasks, so it seems unlikely that they could elect to walk more or less in light of these necessities, especially considering that the SACH foot appeared to work quite well for them. Please provide an explanation or hypothesis for these found differences. Were the questionnaire results statistically significant? Please provide stats. Are the differences clinically significant? Discussion and conclusions: Need to explain the novelty and contributions of this work more clearly. The cost aspect needs to be discussed in comparison to other existing foot technologies designed for LMICs Minor: Line 37 – remove ‘both’ since referring to 3 items Line 73 – spelling of ‘toe’ is incorrect Line 92 – need to specify the models and manufacturers of feet Line 385 – what is 11’142? Caption of Figure 6 stating the 13 and 5 participants is confusing and it is not clear how this relates to the actual sample size of 11 participants. In table 1, the numbers do not make sense. For example, for the pitch angle at 50%, both the P3C and PK4 have the same value (22), yet they are noted as being significantly different. That is not possible. Also, the table is confusing, as it is not clear whether results are better or worse than the PK4. The direction matters and should be clear from the table or other figure. Reviewer \#2: GENERAL COMMENTS: This paper presents a multi-disciplinary approach for designing and testing an affordable passive prosthetic foot suitable for the developing world context. It aims to create an affordable high mobility dynamic prosthetic foot to replace the currently distributed low mobility solid-ankle cushioned heel prostheses. The authors outline the design framework starting with design requirements, set through stakeholder discussions and gait studies of able-bodied walkers. Following with the development of the prosthetic foot design through material testing, structural optimization, manufacturing analysis, and mechanical testing. Lastly, they validated the designed prototype through gait studies, additional mechanical testing and a field trial. The work described here has a lot of significance to the field in terms of the design approach as well as the resulting device. It is definitely worthy of a publication. However the work requires major revisions since additional information would be required to: support all the claims (mainly on the expected level of performance of the prosthetic foot prototype), justify the design and methodology choices (Why were this set of structural optimization target chosen, compared to standard measures such as roll-over shape or push off work? Can a foot that exhibit a certain load displacement behavior be categorized as fulfilling the WHO d4602 or K3 level performance? How close do you need to get to the P_K4 prosthetic foot to fulfill your requirements), and discussions on the gait analysis results would benefit from being put in perspective with existing literature (whole body propulsion, loading rate and peak vertical loads at heel strike, limitations of prosthetic boots usage compared to people with amputation, center of pressure progression, ankle moment etc…). In addition, the title of the work could be misleading as it refers to ‘user-centric’ approach but throughout the design process nor for the design requirements (given by the ICRC and WHO guidelines) were people with amputation consulted nor included in the development. People with amputation were included after the design of the prosthesis was completed. Lastly, minor revisions regarding the spelling, grammar, and phrasing issues would improve on the clarity of the work. SPECIFIC COMMENTS: Abstract: Line 42: What specific aspects about the prosthetic foot performance was improved? By how much? Introduction: The authors should specify the level of performance of the current ICRC SACH foot to put in contrast with the target foot’s performance. What specifically about the current ICRC SACH foot prevents it to meet the listed requirements? Line 54: How about the Niagara foot, which meets the cost requirement, through bolt attachement requirement as well as the K2-K3 level activity? (Ziolo and Bryant 2002, Wellens 2011). Methods Project Framework: Line 94: The selected commercial prosthetic feet models should be specified to allow for reproducibility of the experiments, instead of abbreviated codes (P_K3_C). Line 101: Which part of the in-lab biomechanical and mechanical evaluations where used to improve the model? In-lab characterization of commercial prostheses: Line 116: How long was the habituation period? Was all the tests for each participant with the commercial feet conducted on the same day? Line 141: How were the shank, fore and rear foot frames defined? I assume the rear foot frame was defined using the three markers at the heel, the forefoot frames using the three marker at the forefoot and the shank frame, the four markers at the prosthetic ankle? Line 145: Why were the pitch angle and flexion angle considered instead of the more traditional ankle angle or shank angle? Line 151: Why were the stance phase behavior at 30%, 50% and 100% of BW specifically chosen instead of others gait events/frames? Line 158: the x-axis of Fig 3b seems mislabeled. Design, materials and simulation: Line 165: was the shell included in the optimization/modelling of the prosthesis? Line 168: How were the target performance outlined in d4602 translated into specific design requirements as the d4602 only outlines general walking aspects. Line 170: The term ‘best gait pattern’ should be clarified. Which aspects of the gait pattern were targeted? Does this term refer as getting as close as the K4 foot? How close? Line 182-183: the blade’s elasticity and strength were optimized but what were the optimization targets? Was the blade optimized independently of the foam and ankle part? Line 184: How much eversion/inversion was set as the target? Line 188: It is mentioned here that the foam density was used to accommodate different user weights, but only one foam density is later presented? Was the designed varied for user weights? ‘Different materials’ (line 193) and ‘ specimens’ (line 199) should include the details of the material that were tested/part of the selection process. Line 220: Why was a quasi-static FE model chosen insead of a dynamic FE model that would represent the prosthetic foot loading cycle and loading rate of the ISO tests? Line 224-228: Why was only the compression test of the selected commercial feet chosen as target for the optimization instead of the gait cycle tests? Furthermore, the gait cycle tests included walking activities such as stairs, ramps and side to side stepping. Were these represented too in the compression tests? Was the objective equally weighting the deformation, reaction force and moment ? Is the reaction force prescribed along with the plate orientation or was it an objective? Mechanical compression tests Should this section be presented before the design sections since these results were used as target during the design process? Line 241:’Feet hysteresis were quantified and evaluated’ should these also appear in the results section? Line 245: How were the different pitch angles selected? Field Testing Were these subjects given the same foot designed for an 80kg user as stated in the design section? A copy of the adapted PEQ should be provided as supplemental information. Line 292: How was the threshold of 2% selected? Is that related to a statistical analysis? Results Prototype design Line 319: What was the result of the optimization, how close did the prototype design get to the target objectives? Line 343: How were the stress level/fatigue resistance values set? Line 363: Figure 8a caption refers to P_SACH but the graph represents P_PRO. It is unclear in Fig 8.b what test was conducted? Was the vertical displacement the same for all of the prostheses and the corresponding load was measured? The methods section only refers to loading cases being prescribed and not displacements. Was it from the biomechanical tests? Figure 8 and 9: Why was only one P_K3 represented in these tests? Which one is it? Discussion Line 453: the P_PRO exhibit a more progressive motion of the center of pressure compared to the P_SACH but more abrupt compared to the other commercial feet? Does that allow the P_PRO to meet this requirement as there is a more abrupt change compared to the 3 higher level commercial feet? Line 456: What explains the higher energy restitution after fatigue compared to the new state and how does it compare to the ther prosthetic feet? Was this test conducted with the foam and cosmetic shell? Line 471: Does the difference between the P_SACH and P_PRO in terms of overall satisfaction significant? Does the ankle moment, loading and displacement of the P_PRO enables the device to fulfill the k3 level requirement? What were the limitations of the study? Did the prosthetic boot gait study match the walking pattern of prosthetic users? \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0266656.r002 Author response to Decision Letter 0 26 Jul 2021 Dear reviewers, thank you for your comments which we believe helped us improve our paper. We have updated the manuscript with track changes and answered your review in a different document. Note that we have also provided additional informations about the cost analysis and modelling in a separate.zip file. Sincerely, Mathieu Falbriard 10.1371/journal.pone.0266656.r003 Decision Letter 1 Eshraghi Arezoo Academic Editor 2022 Arezoo Eshraghi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 26 Aug 2021 PONE-D-20-28531R1 A functional approach towards the design, development, and test of an affordable dynamic prosthetic foot PLOS ONE Dear Dr. Falbriard, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. I believe this is an important work that should ultimately be published. Yet as mentioned by the reviewers too, the structure of the paper just does not work well. Therefore, I would invite the authors to submit a reduced paper, or breaking up the work being presented in this manuscript in two manuscripts so that you can apply the necessary level of rigor and details. Please submit your revised manuscript by Oct 10 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Arezoo Eshraghi, Ph.D. Academic Editor PLOS ONE \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: (No Response) Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: First, I would like to acknowledge the importance of the work aimed at addressing a challenging global problem. Secondly, I appreciate the multifaceted nature of the approach, considering a range of aspects relevant to solving the problem. The authors present analyses related to biomechanical, structural, finite element, cost, and various aspects of the design, as well as clinical field-testing results. The main challenge with this paper, is that it tries to present all these findings, and as is evident in many of the responses from the first review, there just is not enough space to treat each of these aspects with the level of detail and rigor that (in my opinion) is needed for a research paper. As such, there continue to be too many gaps in the paper, with important information missing. Unfortunately, the manuscript reads more like a student thesis project, than a research paper that one would expect to find in PLOS. Introduction: “According to the World Health Organization (WHO), 80% of people with disabilities live in low- and middle-income countries (LMICs) \[1\], and the number of lower-limb amputees is likely to globally increase due to several factors such as diabetes, accidents, conflicts, and congenital disabilities \[2\]. Moreover, only 5 to 15 percent of people with disabilities living in LMICs have access to rehabilitation services. Prosthetic feet (further described as "feet" for ease of reading) with advanced biomechanical features..” This section jumps from disability to lower-limb amputees, back to disability, and then very specific info related to feet. I think the flow of the information needs to be improved. Perhaps the information on feet deserves its own paragraph(s). Line 58 ‘usually made of expensive materials and..’ – please be specific and provide an example of the expensive materials in question Line 74 ‘material and production processes (e.g., 3D-printing) have shown promising’ – its not clear why 3D printing is mentioned here, since it does not appear to play a role in the project. Line 79 – Is this manufacturing or the final cost? Most prosthetic devices have high margins, so the manufacturing cost is usually a small fraction of the final cost. In other words, many feet exist that cost under \$100 to make, but are sold for many times that. Noting that it is the final cost that is important to the client. Please clarify and provide more insight about these costs. In addition to K level, why is the weight capacity not considered along with the other design criteria Methods: 3.2.1. – please provide information or references that speak to the validity of using this protocol (able-bodied participants using the brace) to assess foot gait biomechanics. Line 149 – ‘The data from these units were used to empirically confirm that the previously developed algorithms for sound leg gait analysis \[16,17\] were functioning on the prosthetic foot 151 and could be used for field tests.’ – It’s not clear what this statement aims to communicate. ‘However, we used the median, and the interquartile range (IQR) as inter-steps statistics (i.e., calculated over all the steps regardless of the subject) as histograms and the quantile-quantile plots (Q-Q plots) suggested a non-Gaussian distribution’ – Why? Please justify. Line 199 – ‘…minimal use of carbon fiber reinforced material to limit cost, geometrical constraints due to the’ Please elaborate on the above. In particular what is the cost of carbon fibre compared to other materials? This seems like it would be important information if one is trying to optimize costs. Lin 216 – Here and in a number of places the authors rely fully on references that speak to a particular protocol. This is not sufficient as it requires the reader in most cases to seek information elsewhere. The manuscript should possess adequate information/details to stand alone. Line 251 ‘The stiffness of the prototype …overloading.’ Please provide details about the manual optimization. More information is needed about how the optimization was conducted and variables applied. How is injection moulding a design constraint? Line 303 ‘In addition, we used a foot assessment questionnaire inspired from the Prosthesis Evaluation Questionnaire (PEQ) \[24\] to evaluate..’ – The PEQ is a validated questionnaire and unfortunately, using questions from it is not ideal. This is a major limitation that should be clearly recognized. Line 319 – ‘The questionnaire was composed of 14 explicit…’ – which questionnaire is this? Results: Line 329 – ‘This number varied between subjects as invalid steps..’ – it is typical to collect data and exclude invalid step during collection – hence extra steps can be collected to replace the mishits. Why was this not done? Line 364 – ‘Thermoplastic-based materials and processes were considered to limit the cost associated with the complex shape of the ankle’ Please explain what this means. Table 2 – please explain the cost index Table 3 – the numbers are confusing. For example, it seems that the blade failed at 149,451 cycles(?) The data from the PEQ and 6MWT need to be presented and discussed even if the findings were not significant. The fact that the prototype foot did not perform better in the 6MWT needs to be discussed. What have other studies found with this test when comparing SACH to higher end feet? Discussion: Line 433 – ‘..the foot displayed promising results.’ – Such as? Line 434 - ‘the foot proposed in this study can be adapted to a wide range of patients' sizes and weights.’ – it is unclear where in the findings this was demonstrated. Line 454 – ‘This mechanical feature was only observed for PK4 and could explain the feeling of “fluidity” perceived by..’ – Please clarify what data/results demonstrate this perceived fluidity. Line 484 – ‘for PPRO led to a moderate deformation because of a higher stiffness..’ deformation of what? ‘This difference was not perceived as we used a cushioned heel..’ – by what or who was it not perceived? Please refer to the relevant results/data Line 489 – spelling of ‘two’ ‘The pragmatic approach using cost and performance as coupled guidelines allowed the team to step away from a full carbon epoxy composite design..’ reducing the amount of carbon also seems to have reduced the performance compared to the K4 foot. So it in fact appears to be a compromise. These points should be discussed. A carbon ankle (c section) will no doubt play a role in the dynamic performance of the foot. So how much performance is lost be replacing it with an injection moulded component? It would be interesting to discuss the elastic properties in relation to typical carbon fibre ones. ‘PPRO received a better average evaluation for six of the seven most significant questions’ – please include these results in the results section. Reviewer \#2: GENERAL COMMENTS: This paper presents a multi-disciplinary approach for designing and testing an affordable passive prosthetic foot suitable for the developing world context. It aims to create an affordable high mobility dynamic prosthetic foot to replace the currently distributed low mobility solid-ankle cushioned heel prostheses. The work described here has a lot of significance to the field in terms of the design approach as well as the resulting device. It is definitely worthy of a publication. The authors have improved and addressed some of the reviewer’s comments but minor revisions should still be completed before submission. Including additional quantifiable measures to interpret the results would improve on the rigor and clarity of the manuscript. Instead of mentioning that the model is a ‘good match’ with the experiments providing the error in displacement or moments would be beneficial. The supplementary documents should be listed/referred to throughout the manuscript when relevant instead of at the end of the manuscript only. Similarly, the load displacement hysteresis curves, questionnaire information or questionnaire results mentioned in the reviewers’ response to be present in the supplementary document were not found. SPECIFIC COMMENTS: Methods Design, materials and simulation: Line 197: the ‘best’ gait pattern is a subjective assessment, be more specific and accurate by saying for example that feet were designed to replicate the pitch angles at different instances of stance. Materials and process selection methodology: Refer to supplementary material 2 Mechanical simulation: Line 261: typo, either ‘cosmesis’ or ‘cosmetic shell’ Field Testing Refer to the supplemental information for the questionnaire and it was not found in the supporting information listed in section 8. Results Prototype testing Be more specific regarding which figure was collected during gait testing and which one during the mechanical testing. From the manuscript it seems that the (a) and (c) were collected on the mechanical test setup while (b) is data from the gait testing. Line 404: replace ‘gait GRF’ by GRF measured during gait testing to improve clarity. Line 411: ‘good agreement’ should be quantified, model predicted the displacement within..mm Line 420: ‘peak at swing phase’ but the plot shown in Fig 8.a says moment during stance phase. Figure 8a: the title during forefoot simplified stance phase could be clarify to convey the portion of stance phase where the forefoot is loaded. I assume here too that the CoP was back calculated from the moment and loading values in the compression test? ‘the classification of the … by measuring energy restitution as defined in..” can you list the prototype’s energy restitution value compared to the other devices? Field testing: Were the activity monitoring and questionnaire results statistically significant? Please include this information in the result section. In addition, the supplementary data (S4) does not seem to include the questionnaire answers as stated Discussion Line 513: typo ‘toe’ instead of ‘tow’ How do you explain that the CoP for small pitch angles was further down the foot compared to CoP at larger pitch angles for P_SACH? It is counter-intuitive that the CoP progression would be this way in gait testing. Additional comments regarding the results would be helpful. You mentioned that the questionnaire could only be accurately measured up to 2% in the method section (line thickness) but 3 out of the 7 questions had differences of 2% were still considered as an improvement/better answer from the participants. Should the comment regarding the improved feedback/perception only be valid for Q4 or maybe Q6 depending on the statistical analysis? Comments to the reviewers regarding the inversion/eversion testing conducted to validate that the designed prostheses indeed enabled this functionality should be mentioned in the manuscript. Some information about the cost model and cost breakdown included in the reviewer’s documents should be included in the discussion or results to show that indeed the \$100 threshold was achieved. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0266656.r004 Author response to Decision Letter 1 7 Nov 2021 We apologize that during the previous submission to the reviewers we made an error by submitting highly confidential supplementary information ( S2: material characterization) and we kindly ask to revoke those specific documents submitted previously marked confidential that includes a supplementary information word document(SI_confidential_agilis paper) and a detailed cost analysis excel sheet(Cost modelling prostheses). Due to this additional information submitted earlier we understand that the reviewers have many queries about why we have not mentioned more detailed information in our article. The only reason we could not express ourselves more freely is due to the confidential information that we need to protect to ensure the industrialization and wide outreach of this product to all the patients in vulnerable areas worldwide through the ICRC. A revised supplementary information document for the materials and cost is submitted in this version. We hope the reviewers will understand this sensitive situation and support us in this effort. 10.1371/journal.pone.0266656.r005 Decision Letter 2 Eshraghi Arezoo Academic Editor 2022 Arezoo Eshraghi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 14 Dec 2021 PONE-D-20-28531R2A functional approach towards the design, development, and test of an affordable dynamic prosthetic footPLOS ONE Dear Dr. Falbriard, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jan 28 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Arezoo Eshraghi, Ph.D. Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: (No Response) Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: -Line 30 change ‘actors’ to ‘organizations’, also would remove ‘to a growing number of people’. Or improve the flow of the sentence otherwise. -Lines 33-35 – a reference is needed to support the assertion that there is a need for more dynamic feet. Also, it is not the low-income setting that would benefit, but rather the individuals with amputations that live in those settings. Please rephrase -Line 35-36 – this sentence requires references, so that the reader can look up information about these other attempt \- Lines 37-46 should be better organized. Perhaps chronological order. i.e. development of design requirements, design and fabrication of foot, then testing. Also, should information should be included what population testing was done on (gait analysis, field etc) \- line 55 – should say ‘services’ -line 58 -replace ‘like’ with ‘such as’ -line 74 – replace ‘the’ with ‘the feet’. Remove extra period. -Line 75 – ‘needs in biomechanics, durability, aspect, compatibility with the PP technology, and cost.’ This part of the sentence does not make sense. Please reword. -Line 81 – the word ‘and’ is missing before the last point. -Line 82 -which P level in particular is relevant or being targeted here? -Line 85 – replace ‘prevent’ with ‘decrease’ -Line89 ‘ remove word ‘generic’ or replace with ‘main’. Also, where does the testing come in? Line 101- rephrase to ‘To gain insight into the mechanical response of existing feet..’ Line 112 – should say’..was started’ Line 115 – please rephrase this since simulations themselves re not doing the interacting Line 117 – iterations of what? Design interactions? Line 162 – replace ‘over’ with ‘for’ Line 165 – Consider a more objective approach for assessing the normality of the distribution e.g. Shapiro-Wilk test. Please perform the test and provide results in the paper. Line 168 – Should ‘time’ be plural? Line 183 – structural integrity my be a more appropriate term than reliability. Line 213 – a word after patient is missing. Patient characteristics? Line 306- consider replacing the word ‘comfort’ with ‘performance’ to more accurately capture the breadth of the outcome measures used Line 307 – what is a ‘healthy’ amputee. Please be more specific about how they were screened. Line 312 – checking for what? Line 313 – it’s more commonly referred to as the 6 minute walk test. Line 315 – replace ‘inspired’ by ‘adapted’ Line 316 – ‘the feeling of the users’ is awkward wording. Line 328 – be specific about which questionnaire is being referred to. Line 335-36 – need to more specific about the source of the data (i.e. the gait analysis) Line 359 – replace ‘aspect’ with ‘design’ Line 430 – how were instances when participants manipulated the data determined? Line 432 – ‘A total of 11142 walking bouts have..’ is this for all the participants? Line 451 – the word ‘performances’ should be singular. Line 451 –‘ Moreover, the chosen materials and related processing routes allow for performances comparable high-end commercial products..’ this is somewhat of a pointless statement. More useful is to state whether comparable performance was actually achieved. My personal opinion is that there is a compromise that needs to be clearly noted as the design does not perform to the same level of high-end devices. Line 454-5 – The point of the sentence is unclear. What does the transdisciplinary nature of the work have to do with it being reported here? Line 467 – chance ‘inside’ to ‘within’ Line 500- change to ‘during the heel…’ Line 545- ‘prototype's conception objective’ is awkward phrasing Line 547 – change to ISO structural testing standards Reviewer \#2: GENERAL COMMENTS: This paper presents a multi-disciplinary approach for designing and testing an affordable passive prosthetic foot suitable for the developing world context. It aims to create an affordable high mobility dynamic prosthetic foot to replace the currently distributed low mobility solid-ankle cushioned heel prostheses. The work described here has a lot of significance to the field in terms of the design approach as well as the resulting device. It is definitely worthy of a publication. The authors have edited the manuscript with improved the clarity, details, and addressed the reviewer’s comments but minor small revisions should be implemented before submission. SPECIFIC COMMENTS: Introduction Line 53: Typo ‘..’ Line 62: Please explain ‘level 3 activities’ and the remaining mobility levels (Line 74) instead of just adding the reference as they might be confused with the Medicare K levels introduced later. Line 64: ‘ Due to its high profile’, what is considered a high profile and what was the prosthetic height target set for the author’s prosthetic feet? Methods Line 223: Typo ‘ this research, Scalability’ Results Table 3: Typo ‘ Carbon Fibre’ Discussion Line 513: “A closure force of less than 40% vGRF was found for PPRO which is comparable 513 to that of reference prostheses such as PK4.” What was the closure force of PK4 so that the reader can also deem it comparable? Questionnaire supplemental material: In the questionnaire questions, 14 items are listed but only 7 are reported in the results both in the supplemental material as well as in the manuscript. Why were the other 7 questions omitted? The hysteresis curves have should been included in supplemental material as they are mentioned in the test methodology and discussions. In addition, are the energy restitution averaged across the different loading cases or only taken from a single loading case? \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0266656.r006 Author response to Decision Letter 2 23 Mar 2022 Dear reviewers, thank you for taking the time to comment on our study. We believe your feedback and questions helped improve the manuscript. Please find the detailed answers in the "Answer to reviewers" document attached to this submission. Sincerely Mathieu Falbriard, Ph.D. 10.1371/journal.pone.0266656.r007 Decision Letter 3 Eshraghi Arezoo Academic Editor 2022 Arezoo Eshraghi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 25 Mar 2022 A functional approach towards the design, development, and test of an affordable dynamic prosthetic foot PONE-D-20-28531R3 Dear Dr. Falbriard, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Arezoo Eshraghi, Ph.D. Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0266656.r008 Acceptance letter Eshraghi Arezoo Academic Editor 2022 Arezoo Eshraghi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 29 Apr 2022 PONE-D-20-28531R3 A functional approach towards the design, development, and test of an affordable dynamic prosthetic foot Dear Dr. Falbriard: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Arezoo Eshraghi Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction An unbroken trend to urbanization and a growing taste for mass events are among the many reasons why we experience very dense crowds more and more often. The challenge is to ensure safety and comfort for people in such situations. Pedestrian stream simulations are nowadays considered a means to mitigate risk in dense crowds. Buildings and events can be planned with a focus on safety, making use of virtual experience gained from pedestrian streams simulations. With the help of a simulation tool, a trained user can run through multiple “what-if” scenarios to gain experience for situations where it is impossible, uneconomical or even unethical to gather real experience. Moreover, since recently pedestrian stream models aim not only at simulating “what-if” scenarios, but also at short term prediction of real-life scenarios. Short term simulation predictions could be used to warn of critical situations such as danger of high densities. Pedestrian stream simulations are advantageous only if capable of correctly reproducing pedestrian motion. Ideally, the models for social or natural phenomena should be developed on the basis of observations gathered via controlled experiments and field studies. As soon as the basic model is constructed, one should ensure that it reflects the real-life scenario being simulated. An accurate reproduction of known data, without being an exact proof, would indicate that the simulator may be trusted for collecting virtual experience or for realistic predictions. To transform a basic model into a model capable of reproducing real-life scenarios, it is thus necessary to calibrate and validate the model against the relevant real-life data. A number of pedestrian models have been proposed ranging from macroscopic to microscopic simulations. However, construction of reliable models so far has been hindered by a lack of sufficient data from real-life scenarios. Pedestrian motion models neither always come straight from observations nor are rigorously checked against them. Instead most models are calibrated and validated only against literature values or laboratory experiments. They usually focus on a specific phenomenon such as: deceleration with increasing density according to a given fundamental diagram, lane formations in bi-directional pedestrian flows, pedestrian dynamics near a bottleneck, waiting zones, oscillations of flow- direction at doors or dynamics within highly dense crowds in. In the very best case models are calibrated and validated against relatively small and simple real-life scenarios which in most cases again represent some single, isolated phenomenon. The very few examples of simulation calibration against real-life data include scenarios focused on entrances into escalators, queuing at a train station, bottleneck gates at a train station and bi-directional flow in a corridor. Such calibration approach is very helpful and fully justified when one seeks to better understand an isolated phenomenon. However, one cannot expect that calibration solely based on an isolated phenomenon would be sufficient to reproduce a complex real-life scenario. Here, a more holistic attempt is necessary. The authors are not aware of any publications on comprehensive calibration. In this paper we present a method for calibrating a cellular automata-based simulation against a complex real-life scenario observed at a major German railway station. Based on this real-life scenario we also examine the applicability of standard literature suggestions for free-flow velocity values and fundamental diagrams. We calibrate our simulation against pedestrian motion parameters extracted from video data. The success of the proposed approach is demonstrated by validation of our calibrated simulation against the observed real-life scenario. The validation is performed by comparing an aggregated quantity – pedestrian density evolution in the area of observation. This paper is organized as follows: The Results section starts with an *Scenario and observational parameters: a German railway station* subsection on observations at a major German railway station. An overview of assumptions and restrictions of the benchmark model is given in *Simulation tool*. In *Methodology* we identify parameters critical for calibrating simulations and suggest how to feed them into a simulation model. Then the calibrated benchmark simulation is compared with measurements in *Proof of concept: Validation and sensitivity study*. The Discussion part summarizes the results. At the end the *Materials and methods* section gives insight into the video tracking technology we used. # Results ## Scenario and observational parameters: a German railway station Our analysis of pedestrian motion is based on video recordings at the major German railway station. The video data was collected at the station in the morning (7 a.m.) and in the afternoon (between 4 p.m. and 6 p.m.) on a workday. The examined area covers several platforms and a part of the station's main hall. The video recordings show different scenarios at the railway station: A train departure and arrival and pedestrians entering the main hall and heading towards further destinations such as exits and food stalls. In all scenarios bi- directional or multi-directional flow was present. and show a schematic picture of observation area highlighting the fact that one has to deal with hidden areas when using video recordings data. Trajectories of individual pedestrians in time and space were extracted from the video recordings and then analyzed. We used partly automated video tracking to extract pedestrian paths, walking speeds, schedules of pedestrians appearances and disappearances and pedestrian densities. shows trajectories extracted from one of the video recordings. The topology of the area of interest, the schedule of pedestrians appearances and disappearances and the path distributions are scenario-dependent. Walking speeds and flow-density dependencies have a general character and are analyzed in detail below. ### Distributions of free-flow velocities In literature it is mostly assumed that free-flow velocities are normally distributed. With the availability of real-life measurements it makes sense to test the validity of this assumption: Here we test a null-hypothesis on normal distribution of free-flow velocities based on video analysis data. Our analysis shows that, the null hypothesis need not be rejected at the 0.5% level according to the Cramer-van-Mises test. Strictly speaking, assuming a normal distribution is wrong, because we neither allow negative speeds nor speeds above the current men's world record for sprints (which is slightly below 10 m/s). For the purpose of the simulation, however, the error seems to be acceptably small and we assume a normal distribution. Another assumption concerns loiterung and slowly moving people. Since it is difficult to distinguish between the two we exclude outliers in our benchmark simulation. Weidmann claims that pedestrians on their way to work in the morning are 0.2 m/s slower than on their way home in the afternoon. We observed a similar difference between morning and afternoon: In the morning the mean free-flow velocity was 0,97±0.1 m/s compared to 1,04±0.1 m/s in the afternoon. At the same time, even in the afternoon, our measurements showed a much lower mean free-flow velocity than 1.34 m/s generally assumed. To be more precise, in the morning we observed a normal velocity distribution with the standard deviation of 0.29 m/s around the mean of 0.97 m/s. There was only one outlier (a passenger with a velocity below 0.1 m/s or above 4.0 m/s) among the 133 pedestrians with virtually no effect on the statistical outcome. In the afternoon the free-flow velocities were also normally distributed with the standard deviation of 0.51 m/s and the mean of 1,04 m/s. The situation remained unchanged after exclusion of outliers – 203 data points remained with the mean free-flow velocity of 1.10 m/s and the standard deviation of 0.47 m/s. It should be noted that the outliers did not affect the overall dynamics neither in the morning nor in the afternoon as they we were only observed in situations with very low densities and therefore did not act as obstacles. Compared to the benchmark data in with the mean free-flow velocity of 1.34 m/s and a standard deviation of 0.26 m/s, the pedestrians observed at the railway station were slower and more diverse. We would like to point out that these observations may depend on the cultural background and that even qualitative results should not be carried over to other scenarios without verifying. ## Currently prevailing density-flow relationship: fundamental diagram The density-flow relationship observed on the video records also deviated from the fundamental diagram provided by Weidmann which is usually associated with a mean free-flow velocity of 1.34 m/s. The maximum difference in flow was 0.4 persons/ms at a density of a little less than 1 persons/m². Unfortunately higher densities did not occur in that scenario. 90% of the measured data points were within 0.15 persons/ms of the smooth approximation of the data. ## Simulation tool ### Pedestrian stream models There are various models of pedestrian motion, all of them with their own merits,. On [www.ped-net.org](http://www.ped-net.org) alone, 65 different tools are listed, not including the tool described here. For more complete surveys we refer to – with descriptions of a large number of approaches, – for modeling pedestrian movements. Two particularly well established classes of pedestrian motion models, cellular automaton based models – and social force models, both embrace similar ideas: Both use inspiration from similar physical models, such as the idea of repulsive forces between pedestrians to ensure that they keep a distance among each other. The mathematical formulations, however, are very different. Despite the differences, the methodology presented further is suitable for both classes of models and for any other model that is able to calibrate according to the parameters and measurements we describe below, as for example,. ## Cellular automaton model In this section we describe our benchmark simulation tool. It has been introduced by the authors and by their colleagues in earlier publications. Hence, we restrict this description to the minimum necessary to understand the paper. Although the methodology we present here is independent of the specific model, comprehending the main principles of our model helps to identify how empirical results may serve as input. In particular, we aim to identify the data that can be used directly as input from observations and the data that serves as a basis for model calibration. Our goal is to construct a simulation model that is capable of reproducing real- life scenarios and preemptively predicting critical situations, such as life threatening local densities, faster than real-time. That is, it must build on observational data and allow for exceptional computational speed. Our benchmark simulation tool based on a cellular automaton fulfills both requirements. Since we stick to measurable data, many psychological aspects are neglected in our model. Differences between virtual pedestrians are extremely reduced, and the pedestrian model is based on very few simple assumptions: - Pedestrians “know” the shortest path. Limited vision of pedestrians is neglected as well as incomplete knowledge of the terrain. Pedestrians move from their current positions towards individual targets along shortest obstacle-free path, unless such a path is blocked by a fellow pedestrian. - They move at individual preferred speeds – the free-flow velocities – as long as the path is free. - Each individual has a need for private space that depends on his/her current situation. This need is expressed in the distances that individuals try to keep from each other. It also makes people keep distances from obstacles such as walls. - Pedestrians decelerate when the local density in their direction of movement is increased. - All further individuality, such as age or fitness, is captured by the personal free-flow velocity. Clearly, the behavior of pedestrians is strongly simplified with this approach. This has two major advantages: The number of input parameters that must be procured from measurements is low whereas the computational speed of the simulation tool is very high. Both aspects are of extreme practical importance. ### Cellular automaton As in any cellular automaton model, the area of observation is divided into a lattice of cells. Although square cells seem to be the most popular choice, we prefer a hexagonal grid for its two additional natural directions of movement compared to the square grid. The cell diameter is set to 53 cm to accommodate an average sized Caucasian male. This parameter can be adapted to better fit e.g. Asian pedestrians or children. Each cell at each time step has one of the following states: either empty or occupied by either a person, an obstacle or a target. ### Spatial dimensions and topology Pedestrians move on a single plane or on multiple planes such as floors. This allows us to consider two spatial dimensions only. Virtual persons enter and leave the scenario through sources and targets, namely entrances and exits. Sources and targets have three types of parameters – their positions, schedules of pedestrian generation/disappearence and, for the sources, also source-target distributions, that give probabilities for selecting pedestrian destinations when generating pedestrians. All these parameters can be taken directly from measurements and fed into the simulation. ### Potential fields In many aspects, our model is similar to other cellular automaton models based on potentials,. In particular, we borrow the inspiration for the rules from electrostatics: Pedestrians are treated as negatively charged particles, say electrons. Pedestrians are attracted by positive charges, such as exits, and repelled by negative charges such, as other pedestrians or obstacles. The forces between pedestrians, targets and obstacles are expressed through suitable scalar functions, the potentials, which are summed up to form an overall potential field. We choose the potential described in the following, note, however, that different approaches could be used for constructing the potential fields. - Each virtual person carries around his or her own potential given by a radially symmetric function of type that is almost zero a few cells away from the person. - Obstacles like walls are assigned positive potentials to make “people” prefer to keep a distance. - The long–range attracting potential of a target is coded in a floor field that corresponds to the arrival time of a wave front traveling with constant speed from the target through the space formed by the obstacles and boundaries of the scenario. This ensures that each pedestrian moves along the shortest path to his or her target as long as this path is free from other pedestrians. - It is also possible to include clumps of pedestrians in the way to a target in the computation of the floor field thus adding a further dynamic aspect to the potential. When a virtual person steps ahead he or she selects the empty neighbor cell with the steepest descent of the overall potential, thereby obstacles and other persons are successfully skirted. The repulsive potential of a fellow pedestrian on the shortest path leads to either evasion or slowing down. ### Sequential update scheme Simulation dynamics follows a specific kind of sequential update scheme. Each person has an individual speed, the free-flow velocity, which he/she tries to achieve – and indeed does achieve when the path is free. The value of the free- flow velocity prescribes how often the corresponding person will be chosen for an update: At each time step all persons are identified that are allowed to move. Faster persons are chosen more often so that on average each person moves with their prescribed speed as long as the path is free. The positions of the chosen pedestrians are updated in the order of their “life-time” in the simulation – that is, the time that has elapsed since their generation. Using the terminology in our model is: microscopic, discrete and deterministic with stochastic aspects, rule-based but potential-driven. ## Methodology This section describes our methodology of comprehensive calibration against a real-life scenario. The first step towards calibration is to identify the key parameters that must be input into the simulation program. In the previous section we have described the data required for our benchmark model: positions of sources and targets, shapes and positions of obstacles, a pedestrian appearance schedule, source-target distributions, velocity distributions and flow-density dependency. Some of the parameters, such as the location and form of obstacles, can be extracted from the data sources and fed directly into the simulation; Other parameters must be estimated by statistically analyzing the data. Some input even comes in the form of a function such as the density- velocity relation. Most parameters change with time and must be calibrated at suitable intervals. Here, we distinguish between quasi-stationary and dynamic parameters: Dynamic parameters change quickly and have an immediate strong impact on the prediction; Quasi-stationary parameters must be re-calibrated at regular intervals of several prediction periods (whereas stationary input needs to be adapted on demand only). In general, what inputs are stationary, quasi-stationary or dynamic depends on the scenario. Distinguishing among different parameter types is important since it helps to determine how often one should adapt the simulation with respect to a certain type of parameter. In the following we describe our classification which appears suitable for our railway station scenario. Still, the principles listed below have a general character and can be applied to various scenarios. The typical duration of a simulated railway station scenario is several minutes and the classification introduced further is related to this duration. We suggest constructing a “basic scenario” from the observations, and then calibrating the more volatile information in shorter intervals: - *Stationary input* – stationary data can be expected not to change over a period of time significantly longer than the scenario duration. For a railway station, the input is assumed to be stationary if it does not change in a range from several hours, to days, weeks, or even months, etc., – for example, the locations of permanent structures such as platforms. However, even this type of data must be checked in regular and relatively short intervals because any alterations in the infrastructure may dramatically impact the scenario. - *Topology* – the area of observation could be divided into walkable and non- walkable areas. The latter includes obstacles with their types, positions and forms and areas where pedestrians are not allowed to enter, such as railway tracks. This data is fed directly into the simulation and can be assumed to remain stationary. - *Positions of sources and targets* – the topology also determines possible positions of sources and targets for virtual pedestrians, that is, the locations where people come from and where they go to. For a railway station these can be: Departure and arrival platforms, entrances and exists. Again, this is direct input data that is mostly stationary. One should, however, check this information at some regular intervals since entries and exits can get closed at any time. ### Quasi-stationary input By quasi-stationary parameters we mean parameters which gradually change over time but can be considered stationary for several prediction periods. On the scale of our simulation scenario, they can be considered stationary for a few minutes. Such parameters often depend on the time of day or type of scenario. We recommend to store typical values in a scenario data base and to use them as default starting values for fast and yet accurate calibration: - The *Distribution of free-flow velocities* is very often taken from literature following Weidmann's suggestions of a normal distribution with the mean velocity 1,34 m/s and the standard deviation of 0,26 m/s. In *Distributions of free-flow velocities* we described how the results of our free-flow velocity measurements differ from literature values. Therefore, in contrast to the usually assumed distribution, we propose to extract the distribution of the free-flow velocities from observations and to generate virtual pedestrian velocities according to the empiric distribution from the last measurement. Note, if a scenario contains stairs, escalators or other types of non-standard planes, the free-flow velocities should be measured additionally for these surfaces. For example, for stairs and escalators up and down the free-flow velocity should be measured separately. - *Density-flow relation* – often Weidmann's diagram is used as fundamental diagram to describe the density-flow relation. However, our measurements suggest that the real behavior of pedestrians may differ. Therefore, we propose to use a measured flow-density relation instead. In particular, to deal with measured scattered values, we propose to use a smooth approximation or piecewise interpolation of the measured density-flow relation as a reference curve, or objective function, for calibration. This can be achieved robustly with standard optimization methods. We expect the relationship to be quasi-stationary which is a very desirable quality to speed up the calibration process: The last set of calibrated parameters gives an excellent starting position for the next calibration; thus, calibration time would not become an issue. However, the assumption of quasi-stationarity of the density-flow relationship remains to be further substantiated through more extensive measurements. Note that similar to the case of the free-flow velocity, the density-flow relationship should be measured separately for stairs, escalators, etc. - *Distance kept from walls* – when moving, pedestrians keep a certain distance from obstacles. In movement of a single person in the presence of walls is examined. We propose to investigate multiple trajectories to derive a distribution of the distances kept to the walls. This distribution can be taken as a reference curve according to which the simulation tool can be calibrated. The statistical data on the distances to walls in our video footage is under investigation at the moment. First observations indicate that the influence range of a wall is about 2 m. - *Source-target relationship* – to direct virtual pedestrians from sources to targets in a way that fits the scenario, statistical information on mapping between sources and targets must be extracted from video footage, or any other type of suitable sensor. The analysis of the video data for our railway station suggests that the distribution differs from scenario to scenario and depends on time, however, usually it changes on time intervals larger than the simulated period of time. Hence we propose to re-evaluate the target-source distribution at relatively short intervals to direct virtual pedestrians from sources to targets in a way that fits the scenario. ### Dynamic input Dynamic parameters change very quickly and therefore demand constant recalibration. An obvious example are pedestrian appearances following, e.g. train arrivals. Neither arrival times nor train occupancies are predictable with precision, so pedestrian appearances are volatile. Since this kind of information changes every few minutes, it limits the maximum span of predictions intervals accordingly. A permanent readjustment process is necessary here. *Schedule of pedestrian appearances and disappearances* – how many pedestrians per second appear from a source? How many disappear at a sink or a target? This data is necessary to feed virtual pedestrians into the simulation and remove them in accordance with the scenario. At the same time this schedule is very volatile. ## Proof of concept: validation and sensitivity study In this section we describe the validation of the model by comparing simulation results to measurements. We also present the results of a sensitivity analysis conducted to determine the parameters, whose small changes have a significant impact on the simulation results. For sensitive parameters, accurate measurements are critical if one aims at reproducing real-life scenario and performing predictive simulations. ### Validation goal The goal of validation is to make certain that a model used to describe some complex real system well matches the characteristics of this system. Our benchmark model was first validated with respect to a number of phenomena and data from laboratory experiments following the suggestions for tests of the RiMEA validation initiative. Then we conducted our own validation tests described in and. In this section we describe the validation of our benchmark model for the real-life railway station scenario. ### Quantities suitable for comparison The simplest way to validate a model against real-life data is to compare video recordings with simulation visually. This is a valuable plausibility check to which validation has been largely restricted, so far. As far as quantitative validation has been attempted in previous investigations, it was performed based on trajectories, flow or velocity comparisons. Here, we would like to go further and identify quantities suitable for forecasting critical situations such as critical densities. For this, we pick the crowd density as it evolves with time in the area of observation: It can be measured quite easily in both, real scenario and its virtual reenactment and is of high practical interest since local high densities indicate *hot spots* where the risk for accidents is elevated. ### Comparison of the real-life scenario with the virtual reenactment As soon as adjustments and calibration against measured data have been performed, a predictive simulation can be started. The velocity distribution, the density-flow relation, the schedule of appearances and disappearances of pedestrians, the source-target distribution and the positions and shapes of sources and targets (s. *Methodology*) were extracted from video recordings and used as input. For the model validation we choose the most challenging scenario where the highest densities together with multi-directional flow were observed: A train arrives, passengers disembark and walk towards exits. There are also other pedestrians walking in the main hall. At first, the pedestrian density is relatively low until the bulk of passengers passes through the area of observation. Then the density decreases again until all passengers from that train are gone. To compare the simulation and the data extracted from the video recordings let us first look at the density evolution in the most busy and complicated area in the scenario: the rectangle covered by a grid as shown in. shows the comparison of densities (simulated to measured) in a time span of 3 minutes: The solid line corresponds to the video footage, the dashed line to the simulation. Consider The simulation reproduces the scenario quite well: The pedestrian density peak occurs at the proper area and time, and it has the correct duration and order of magnitude. However, the simulation somewhat overestimates the density. Part of the difference can be explained by the influence of chance: The simulation is subject to random input. For each new seed the results differ so that individual velocities and trajectories cannot be expected to match. Only a statistical match of measurements and simulation results is possible. We also suspect that real pedestrians coordinate their movements better than virtual pedestrians do: Our virtual pedestrians are quite “short-sighted” and take steps to avoid collision only when they actually “feel” the potential of other pedestrians; Real pedestrians are more likely to plan ahead. This is a typical disadvantage of so-called greedy algorithms that rely on locally optimal choices to enable high-speed simulations. The important question is whether such systematic overestimation is acceptable. In our case we are interested in a warning system for potentially dangerous densities. Therefore we believe that a slight overestimation can be tolerated, whereas any underestimation would render the model unusable. So far the comparison was restricted to a single rectangle in the area of observation. As a next step we look at the whole area of observation. The results are represented in. We observe that the density magnitudes are very similar for both, the video recordings and the calibrated simulation, in the whole area of observation. The slight density overestimation at the hot spot is also visible (calibrated simulation). Most importantly, the position of the density peak in the simulation is very close to the one in the video recordings. In contrast to that, the uncalibrated simulation is not capable of reproducing the observed data. No high densities occur and the small spikes appear randomly. ### Sensitivity analysis We finally investigate how the uncertainty in our model input affects its output: For which parameters does a small change cause a significant change in the simulation results? These sensitive parameters should be measured very carefully whereas non-sensitive ones require only a rough estimation. We introduce a target function, the *fitness*, that quantifies the precision of our simulation. The higher the fitness the better the match between observation and simulation is supposed to be. To calculate the fitness, we cover the scenario area with *N* measurement tiles of unit size 1 m×1 m. In each tile we measure the distance between the observed and simulated densities by counting the number of pedestrians in each unit tile every two seconds (the two second interval is chosen in accordance with the observed velocities). The fitness at time *t* is given by the inverse of the sum over all measurement tiles: Here, *N* is the number of tiles and and are the pedestrian densities observed on the video and in the simulation at tile *k* at sample time *t*. Overall fitness is obtained by summing over all samples in time. This approach quantifies how closely the local densities in the video recordings and in the simulation evolve. The calibrated parameters gathered when applying our methodology yield the benchmark fitness. To conduct the sensitivity analysis, we compare the fitness of the simulation outcome when varying the parameter values by 10%. illustrates the influence of the crucial parameters on the quality of the match. The source-target distribution seems to be the most sensitive parameter and should be measured very precisely for a scenario. Also the schedule of pedestrian appearances on the scenario and the velocity distribution play a very important role for the predictive power of the simulation. # Discussion and Conclusion Pedestrian stream simulations are only helpful for gaining virtual experience and for real-time prediction when they are able to reproduce the corresponding real-life scenario. A vital step to achieve this is calibration and validation of a model against the real-life scenario. So far, the data on realistic behavior of pedestrians under natural conditions was very limited. Calibration and validation in many models was mostly performed for an isolated phenomenon observed in a controlled experiment or for a relative small observation area in case of real-life scenarios as in etc. Therefore, the question was open: How to calibrate and validate a pedestrian stream simulation against complex real-life scenarios? To answer this question we addressed a number of subquestions: - Are standard values for free-flow velocities and density-flow relations always valid? - What characteristics of pedestrian dynamics should be extracted from real- life scenario for simulation calibration? - Which characteristics have a high impact on simulation accuracy? - How to calibrate a simulation based on characteristics of pedestrian dynamics extracted from video recordings? - How to validate a model quantitatively after calibration? To answer these questions we first analyzed video recordings of a real-life scenario at a major German railway station. We provided experimental evidence that people at the railway station walk significantly slower than standard literature suggests. These results support the idea that free-flow velocities depend on the environment. Pedestrians also decelerate more strongly than Weidmann′s fundamental diagram indicates. This underlines the importance of scenario-specific measurements as input data and calibration to measured data, especially if predictive simulations are attempted. Based on data extracted from video recordings we suggested and applied a methodology for model calibration against our real-life scenario. The benchmark model was based on a cellular automaton. The most important aspects and characteristics taken into account are: scenario topology, sources and targets positions, statistical distribution of trajectories between sources and targets, schedule of pedestrian appearances and disappearances (in the scenario), distribution of free-flow velocities, prevailing density-flow relationship, and distances that pedestrians keep from obstacles. Some of the parameters are direct input parameters while others, like the density-flow relationship, are expressed through target functions for parameter adjustments. We demonstrated that the proposed method significantly improves the quality of simulation and the potential prediction accuracy: The success of the proposed approach was tested by comparing evolutions of simulated and observed pedestrian densities. The simulation predicts the density evolution correctly, both qualitatively and quantitatively: Maximum density is somewhat overestimated but never underestimated, so that the simulation can be used as a predictive warning system for potentially dangerous local densities. We also found that the most sensitive parameters are the source-target distribution, the pedestrian appearance schedule and the free-flow velocities. These core parameters should be measured with high precision as they strongly influence the accuracy of the pedestrian simulations. The proposed method is a first step towards comprehensive calibration and adjustment of simulations to complex real-life scenarios. The proof of concept with a benchmark simulation tool shows that short term predictive simulations are possible as long as the crowd is sufficiently dense to allow statistical interpretation. This also defines the limits of pedestrian simulation: Individual behavior can only be captured and reproduced in a statistical sense. Dangers that are triggered by highly individual behavior may be investigated as far as their effect on the crowd is concerned, but one cannot predict when they occur or whether they will occur at all. Furthermore, fast changes – sometimes within seconds – of very sensitive input data, such as pedestrians appearance, make clear that longer-term forecasts can only be understood as possible outcomes among other conceivable scenarios. The only useful procedure here is to simulate scenarios and observe possible outcomes to gain virtual experience. Hence, longer term simulations can be seen as valuable contributions to risk analysis, while short-term simulations have the potential to support immediate decisions on safety issues. For example, a short-term prediction could help to decide whether train passengers should be allowed to disembark when a train station is already crowded. Further steps to improve the method can include considering group behavior: In our benchmark scenario single commuters dominated the crowd so that there was no need to incorporate group behavior as in and. However, if groups are present in a crowd, their effect cannot be neglected. # Materials and Methods ## Data extraction from videos All trajectories were extracted from the video recordings using a partly automated tool that allowed to “click” positions on the video recordings. We analyzed five recordings in total, each of which had a duration of at least 1.5 minutes. The number of pedestrians on each recording was about 400. One video recording was made at 7 a.m. in the morning of a workday, other recordings were made in the afternoon between 4 p.m. and 6 p.m., also on a workday. For each trajectory detected on video recordings, its source and its target were identified: usually the first and the last detectable position. Some meaningful sources and targets may have been obscured by obstacles or hardly visible from the camera angle. In such cases we used additional information like the direction of pedestrian movement to identify, for example, a likely exit. Based on this information we identified how likely a person coming from source A is to choose target B thus collecting the source-target statistics. Also using the extracted trajectories resolved in time, we were able to derive the schedule of pedestrian appearance at each source and disappearance at each target on the scenario. Trajectories that were partially obscured from the camera view or where the view was distorted by distance were only used to gather source-target statistics. Detailed analyses of velocities, flows and distances to walls were conducted exclusively on the visible and undistorted parts of the trajectories to keep measurement errors small. While the programs we used also introduced errors, we estimate that they were small compared to the error committed by manually pinning down the center position of each person's head when tracking pedestrian trajectories. We had no cost-effective way to systematically investigate this error. However, we think it is safe to assume that the head was always correctly identified, but with a deviation from the center. In this case, an error in position would not exceed 9 cm – the radius of a circular approximation of a human head. ## Ethics No ethics statement is required for this work. The video footage was recorded by the train station operator, Die Bahn, in accordance with § 6b (1) of the German Data Protection Act (DPA) and provided to us already anonymized (in very low resolution making no personal identification possible, § 3 (6) of the DPA). Usage of such anonymized data is exempt from further regulation under DPA as per § 1 (2). We would like to thank Lev Olkhovich and Florian Wilhelm Geiss for their support. [^1]: Dr. Maria Davidich is a Siemens AG employee. Siemens AG has provided the clearance to publish the paper as it is. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials. The methods described in the following patent were used to assign sources and targets to pedestrians in this study. However, the patent is unrelated to the core subject of the study: M. Davidich “Method and device for assigning sources and sinks to routes of individuals” EP Patent 2,447,882//US Patent App. 13/285,759, 2011, Publication date 2011/10/31. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: MD. Performed the experiments: MD. Analyzed the data: MD GK. Contributed reagents/materials/analysis tools: MD. Wrote the paper: MD GK.

# Introduction Schizophrenia is a severe, chronic, often recurring mental disorder affecting 1% of the general population, and it is associated with a relevant long-term impact on patients’ social and occupational functioning. It is treated with a combination of medical, psychological and psychosocial interventions, with varying degrees of success. The economic consequences of schizophrenia, defined as costs of illness generated by the aggregation of direct and indirect costs, are considerable. Direct costs include those associated with inpatient (i.e., hospitalizations) and outpatient treatments, long-term care, costs of medications, and justice costs. Indirect costs arising from loss of productivity suffered by individuals with schizophrenia and their family members. The main reasons for such a high economic burden of this disorder are complex clinical processes related to the early onset, the chronic nature with frequent relapses and high rates of hospitalizations. Complex clinical processes include all activities provided by healthcare professionals addressing patients’ healthcare issues, that refer not only to hospitalizations but also to emergency and planned outpatient visits. In order to increase the quality of care and to reduce treatment costs, it is of paramount importance to optimize those clinical processes. Non-adherence to antipsychotic medications is one of the most important factors increasing relapses in schizophrenia. About 60% of patients with schizophrenia are non-adherent to antipsychotic medications already in the first phases of the illness and are less likely to be compliant later on. Most guidelines for the management of schizophrenia recommend improving medication adherence as a strategy to reduce hospitalization rates and costs. A systematic review and meta-analysis of 25 mirror-image studies in patients eligible for clinical use of LAIs showed strong superiority of LAIs compared to oral antipsychotics in preventing hospitalization. These results are in contrast to the meta-analysis of randomized controlled trials (RCTs), which showed no superiority of LAIs in preventing relapse and hospitalizations. Another recent study, carried on an administrative database analysis in Japan, showed that LAIs, compared to oral antipsychotics, reduce rehospitalizations and emergency visits. A very recent Swedish study analyzed 29,823 patients with schizophrenia from nation-wide register-based data to evaluate the risk of rehospitalization and treatment failure. The authors found that clozapine and LAIs are the pharmacological treatments with the highest rate of relapse prevention. However, most of the studies have been conducted under controlled conditions and have evaluated rehospitalization rates only. Given the possible biases in mirror-image studies, such as expectation biases, natural illness course, and time-effect, a cautious interpretation is required. Nevertheless, the population in mirror–image studies better reflects the population receiving LAIs in clinical practice. In this study, our goal is not to test the efficacy of LAIs compared to oral medications, but we aim to evaluate the effectiveness (i.e., efficacy under ordinary circumstances) in terms of clinical process management in specific patients, with a diagnosis of schizophrenia spectrum disorder, who needed to switch from oral to LAI therapy in real-life conditions. In order to obtain real-life measures, patients had to be treated in community mental health centers. The effectiveness of antipsychotic medications was evaluated through means of hospitalizations, emergency and planned visits. # Material and methods ## Study design An observational, retrospective, naturalistic, mirror-image study was designed to determine the efficacy of LAIs compared to oral antipsychotics. The use of mirror-image study design does not include a parallel active control group; instead, each patient serves as their own control. As a result, it cannot be determined whether other treatments may have had similar effects. We defined Time 0 (T0) as the time in which each patient switched from oral to LAI antipsychotic medication. Patients were recruited in 5 community mental health services of the Department of Mental Health of Bari. We informed the local ethical committee prior to initiating the study, in line with Istituto Superiore di Sanità protocol. Each patient was assigned with an ID code to guarantee anonymity. All the patients whose data were collected had previously signed the informed consent, present in the medical record, to the processing of personal data and the use of the data for research purposes. Given the naturalistic design of the study, the results remained purely observational and researchers did not influence the results in any way. The study design is detailed in. ## Study sample The clinical and electronic (SISM Experia, Italy) files of all patients attending five community mental health services of the Department of Mental Health of Bari (ASL BA) and receiving LAI antipsychotic medications from July 2007 to June 2017 were analyzed. Exclusion criteria were: a) diagnosis of schizophrenia spectrum disorder according to the DSM-5 criteria for less than one year before T0; b) LAI concomitant antipsychotic medication; c) substance use disorder or of intellectual disability disorder; d) a major change in life situation (i.e., admission in residential facilities programs) in the year before and after T0. All patients had been treated with oral antipsychotics one year before T0 and with LAIs for one year after T0. Patients with illegible medical records were excluded. ## Study measures and end-points Patients’ demographic characteristics, including age, gender, educational level, diagnosis and sub-diagnosis of schizophrenia spectrum disorder, duration of illness at T0 and oral psychopharmacological medications before and after T0, were registered for all patients included in the analyses. For all patients, the following information was collected one year before and one year after T0: type of LAI antipsychotic treatment (in particular first or second generation antipsychotic); number of hospitalizations; number of emergency visits; number of planned visits. We defined the primary end-points of the study: (i) hospitalization rates, (ii) total number of hospitalizations, (iii) emergency rates, and (iv) total number of emergency visits, as associated with the severity of the condition of the patient. As secondary end-point, we considered the (v) total number of planned outpatient visits, as associated with the therapeutic compliance and alliance of the patient. In the number of planned outpatient visits, we excluded the planned contacts with nurses for injection administration. Hospitalization rates were calculated as the proportion of patients with ≥ one psychiatric hospitalization. Similarly, emergency rates were calculated as the proportion of patients with ≥ one emergency visits. First and second-generation LAI antipsychotics were compared one year before and one year after T0 on all assessed outcome measures. ## Statistical analyses Effects of treatment (before and after T0), LAI generation, age, gender, educational level, diagnosis, and illness duration on the end-points were measured. We used GEE (Generalized Estimating Equations) models to account for within-subject correlations. These preliminary analyses revealed that the treatment and LAI generation seems to have major effects. Therefore, we studied in deep with specific tests the effects of treatment and LAI generation. Non-parametric tests were used since the distribution of the dependent variables (hospitalization rates, total number of hospitalizations, total number of planned outpatient visits, emergency rates, and total number of emergency visits) was non-normal. The McNemar test was used to study the effect of LAIs on hospitalization and emergency rates in order to determine if the number of patients that were hospitalized / required emercency visits before T0 (dependent variables: “hospitalization”/ “emergency”; “yes” or “no” categories) decreased after the introduction of LAIs. Wilcoxon test was used to study the effect of LAIs on total number of hospitalizations, total number of planned outpatient visits, and total number of emergency visits before and after T0. The Mann- Whitney U test was used to compare the differences between first and second- generation antipsychotics on the four analyzed dependent variables. To establish the real-world benefit, we plan to test the hypothesis that it is necessary to demonstrate the LAI effects on “all” the primary endpoints. If this hypothesis is rejected, we can test the weaker hypothesis that the demonstration of the LAI effect on at least one of several primary endpoints is sufficient. In this case, correction for multiple comparisons should be performed to control the Type I error. We chose to apply the Holm–Bonferroni method, if the case. # Results ## Subjects Data from 207 patient records were collected. Sixty percent of them were male, with a mean age of 47.9 (SD 12.0) years, a mean duration of illness of 15.8 (SD 8.7) years, a mean educational level of 9.2 (SD 3.6) years. They had a diagnosis of schizophrenia (48%), schizoaffective disorder (30%), or other specified schizophrenia spectrum and other psychotic disorders (22%). 49% of patients were in monotherapy with LAIs. At T0, 68% of patients were treated by second generation LAIs (paliperidone: 27.5%; risperidone: 22.2%; aripiprazole: 13.5%; olanzapine: 4.8%) and 32% by first generation (haloperidol: 14.0%; fluphenazine: 11.6%; zuclopenthixol: 4.8%; perphenazine: 1.4%). ## GEE models GEE model analyses, whose results are reported in, revealed that the treatment has an effect on all the end-points. LAI generation seems to have an effect on hospitalization rates, emergency rates, and total number of emergency visits; it has a lower effect on total number of hospitalizations and no effect on total number of planned outpatient visits. As to the other characteristics, there is not enough evidence to conclude that they have an effect on the end-points. ## LAI versus oral antipsychotic treatment After switching to LAIs, the number of hospitalizations was drastically reduced from 187 (0.90 hospitalizations per patient/year) to 20 (0.10 hospitalizations per patient/year) (Wilcoxon test; N = 207; Z = -9.769; p\<0.001). Hospitalization rates (from 61.8% to 5.3%; McNemar Test; N = 207; χ2 = 113.076; p\<0.001), emergency rates (from 66.7% to 24.2%; McNemar Test; N = 207; χ² = 77.235; p\<0.001), and emergency visits (from 337 to 106; Wilcoxon test; N = 205; Z = -9.109; p\<0.001) also significantly decreased after the introduction of LAIs. On the contrary, planned outpatient visits significantly increased (from 6.4 to 9.1; Wilcoxon test; N = 205; Z = -6.125; p\<0.001). The number of hospitalizations was significantly reduced by both first and second generation LAIs compared to oral antipsychotic treatment (Wilcoxon test; first generation LAI; N = 66; Z = -4.965; p\<0.001; second-generation LAI; N = 141; Z = -8.427; p\<0.001). Hospitalization rates were significantly reduced by both first and second generation LAI compared to oral antipsychotic treatment (first-generation LAI; N = 66; χ² = 27.034; p\<0.001; second-generation LAI; N = 141; χ² = 84.100; p\<0.001). Again, the emergency visits were significantly reduced by both first and second generation LAIs compared to oral antipsychotic treatment (Wilcoxon test; first- generation LAI; N = 66; Z = -5.214; p\<0.001; second-generation LAI; N = 139; Z = -7.565, p\<0.001). The emergency rates were significantly reduced by both first and second generation LAIs compared to oral antipsychotic treatment (first-generation LAI; N = 66; χ² = 28.033; p\<0.001; second-generation LAI; N = 141; χ² = 47.779; p\<0.001). Planned outpatient visits significantly increased with both first and second generation LAIs compared to oral antipsychotic treatment (Wilcoxon test; first- generation LAI from 5.0 to 10.4 per patient per year; N = 66; Z = -5.191; p\<0.001; second-generation LAI from 7.0 to 8.5 per patient per year; N = 139; Z = -3.563; p\<0.001). ## First vs. second generation LAI antipsychotic treatment Second generation LAIs were significantly more effective than first-generation LAIs on all primary endpoints, except emergency rates. The decrease of number of hospitalizations was significantly higher for second generation LAIs (first- generation LAI reduced from 50 to 4, second-generation LAI reduced from 137 to 16; Mann-Whitney U Test; N = 207; U = 3860; p = 0.033). The decrease of hospitalization rates is also significantly higher for second generation LAIs (first-generation LAI reduced from 47.0% to 3.0%, second-generation LAI reduced from 68.8% to 6.4%; Mann-Whitney U Test; N = 207; U = 3779; p = 0.011). The decrease of emergency visits is significantly higher for second generation LAIs (first-generation LAI reduced from 74 to 16, second-generation LAI reduced from 263 to 90; Mann-Whitney U Test; N = 205; U = 5347; p = 0.046). The decrease of emergency rates is not significant (first-generation LAI reduced from 59.1% to 13.6%, second-generation LAI reduced from 70.2% to 29.1%; Mann-Whitney U Test; N = 207; U = 4527; p = 0.720). As to planned outpatient visits, after switching to LAIs, there is no statistically significant difference between the two generations of LAIs (Mann-Whitney U Test; N = 205; U = 3945; p = 0.105). However, since the number of planned outpatient visits was significantly lower for patients treated with first generation LAIs (Mann-Whitney U Test; N = 205; U = 3048.5; p\<0.001), this leads to a higher statistically significant increase for first-generation LAIs (Mann-Whitney U Test; N = 205; U = 3115; p\<0.001). # Discussion In our study, we considered the number of hospitalizations, the number of emergency and planned visits one year before and after the change from oral to LAI antipsychotics. For each patient, the effectiveness (i.e., the efficacy under ordinary circumstances and not under controlled circumstances) of LAIs was measured and compared. Our findings, consistently with those reported in the literature, show that the efficacy of antipsychotic medications in enhanced by the use of LAI formulations. The main novelty of our study is the focus on the optimization of clinical processes and healthcare resources with LAIs, not only in terms of hospitalization rates but also of emergency and planned visits, which are rarely considered in studies on the effectiveness of antipsychotic medications. Consistently with previous studies, all our primary endpoints show that the use of LAIs is significantly associated with a reduction of hospitalization rates and emergency visits. Moreover, the number of planned outpatient visits with physicians significantly increases after the introduction of LAIs, although we excluded the planned contacts with nurses for injection administration. This finding clearly indicates that the use of LAIs is associated with a better distribution of resources, which should be taken in serious consideration by clinicians, policy makers and all stakeholders involved in the mental health field. Contrary to what we could anticipate, the increase in the number of planned outpatient visits from 6.4 to 9.1 per patient per year suggests that the introduction of the monthly injection increases the number of contacts of patients with the local mental health center. This means that the use of LAI antipsychotics is associated with a more focus on patients’ real-life needs and more time dedicated to psychosocial interventions, as suggested by most international guidelines. We believe that, in case of infinite available resources, outpatient visits should be planned at least every six weeks for a better therapeutic alliance and for a better patient’s motivation to join integrated treatments. The increase in planned visits correlates with better pharmacological adherence and rehospitalization prevention. A better therapeutic alliance, due to increased planned visits, could be itself a more successful approach to relapse prevention. As regards the possible effect of the hospitalization length of stay, the average duration of the hospitalization for schizophrenia spectrum disorder is 18.1 days in our Department of Mental Health. This value combined with the average number of hospitalizations per patient/year (0.903 for oral, 0.097 for LAI treatment) makes quite small the possibility that a long period hospitalization may preclude a patient from readmission or emergency visits. Moreover, these rare events would contribute to reducing the evidence of a significant reduction of readmission or emergency visits, which our results show.” Furthermore, we analyzed the effect of first versus second generation LAI antipsychotics on our endpoints. Several recent studies showed that second- generation LAI antipsychotics are superior to first-generation LAIs on treatment adherence (defined as the number of non-overlapping days of supply divided by the number of days in the observational period of 365 days) and rehospitalization risk. To our knowledge, for the first time, our findings show that second-generation LAIs are more effective also in reducing the number of hospitalizations and emergency visits, although the same number of planned outpatient visits with first generation LAIs. These data suggest that second- generation antipsychotic LAIs improve more effectively the clinical management of psychosis also when compared with first generation LAIs. Of course, our study has some important limitations. A first important limitation is the choice of a mirror-image study design without a control group. The choice of a control group, i.e. patients with the same propensity score at T0 who continued on oral medication, is quite difficult because the true propensity score is never known in observational studies. Moreover, RCTs also present selection bias due to the enrolment of patients with different therapy adherence from real-world settings and, furthermore, in such design the trial itself could affect patient outcomes (Hawthorne effect), because of the social treatment and the increased personal attention often associated with participating in trials. Our design choice is supported by several other authors who used studies designed without control groups. A second important limitation is that we took into account only one-way of switching for two reasons. We were not able to collect from medical records the number of patients discontinuing LAIs. Indeed this limitation in the collected data may bias the results positively. However, with a specific focus group, we estimated, between 10% and 15%, the number of participants who discontinued LAIs in our department. This data is compatible with what reported in the literature with similar study designs. These selection biases represent limitations for nearly all pragmatic studies. Nevertheless, even with their imperfections, these studies better reflect the broad range of patients in the “real-life” management of schizophrenia spectrum disorders which is fundamental for the assessment of clinical process management. A third limitation is that we did not collect any measure on patients’ psychiatric symptoms and functionality. No subjective questionnaires on quality of life or satisfaction with therapy were given to patients, as seen in other studies. Given the naturalistic and retrospective design of this study, we only used illness duration as an indirect measure of illness severity. However, this study aims to evaluate clinical process management and not clinical outcomes. Moreover, better outcomes in clinical processes could indirectly suggest also better clinical response at least in terms of symptoms. Another limitation is the retrospective design of the study and the fact that the observation period was limited to one year, whereas a longer period could have provided more information. This methodological choice was due to the fact that we wanted to analyze the effectiveness of LAI medications in the real world and one-year observation, albeit short for sophisticated trials, may be considered adequate for this type of studies. Due to these limitations, we are aware that the study design is inadequate to draw causal conclusions about the effectiveness of LAI as opposed to oral antipsychotics, but an alternative interpretation of our results can be that LAIs are definitively more effective on the population of patients who need to switch to LAI in their clinical history. We can only suppose, because of the lack of data on the motivations for the switch, that the subjects of our study had shown bad therapeutic adherence, and therefore had been switched to LAI. In conclusion, our study combines for the first time: (i) a retrospective, naturalistic and mirror design; (ii) data analysis from medical records using each patient as control of her/himself, i.e. for each patient, the medical history before LAI is compared with her/his history after LAI; and (iii) the analysis of the total number of emergency visits and the of planned outpatient visits. A key point is that data were collected from community health records, in order to control for confounding variables (e.g. chances in life context) that may influence the outcomes we considered in the analyses (i.e., treatment adherence, hospitalization rates, emergency visits). These results suggest a relevant advantage by using LAIs for all stakeholders involved in the mental health field, including policy makers, mental health professionals, patients and caregivers. In particular, from the perspective of policy makers, the use of LAIs may result in marked savings given the significant reduction in the hospitalization rate. Mental health professionals could optimize their work considering that emergency interventions require more resources in terms of time, employed staff and risk factors for professionals’ and patients’ safety (e.g., accidents). In fact, a more “virtuous” longitudinal observation of patients may reduce the risk of burnout in professionals. As regards the patients, a reduction in relapse rate improves prognosis and psychological personal burden, related to hospitalization treatments and may prevent family burden. Finally, LAI antipsychotics actually reduce the severe economic burden of schizophrenia spectrum disorders, not only in terms of direct and indirect costs, but they can also improve other costs (e.g. the costs of justice and of law enforcement interventions), which are mainly related to emergency visits. We are currently evaluating the effect of LAI treatment on the economic costs of this mental disorder on society, and we will try to assess it in future work. # Supporting information 10.1371/journal.pone.0230051.r001 Decision Letter 0 Lu Kevin Academic Editor 2020 Kevin Lu This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 2 Dec 2019 PONE-D-19-20664 Improving the “real life” management of schizophrenia spectrum disorders by LAI antipsychotics: a one-year mirror-image retrospective study in community mental health services PLOS ONE Dear Dr latorre, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. We would appreciate receiving your revised manuscript by Jan 16 2020 11:59PM. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Kevin Lu, PhD Academic Editor PLOS ONE Journal Requirements: 1\. When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_m ain_body.pdf> and <http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_f ormatting_sample_title_authors_affiliations.pdf> 2.  Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: <http://journals.plos.org/plosone/s/supporting-information>. 3\. Thank you for providing the additional information regarding your ethics review at submission. In your methods section, please ensure that you discuss the ethical considerations. Based on the information you provided we would suggest the following: 'We informed the \[INSERT SPECIFIC NAME OF ETHICS COMMITTEE\] prior to initiating the study, in line with Istituto Superiore di Sanità protocol.  Each patient was assigned with an ID code to guarantee anonymity. All the patients whose data were collected had previously signed the informed consent, present in the medical record, to the processing of personal data and the use of the data for research purposes. Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors examined the effectiveness of LAIs compared to oral medications in patients with schizophrenia spectrum disorder. Their primary outcome measures were total number of hospitalizations, hospitalization rates, and total number of emergency visits. As a secondary outcome, the authors examined the number of planned outpatient visits as a measure of therapeutic compliance and alliance. These outcome measures were also examined between first and second-generation LAIs. The authors found that LAIs compared to oral medications significantly reduced hospitalizations (total number and rate) and increased the number of planned visits. They also report that second-generation LAIs are superior to first-generation LAIs in reducing the rates of hospitalization and emergency visits. Overall, the manuscript is well-written and concise. However, the following issues should be addressed: 1\. The authors included 207 patients. Did the authors include all patients who switched from oral medication to LAIs or only patients who have 1-year follow-up data available? If the authors are only selecting cases with 1-year follow-up information available, the authors may be biasing the results by only selecting cases with good outcomes. 2\. Related to the above comment, do the authors have information on the number of participants who switched to LAIs, but subsequently discontinued? 3\. Introduction: Should it be 29,823 instead of 29.823? 4\. Introduction: “These results are in contrast to meta-analysis of RCTs, which showed no superiority of LAIs” – the authors may want to clarify what they mean by superiority (e.g., in preventing relapse, hospitalizations, adherence, clinical outcomes?) 5\. Methods, Study Sample: “Patients with inadequate data were excluded”. The authors may want to clarify what they mean by “inadequate”. 6\. Methods: The authors considered hospitalization rates, total number of hospitalizations and total number of emergency visits as their main outcome measures. Did the authors also look at rates of emergency visits? 7\. Methods/Results: In comparing first- and second-generation LAIs, I am wondering why the authors used Mann-Whitney U test for hospitalization rates instead of McNemar test. 8\. I apologize if I missed this, but were the patients on any concomitant medications? 9\. Results: The authors report total number of hospitalizations. Can the authors also provide the mean number of hospitalizations before and after T0? 10\. Results: It will be helpful to include a table with the demographic and clinical characteristics of the included participants. 11\. Results: Did the authors find any differences in demographic or clinical characteristics between patients who had first compared to second-generation LAIs? Reviewer \#2: Dear Dr. Lu, Thank you for the opportunity reviewing the manuscript “Improving the “real life” management of schizophrenia spectrum disorders by LAI antipsychotics: a one-year mirror-image retrospective study in community mental health services” submitted to the Plos One Journal. In this paper, the authors examined the effectiveness of long-acting injectable antipsychotics (LAIs) compared to oral medications among patients with schizophrenia over a 10-year period. The authors found that LAIs, in particular, second generation ones, were associated with reduced hospitalization rates and emergency visits, at the same time improved the economic burden of schizophrenia. While the study design is relatively robust and findings important to clinical practices, there are a few limitations that need to be addressed before accepted for publication. Major issues: • The study design required subjects to be followed for a year before switching to LAIs and at least a year after T0. This could exclude patients who died or lost to follow-up over the post-T0 period. This immortal bias of a year could make the eligible subjects artificially better outcomes due to selection bias. • Number of hospital admissions alone may not be sufficient in demonstrating the comparative effectiveness of LAIs and oral drugs. A person can be hospitalized for long periods, and technically preclude him from readmission/ emergency room visits. The total length of stay is an important outcome and should be interpreted along with other healthcare utilization patterns. • Using non-parametric tests is insufficient to prove statistical differences in the treated group compared to themselves before switching to LAIs. One should consider the effects of other comorbidities which could affect hospitalization/emergency room visits. Also, due to within subject correlation, events from the same subjects should be considered in comparison. I would suggest using GEE models to account for within subject correlations. Minor issues: • The paper speaks of costs but without mentioning any measurement of costs in monetary terms. In order to be cost-effective, comparing number of hospitalization and ER visits are not enough. Per month per person costs and per ER visits costs should be analuzed, while accounting for differences in baseline medical needs. • The terms of “hidden costs” is being used loosely and best avoid or specified. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0230051.r002 Author response to Decision Letter 0 16 Jan 2020 REVIEWERS' COMMENTS: \*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\*\* \*\*\* Reviewer: 1 Overall, the manuscript is well-written and concise. However, the following issues should be addressed: 1\. The authors included 207 patients. Did the authors include all patients who switched from oral medication to LAIs or only patients who have 1-year follow-up data available? If the authors are only selecting cases with 1-year follow-up information available, the authors may be biasing the results by only selecting cases with good outcomes. 2\. Related to the above comment, do the authors have information on the number of participants who switched to LAIs, but subsequently discontinued? Response We thank the reviewer for this comment. As we already stated in the study limitations in the discussion section, we could not collect information on patients who discontinued LAI therapy. However, with a specific focus group, we estimated, between 10% and 15%, the number of participants who discontinued LAIs in our department, This data is compatible with what reported in the literature with similar study designs \[20, 21\]. We added and clarified in the revised version of the paper in the Discussion Section: “We were not able to collect from medical records the number of patients discontinuing LAIs. Indeed this limitation in the collected data may bias the results positively. However, with a specific focus group, we estimated, between 10% and 15%, the number of participants who discontinued LAIs in our department. This data is compatible with what reported in the literature with similar study designs \[20, 21\]. These selection biases represent limitations for nearly all pragmatic studies \[11\]. Nevertheless, even with their imperfections, these studies better reflect the broad range of patients in the “real-life” management of schizophrenia spectrum disorders which is fundamental for the assessment of clinical process management.” 3\. Introduction: Should it be 29,823 instead of 29.823? Response We corrected the number format. 4\. Introduction: “These results are in contrast to meta-analysis of RCTs, which showed no superiority of LAIs” – the authors may want to clarify what they mean by superiority (e.g., in preventing relapse, hospitalizations, adherence, clinical outcomes?) Response We clarified, in the revised version of the paper “…which showed no superiority of LAIs in preventing relapse and hospisalizations \[8\].” 5\. Methods, Study Sample: “Patients with inadequate data were excluded”. The authors may want to clarify what they mean by “inadequate”. Response We clarified, in the revised version of the paper: ” Patients with illegible medical records were excluded” 6\. Methods: The authors considered hospitalization rates, total number of hospitalizations and total number of emergency visits as their main outcome measures. Did the authors also look at rates of emergency visits? Response We added it as a primary end-point as suggested by the reviewer. We report the text added in the revised paper and we also modified Figure 3 accordingly. Figure 4 was not modified because the difference in the variation of emergency rates between first- and second-generation LAI is not significant. “We defined the primary end-points of the study: (i) hospitalization rates, (ii) total number of hospitalizations, (iii) emergency rates, and (iv) total number of emergency visits, as associated with the severity of the condition of the patient.” “\[…\] emergency rates (from 66.7% to 24.2%; McNemar Test; N=207; χ2=77.235; p\<0.001) also significantly decreased after the introduction of LAIs.” “The emergency rates were significantly reduced by both first and second- generation LAIs compared to oral antipsychotic treatment (first-generation LAI; N=66; χ2=28.033; p\<0.001; second-generation LAI; N=141; χ2=47.779; p\<0.001).” Comparing first- and second-generation LAIs “The decrease of emergency rates is not significant (first-generation LAI reduced from 59.1% to 13.6%, second- generation LAI reduced from 70.2% to 29.1%; Mann-Whitney U Test; N=207; U=4527; p=0.720).” 7\. Methods/Results: In comparing first- and second-generation LAIs, I am wondering why the authors used Mann-Whitney U test for hospitalization rates instead of McNemar test. Response In these analyses, for each end-point, we considered as dependent variable the difference “d” between the value of the end-point before and after T0, and then we compared the two samples first and second-generation LAI. As to the hospitalization and emergency rate, we can have d=-1 if a patient was not hospedalized before T0 and hospedalized after T0, d=1 if a patient was hospedalized before T0 and not hospedalized after T0, d=0 in the other cases. Thus, we have no more two endpoints (hospitalization: yes or no) to analyze and then we cannot use McNemar test. 8\. I apologize if I missed this, but were the patients on any concomitant medications? Response We thank the reviewer for this comment. We explicitly excluded patients with LAI concomitant use of psychiatric oral therapy, but we did not report in the exclusion criteria. We had no information on other medical treatments. We clarified, in the revised version of the paper in the exclusion criteria “b) LAI concomitant antipsychotic medication;” 9\. Results: The authors report total number of hospitalizations. Can the authors also provide the mean number of hospitalizations before and after T0? Response We added, in the revised version of the paper: “the number of hospitalizations was drastically reduced from 187 (0.903 hospitalizations per patient/year) to 20 (0.097 hospitalizations per patient/year)” 10\. Results: It will be helpful to include a table with the demographic and clinical characteristics of the included participants. Response We added the Table in the revised version of the paper “Table 1: Sociodemographic and clinical characteristics of our sample” 11\. Results: Did the authors find any differences in demographic or clinical characteristics between patients who had first compared to second-generation LAIs? Response We clarified, in the revised version of the paper that, using GEE, we found there is not enough evidence to conclude that age, gender, educational level, diagnosis, and illness duration have an effect on the end-points. See response to comment n.3 of reviewer 2. Reviewer: 2 While the study design is relatively robust and findings important to clinical practices, there are a few limitations that need to be addressed before accepted for publication. Major issues: 1 The study design required subjects to be followed for a year before switching to LAIs and at least a year after T0. This could exclude patients who died or lost to follow-up over the post-T0 period. This immortal bias of a year could make the eligible subjects artificially better outcomes due to selection bias. Response Please see response to Reviewer 1, point 1 and 2 2 Number of hospital admissions alone may not be sufficient in demonstrating the comparative effectiveness of LAIs and oral drugs. A person can be hospitalized for long periods, and technically preclude him from readmission/ emergency room visits. The total length of stay is an important outcome and should be interpreted along with other healthcare utilization patterns. Response We thank the reviewer for raising this issue. We agree that the total length of stay is an important factor. We discussed and clarified this effect, in the revised version of the paper in the discussion section: “As regards the possible effect of the hospitalization length of stay, the average duration of the hospitalization for schizophrenia spectrum disorder is 18.1 days in our Department of Mental Health. This value combined with the average number of hospitalizations per patient/year (0.903 for oral, 0.097 for LAI treatment) makes quite small the possibility that a long period hospitalization may preclude a patient from readmission or emergency visits. Moreover, these rare events would contribute to reducing the evidence of a significant reduction of readmission or emergency visits, which our results show.” 3 Using non-parametric tests is insufficient to prove statistical differences in the treated group compared to themselves before switching to LAIs. One should consider the effects of other comorbidities which could affect hospitalization/emergency room visits. Also, due to within-subject correlation, events from the same subjects should be considered in comparison. I would suggest using GEE models to account for within-subject correlations. Response We thank the reviewer for having pointed out this issue. As suggested, we used GEE models to account for within-subject correlations considering the effects of all the variables on the end-points: treatment (before and after T0), LAI generation, age, gender, educational level, diagnosis, and illness duration. We added these results in Statistical Analyses: “Effects of treatment (before and after T0), LAI generation, age, gender, educational level, diagnosis, and illness duration on the end-points were measured. We used GEE (Generalized Estimating Equations) models to account for within-subject correlations. These preliminary analyses revealed that the treatment and LAI generation seems to have major effects. Therefore, we studied in deep with specific tests the effects of treatment and LAI generation.” and a new sub-section in Results: “GEE models GEE model analyses, whose results are reported in Table 2, revealed that the treatment has an effect on all the end-points. LAI generation seems to have an effect on hospitalization rates, emergency rates, and total number of emergency visits; it has a lower effect on total number of hospitalizations and no effect on total number of planned outpatient visits. As to the other characteristics, there is not enough evidence to conclude that they have an effect on the end- points. Table 2: GEE (Generalized Estimating Equations) models for within-subject correlations.” Minor issues: 4 The paper speaks of costs but without mentioning any measurement of costs in monetary terms. In order to be cost-effective, comparing number of hospitalization and ER visits are not enough. Per month per person costs and per ER visits costs should be analyzed, while accounting for differences in baseline medical needs. Response We are currently working on a specific paper on the cost assessment of LAI treatment. We added in the conclusions: “We are currently evaluating the effect of LAI treatment on economic costs of this mental disorder on society, and we will try to assess it in future work.” 5 The terms of “hidden costs” is being used loosely and best avoid or specified. Response We thank the reviewer for this comment. We remove the term “hidden” and specify the type of costs in our context. “…they can also improve other costs (e.g. the costs of justice and of law enforcement interventions), which are mainly related to emergency visits.” EDITOR’S COMMENTS: We added in the Study Design: “We informed the local ethical committee prior to initiating the study, in line with Istituto Superiore di Sanità protocol. Each patient was assigned with an ID code to guarantee anonymity. All the patients whose data were collected had previously signed the informed consent, present in the medical record, to the processing of personal data and the use of the data for research purposes. Given the naturalistic design of the study, the results remained purely observational and researchers did not influence the results in any way.” 10.1371/journal.pone.0230051.r003 Decision Letter 1 Lu Kevin Academic Editor 2020 Kevin Lu This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 21 Feb 2020 Improving the “real life” management of schizophrenia spectrum disorders by LAI antipsychotics: a one-year mirror-image retrospective study in community mental health services PONE-D-19-20664R1 Dear Dr. latorre, We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements. Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication. Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at <https://www.editorialmanager.com/pone/>, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. With kind regards, Kevin Lu, PhD Academic Editor PLOS ONE 10.1371/journal.pone.0230051.r004 Acceptance letter Lu Kevin Academic Editor 2020 Kevin Lu This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 26 Feb 2020 PONE-D-19-20664R1 Improving the “real life” management of schizophrenia spectrum disorders by LAI antipsychotics: a one-year mirror-image retrospective study in community mental health services Dear Dr. latorre: I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. For any other questions or concerns, please email <plosone@plos.org>. Thank you for submitting your work to PLOS ONE. With kind regards, PLOS ONE Editorial Office Staff on behalf of Professor Kevin Lu Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Random Forest (RF) has become a widely-used method for classification and regression analysis of biological data. It often achieves good prediction performance on datasets that are characterized by a large number of features and a relatively small number of samples. For example, Random Forest consistently performs very well in the DREAM prediction challenges. Importantly, the best performance on biological data is generally not achieved with the standard Random Forest implementation. Specific extensions and adaptations have been developed to handle the intricacies of certain biological datasets and associated research questions. These include, but are not limited to: unbalanced classes, heterogeneous feature types, alternative notions of feature importance, methods for feature selection, and robustness against noisy and missing data. Additionally, the huge number of features in biological datasets that are derived from high-throughput genome-wide measurement technologies, such as microarrays and sequencing platforms, necessitates fast RF implementations. We developed CloudForest, a well-documented RF package with a flexible design that enables straightforward implementation of extensions, many of which are already present in the current version. Here, we describe the underlying structure and features of CloudForest and compare it to the most widely used RF packages in terms of prediction performance and computation time. # 1 Methods CloudForest has been written in Go (<http://golang.org/>), a language developed at Google that strives to balance the speed of a low-level compiled language with the ease of development of a higher level language. Specifically, Go code can achieve speeds near that of compiled languages like C while allowing relatively terse code, which is familiar to programmers accustomed to scripting languages like Python or R. Go supports functional programing paradigms that map well to operations with and on decision trees. See for some code snippets to illustrate this point. CloudForest implements rigorously defined interfaces to represent splitting criteria and operations needed for split searching without restricting how the underlying data is represented. This enabled us to represent categorical and numerical data using the data type that is most efficient. As a result, CloudForest can natively handle categorical data types and missing values (Section 1.1). Additionally, CloudForest’s design accommodates rapid implementation of new extensions using the common core functionality. Some of the most useful extensions, such as dealing with unbalanced classes (Section 1.2) and alternative feature importance scores (Section 1.3), are already available. CloudForest’s design was prioritized to minimize training time by smart use of the CPU cache (Section 1.4) and efficient multi-threading (Section 1.5). CloudForest can be run from the command line and from a wrapper script written in an arbitrary programming language that pipes CloudForest commands directly to the terminal. In this way, CloudForest accommodates users that want to implement or experiment with RF extensions as well as users that simply want to use the existing functionalities in CloudForest. links to additional documentation to facilitate users that want to develop novel functionalities in CloudForest. shows a comparison between CloudForest and R’s randomForest package and scikit- learn’s RandomForestClassifier, two of the most widely use RF implementations. ## 1.1 Feature heterogeneity Variables used in the RF can be of different types, i.e. numerical (either discrete or continuous) or categorical (ordinal or nominal), and features derived from biological measurements can fall into any of these types. For example, gene expression values are numerical variables. Features derived from sequencing data are often ordinal categorical, such as binary variant or mutations calls, or nominal categorical, such as zygosity calls. The distribution of values of these features in terms of sparsity and cardinality as well the number of missing values may differ dramatically across the feature set. Features used in CloudForest can be encoded as numerical, (nominal) categorical and binary. The latter is a special case of ordinal categorical features. All other ordinal categorical features are considered numerical. CloudForest offers native support for fast split searching in categorical features via bits packed into integers as in Brieman’s Fortran implementation. To support datasets with a large number of features CloudForest provides an alternative split search that uses Go’s big.Int to efficiently perform arithmetic with big integer numbers. Additionally, CloudForest implements a dedicated optimization for binary features and categorical features with a small cardinality. This enables faster and more efficient computation. CloudForest handles missing values natively using either a bias correction or ‘three-way-splitting’. This is especially important for biological data, where missing values are very common. ## 1.2 Unbalanced classes Many datasets in biology are unbalanced, meaning that there is a considerable difference in the number of samples per class. CloudForest implements several widely used approaches to deal with unbalanced classes, such as roughly balanced bagging of samples and class-specific weighting of errors. Additionally, experimental boosting approaches, such as adaptive boosting, which often increase performance, are implemented as well. ## 1.3 Feature importance Feature importance scores in RF are used as splitting criteria in the decision tree and are often useful as output to the user. Besides the commonly used Gini impurity and squared/L2 error, alternative scores can be quickly implemented. CloudForest includes additional impurity measures including entropy, weighted entropy, weighted Gini, absolute/L1 error and impurities based directly on misclassification cost. Additionally, CloudForest can output the global feature importance scores, mean-minimal-depth-used scores, the local importance scores and sample proximities. Finally, CloudForest enables internal feature selection and feature importance significance testing via Artificial Contrasts with Ensembles (ACE), which provides p-values for importance scores based on the comparison of feature performance to artificial contrasts features. This method has been shown to reduce bias towards high cardinality features. CloudForest further implements optional methods for vetting features during feature selection using out-of-bag samples or artificial contrasts. ## 1.4 Effective use of CPU cache Training the multitude of decision trees that make up the RF requires, for each node in each tree, the selection of the best feature (from a randomly chosen set of features) and the value at which to split this feature. CloudForest employs data layouts and algorithms designed to maximize the efficacy of a modern CPU cache during split selection. The data for each individual feature within the data set is stored in a separate continuous array and all index and accumulator arrays are preallocated and reused within each thread. Inspired by an optimization to the internal sort used in split selection first introduced in scikit-learn v.15, numerical values are copied into a preallocated, ordered array thereby accelerating performance by reducing value lookup time. Specifically, the sorting algorithm in CloudForest was optimized to minimize cache misses by ensuring that the data that need to be accessed during sorting are stored in continuous sections of memory that the operating system will recognize and place on the cache. ## 1.5 Multi-core CPU optimizations RF training is embarrassingly parallel, since individual decision trees are learned independently. To support efficient multi-threading, CloudForest uses lock-free concurrent access to the underlying data where possible. It overcomes one of the main memory bottlenecks for RF by allowing forests to be written to disk as they are grown without needing to store the entire forest in memory. Additionally, the text based format used to store the trees can be easily concatenated, which allows the training of a single large forest to be parallelized across multiple machines. Importance scores and other statistics that must be gathered across the trees are calculated using thread safe data structures. ## 1.6 Ethics statement Concerning the datasets described in Section 2.1: Individuals were recruited at Inova Fairfax Hospital during 2011–2013 and enrolled in the Inova Translational Medicine Institute’s clinical study entitled “Molecular Study of Pre-term Birth.” All study participants provided written informed consent for use of their genome sequences and medical records for research purposes. The “Molecular Study of Pre-term Birth” was approved by the Institutional Review Board of Inova Health System and the Western Institutional Review Board (#1124761). All patient data and information were anonymized and de-identified prior to analysis. # 2 Results We compared CloudForest to R’s randomForest package and scikit-learn’s RandomForestClassifier. R’s randomForest is based on Brieman’s original Fortran code and is the most established implementation. Scikit-learn’s implementation is in Cython (a compiled Python-like language). Scikit-learn v.15 offers one of the fastest available implementations on large numerical datasets. ## 2.1 Computation time and prediction performance We applied these RF implementations on two large biomedical datasets. The first dataset contains 738 clinical features derived from electronic medical records and patient surveys. They include numerical (44), binary (673) and categorical (21) features across a set of 659 samples (patients). The second dataset contains 167,698 genomic features for the same set of patient samples. These features are derived from whole-genome sequencing data, and include binary (87,084) and categorical (80,614) calls of homozygous and heterozygous minor allele variants. Both these datasets are part of an ongoing study on the causes of preterm birth. These datasets, initially described in, are examples of large and heterogeneous datasets, which characterize current data in biology. In particular, clinical features derived from electronic medical records and genomic features derived from sequence data often contain binary or categorical data. The RF implementations were used to classify the 439 cases of fullterm birth and 220 cases of preterm birth, i.e. predict the class labels of these 659 samples. The clinical dataset was used to evaluate classification performance. To estimate the error, we employed a stratified 10-fold cross-validation scheme. The genomic dataset was used to evaluate the speed of the RF implementations. Because R’s randomForest and scikit-learn do not handle missing values, we imputed all missing values to the feature mean for numerical features and feature mode for categorical features before analysis. Additionally, categorical data were encoded as numerical features for scikit-learn, as it does not support categorical data. This encoding was performed by creating a binary feature for each of the categories in the categorical feature. All implementations were set to grow 500 trees with their default parameters. Tests were conducted using a single thread, though both scikit-learn and CloudForest can efficiently do multi-threading. The experiment demonstrated that CloudForest offers a classification performance equivalent to R and scikit-learn on the clinical dataset. However, it is faster than either implementation on the large heterogeneous genomic data set. Noteworthy, R is much slower than CloudForest and scikit-learn. The 24% speed-up with respect to scikit-learn is mainly due to CloudForest’s native ability to handle categorical features. If categorical features are made numerical, both approaches are comparable in terms of running time (Figure A). The extensions available in CloudForest, offer the user the ability to quickly evaluate various algorithmic variations on the RF. We observed that using roughly balanced bagging to handle the class unbalancedness substantially improved performance on the clinical dataset (Figure B in). Additional experiments on benchmark datasets from the LIBSVM data repository demonstrated that in some cases these extensions substantially lower error rates on independent test sets (Figure C). These experiments also demonstrated that the training speed of scikit-learn and CloudForest are comparable. ## 2.2 Missing values One prominent feature of CloudForest is its native support for missing values. We employed several datasets from The Cancer Genome Atlas (TCGA) to demonstrate the usefulness of this feature. Specifically, for each of 6 different tumor types, we created a large dataset containing numerical gene expression features, categorical and numerical copy number features and binary gene mutation features. Each dataset contains hundreds of samples (cancer patients) and thousands of features. We defined the classification task of predicting the mutation status of the often mutated tumor suppressor gene *TP53* using half of the samples for training and the other half for testing. In the training datasets, we artificially introduced missing values using a randomization scheme that preferentially puts missing values in features that are highly correlated with the target, i.e. the mutation status of *TP53* (See). This procedure was performed for increasing percentages of missing values in the dataset, and repeated 10 times. We employed CloudForest and scikit-learn to train RFs on these training sets and test on the independent test set. For scikit-learn, which does not handle missing values, we imputed all missing values to the feature mean for numerical features and feature mode for categorical and binary features before analysis. depicts the results of this exercise for colorectal cancer (CRC), one of the six TCGA datasets. It is clear that CloudForest retains a consistently better prediction performance than scikit-learn, especially for the case of many missing values. We found this pattern across most TCGA datasets, although not always as pronounced as for CRC (Figure D). Moreover, the computation time to train the decision trees in CloudForest decreases with more missing values. This is not the case for scikit-learn, where missing values are imputed before RF analysis, leading to a dataset of the same size. Although the training speed of CloudForest and scikit-learn is similar for datasets without missing values, CloudForest is faster than scikit-learn for datasets with many missing values (Figure E). # Conclusion CloudForest is a high-performance, extensible, and feature-rich implementation of the Random Forest algorithm. It supports classification and regression directly on common data types, such as binary, numerical and categorical features. Custom modification of the algorithm and data handling is provided through a set of general programmatic interfaces. CloudForest is an open source project released under a three clause BSD-style license. The stable implementation released as part of this publication is found here: <https://github.com/IlyaLab/CloudForest>. # Supporting Information This work was supported by the Inova Health System, a non-profit healthcare system in Northern Virginia, and the National Cancer Institute U24CA143835. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: RB RBK. Performed the experiments: RB RBK TAK. Analyzed the data: RB TAK BB. Wrote the paper: RB RBK BB JEN JGV IS TAK. Implementation and design of CloudForest Software: RB RBK.

# Introduction Access to HIV care services, including antiretroviral therapy (ART), is essential for the improvement of health outcomes of people living with HIV (PLHIV). During the period from 2010 to 2019, ART was reported effective in reducing HIV infections by 23% and AIDS-related deaths by 39% globally. However, during the same period, Indonesia experienced a significant increase in the burden of HIV, including an increase of up to 132% infections and 60% AIDS- related deaths, a reflection of the inequitable distribution or limited coverage of ART, late diagnosis, and poor access and adherence to ART. The 2021 national AIDS report shows that of the total number of 427,201 PLHIV in the country, only 63% ever started ART. Of the ones who have started ART, 26.9% have failed to follow up or have stopped the therapy, 18.3% have died, and 53.7% are currently on ART. Globally, previous studies have reported a range of barriers to access to HIV care services among PLHIV. These include limited availability of HIV care services, the shortage of qualified healthcare professionals (HCPs) to deliver the services to PLHIV, long-distance travel to healthcare facilities or HIV clinics and the lack of public transportation A limited approachability of HIV care services reflected in poor dissemination of information about the services and the poor health literacy of PLHIV about both HIV and HIV care services, are also barriers for them to perceive their needs for care and access the services. Poor health literacy is also reported to influence them in making critical decisions about their health, including seeking and accessing available and appropriate healthcare services. The unaffordability of HIV care service-related costs and the inability of PLHIV to afford them due to poor economic conditions are also barriers to their access to the services. Concerns about the confidentiality of HIV status, fear of losing a job if HIV status is known to others, perceived healthy status, lack of time, and psychological burden of undergoing HIV care, are reported as demotivating factors for their access to HIV care services. Similarly, perceived, anticipated and external stigma from families, community members, and healthcare professionals are also barriers to their access to the services. Despite a range of barriers as reported in the aforementioned studies, the literature suggests that evidence on the influence of cultural practices (e.g., the use of traditional medicines (TMs)), family and social factors on access to HIV care services among PLHIV is still limited. TMs in the context of Belu refer to the roots, leaves, and barks of plants that are prepared and mixed by traditional healers (THs) or family members and used for the treatment of HIV in PLHIV. THs in this context refer to people who do not attend formal medical training but are considered by family or community members as competent to provide HIV care using roots, leaves, and barks of plants based on traditional practices. Previous studies in Africa have reported the use of TMs and their influence on ART adherence among PLHIV. However, none of these studies specifically focused on exploring the use of TMs as a barrier to access to HIV care services among PLHIV. For example, studies in Ethiopia and Uganda reported on the common concomitant use of TMs and ART as a barrier to ART adherence. Similarly, a study in Tanzania, Uganda, and Zambia reported on consultation with a TH as a factor associated with poor ART adherence. Only a recent study in Ethiopia by Gesesew and colleagues, which explored the access of women living with HIV (WLHIV) to HIV care services, reported on the restrictions of THs towards the use of ART as barriers for the women to accessing HIV care services. This paper aimed to fill this gap in knowledge by exploring the cultural practice of the use of TMs for HIV treatments in Belu, and family and social influence on HIV treatment of PLHIV. It also explored how these barriers to HIV care access had impacted health outcomes and the life of PLHIV. Our findings will be useful information for government and non-governmental institutions, policy makers, and program planners to address the issue through policies and interventions for better health outcomes of PLHIV in Indonesia and globally. # Methods The report of the methods section was guided by consolidated criteria for reporting qualitative studies (COREQ) checklist. The checklist contains 32 items that need to be covered to support the explicit and comprehensive reporting of qualitative studies. ## Study setting Belu is a district in East Nusa Tenggara province, located in the eastern part of Indonesia. It shares the border with East Timor with a total population of 204,541 people. The district comprises 12 sub-districts and had three hospitals, 17 public health centres located in each sub-district, 21 sub-public health centres, 23 village health posts, 48 village maternity posts, five private clinics, and one HIV clinic where ART is provided. It has a total number of 1,200 PLHIV, but just over half of them (637 people) had accessed HIV care service or ART. Of the ones who had accessed ART, only half regularly accessed and adhered to ART at the time this study was conducted. ART was the only HIV care service available for PLHIV in Belu. The use of TMs for the treatment of any health issues, including HIV, is also a common practice within families and communities in this area, which might have been a barrier to the access of PLHIV to HIV care services. To the best of our knowledge, there has not been any study exploring the influence of TMs and the social influence of families and friends on access to HIV care services among PLHIV in the context of Belu and Indonesia. Belu was selected as the study setting due to the feasibility, familiarity, and potential of undertaking the current study successfully. ## Recruitment and data collection The participants were PLHIV and HCPs recruited using the snowball sampling technique. PLHIV and HCPs were respectively recruited from an HIV clinic and public health centres providing HIV care services in Belu. Following the initial talk and agreement with the heads of these healthcare facilities, the study information sheets for PLHIV and HCPs were initially distributed through the receptionists in these healthcare facilities. The information sheets which contained a brief explanation about the study and contact details of the field researcher (NKF) were posted by the receptionists on their information boards. Potential participants who called to confirm their participation were recruited and asked to suggest a preferred time and place for an interview. The initially interviewed participants were also asked for help to distribute the study information sheets to their eligible friends and colleagues who might be willing to participate. This process was iterative, leading to a total of 46 PLHIV (26 women and 20 men) and 10 HCPs participating in this study. Two people withdrew their participation after a few minutes of the start of the interview due to personal reasons and their information was excluded from this report. One-on-one in-depth interviews were used to collect data from the participants. Interviews with PLHIV and HCPs were respectively conducted in a private room at the HIV clinic and at healthcare facilities where the HCPs worked, which was mutually agreed upon by the field researcher (NKF, male) and each participant. Only the researcher and each participant were present in the interview room. Interviews were conducted in Bahasa for a duration of 35 to 87 minutes and audio recorded. Notes were also taken during the interviews. Pertaining to the topic of traditional treatment, the interviews focused on several key areas, such as PLHIV’s experiences with access to HIV care services; whether or not they regularly access the services; factors that influenced their access to the services; whether or not they used other treatment (e.g., TMs) in addition to ART; the influence of their family members, relatives, friends and neighbours in their HIV treatment; and HCPs’ knowledge about barriers to access to HIV care service by PLHIV. Recruitment of the participants and interviews ceased when the researchers felt that no new added information or data in the responses of the last participants, an indication of data saturation. Interviews were conducted in Bahasa Indonesia, the primary language of the participants and the field researcher who is from Belu and who can also speak English fluently. The field researcher is a PhD student in public health and had attended training on qualitative research through formal education. The participants (PLHIV) were not offered an opportunity to review their transcript to prevent the possibility of the transcript being received by their family members once it was sent to them in hard copies. This could lead to a breach of the confidentiality of their HIV status, in case they had not disclosed it to their family members. No repeated interview was conducted and no established relationship between the researcher and any participants prior to the interviews. ## Data analysis The audio recordings were transcribed verbatim and manually using a laptop by the first author (NKF). Data analysis was performed using NVivo 12 software and guided by the five steps introduced in the qualitative data analysis framework by Ritchie and Spencer. These steps are (i) familiarisation with the data or transcripts through reading the transcripts line by line and repeatedly, providing comments and labels, and highlighting ideas related to barriers to access to HIV care services; (ii) identification of a thematic framework by writing down key issues and concepts; (iii) indexing the data by creating open coding to each transcript, where data extracts of each transcript were given a code or node. This was followed by close coding to identify similar or redundant codes and group them together to reduce the length of the coding list. Codes that formed the same themes or sub-themes were grouped together; (iv) charting data by arranging a thematic framework in a summary of chart; and (v) mapping and interpretation of the data. Data analysis was primarily performed by the field researcher, although the team-based analysis was carried out at regular supervision meetings whereby comments and suggestions were provided and discussed, and then team decisions were made about the validity of the final themes and interpretation. ## Ethical consideration The study was approved by the Social and Behavioural Research Ethics Committee, Flinders University, Australia (No. 8286), and Health Research Ethics Committee, Duta Wacana Christian University, Indonesia (No. 1005/C.16/FK/2019). In addition, a permission letter from the local government of Belu district was obtained prior to the data collection (No. RSU/890/Diklat/423/VIII/2019). All the study participants were informed about these ethics approvals and permission from the government prior to commencing the interviews. They were informed about the purpose of the study through the study information packs and by the research prior to the interview. They were also advised that their participation was voluntary and that they had the right to withdraw their participation for any reasons before or during the interview without any consequences. They were informed that the interviews will be digitally audio recorded and assured that the information they provided during the interview will be treated confidentially and anonymously by assigning each participant a specific letter and number. This helps to prevent the possibility of linking back the data to any individual in the future. Before commencing the interviews, each participant was provided with informed consent to sign and return to the research. # Results ## Demographic profile of the participants The participants’ (PLHIV) age ranged from 18 to 60 years old, and half of them were married, while the others were unmarried. The majority of them (33 people) have been diagnosed with HIV for 1 to 5 years at the time this was conducted, while the rest have been diagnosed with HIV for 6 to 10 years (11 people) and 11 to 15 years (2 people). Several of them have also been diagnosed with other infections, such as herpes, candidiasis, gonorrhoea, and tuberculosis. Most participants graduated from either elementary school or high school (38 people), and only a few graduated from university (8 people). The majority of them had different kinds of jobs, a few reported being housewives (11 women) and some were unemployed (3 women). The age of the healthcare professionals ranged from 30 to 49 (8 people) and the majority of them were married (7 people). All of them graduated from university, eight people were nurses and two were medical doctors. ## Traditional treatment of HIV ### Traditional treatment as a cultural practice The use of traditional medicines to treat any kind of diseases, including HIV/AIDS was common among community members in Belu. These medicines were made of a combination of roots, leaves, and the bark of plants, as one of the participants said “*There are various common traditional medicines which are made of the roots*, *leaves*, *and barks of plants*, *and those plants can be found around us*” (MP9, married). PLHIV acknowledged that the use of traditional medicines provided had been a well-known cultural practice within communities, passed down from one generation to another: > *“It has been our culture since a long time ago*, *from our ancestors > that if people get sick then they would firstly seek for traditional > medicines”* > > *(FP15*, *widowed)*. > *“It (the use of TMs) is a common practice here (in Belu)*. *People > use different types of traditional medicines to treat their health > problems*. *My parents*, *grandparents and so on have used traditional > medicines for years even before I was born…*.*”* > > *(MP7*, *married)*. The practice of traditional medicine use for the treatment of health issues was also recognised by healthcare professionals interviewed in this study. They acknowledged the existence of such practice within communities in the study setting and even within their own families. All healthcare professionals described that the majority of PLHIV in the district who did not access HIV care services or medical treatment were still using traditional medicines for their HIV treatment: > *“The use of traditional medicines is common and has been passed down > from one generation to another until now*. *Those medicines exist > within communities and families*. *Even my family members and > relatives still use traditional medicines (for other health issues)*. > *So*, *I am not surprised that there are also traditional medicines > for HIV treatment and many PLHIV are using those medicines”* > > *(HCP10*, *nurse and counsellor)*. > *“The use of traditional medicines for HIV treatment is very common > among PLHIV in Belu*. *I am sure that most PLHIV who have known their > HIV status but do not start antiretroviral therapy are taking > traditional medicines”* > > *(HCP1*, *medical doctor)*. ### The impact of traditional treatment use on HIV care access and the health of PLHIV The availability and common use of traditional medicines for the treatment of health issues within communities in this area seemed to have a significant influence on the participants’ and other community members’ health-seeking behaviours. The narratives of the participants revealed that traditional medicines had often been used in the first place for the treatment of health issues by many community members in the district. The following quote reflects the availability of traditional medicines and how it influenced health-seeking behaviours of community members, including himself: > *“The use of traditional medicines is a common practice here*. *…*. > *People use traditional medicines to treat HIV too*. *Once I was > tested positive*, *the first treatment my parents thought of and > prepared for me to use was traditional medicines*. *They searched for > those roots*, *leaves*, *and barks of plants*, *and prepared for me*. > *It is because traditional medicines are easy to make and are commonly > used in our community*. *My parents have also been taking those kinds > of traditional medicines…*.*”* > > *(MP4*, *Belu)*. It was apparent that the cultural practice of traditional medicine use influenced or delayed the acceptability of and access to HIV care services among both women and men living with HIV in Belu. It influenced the initiation of and retention in HIV treatment. Nearly half of the women and men interviewed described that they did not access the HIV care service straightaway following their HIV diagnosis or stopped the biomedical therapy due to undergoing traditional treatment using traditional medicines: > *“After the HIV diagnosis*, *the doctor told us (the woman and her > husband) that the medicines for HIV treatment are available at this > hospital (HIV clinic) but we did not access the medicines directly*. > *We did the treatment using traditional medicines provided by a > traditional healer in XX (name of a place)”* > > *(FP8*, *widowed)*. > *“After the diagnosis*, *I accessed the therapy (ART) and then once I > finished the medicines (first month)*, *I switched to traditional > medicines*. *I came back here (to restart ART) because my physical > condition was getting weaker”* > > *(MP9*, *married)*. Similar stories were also echoed by all the healthcare professionals interviewed in this study. They described how the use of traditional medicines for HIV treatment influenced access to HIV care services by PLHIV in the district: > *“There are patients who do not use traditional medicines straightaway > after the HIV diagnosis even though we (HCPs) have talked to them and > encouraged them to start ART*. *Some patients have started ART for a > few months and then quitted and switched to traditional medicines”* > > *(HCP4*, *nurse)*. The use of traditional medicines for HIV treatment caused negative impacts on the health of PLHIV. Both women and men living with HIV, who previously used traditional medicines, described that the use of traditional medicines for HIV treatment worsened their health. Similarly, the healthcare professionals commented the use of traditional medicines for the treatment of HIV or switching from antiretroviral medicines to traditional medicines often deteriorated the health condition of PLHIV: > *“My husband and I took a traditional medicine*, *but my husband’s > condition got worse*, *so we went back to the hospital (to do continue > treatment with ART*, *her husband died from AIDS)”* > > *(FP4*, *widowed)*. > *“There were also patients who have started the therapy (ART) for a > few months and then quitted and switched to traditional medicines*. > *Some of these patients came back to this clinic to restart ART once > their physical and health conditions gradually declined*. *I have seen > many HIV-patients whose (physical) condition got worse and HIV status > progressed to AIDS level due to using traditional medicines instead of > ART”* > > *(HCP3*, *nurse and counsellor)*. Furthermore, the use of traditional medicines for HIV treatment seemed to also lead to worse consequences such as the death of PLHIV. Some women and men living with HIV described the situation where their spouses who were also HIV-positive passed away during the treatment using traditional medicines, which in turn became the underlying reason for their return to access HIV care services at HIV clinics. Similarly, healthcare professionals acknowledged such fatal consequence of the use of traditional medicines on PLHIV in the district: > *“After the HIV diagnosis*, *the doctor told us (the woman and her > husband) that the medicines for HIV treatment are available at this > hospital (HIV clinic) but we did not access the medicines directly*. > *We did the treatment using traditional medicines provided by a > traditional healer in XX (name of a place)*. *…*. *As I said before*, > *traditional medicines did not help*, *and our health condition > declined*. *That was the reason why we decided to come here (HIV > clinic) to access these medicines (ART) but my husband could not make > it*, *he passed away”* > > *(FP8*, *widowed*, *Belu)*. > *“There have been many patients with HIV who died from AIDS because > they did not want to comply with what we (HCPs) told them to do*. *We > keep encouraging each of them until now to undergo ART and leave > traditional medicines but there are still many who are holding on to > their traditional medicines*. *You can see from the data*, *of more > than one thousand people diagnosed with HIV*, *there are only around > 300 who access and adhere to ART*. *Nearly every month we get the > report from nurse or counsellor from public health centres at > sub-district levels that patient A or B (PLHIV) has passed away”* > > *(HCP1*, *medical doctor)*. ### Economic consequences of traditional treatment of HIV The use of traditional medicines was reported to have financial consequences for PLHIV in Belu as reported by both PLHIV and HCPs in this study. Some participants, both women and men living with HIV, acknowledged that they had to pay a certain amount of money and give a sacrificial animal to traditional healers who provided the traditional medicines. The amount of money and the animal seemed to differ from one traditional healer to another and had to be provided prior to the commencement of the treatment. Meanwhile, some other participants revealed that they had the traditional treatment for free as the traditional healers were in their families: > *“The use of traditional medicines for the treatment of any kinds of > diseases is very common here*, *but it is costly*. *Patients have to > pay some amount of money and take with them a chicken or pig or goat > as a sacrificial animal*. *I heard that to get traditional medicines > from Marry (pseudonym of the traditional healer)*, *a patient has to > pay five million rupiahs and give a chicken or pig to her”* > > *(MP16*, *married)*. > *“I used traditional medicines from my relatives*, *they (husband and > wife) knew that I am infected with HIV*, *and offered the medicines to > me*. *…*. *I did not pay them*, *they are my family*, *they just > wanted to help me and did not ask for money”* > > *(FP7*, *remarried)*. The desire to recover faster and be completely cured of HIV, and distrust of HIV test results, were reported as barriers to the acceptance of medical treatment using antiretroviral medicines among PLHIV. These were acknowledged by the participants from both groups as supporting factors for the decision of PLHIV and their families not to access HIV care services or use medical treatment or switch from ART to traditional medicines: > *“To get the traditional medicines and undergo the treatment with the > healer*, *we (the man and his wife who is also HIV-positive) had to > pay a lot of money to him (the healer) and give him a goat*. *It was > very expensive*, *but we had to do it because we wanted to get better > faster*. *We spent about fifteen million rupiahs on this treatment*. > *This amount had to be paid at once before the start of the > treatment*, *only once*. *My family*: *parents and siblings helped me > with the payment”* > > *(MP11*, *married)*. > *“One of the reasons they or their families decide to use traditional > medicines or switch from antiretroviral therapy to traditional > treatment of HIV is because of the desire to get better faster or full > recovery*. *Some patients told me that they wanted to get cured and > strong faster*, *that is why they switched to traditional medicines*, > *even though we have told them that they will not be completely cured > of HIV and therefore they have to take antiretroviral medicines every > day for the rest of their lives”* > > *(HCP9*, *nurse and counsellor)*. ## The influence of family on the use of traditional medicines for HIV treatment ### Family decisions for traditional treatment of HIV The participants, both PLHIV and HCPs, reported that family members had a crucial role in determining the treatment for PLHIV. Several female and male participants (PLHIV) commented that they underwent HIV treatment using traditional medicines due to being asked by their family members, such as parents or in-laws. Such an influence of family members was also reported by HCPs: > *“After he (her late husband) was diagnosed with HIV*, *at first the > doctor gave him cotrimoxazole for two weeks*, *but they (her husband’s > family) told him to take traditional medicines*. *…*. *They asked him > and me to use the traditional medicines for bathing as well*. *My > husband and I took the traditional medicines*, *but after a while*, > *my husband’s condition got worse*, *so we went back to the hospital > (to start ART but her husband died)”* > > *(FP4*, *widowed)*. > *“I see that family members have a dominant role in determining > treatment for the sick ones (HIV-positive)*. *Often people are > diagnosed with HIV while they are in a severe condition and being > admitted to hospital*. *So*, *once they left the hospital*, *many of > them switched to traditional medicines because their family members > asked them to do so and then stopped taking antiretroviral medicines*. > *Family members are the ones who look for traditional medicines for > them (PLHIV)”* > > *(HCP10*, *nurse*, *and counsellor)*. Both PLHIV and HCPs described that family members’ decisions to use traditional medicines for HIV treatment of their sick family members were influenced by their experience with the effectiveness of traditional medicines in treating other health issues. Several participants (PLHIV) described that their family members were regular, long-term users of traditional medicines, and therefore asked them to use traditional medicines for HIV treatment. This also seemed to be supported by the lack of knowledge among their family members about HIV care services provided in the HIV clinic: > *“My parents and grandparents are very familiar with traditional > medicines and these are the number one medicines for them*. *Every > time they feel sick or unwell*, *they use traditional medicines to > treat their body*. *So*, *they recommended the use of traditional > medicines to treat HIV as well*. *They do not know anything about > medical treatment like these (showing her antiretroviral medicines she > just collected)”* > > *(FP5*, *widowed)*. > *“In general*, *(HIV) patients come from families with a low level of > education*. *Their family members do not know about ART and do not > understand the function of the therapy to suppress viral load*. > *Besides*, *I believe their family members have seen and experienced > healing from certain ailments or health issues due to the use of > traditional medicines*. *I think*, *these are also the reasons why > many families rely on traditional medicines for the treatment of their > HIV-positive family member”* > > *(HCP3*, *nurse and counsellor)*. Besides, the stories of some HCPs and PLHIV interviewed in this study showed that being physically weak and taken care of by their family members, and a lack of knowledge about HIV care services, especially ART, appeared to be personal- related supporting factors that made them accept and follow the recommendations of their family members for the use of traditional medicines: > *“Once we (the woman and her husband) were tested positive with HIV*, > *our physical conditions were weak already*, *our families took care > of us*. *My parents-in-law asked us to use traditional medicines*. > *So*, *we just used them*, *my mother-in-law sent them to us every > month*. *We just listened to what they said and used the traditional > medicines because we wanted this HIV to go away”* > > *(FP23*, *married)*. > *“It is true that family members play a very important role in the > treatment of HIV patients*. *If their family members provide > traditional medicines*, *then they (PLHIV) will definitely take the > medicines*. *There is no way for them to refuse because they are taken > care of by family members*, *and they are sick”* > > *(HCP7*, *nurse and counsellor)*. ### Social influence on the use of traditional treatment of HIV Extended family members and neighbours were also reported to have an influence on the decision of the participants’ family members regarding the use of traditional medicines to treat HIV infection. Providing information about traditional medicines for HIV treatment, and encouraging the participants’ family members about the effectiveness of traditional medicines to treat HIV infection, were in some instances reflecting the influences of others on the participants’ family members regarding the use of traditional medicines: > “*My family members encouraged us (the woman and her late husband who > died from AIDS) to use traditional medicines because they were > encouraged by our extended family members and neighbours that > traditional medicines can cure HIV* > > *(FP8*, *widowed)*. > *“We (the man and his wife) know about a traditional medicine for HIV > from a friend of mine*. *He came to my house and talked to my wife and > me about it*. *My wife was convinced by his story that traditional > medicine is very good and has cured people of this disease (HIV > infection)*. *Then my wife encouraged me to switch (from ART) to that > traditional medicine”* > > *(MP9*, *married)*. # Discussion Increased HIV transmissions and AIDS-related deaths in many countries and regions globally have been associated with poor access to HIV care services and poor adherence to ART. This paper describes the influence of the use of traditional medicines for HIV treatment and the role of families, friends, and neighbours in determining HIV treatment for PLHIV, which have not been addressed in previous literature on barriers to access to HIV care services. Previous studies have reported the existence of traditional medicines within communities in some African countries and their influence on ART adherence for PLHIV. These were also apparent in our data from both women and men living with HIV and HCPs. What our data add is the concept of the use of traditional medicines for the treatment of HIV and other health issues as a cultural practice that has been passed down from one generation to another within communities in Belu. This concept also seemed to indicate how influential this practice is on the health-seeking behaviours and the preference for treatment by family members of PLHIV and other community members in the study setting. Our study adds further evidence suggesting that the use of traditional medicines for HIV treatment influenced or delayed acceptability and access to biomedical HIV care services among both women and men living with HIV. Such an influence seemed to be strongly supported by the availability and approachability of traditional medicines within communities in the area. These support the constructs of access to healthcare service framework, suggesting the availability of a healthcare service and approachability or how well-known the information about the service is among people in health needs as some of the dimensions determining the accessibility of the service. The current study also suggests that the use of traditional medicines for HIV treatment had a negative financial impact on some participants as the service costed large amounts of money and required sacrificing animals. Previous studies have found that support from family members played an important role in the successful access to HIV care services by PLHIV and their retention in medical treatment or ART. However, our findings suggest that the role of family members in determining the use of traditional medicines for HIV treatment was a significant barrier to access to HIV care services and the initiation of ART by PLHIV in this study. Factors such as the regular use of traditional medicines, the effectiveness of traditional medicines in treating other health issues and a lack of understanding about ART and its function were the underlying reasons for family decisions supporting the use of traditional medicines over ART. These are similar to previous findings which report insufficient or incorrect knowledge about ART as a barrier to access and adherence to ART among PLHIV, and positive relationships between a lack of ART knowledge and less family support for ART access and adherence. In short, consistent with previous findings reported elsewhere, the current study suggests a lack of family support for the participants (PLHIV) prior to or in the early stages of their HIV treatment as a barrier to their access to HIV care services and treatment adherence. The social influence of relatives, neighbours, and friends on family decisions regarding the use of traditional medicines for HIV treatment was another access barrier for PLHIV. These support the findings of a recent study suggesting that social influences of sexual partners, peers, and healthcare professionals are a hindrance to PLHIV seeking HIV treatment and care services. Thus, our findings indicate that the role of good social support from people surrounding PLHIV is crucial for prompt and sustained access to HIV care services and treatment adherence, as has been reported elsewhere. In addition, evidence from the current study also demonstrates that the poor physical and health conditions of PLHIV, and their dependency on family support for their daily needs and healthcare influenced their ability to make decisions about their health treatment. This often led to them deferring to their family about treatment decisions, including the use of traditional medicines which negatively affected their willingness to access and adhere to ART. ## Study limitations and strengths PLHIV who participated in this study were recruited from an HIV clinic in Belu, whilst some of them may have stopped ART at some point, they are still engaged in care. We did not include PLHIV who are disengaged from care, who may have had different stories to tell about the impact of family or traditional medicines on their use of ART. However, to our knowledge, this is the first qualitative inquiry exploring the influence of traditional medicine use and the role of families on the access to HIV care services among PLHIV in the context of Indonesia. Our findings have important implications for the health sector in Belu and other similar settings in Indonesia. As new HIV infections and AIDS- related deaths significantly increase every year, there is a need for responses at the policy level and the development of interventions that address barriers to access to HIV care services experienced by PLHIV. Such efforts may lead to an expansion of the coverage of HIV care services especially ART and increased access to the services. The current findings also have an implication for policy and practice in Belu district and other similar settings in Indonesia and globally to address and develop HIV information and education programs for family and community members to improve their health literacy about HIV, ART and its function. As family members have an important role in the HIV treatment for PLHIV, HIV education programs may result in better understanding and access to HIV care services and retention in HIV treatment among PLHIV. # Conclusions The paper reports the use of traditional medicines for HIV treatments and the role of family members in determining treatment for their HIV-positive family members as barriers to access to HIV care services among PLHIV in Belu. It shows that the use of traditional medicines in treating any kind of health issues, including HIV is a well-known cultural practice within communities in Belu, which influenced participants’ (PLHIV) access to HIV care services and ART adherence. It also reports a strong family role in determining the use of traditional treatment for an HIV-positive family member, which was supported by a lack of knowledge about ART, positive experience of the effectiveness of traditional medicines in treating other health issues and the social influence of other. Future large-scale studies are recommended to obtain a comprehensive understanding of the influence of cultural, family, and social factors on access to HIV care services among PLHIV in Belu and other settings in Indonesia and globally. # Supporting information AIDS Acquired immune deficiency syndrome ART Antiretroviral therapy HCPs Healthcare professionals HIV Human Immunodeficiency Virus PLHIV People living with HIV THs Traditional healers TMs Traditional medicines 10.1371/journal.pone.0264462.r001 Decision Letter 0 Myers Bronwyn Academic Editor 2022 Bronwyn Myers This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 21 Apr 2022 PONE-D-22-03726Traditional treatment of HIV and the role of family members as barriers to access to HIV care service or antiretroviral therapy among people living with HIVPLOS ONE Dear Dr. Fauk, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jun 05 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Bronwyn Myers Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf>  and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. Please include a complete copy of PLOS’ questionnaire on inclusivity in global research in your revised manuscript. Our policy for research in this area aims to improve transparency in the reporting of research performed outside of researchers’ own country or community. The policy applies to researchers who have travelled to a different country to conduct research, research with Indigenous populations or their lands, and research on cultural artefacts. The questionnaire can also be requested at the journal’s discretion for any other submissions, even if these conditions are not met.  Please find more information on the policy and a link to download a blank copy of the questionnaire here: <https://journals.plos.org/plosone/s/best-practices-in-research-reporting>. Please upload a completed version of your questionnaire as Supporting Information when you resubmit your manuscript. 3\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: N/A Reviewer \#2: N/A \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: No Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Traditional treatment of HIV and the role of family members as barriers to access to HIV care service or antiretroviral therapy among people living with HIV Use of abbreviations like HIV in the title is not recommended and the title is not well articulated. \- Possibly “Traditional treatment of Human Immunodeficiency Virus as barriers to access HIV care and treatment in Belu, Indonesia”. \- Still can well articulate the title… General: The paper raised an important issue that is the use of traditional treatments in Belu is a barrier to HIV care and ART. The role players for decision-makers for use of traditional medicine (TM) were family members and the culture. The impact of this practice is to the extent that the patients loss their life. However, this important finding is not well written and elaborated to show how much the use of traditional medicine in these specific areas is affecting the life of HIV patients. The paper is poorly written, I recommend this better be rewritten clearly organizing the drivers for the use TM and its impacts on the life of patients. In writing the manuscript please avoid unnecessary long statements (some of which seems desire to extend the statement – better use short sentences) \- As a researcher you seem biased, because you started by having negative belief on traditional medicine use. It would have been better if the impact on traditional medicine use on access to biomedical HIV care & treatment was assessed. \- Lacks the list of abbreviations/acronyms \- Good to have operational definitions for some terminologies. (like TM, traditional healers … in your context) Abstract: \- “As a part of a qualitative study in Belu, this paper describes the use of traditional treatment and the role of families in determining traditional treatment for their HIV-positive family member as barriers to access to HIV care service or ART among PLHIV.” This statement not clear better to rewrite in more clear way. \- Snow ball technique you used is not clear – what is the information pack? Background \- Good coverage, however most sentences are too long that need to be rephrased. Last paragraph: In this aim statement and throughout the manuscript you have used terms in different words like “HIV care services or ART”, “traditional treatment or medicines” which you could use either of them or better expressive phrase. \- Why you did not discuss on the barriers to access ART, rather than having too narrow scope that focus only on family impact. Methods: Study Setting: Recruitment and Data Collection: • P1. “As part of a larger qualitative study to understand HIV risk factors, impacts and determinants of access to HIV care services among women and men living with HIV in Belu, this paper describes the use of traditional treatment for HIV and the role of family members, friends and neighbours in supporting the use of traditional treatment as barriers to the access of PLHIV to HIV care service or ART”. \- The aim of this study was already described in the background section; not necessary to repeat it here (remove this statement). • The snowball technique used is not clear. I do not think this is a snowball technique? • What was the study information pack given to the patients and HCPs? • … “HIV clinic receptionist and healthcare facilities” not clear. P1L11: “The initially interviewed participants were also asked to distribute the study information packs to their friends and colleagues who might be willing to participate.” \- I think this information sharing between participants makes the second group of participants affected (influenced) by others resulting in biased information. • The method section should have clearly shown the positionality of the researcher (criticality), validity… quality measure of a qualitative study. Result: Generally good. Rephrase the statements to sound scientific. • Table 1: The occupation for HCPs and PLHIV need to be separate. Also, the list of occupation needs to be re-categorized. Discussion: • Almost the result is rewritten. It is better to make well-referenced and explained comparing with national or international policies and research findings. • P1L1: different form referencing (UNAIDS 2020). Make uniform. Reviewer \#2: This was an interesting article. There are some aspects that require revision. First, the entire paper needs to be professionally edited for English language. Many of the sentences are long and the english is hard to follow. Introduction: I recommend describing barriers to use of ART rather than barriers to accessing HIV care services or use of ART- this is just too wordy and hard to follow. Please describe the role and use of traditional medicines in Indonesia- more information is needed on the context of these medicines and how traditional and western systems of care for HIV interact if at all. Methods: The sampling approach you mention here is not snowball sampling- rather convenience sampling Data analysis section- it very long and can be abbreviated All the COREQ guidelines have not been followed- please check and report back for example on member checking Results: You only need 1-2 illustrative quotes per section. Three is too many. There seems to be some overlap between the themes - eg family influence permeates across several themes Discussion. This needs the most work. You report the findings but do not discuss the implications for policy and practice. If family is so influential what can be done to improve HIV health literacy, overcome medication myths etc, can traditional healers be engaged in the HIV care system to advocate for use of ART? Needs to move beyond mere reporting of the results. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0264462.r002 Author response to Decision Letter 0 22 May 2022 Response to reviewer file is attached. 10.1371/journal.pone.0264462.r003 Decision Letter 1 Myers Bronwyn Academic Editor 2022 Bronwyn Myers This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 12 Jul 2022 Traditional Human Immunodeficiency Virus treatment and family and social influence as barriers to accessing HIV care services in Belu, Indonesia PONE-D-22-03726R1 Dear Dr. Fauk, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Bronwyn Myers Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0264462.r004 Acceptance letter Myers Bronwyn Academic Editor 2022 Bronwyn Myers This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 15 Jul 2022 PONE-D-22-03726R1 Traditional Human Immunodeficiency Virus treatment and family and social influence as barriers to accessing HIV care services in Belu, Indonesia Dear Dr. Fauk: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Bronwyn Myers Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Most industrialized countries are aging fast due to an increase of life expectancy and a reduction of child birth rate. In this aging society, it is expected that there will be a growing need for home, medical and nursing care services. For this purpose, robots, and in particular robots with appearance based on the human body, are expected to perform human tasks such as provide personal assistance, social care for the elderly or cognitive therapy, and be used in entertainment and education. Just as over the last 30 years the computer business has become an integral part of our daily life, so is robotic technology expected to follow a similar development in the near future. These prospects bring into consideration issues related to natural social interactions with these artificial agents. To become part of our everyday environment, personal robots need to be capable of smooth and natural interactions with humans. It has been proposed that consumer product humanoids should be designed to balance human-ness (to facilitate social interaction) and robot-ness (to avoid false expectations about the robots' abilities). Already several robots have been developed to investigate the socio-emotional aspects of human-robot interactions: animaloid robots like the therapeutic robot PARO and SONY AIBO elicit emotional attachment; humanoid robots like Honda ASIMO and Kawada HRP-2 cooperate with humans; android robots like Actroid and Geminoid explore face-to-face interactions. The humanoid robot WE4-RII (Waseda Eye No.4 Refined II) was designed to expresses human-like emotions in order to improve the social competence of human-robot interactions. The current study was designed to assess how the neural substrates involved in the perception of human emotions respond to the same gestures impersonated by this anthropomorphic yet clearly mechanical robot, in an endeavour to describe how the agent's appearance modulates brain responses to the perception of emotional facial actions. This research is theoretically grounded in the hypothesis that resonance is pivotal in natural human social interactions. Resonance describes the mechanism by which the neural substrates involved in the internal representation of actions, as well as emotions and sensations, are also recruited when perceiving another individual experiencing the same action, emotion or sensation. While this hypothesis can be traced back as far as William James, its interest has been renewed by the discovery of ‘mirror neurons’ in the ventral premotor cortex of the macaque monkey. Mirror neurons fire both when monkeys perform a goal-directed action and when they perceive (see or hear) or infer the same action performed by an experimenter. Neuroimaging studies have identified brain regions, in premotor and parietal cortices, in which action execution and observation overlap in the human brain (for review see). The ventral premotor cortex, in particular, constitutes a major locus of motor resonance in humans. Furthermore, the somatosensory cortex responds to the observation and feeling of touch, and the insula responds to the observation and feeling of disgust. These examples support a generalization of resonance to multiple domains of cognition including emotions. Artificial agents such as the humanoid robot used in this experiment can participate to a better understanding of factors affecting this resonance, and in particular the role of anthropomorphism. Neuroimaging experiments comparing the observation of humans to artificial agents have yielded mixed results in the inferior premotor and posterior parietal regions of the human motor resonance mechanism. In a PET study, the left ventral premotor activity found in previous experiments of action observation responded to human, but not robot, actions. However, a more recent fMRI study indicated that motor resonance is elicited by a robotic arm and hand. While activity in a neural marker of motor resonance was not significantly related to the anthropomorphism of computer-animated avatars, it decreased with the bias to perceive their actions as biological, raising questions about the interaction between perceptual processes related to anthropomorphism, and subjective perception of artificial agents' actions as natural. To address this question, we investigated whether facial emotions expressed by a humanoid robot activate brain regions involved in the perception of human emotions, in particular those engaged in motor and emotional resonance. We used the humanoid robot WE-4RII (Waseda Eye No.4 Refined II), developed by Takanishi Laboratory at Waseda, to express emotions by using facial expressions and the movement of the upper-half of the body including neck, shoulders, trunk, waist, as well as arms and hands. Short videos of the humanoid robot and human actors expressing three emotions (Joy, Anger, Disgust), and silent speech were presented to participants, who were asked to rate either the emotional content or the motion, in order to orient their attention either explicitly to the mental state conveyed by the gesture, or to a purely visual feature, thus privileging an implicit processing of the intentional gesture. On the basis of the mechanical appearance of the anthropomorphic robot, we hypothesized a reduced activity in brain regions involved in motor (ventral premotor and inferior frontal gyrus) and emotional (in particular amygdala and insula) resonance during the observation of the robotic agent compared with the observation of a human agent. # Methods ## Participants 13 right-handed participants (4 males; aged 29.4+/−7 years) with no history of neurological disorder and normal or corrected-to-normal vision gave their informed consent in writing to take part in this experiment. The study was approved by UCL National Hospital for Neurology and Neurosurgery and Institute of Neurology joint Ethics Committee. ## Stimuli The humanoid robot used in this experiment, WE-4RII, has 59 degrees of freedom (DOFs), 26 of which are specifically used for controlling facial expression (eyebrows: 8; eyelids: 6; eyes: 3; lips: 4; jaw: 1; neck: 4). A subset of the facial Action Units (AU, described) was chosen for a simplified but realistic impersonation of the facial gestures used in the experiment - Eyebrows: AU 1, 2, 4; eyelids: AU 7, 42, 43; eyes: AU 5, 6, 43, 44, 45, 46; Mouth: AU 15, 17, 20, 25, 27; Lips: AU 12, 15, 16, 20, 23. The shoulders have 3 DOFs, plus 2 additional DOFs used for squaring or shrugging gestures. Both the posture and the motion velocity are controlled to realize an effective execution of each gesture. Stimuli consist of 1.5-second greyscale video clips (38 frames at 25 frames per second) showing the agent face and upper body starting from a neutral pose and depicting one of the following gestures: expression of Joy, of Anger, of Disgust and silent Speech. Two different actors were recorded for human stimuli while two versions of the humanoid robot were obtained by the addition of a wig, and four different versions of each type of stimulus were prepared, leading to a total of 64 different stimuli (4 gestures, 2 agents, 2 versions of each agent, 4 versions of each type of stimulus). The greyscale was digitally modified to match the background luminosity and the overall contrast between the human and robot stimuli (see, top). Great care was taken to match the dynamics of the human and robot stimuli pairwise. ## Experimental paradigm There was a total of 16 experimental conditions: across the eight types of stimuli defined by four gestures (Joy, Anger, Disgust and Speech) impersonated by two agents (Human, Robot), participants to the experiment were asked, after each stimulus, to rate the emotional content (“How much EMOTION did the face show?”) or the amount of motion in the stimuli (“How much MOVEMENT did the face show?”). Participants underwent four sessions of fMRI scanning. Each session contained eight blocks, four in which emotion was rated and four in which motion was rated, presented in a fully randomized order. Participants were informed of the object they rated by a one-word description presented for 1.5 second at the onset of each block (“EMOTION” or “MOVEMENT”, see). There were eight stimuli presented in each block in a pseudorandomized order so that each stimulus was seen once in each session and twice for each rating over the course of the experiment. Inter-stimuli onsets were jittered based on a normal distribution of mean 4.5 (+/− SD 0.5) seconds. After each stimulus, the participant's rating was recorded using an analogue scale that ranged from “None” to the target emotion (e.g. “Anger”) to rate emotion, and from “None” to “A lot” to rate motion. The direction of the scale was assigned randomly, and at the onset the response bar was located close to the centre of the scale; the participants pressed a left or right key on their keypad to move the response bar towards the left or the right respectively, and released the key when the response bar reached the desired rate. These characteristics were selected to avoid motor preparation of the response prior to the appearance of the response screen. The duration of the response screen was 1.5 seconds. Prior to scanning subjects were trained with a limited subset of stimuli (3 blocks of 3 stimuli) outside the scanner to become acquainted with the response procedure. Presentation of stimuli and recording of participants' responses were carried out using Cogent (<http://www.vislab.ucl.ac.uk/CogentGraphics/index.html>) running in Matlab 6.5 (MathWorks™) and analysis of ratings using the statistical program SPSS (SPSS Inc.) ## fMRI data acquisition Scanning was performed using a 1.5T Siemens Sonata MRI scanner. High-resolution anatomical images were acquired using a T1-weighted 3D MPRAGE sequence. In each of the four experimental sessions, T2\*-weighted, gradient-echo, echo-planar imaging sequence was used to acquire 116 volumes containing 48 slices (2 mm thickness and 1 mm gap) covering the whole brain and cerebellum with an in-plane resolution of 3×3 mm (64×64 matrix, fov 192×192×144 mm³). The sequence was optimized for blood-oxygen-level dependent signal sensitivity in the ventral cortical areas (orbitofrontal, inferotemporal and amygdala regions) by the use of a tilt angle of −30 degrees and negative phase encoding. The first 4 volumes of each time-series were discarded prior to the analysis to allow for T1 equilibrium. Field maps were also acquired to correct for geometric distortions in EPI images caused by magnetic field inhomogeneities. ## fMRI data analysis fMRI data were analyzed using SPM5 (<http://www.fil.ion.ucl.ac.uk/spm>), running in Matlab 6.5 (MathWorks™). Slice timing correction was applied to correct for offsets of slice acquisition. EPI volumes were realigned to the first volume for each subject to correct for interscan movement, and unwarped for static magnetic field inhomogeneities using field maps and for movement-induced inhomogeneities using realignment parameters. The high-resolution structural image was co- registered with the mean image of the EPI series, and stereotactically normalised to the Montreal Neurological Institute (MNI) template using sinc interpolation. The normalisation parameters were applied to the EPI time-series, achieving an anatomically informed normalisation. EPI volumes were finally smoothed using an 8mm isometric Gaussian kernel to account for residual inter- subject differences in functional anatomy. The analysis of the functional imaging data entailed the creation of statistical parametric maps representing a statistical assessment of hypothesized condition- specific effects. A random effects procedure was adopted for data analysis. The 1.5-second response periods, and, separately for each of the 16 experimental conditions, the 1.5-second stimulus periods, were modelled at the subject level. These condition-specific effects were estimated with the General Linear Model, with each condition being defined with a boxcar function convolved with the canonical hemodynamic response function. Low-frequency sine and cosine waves modelled and removed subject-specific low-frequency drifts in signal, and global changes in activity were removed by proportional scaling. Each component of the model served as a regressor in a multiple regression analysis. The brain response to the human stimuli irrespective of the gesture was investigated by contrasting human stimuli presentation, across the four gestures and the two ratings, against the global mean. The resulting statistical maps were entered in a second-level one-sample *t*-test. Similarly, brain response to the human stimuli for each gesture was investigated by contrasting human stimuli presentation, for each gesture and across the two ratings, against the global mean, and entering these contrasts in four second-level one-sample t-tests. All contrasts were thresholded at *p*\<0.05 FDR-corrected with an extent threshold of 20 voxels. Anatomical localization was performed using a brain atlas and, when possible, statistical localization relied on probabilistic cytoarchitectonic maps. Other functional attributions relied on comparisons with the literature. To address specifically the scientific hypothesis, regions responding to the perception of human gestures were further explored to assess their response to robot gestures using a Region Of Interest (ROI) approach. The SPM extension toolbox MarsBar (<http://marsbar.sourceforge.net/>) was used to extract percentage signal change in 5-mm radius spherical ROI centred on the maximum of the cluster under investigation. Percent signal changes were further analyzed using ANOVA and *t*-tests implemented in the statistical program SPSS (SPSS Inc.), with a significance threshold of 0.05. Regressions (reported at *p*\<0.05) between percent signal change and emotional ratings of robot and human stimuli were assessed in brain areas responding specifically to single gestures. # Results ## Behavioural data It was shown in a separate experiment, and confirmed in preliminary tests with the stimuli used in the present experiment, that the robot depictions of the three emotions used in this experiment (Anger, Joy and Disgust) were correctly recognized above chance levels (all \>75% correct recognition). Repeated-measures ANOVA indicated a significant effect of the Agent (F_1,12 = 16.1; *p* = 0.002) and the Gesture (F_3,36 = 57.0; *p*\<0.001) on the emotional ratings recorded during the fMRI experiment, as well as a significant interaction between the two factors (F_3,36 = 12.2; *p*\<0.001). As expected given the lack of emotions for the gesture Speech, contrasts revealed significantly increased ratings for Joy, Disgust and Anger compared to Speech (*p*\<0.001) irrespective of the agent. Repeated-measures ANOVAs assessed the effect of Agent on subjects' emotional rating for each gesture separately. Their results indicated significantly higher ratings for human than for robot videos for Anger (F_1,12 = 31.0, *p*\<0.001) and Disgust (F_1,12 = 7.8, *p* = 0.02, see). Speech was rated as significantly more emotional (i.e. less neutral) for the Robot than the Human videos (F_1,12 = 14.7, *p* = 0.003). Differences between ratings of Joy expressed by Human and Robot were not significant (F_1,12 = 1.4, *p* = 0.262). ## fMRI data ### Main effect of human stimulus presentation The main effect of watching human visual stimuli against the global mean irrespective of the gesture and independent of the rating, yielded bilateral activity in occipital, temporal, parietal and frontal cortices. A large cluster (#1, k = 4001 voxels) extended from extrastriate cortices to ventral and lateral temporal cortices bilaterally and to the inferior parietal lobule in the right hemisphere. Extrastriate maxima were attributed to Brodmann areas 17 and 18 bilaterally as well as to the right hemisphere functional areas V3v, V4 and V5. In the right temporal cortices, maxima were reported at the junction between the occipital and temporal lobes, a region responding to the perception of faces (MNI coordinates 42, −68, −6, compared to 43, −67, −9) referred to as the lateral face area (LFA) hereafter (see also), in the fusiform gyrus at the vicinity of the fusiform face area, or FFA, (MNI coordinates 42, −62, −20 compared to 40, −56, −15), and in the posterior superior temporal gyrus (MNI coordinates 58, −36, 10 compared to 50, −34, 4). In the left hemisphere, clusters were found in V3v (#2), V4 and V5 (#3), as well as in the left- hemisphere FFA (MNI coordinates −34, −62, −18 compared to −35, −64, −16), but not in the lateral temporal cortex. Extracted signal changes collapsed across the 4 gestures, in 5-mm radius spheres centred on the maxima localized in V3v, V4, V5 and FFA bilaterally as well as in LFA and STS in the right hemisphere were submitted to 2 (Agent) by 2 (Rating) repeated measures ANOVA. Results illustrated in illustrate the significant effect of Agent in all ROI but the STS, corresponding to an increase of the response to Robot compared to Human agents (V3v and V4 bilaterally p\<0.001; V5, FFA bilaterally and right LFA p\<0.05), without significant effect of the object of Rating (all p\>0.05) nor a significant interaction between Agent and the Rating. There were no significant effects of Agent or Rating nor an interaction between Agent and Rating (all p\>0.1) in the right STS. ### Main effect of human stimulus presentation: frontal cortices Because of our *a priori* hypothesis on the role of inferior frontal cortices in motor resonance, percent signal change was extracted in 5mm radius spheres centred on the maxima of inferior frontal gyrii activated clusters, localized in three Brodmann areas (BA) according to the cytoarchitectonic probabilistic maps : BA 6 in the right hemisphere, and bilateral BAs 44 and 45, located in the vicinity of clusters reported during the perception of a human face performing intransitive mouth gestures. Signal extracted in these ROIs, collapsed across the 4 gestures, was submitted to 2 (Agent) by 2 (Rating) repeated measures ANOVAs. There was no significant main effect or interaction (all *p*\>0.5) affecting signal in the right BA6. In the left BA44, there was a significant interaction between Agent and Rating (*p* = 0.02), with no main effect of Agent (*p* = 0.4) or Rating (*p* = 0.8). Paired *t*-tests revealed that response to the robot was not significantly affected by the Rating, while response to human stimuli was significantly increased for the Movement compared to Emotion rating (*p* = 0.04). A similar profile in the right BA44 did not reach significance (all *p*\>0.1). In the left BA45, there is a significant effect of Rating *(p* = 0.05), and a trend in the interaction between Rating and Agent (*p* = 0.06), with no main effect of the Agent (*p* = 0.8). As with BA44, a similar profile in the right hemisphere BA45 did not reach significance (all *p*\>0.1). The only significant *t*-test showed that signal change for robot stimuli was significantly increased during rating of the emotional content of the stimulus compared to its motion (left *p* = 0.01, note than on the right *p* = 0.1). The same contrast did not reach significance for human stimuli. ### Action-specific brain responses Brain response to human stimuli was investigated for the four gestures independently at the second level to isolate brain areas responding to individual gestures. Areas responding specifically to each of the four types of facial action against the global mean are provided in and illustrated on. The left inferior frontal gyrus activity associated with perception of Speech gestures was localized in *Pars Triangularis*, and attributed to Brodmann area 44. Its location falls into in a subdivision of Broca's region putatively involved in syntactic aspects of speech execution and perception (reviewed). A similar region was reported for the auditory perception of language coordinates (−46, 12, 24 compared to −40, 14, 28). In the present experiment, this area responded to the perception of human speech gestures and was not found in the other types of action, supporting the specificity of its response to language- related actions. Signal change for Speech stimuli extracted in a 5-mm sphere centred at −46, 12, 24 was submitted to 2 (Agent) by 2 (Rating) ANOVA. There is a significant effect of Agent (*p* = 0.05) corresponding to increased signal to human compared to robot stimuli. There was a trend (*p* = 0.09) towards increased response when rating emotion compared to movement. The left anterior insula, a mirror region for this emotion (−30, 22, 4 compared to −34, 28, 6) was associated with the perception of Disgust gestures. In the ROI associated with this activity, only the main effect of agent showed a trend (*p* = 0.1), corresponding to an increased response to human expressions of disgust compared to robot's expression of the same emotion (paired *t*-test *p* = 0.1). There was no significant effect of the object of Rating, or correlation between emotional rating and activity in this ROI. The right orbitofrontal cortex was associated with the perception of human expression of Anger. Repeated measure ANOVA indicated a significant main effect of Agent (*p* = 0.01) in the signal extracted in this region, corresponding to an increased response to human compared to robot stimuli. In addition, one- sample *t*-test reveals that response to the robot's expression of anger in this region was not significantly different from the global mean (*p* = 0.3). Finally, the right putamen, part of the ventral striatum associated with the perception of human gestures of Joy, was the only non-cortical region reported in this section. There was no significant main effect of Agent or Rating on the signal extracted in the putamen, but a trend (*p* = 0.1) towards an increase of response to human compared to robot stimuli. There was a significant correlation between extracted percent signal change during perception of human stimuli of joy and the emotional rating (R² = 0.461, *p* = 0.04), but not for robot stimuli of joy (R² = 0.174, *p* = 0.16). No other correlations between action-specific brain regions and emotional ratings were significant for the human or the robot stimuli. # Discussion In the current fMRI study, participants observed short videos depicting emotional (Anger, Joy and Disgust) or emotionally neutral (Speech) facial gestures expressed by real humans or by the robotic humanoid platform WE-4RII, designed to resemble a human face. WE-4RII can reproduce a subset of the facial Action Units, by movements of its eyebrows, eyes, eyelids, lips, mouth, neck, shoulder and upper torso, so as to express in a recognizable manner the four gestures used in this experiment while at the same time being perceived as an artificial, i.e. non-human and non-intentional, embodied agent. Analysis of the ratings of the emotional content by the participants of the current experiment indicated that emotional gestures were perceived as more emotional (and the emotionally neutral speech gestures, less emotional) when expressed by the humans than by the robot. The use of stimuli derived from this robotic platform in an fMRI experiment provided a unique opportunity to test whether the reduction of perceived emotionality of the artificial agent is associated with reduced activity in brain areas involved in the feeling or the perception of the same emotions depicted by human agents. Note that because the robot is clearly mechanical compared to human actors, it is not possible to dissociate, in the present experiment, differences in activity related to the appearance and to the artificial nature of the robot. In addition, stimuli were grouped into fMRI blocks during which participants were asked to rate either the emotional content or the movement depicted, as a proxy to orient their attention either towards the intention underlying the gestures (the emotion) or toward a purely visual feature of the stimuli (the amount of movement) so that processing of the mental state causing the action (the emotion being displayed in Joy, Anger and Disgust, the will to communicate in Speech) is implicit. This manipulation was chosen to disentangle bottom-up processes, influenced by the nature of the stimuli, and top-down processes, influenced by the instruction to attend the emotion or the motion of the stimulus. fMRI analysis consisted of, first, isolating regions of interest on the basis of their response to human stimuli, and second, assessing the modulation of their activity by the agent depicting the gestures and by the object of attention. Discussion of the data focuses on regions of the visual association areas in the occipital and temporal cortices involved in the perception of faces and objects; regions found to be specifically associated with the perception of the different types of basic emotions, insula for disgust, putamen for joy and orbitofrontal cortex for anger, and silent speech in the left inferior frontal cortex; and the inferior frontal cortices, which were predicted on the basis of their contribution to motor resonance. ## Visual cortices Responses to human stimuli are reported in visual areas V3, V4 and V5, and in temporal areas responding to the perception of faces (fusiform face area FFA, lateral face area LFA) and actions (superior temporal gyrus). Activations in these occipital and posterior temporal cortices when perceiving human gestures was predicted on the basis of their essential role in visual perception of biological motion and body parts. In terms of the effect of robotic stimuli on activity in occipital and posterior temporal visual cortices, the main finding was that all regions, with the notable exception of the superior temporal gyrus cluster, showed an increased response for robot compared with human stimuli. This increase appears at odd with their proposed human face-specificity of FFA bilaterally and right LFA. Already, a bilateral fusiform gyrus activity was reported in response to animal faces depicting actions. Another fMRI study found similar responses when perceiving human faces and animals with or without faces in the same fusiform region, suggesting that perception of animals relies on the same substrates of perception of human faces. Explaining this increased response to the robot's face entails discussing mechanisms involved in the domain-specificity of perception in the FFA. Face perception is holistic, and deficits of prosopagnosic patients support that the FFA is crucial for this holistic perception. According to Pinker, a perceptual process must be characterized by the type of geometry it pays attention to, and the geometry the human face recognition system is sensitive to can be demonstrated in newborns. Pinker argues any object that shares these geometric features, as the robotic face used here, will be automatically processed by the “face module”. This automatic processing might explain activity in the FFA bilaterally and in the right LFA normally activated by human gestures in response to robot stimuli. It has been proposed that in the FFA, features of the presented face are compared to an average “face template”. Because the robot face was clearly distinguishable from a human face, this comparison could lead to a reduction of signal, as was the case for the perception of animals or of cartoon faces. Alternatively, this comparison could require additional processing of the visual input in order to recognize the robot as a face. This interpretation is supported by the significant increase of response in extrastriate areas V3, V4 and V5, implied in the processing of low-level aspect of visual stimuli such as form, colour and motion. Furthermore, a similar increase of response has been reported in the visual word form area of the ventral occipital cortex when the visual appearance of a written word is degraded, Altogether, increased response to robot compared to human gestures in visual areas implicated in the perception of faces and actions is likely to reflect additional processing of the unfamiliar stimulus. There is no significant difference in responses to robot and human stimuli in the right superior temporal gyrus. The posterior temporal cortex responds to a large range of stimuli. It is particularly respondent to visual depictions of actions across a variety of presentations (full body or body parts actions, point-light displays, as well as animal actions and scripted geometrical shapes movements). The finding of a similar response to robot and to human stimuli in this region argues in favour of a fully integrated representation of gestures, as both types of stimuli are similar in most respects but the appearance of the agent depicting the gesture. ## Regions responding to only one type of human gesture Aside the occipital and temporal regions involved in processing all gestures, some brain areas respond only to one of the human gestures used in this experiment. We are particularly interested in regions known to be involved in the processing (either in execution or in perception) of the specific gesture they were found associated with, namely the insula for disgust and Broca's region for speech. Activity in the left insula was predicted on the basis of its participation in emotional resonance during the perception of disgust gestures. The short insular gyrus cluster associated with the perception of disgust gestures (−30, 22, 4) was in the vicinity of a left anterior insula cluster in which overlap between observation and feeling of disgust has been reported. This region was activated in response to the humanoid robot's expression of disgust in comparison to baseline, and the trend showing a reduction of its response in comparison to human stimuli did not reach significance (*p* = 0.1). This finding demonstrates emotional resonance towards an anthropomorphic robot in the case of disgust gestures. Perception of human joy was associated with activity in the right putamen, a brain area repeatedly associated with the induction of happy mood (see meta- analysis). This can be attributed to its role in reward-processing following the suggestion that dopaminergic signalling in these regions is important to elicit internal rewarding response. Such interpretation supports its involvement in the emotional resonance for Joy. As was the case for the insular cluster associated with Disgust, results indicated that there was a trend towards decreased response to robot compared to human stimuli. In addition, the correlation between emotional ratings and brain activity, significant for human stimuli, was not significant in the case of robot stimuli. Altogether, our data support a reduced emotional resonance towards robotic expressions of Joy in the striatal structure, extending the results from Disgust to a non-cortical area. The involvement of the orbitofrontal cortex in emotions has been demonstrated by lesion studies in humans. The right orbitofrontal region found here has already been shown to respond to angry faces. Activity was significantly larger in the OFC for human than for robot angry gestures, and the response to robot stimuli was not significantly different from the baseline, suggesting that response of this region was limited to human stimuli. An explanation based on the large difference in perceived emotion of the two agents depicting anger can be excluded by the absence of significant correlation between orbitofrontal activity and emotional ratings for either agent. An alternative explanation, according to which the orbitofrontal cortex is involved in top-down aspects of emotional evaluation is contradicted by the absence of effect by the manipulation of attention through rating instructions. The absence of significant response to robot stimuli might result from the role of the orbitofrontal cortex in social cognition. Orbitofrontal lesions have been associated with disinhibited social behaviours, putatively by lack of anticipation of their negative outcomes. We suggest that because of its clearly artificial nature, the robot did not elicit a desire for social contact sufficient to be reflected in orbitofrontal activity. Further investigations including socially rewarding interactions with artificial agents, for example interactions with androids, will be necessary to confirm this interpretation. A cluster associated with the perception of human speech only was attributed to Brodmann area 44, a part of Broca's region associated with speech. This activation was similar to clusters reported for auditory, visual and visuo- auditory processing of speech. More generally, Broca's region involvement in language production and comprehension supports a role of motor resonance in the domain of speech perception that was hypothesized prior to the discovery of mirror neurons as the “motor theory of speech perception”. Activity in this region was reduced when speech was impersonated by the humanoid robot, compared with human agents, but significantly activated compared to baseline, suggesting robot stimuli elicited reduced motor resonance compared to human stimuli. In contrast to the inferior frontal activities described in the next section, the absence of a significant interaction between Agent and Rating suggests that this reduced activity was caused by the unrealistic appearance of the humanoid robot. ## Inferior frontal cortices The inferior frontal gyrii and ventral premotor cortices were scrutinized because of their involvement in motor resonance, important for the perception of actions, and by extension of emotions, expressed by facial and body gestures. Five clusters were isolated, in the left lateral premotor cortex (BA 6), and bilaterally in the posterior (BA 44) and anterior (BA45) *pars triangularis* of the inferior frontal gyrus. This region of the cortex, which has been implicated in the perception of human actions and imitation, is likely homologous to frontal regions responding to action observation in macaque monkeys. The agent displaying the emotion had no effect on activity in these regions of interest, in keeping with the responses to the observation of human and robot, hand actions that have been reported in this region. Both previous studies and in the present experiment, mechanical robot effectors, respectively a “hand” and a “face”, were clearly associated with a bilateral increase of activity in the inferior frontal cortex, with no significant difference in activity between the robotic and human agents. This supports that motor resonance is recruited irrespective of the agent executing the action. Even point light displays of human body motions evoke motor resonance within Broca's region. Mere resemblance of the body shape is thus sufficient to elicit motor resonance: while mirror neurons in monkeys have been reported anecdotally to respond to conspecifics' actions, most of their recordings have been made when monkeys observed human actions; while there is a generic correspondence between the body shapes and degrees of freedom of the two species, the match is not perfect, implying that mirror neurons can generalize across species. Human neuroimaging experiments presenting human, monkey and dog facial movements suggest that even for the less anthropomorphic agent, the dog, motor resonance can be observed provided the action is part of the observer motor repertoire \[biting in contrast to barking; 56\]. Recent results using robots, including the present data, support that motor resonance generalizes to anthropomorphic artefacts. This conclusion is consistent with behavioural experiments investigating motor resonance, that demonstrated that the observation of humanoid, but not industrial, non anthropomorphic, robotic gestures cause a motor interference effect. In another line of research using hand action imitation, both real and robotic hands had an action priming effect. In both BA44 and BA45 of the left hemisphere, an interaction between the effect of Agent and of Rating was identified, with a main effect of rating in BA45 corresponding to increased response when attention was explicitly directed towards the emotion. BA44 response to the robot was not influenced by the object of attention, while response to human increased when attention was directed towards the gesture's movement compared to its emotion. In contrast, response of the anterior BA45 to human stimuli was not influenced by the direction of attention, but response to robot stimuli was increased when participants were required to rate the emotion of the stimuli, compared to its movements. Altogether, these results suggest a modulatory influence of task on the activity of both left inferior frontal areas. One interpretation of our results is the preference for representation of actions' intentions in BA45, similar to the response to abstract actions in the more rostral region of macaque monkey's arcuate sulcus. The main effect of rating in the current experiment corroborated BA45's preference for the representation of intentions underlying the depicted gestures when attention is explicitly directed towards emotion. The pattern of activity in BA45 could thus be explained by the interaction between bottom-up and top-down processes. Bottom-up processes of intention understanding could be automatic for human stimuli, and therefore not sensitive to modulation by attention. In contrast, because the system has no prior representation of robots' actions, robot stimuli would not be processed automatically. Response to robot stimuli would be modulated by the object of attention: stimuli would be processed as intentional actions when the task required assessing the emotion, but as artefact movements when the task did not require processing the emotion. The interaction between Task and Agent in BA45 could thus derive from an interaction between bottom-up processes, influenced by the nature of the agent, and top-down processes, depending on the object of attention. ## Conclusion Using fMRI, we investigated whether regions responding to human basic facial emotions and silent speech were also activated when a humanoid robot impersonated the same gestures. While robot stimuli elicited larger responses in occipital and posterior ·temporal areas, a reverse pattern was observed in regions responding specifically to one type of human gesture only, namely the left inferior frontal cortex for motor resonance in speech perception and insula for emotion resonance in disgust. We suggest that the clearly artificial appearance of the humanoid robot used in this experiment, WE-4RII, together with the limited number of degrees of freedom available in comparison to a real human, precluded high levels of resonance towards this agent's gestures. While none of the subjects had previous experience with an emotional robot, it is possible that experience leading to the establishment of real relationships with a robot could create a sense of social bonding. Further work should investigate the relation between familiarity with robots and the activity of neural markers of motor and emotion resonance. This first study paves the way for further exploration of perception of robotic actions. # Supporting Information MZ and AT would like to express their gratitude to Okino Industries LTD, STMicroelectronics, Japan ROBOTECH LTD, SolidWorks Corp and the Advanced Research Institute for Science and Engineering of Waseda University. MZ and AT would also like to thank Dr. K. Itoh, Mr. M. Saito, and Mr. Y. Mizoguchi for their help in the preparation of the emotional patterns of the robot, and Prof. H. Kimura for his support to the research. [^1]: Conceived and designed the experiments: TC SJB AT CDF SM GR VG MAU. Performed the experiments: TC. Analyzed the data: TC SJB CDF. Contributed reagents/materials/analysis tools: MZ AT VG MAU. Wrote the paper: TC SJB CDF VG MAU. Initiated the project: AT SM PD GR VG MAU. [^2]: The authors have declared that no competing interests exist.

# Introduction Cancer stem-like cells (CSCs) are the most malignant subpopulations in tumors because of their resistance to drug therapy and association with poor outcomes. As in adult tissue-specific stem cells, which have extensive self-renewal abilities that ensure the continuous turnover of normal tissues, CSCs are required for continuous tumor growth. Great progress has been made in the identification and characterization of CSCs in several types of solid tumors, such as those of the brain, breast, colon and liver. In addition, a few key signaling pathways, including Wnt/β-catenin, AKT and TGF-β, have been implicated in the maintenance of CSCs. However, the molecular events that preserve the pool size and stem cell properties of CSCs by maintaining the balance of proliferation, differentiation and self-renewal are still poorly understood. Elucidating the mechanistic differences between CSCs and conventional tumor cells may be valuable for developing strategies for cancer therapy. The mammalian target of rapamycin (mTOR) is an essential serine/threonine kinase that regulates cell growth by controlling protein synthesis, autophagy, endocytosis, and metabolism in response to growth factors, nutrients, energy, and stress. Increasing evidence indicates that the signaling pathways that activate mTOR are frequently improperly regulated in most human cancers. Rapamycin, an inhibitor of mTOR, can block tumor growth and inhibit tumor cell motility. These findings have promoted the clinical use of mTOR inhibitors for cancer therapy. However, the use of mTOR inhibitors alone has had limited clinical success. The role of mTOR signaling in the maintenance of CSCs has been addressed recently, but the conclusions of these reports are controversial. In embryonic and adult stem cells, mTOR hyperactivation resulted in the differentiation and exhaustion of stem cells. In the tumor development, mTOR signaling has been shown to enhance the survival of dormant tumor cells and maintain the self-renewal and tumorigenicity of glioblastoma stem-like cells and breast cancer stem-like cells. In sharp contrast, mTOR inhibition by rapamycin has been shown to significantly increase CD133 expression in gastrointestinal cancer cells via down-regulation of HIF-1α. However, there is a lack of *in vivo* data regarding the changes in CSC maintenance and tumorigenicity in tumor cells after transient exposure to rapamycin. In this study, we examined the role of mTOR in the regulation of CD133+ CSCs differentiation, the conversion of CD133- conventional cancer cells to CD133+ CSCs, and the repopulation *in vivo* of mTOR-manipulated tumor cells. We found that mTOR inhibition increased the CD133+ subpopulation in liver tumor cells and potentiated the continuous growth of tumor cells *in vivo* via preventing differentiation, biased insensitivity of CD133+ subpopulation to rapamycin- induced apoptosis, and increasing the retrodifferentiation of CD133- to CD133+ cells. # Results ## mTOR inhibition increases CD133+ subpopulations and retain stemness properties in liver tumor cells To elucidate whether mTOR is involved in the maintenance of CSCs, H-Ras- transformed mouse liver tumor cells were employed in the study. We found that the percentage of CD133+ cells, which have been referred as a subpopulation containing CSCs, was approximately 0.2 to 1% in H-Ras-transformed mouse liver tumor cells (LPC-H) and 10 to 20% in the derived clone LPC-H12, both of which expressed markers of live progenitor cells, and the latter displayed much high potential of tumorigenicity. To examine whether there was a significant difference in mTOR activation between CD133- and CD133+ LPC-H12 cells, we purified these two population by FASC sorting and found that the CD133+ subsets exhibited lower activation of mTOR signaling compared with CD133- cells, as indicated by the levels of phosphorylated p70 S6 kinase (p-p70 S6). We next examined whether mTOR inhibition could lead to a change in the proportion of CD133+ LPCs. LPC-H and LPC-H12 cells were treated with the mTOR inhibitor rapamycin and measured CD133 expression by FACS. Intriguingly, the CD133+ fraction of the cells treated with rapamycin was significantly greater than that of the cells without treatment. The influence of rapamycin on the maintenance of CD133+ subpopulations was dose- and time-dependent. Expression of p-p70 S6 was undetectable in cells after rapamycin treatment, demonstrating the efficiency of rapamycin in eliminating mTOR activity. Notably, we found that the expression of several stem cell-associated genes, such as Nanog, Klf4 and SOX2, were increased significantly after treatment with rapamycin. To further identify the role of mTOR in the CD133+ subpopulation, we knocked down mTOR expression in LPC-H cells using the retrovirus expressing short hairpin RNA (shRNA). Downregulation of mTOR expression via mTOR shRNA-2 (mTOR-sh2) dramatically increased the CD133+ fraction in cells in a time-dependent manner. Similar results were obtained in LPC-H12 cells. In some human liver tumor cell lines, such as Hep3B, Huh7 and PLC/PRF/5, the enhanced lever of CD133 expression after treatment with rapamycin was also detected. Similarly, the expression of stem cell-associated genes, including Bmi1, Oct4 amd Nanog, were significantly increased in the human liver tumor cell lines treated with rapamycin, even in MHCC-97L cells that are no CD133 expression. Collectively, these data indicate that inhibition of mTOR could enrich CSCs in both mouse and human liver tumor cells. To investigate whether the disparities exist in the sensitivity to rapamycin between CD133- and CD133+ cells, we directly examined the effect of rapamycin on cell growth of the two subsets among LPC-H12 cells. We found that rapamycin- mediated inhibition in CD133- cells was greater than that in CD133+ cells, indicating that a greater impairment of the CD133- subset upon rapamycin treatment might partially contribute to the enrichment of CD133+ cells. ## mTOR activation by Rheb overexpression decreases the CD133+ subpopulations To test whether activation of mTOR signaling could decrease the CD133+ subpopulations, Rheb, an activator of mTOR signaling, was overexpressed in LPC-H and LPC-H12 cells by retroviral infection. We found that Rheb expression significantly enhanced mTOR activity, as determined by increased phosphorylation of p70 S6. As expected, mTOR activation resulted in significant exhaustion of the CD133+ subsets. A time-course analysis of CD133 expression by FACS showed that Rheb overexpression gradually decreased the percentage of CD133+ cells in LPC-H12 cells (from 19.22±2.09% to 3.94±0.86%), whereas cells infected with empty retrovirus showed only a slight decrease in CD133 expression (from 22.85±2.17% to 16.99±1.28%). Furthermore, the expression of all detected stem cell-associated genes were significantly reduced in Rheb-overexpressed cells. These results suggested that mTOR activation might enhance the differentiation of CD133+ to CD133- liver tumor cells. ## Single-cell culture experiments define a critical role for mTOR in the conversion of CD133- to CD133+ liver tumor cells Previous studies have shown that conventional cancer cells can be induced by certain events to express stem cell markers and acquire properties of CSCs. To address whether CD133- cells can be converted to CD133+ tumor cells without adding exogenous TGF-β, and if so, whether mTOR signaling is involved in the process, we employed a single-cell long-term culture system. CD133- cells of LPC-H12 were FACS-sorted, and each single cell was robotically plated into individual wells of 96-well plates. After 10 days in 96-well plates, single-cell derived clones were individually trypsinized and seeded into 24-well plates for expansion. After 11 days, the cells were harvested for FACS analysis of CD133 expression. Of significant interest, 46.7% of colonies derived from single CD133- LPC-H12 cells expressed CD133, and the percentage of CD133+ cells in the positive colonies was approximately 1 to 7%. Notably, we found that rapamycin treatment significantly increased both the formation of CD133+ cell colonies and the percentage of CD133+ cells in positive colonies. Accordingly, using the same protocol, downregulation of mTOR expression by shRNA in the LPC-H12 cells dramatically increased the number of colonies that contained CD133+ cells and the proportion of CD133+ cells in the positive colonies. Similar results were obtained using LPC-H cells (data not shown). In agreement with the FACS results, the level of *CD133* mRNA in the single-cell-derived colonies expressing mTOR- sh2 was significantly higher than that in the colonies containing control shRNA. ## Repression of mTOR inhibits differentiation of CD133+ liver tumor cells On the basis of recent findings that mTOR activity mediates the exhaustion and differentiation of stem cells, we hypothesized that the increase in the proportion of CD133+ cells in the presence of rapamycin might be partially due to blockage of differentiation of CD133+ cells. To test this hypothesis, we isolated CD133+ cells from LPC-H12 by FACS sorting, and further treated cells with rapamycin for 10 days. Strikingly, we found that CD133+ expression declined substantially in the control group, whereas the proportion of CD133+ cells in the rapamycin-treated group was maintained. Moreover, we sorted CD133+ cells and seeded them for single-cell culture. At day 10 of culture, 32.8% (63 of 192 wells) and 54.6% (105 of 192 wells) single-cell-derived colonies were formed in the culture medium with or without rapamycin, respectively. These colonies were then transferred into 24-well plates. After one week of culture, we found high levels of CD133+ cells in the colonies that had been treated with rapamycin. In agreement with these results, high CD133 expression was associated with colonies derived from single CD133+ cells with expression of shRNAs against mTOR. Collectively, these results illustrate that inhibition of mTOR can maintain CD133+ subpopulations by suppressing the differentiation of CD133+ cells. ## mTOR inhibition maintains CD133+ subpopulations *in vivo* and enhances re-propagation of Ras-dependent liver tumors To assess the maintenance of CD133+ subpopulations in tumor cells *in vivo* by inhibition of mTOR, we injected LPC-H cells expressing mTOR-sh2 or control shRNA into nude mice. After two weeks, both types of cells resulted in tumor formation. However, using FACS analysis to detect GFP+ tumor cells in the xenotransplants on day 24, a high percentage of CD133+ cells were preserved in tumors derived from LPC-H cells expressing shRNA against mTOR. No significant difference was found in tumor size between the two types of cells. Increased proportion of CD133+ tumor cells was also found in the tumor tissues after treatment of the tumor-bearing mice with rapamycin for 10 days. The increase in CD133+ subsets induced by mTOR inhibition led us to speculate that transient treatment with rapamycin might result in the enhanced continuous growth of Ras-dependent mouse liver tumors *in vivo*. We tested this hypothesis by taking advantage of the secondary tumor xenograft model. GFP+ LPC-H cells were subcutaneously injected into nude mice. The nude mice bearing xenografts were then treated with rapamycin or control diluents beginning 3 weeks after tumor implantation. After 10 days of rapamycin treatment, tumor growth was significantly reduced. The tumor tissues were enzymatically dissociated into single-cell suspensions, and CD133 expression was analyzed in the GFP+ LPC-H cells. The significant increase in CD133 expression in tumor cells from the first implantation after rapamycin treatment was similar to the effect of downregulation of mTOR *in vitro*. The tumor cells, harvested by FACS sorting of GFP+ population, were then transplanted into the secondary recipients. We found that as few as 2.5×10³ tumor cells with rapamycin treatment were sufficient to initiate tumor development, whereas up to 1×10⁵ tumor cells without rapamycin treatment failed to initiate subcutaneous tumors in secondary recipient mice. Thus, mTOR inhibition by rapamycin could block tumor growth from the first implantation but enhance continuous growth of the secondary tumors derived from H-Ras-transformed mouse liver cells. # Discussion Aberrant activation of the mTOR pathway has been implicated in the growth of various human cancers. mTOR signaling promotes cell proliferation, and, in the context of cancer development, it is highly correlated with tumor expansion and rate of tumor growth. Remarkably, several reports have highlighted the critical importance of mTOR signaling in driving differentiation of adult progenitor cells accompanied by rapid cell expansion. Recently, it has been established that the preservation of small subpopulations of CSCs within tumors is critical for tumors to assume malignant behavior and resistance to therapy. To date, whether mTOR signaling is critically involved in maintaining the balance between proliferation and differentiation is still poorly defined. Because mTOR inhibitors have been utilized for cancer therapy, and only modest or unremarkable success has been achieved in some clinical trials, it is urgent to investigate the long-term effects of transient mTOR treatment on tumor re- emergence and malignant outcome. We demonstrated here that targeted inhibition of mTOR significantly enhanced the generation and maintenance of CD133+ subpopulations and promoted secondary tumor re-propagation of H-Ras-transformed mouse liver tumor cells *in vivo*. CD133 is considered a universal marker of stem cells and CSCs. CD133+ tumor cells exhibit properties of CSCs in human hepatocellular carcinomas and mouse liver cancer. In this study, we found that H-Ras-transformed mouse liver tumor cells expressed the surface marker CD133. To our surprise, by exploiting the difference in the status of mTOR activation between CD133+ and CD133- populations, we found elevated mTOR signaling in the CD133- subset. Although the mTOR pathway has been reported to be important for the viability and maintenance of breast and glioblastoma CSCs, inhibition of mTOR with rapamycin in H-Ras- transformed mouse liver cancer cells significantly increased the proportion of CD133+ cells and retained the stemness properties. Using RNAi-mediated knockdown of mTOR and overexpression of Rheb, an activator of the mTOR kinase, we confirmed a critical role for mTOR in exhaustion and differentiation of the CD133+ subsets. Interestingly, these data are consistent with the notion that elevated p70 S6 kinase activation induces differentiation of human embryonic stem cells (hESCs). Furthermore, a previous study has shown that persistent activation of mTOR in normal epithelial stem cells results in exhaustion of these stem cells. Notably, mTOR pathway is also significantly important in the process of cellular senescence in multiple human and rodent cells. Rapamycin and MEK inhibitor that can target mTOR/S6 pathway, decelerate cellular senescence. Moreover, mTOR inhibition attenuates cellular senescence but favors quiescence in p53- arrested cells. These important findings are principally supportive to our observation since the impaired capacity in proliferation and the great potential to regain propagation (rejuvenation) are basically the features of CSCs. We must note that the similar effect of rapamycin was also found in multiple human liver tumor cell lines, such as Huh7, PLC/PRF/5, and Hep3B, suggesting that, even it was able to inhibit growth of tumor, rapamycin might enrich liver cancer stem-like cells in a clinical setting. Several studies have revealed that cellular stress (such as hypoxia), cytokines (such as TGF-β), or activation of the mTOR pathway are able to increase the expression of CSC surface markers and phenotypes in certain bulk tumor cells, indicating that cell-extrinsic environmental factors may reprogram conventional tumor cells to cells with stem cell-like properties in the course of cancer development. In the present study, using single-cell culture experiments that can bring out conclusive results, we unexpectedly revealed that CD133- liver tumor cells were capable of converting to CD133+ cells, and blocking intracellular mTOR activity with rapamycin or RNAi against mTOR indeed promoted the conversion of CD133- to CD133+ cells. Notably, certain stem cell-associated genes are upregulated by rapamycin prior to the increasing of CD133 expression in LPCs, suggesting that these genes might function in the rapamycin-mediated enrichment of CD133+ subpopulations. Previous studies have shown that HIF-1α is a positive downstream target of the mTOR signaling pathway. In human glioma cells, the activation of HIF-1α enhances CD133+ glioma-derived cancer stem cell expansion by increasing self-renewal activity and inhibiting cell differentiation. Thus, the relationship between HIF-1α and CD133 expression in the process of conversion requires further investigation. In human liver tumor cells, the induced CD133+ cells display significant potential for tumorigenesis. Indeed, we also found that CD133 expression and certain stemness genes were upregulated with the treatment of rapamycin. The relevant significance of the regulation mediated by rapamycin on CSCs maintenance was illustrated by the *in vivo* study, in which we demonstrated that the tumor cells harvested from the first implantation with treatment of rapamycin were significantly higher in secondary tumorigenicity than those from the control, as manifested by the increased CD133 expression. Notably, these results are in agreement with the evidence derived from RNAi-mediated downregulation of mTOR expression. These data suggest that mTOR signaling is involved in regulation of the balance of proliferation and differentiation of CSCs in Ras-dependent tumorigenesis and that transient inhibition of mTOR promotes tumor re-emergence via an increased CD133+ subpopulation. Taken together, our observations show that mTOR inhibition enriches CD133+ subpopulations. This enrichment is most likely achieved through blocking differentiation of the CD133+ subpopulations, enhancing apoptosis in the CD133- subsets, and triggering the conversion of CD133- to CD133+ cells. The maintenance of CD133+ cells *in vivo* by rapamycin leads to high continuous tumorigenic potential. Therefore, therapeutic design strategies should consider the possibility that complex regulatory events in CSC generation may be triggered by mTOR inhibition. # Materials and Methods ## Ethics statement All animal experiments were conducted in accordance with protocols approved by the Shanghai Medical Experimental Animal Care Commission (approval ID. ShCI-11-020). ## Cell culture Isolation, culturing and retroviral infection of purified p53-/- mouse fetal liver progenitor cells were performed as described by Zender et al.. Retroviruses were based on MSCV-IRES-GFP vectors containing cDNAs encoding H-Ras and used for infection of the purified cells. H-Ras-transformed mouse liver tumor cells were cultured in a maintenance medium containing 1∶1 DMEM:F12 (GIBCO, Grand Island, NY) with 10% fetal bovine serum (GIBCO) and the following additives: dexamethasone (1×10⁻⁷ M; Sigma-Aldrich, St. Louis, MO), nicotinic acid (0.01 M; Sigma-Aldrich), 2-mercaptoethanol (50 mM; Sigma- Aldrich), 1× ITS liquid media supplement (Sigma-Aldrich), EGF (20 ng/ml; R&D Systems, Minneapolis, MN), HGF (50 ng/ml; R&D Systems), and penicillin/streptomycin (1% \[v/v\]; GIBCO). For treatment studies, cells were plated and grown overnight in 1∶1 DMEM:F12 with 10% fetal bovine serum and penicillin/streptomycin (1% \[vol/vol\]). The medium was then replaced the following morning with medium containing rapamycin (Cell Signaling Technology, Beverly, MA) at concentrations of 0 to 100 nM. To assess the inhibition of mTOR by rapamycin, cells were treated with rapamycin (10 nM) for 24 h. These cells were lysed for immunoblot analysis. Human liver tumor cell lines, Huh7 (purchased from Riken Cell Bank, Tsukuba Science City, Japan), MHCC-97L (purchased from Liver Cancer Institute, Fudan University, Shanghai), Hep3B and PLC/PRF/5 (purchased from American Type Culture Collection, Manassas, VA) were maintained in DMEM with high glucose (GIBCO) supplemented with 10% fetal bovine serum and penicillin/streptomycin (1% \[vol/vol\]). ## Retroviral transfection A Rheb cDNA was Flag-tagged and cloned into the retroviral expression vector MSCV (MSCV-Rheb). Recombinant retroviral particles were generated from the Phoenix packaging cell line and harvested 2 consecutive times for infection. Transfected cells were pooled for further analysis after selection with 10 µg/ml of puromycin (Sigma-Aldrich) for 1 week and then with 5 µg/ml for additional 1–2 weeks. Short-hairpin RNAs targeting *mTOR* (mTOR-sh1 and mTOR-sh2), based on *mTOR* coding sequences 5'-AAGAATGTTGACCAATGCT-3' and 5'-TGATGGACACAAATACCAA-3' respectively, were expressed from the LTR promoter of MSCV retroviruses. ## Immunoblotting Cell lysates were prepared with RIPA lysis buffer (Beyotime, Nantong, China) with Protease Inhibitor Cocktail and PhosSTOP Phosphatase Inhibitor Cocktail (Roche, Monza, IT) added. Specific antibodies were used for analysis: anti- phospho-mTOR (Ser2448), anti-mTOR, anti-phospho-p70 S6 kinase (Thr389), anti-p70 S6 kinase (all from Cell Signaling Technology); anti-Flag (Sigma); and anti-α- tubulin (Santa Cruz Biotechnology, Santa Cruz, CA). Images were obtained with the ChemiDoc^TM XRS system (Bio-Rad Laboratories, Hercules, CA). ## RNA extraction, reverse-transcription, and RT-qPCR Total RNA was extracted from cultured cells with RNAiso Reagent (TaKaRa, Dalian, China) and converted to cDNA using the PrimeScript^TM RT reagent kit (TaKaRa) according to the manufacturer’s protocol. Gene expression analysis was determined using primers for *CD133*, *AFP*, *albumin* (*Alb*), *cytokeratin 19* (*CK19*), and *GAPDH*, as listed in. For detection of *CD133* expression, relative quantitative PCR (qPCR) was performed on a 7300 Real-Time PCR System (Applied Biosystems, Foster City, CA) using Premix Ex TaqTM (TaKaRa). Quantitative real-time PCR was performed using SYBR green mix from TaKaRa for *Bmi1*, *Klf4*, *Nanog*, *Oct4*, and *Sox2*. All samples were normalized to the level of *GAPDH*. The primers are listed in. ## Fluorescence-activated cell sorting and single-cell analysis Cells were dissociated and resuspended in PBS containing 0.5% BSA. Flow cytometry was performed with PE-conjugated anti-mouse CD133 antibody (eBioscience, San Diego, CA) or PE-conjugated anti-human CD133 antibody (Miltenyi Biotec, Bergisch Gladbach, Germany) incubated at 4°C for 30 min. Rat IgG1/κ antibody conjugated to phycoerythrin was used as an isotype control. Preincubation with 7-AAD (Sigma-Aldrich) was used to exclude dead cells. For cell sorting, samples were stained with PE-conjugated anti-mouse CD133 antibody and analyzed using a MoFlo XDP (Beckman Coulter, Fullerton, CA) with rat IgG1/κ antibody conjugated to phycoerythrin as isotype controls. Isolated CD133- or CD133+ cells were plated in 6-well plates. For single-cell experiments, single CD133- or CD133+ cells were stringently gated, isolated using a FACSVantage set for Single Cell Purity, and robotically plated into a 96-well plate with 150 µl medium/well. After 24 h, an additional 50 µl medium with or without rapamycin was added to each well. After 1–2 weeks in culture, robust single-cell-derived colonies that filled more than 50% of the well area were transferred into 24-well plates for expansion. ## Tumor cell preparation Tumor tissue was mechanically cut into small pieces with sterile scalpel blades, then enzymatically dissociated for 1 h with a mixture of collagenase type I (0.05 mg/ml; Sigma-Aldrich), collagenase type IV (0.05 mg/ml; Sigma-Aldrich), hyaluronidase (0.025 mg/ml; Sigma-Aldrich), and DNase I (0.01 mg/ml; Roche) at 37°C and shaken repeatedly. The dissociated sample was then filtered (70-µm pore size) and washed with PBS. Cells were then injected into nude BALB/c mice for secondary tumor propagation or stained with mouse CD133-PE antibody for assessing CD133 expression using a FACSCalibur flow cytometer (BD Biosciences, San Jose, CA). ## Subcutaneous implantation into nude BALB/c mice and rapamycin treatment H-Ras-transformed mouse liver tumor cells (1×10⁶) were suspended in 100 µl PBS and injected subcutaneously into male nude BALB/c mice (5 to 6 weeks of age). Tumor size was serially measured using calipers. Tumor volume was estimated using the following formula: (*w*₁×*w*₂×*w*₂)/2, where *w*₁ represents the length and *w*₂ represents the width of the tumor. In some experiments, rapamycin (LC Laboratories, Woburn, MA) was reconstituted in absolute ethanol at 10 mg/ml and diluted in 5% PEG-400 (Sigma-Aldrich) and 5% Tween-80 (Sigma-Aldrich) for treatment of mice. Each treatment group consisted of 4 or 5 animals. Nude mice bearing xenografts received 1.5 mg/kg rapamycin by i.p. injection daily starting 3 weeks after tumor transplantation. Control animals bearing tumors were treated with diluent at equivalent doses and schedules to the experimental animals. Tumors were removed and injected into secondary recipients by limiting dilution experiments. ## Cell viability assay A total of 500 cells/well were plated in 96-well plates. After 24 h, rapamycin (10 nM) was added to the media. Cell viability was performed daily by adding 3-(4,5-dimethyl-2-thiazolyl)-2,5-diphenyl-2H-tetrazolium bromide (MTT) (Sigma- Aldrich) to each well and measuring the absorbance at 540 nm. ## Apoptosis assay For quantification of apoptosis, Pharmingen™ PE Aneexin V Apoptosis Detection Kit (BD Biosciences) was performed according to the manufacturer’s instructions. ## Statistical analysis Statistical analyses were performed using SPSS 11.5.0. A paired, two-tailed Student’s *t* test was used to determine the significance between two groups. A *p* value less than 0.05 was considered statistically significant. # Supporting Information [^1]: Conceived and designed the experiments: YL ZY JL. Performed the experiments: ZY LZ AM LL. Analyzed the data: ZY YL. Contributed reagents/materials/analysis tools: YL ZY. Wrote the paper: ZY YL JG. [^2]: The authors have declared that no competing interests exist.

# Introduction Bipolar disorder (BD) is a severe psychiatric disease generally characterized by repeated manic and depressive episodes. The time of onset varies, but it usually begins between the ages of 15 and 25 years, affecting men and women equally with a lifetime prevalence of 1.5%. Family, twin, and adoption studies have shown that genetic factors contribute to BD, but the involved genes have not yet been identified. This may partly be explained by a genetic architecture characterized by both genetic heterogeneity (at the population level) and polygenic interactions (at the level of the individual). Locus heterogeneity in a disease may occur when dysfunctions in different mechanisms each produce the symptoms characterising the phenotype. In the search for disease-associated genes, high-throughput genotyping methods have allowed analysis of a large number of single nucleotide polymorphisms (SNPs), but most of these studies have used single-locus analysis strategies. Yet, many common diseases have complex aetiologies that may involve combinations of SNPs from different genes and, possibly, different combinations within the population of affected individuals. The growing interest in interactions and their contribution to risk for complex diseases has resulted in a search for methods of calculating disease-related interactions of two or more SNPs. Most of these methods are theoretical, and many concentrate on finding combinations of two SNPs. Only a few studies have looked at combinations of more SNP genotypes; for example, Xie et al., in a study of oesophageal cancer, analysed SNP genotypes from genes related to DNA repair mechanisms and liver function and identified combinations of genotypes relevant to the disease. In the quest for alternative approaches in the search for interacting SNPs, interest has grown in pathways with an enrichment of associated signals. These methods are robust to the detection of enrichment that derives from genetic heterogeneity at the population level or from gene- or protein-interactions at the individual level. In BD, hyperactivity is the main symptom of the manic phase, and this clinical phenotype may reflect altered activity at the cellular or molecular level that leads to faster signal transmission in the brain. Numerous mechanisms are involved in signal transmission, suggesting that the mania phenotype may result from many different dysfunctions in mechanisms relevant to signal transmission velocity in the nervous system; this concept is in agreement with the genetic heterogeneity of BD. We previously suggested that propagation of nerve impulses may be faster in mania than in a normal affective state, which is in agreement with a number of similar hypotheses proposing mania as a disorder of ionic conductance of nerve cell membranes, nerve cell excitability, action potential firing, neuronal hypersynchronization, changes in sodium pump number, membrane abnormalities, and cortical instability. Furthermore, the suggestion that nervous system hyperactivity may play a role in BD is in accordance with the fact that several drugs used to treat epilepsy are effective in the treatment of mania. We have investigated genes (1) related to the action potential, where we focused on genes related to the architecture of the sodium channels in the nodes of ranvier and CNS myelination,. Among these genes, we have selected those (2) that occur in loci considered to be of interest with respect to BD, as indicated by linkage analyses, ; (3) that encode proteins that are targets for mood- stabilizing drugs, ; or (4) that previous studies have shown are associated with BD. Many genes in this group are related to ion channels, an observation that agrees with the finding of Askland et al. that variation in ion channel genes may contribute to susceptibility to BD. In the present study, we analysed 803 SNPs in 55 genes related to aspects of signal transmission and calculated all combinations of three genotypes (the normal homozygote, the heterozygote, and the variant homozygote, respectively) from the 3×803 SNP genotypes. We likewise calculated all combinations with three SNP genotypes from the genome wide association study (GWAS) performed by the Wellcome Trust Case Control Consortium (WTCCC) using SNPs analysed in both datasets. We found 132 SNPs that were genotyped in both studies, by imputation we obtained additional 337 SNPs with a good quality, thus 469 SNPs were among our 803 SNPs and available for validation. # Results ## Calculation of single SNP association 803 SNPs in 55 genes were analysed and a Chi-square test (or Fishers exact test when appropriate) was performed for each SNP. The genotype distribution was significantly different (p\<0.05) between control persons and patients for 86 SNPs, but none remained significant after a Bonferroni correction. ## Combination of two SNPs Combinations of two SNPs from 803 SNPs results theoretically in 803!/2!(803−2)!×9 = 2,898,027 combinations, whereas the actual number in the material was 2,770,033 combinations. Chi-square test for each of these combinations was not performed, but 1000 permutation tests showed that combinations found exclusively in the patients (161,070 combinations) could be random. ## Combinations with three SNPs Combinations of three SNP genotypes from the 803 SNPs results theoretically in 803!/3!(803−3)!×3³ = 2,321,319,627 combinations of three-SNP genotypes (in the following to be termed 3-combinations). The actual number of 3-combinations of genotypes found in our material (1962 individuals) was 1,985,613,130. Most, namely 1,719,002,329 of these 3-combinations (87%) were common for both control persons and patients, whereas 208,699,590 3-combinations were found in control persons only, and 57,911,211 3-combinations were found in patients only. The number of patient-specific 3-combinations shared by several patients decreases as the number of patients in a group increases. When the number of patients was nine or more, only 1181 3-combinations of the 57,911,211 patient-specific 3-combinations remained. In order to see if the subgroup of 1181 3-combinations may be of importance for BD, 1000 permutation tests of the complete material were calculated, and it was found that 1181 3-combinations shared by nine or more patients might be a random finding (found 113 times; p = 0.11). Among the 1181 3-combinations, many genotypes were part of one or two 3-combinations and only a few genotypes occurred in more than ten 3-combinations. However, we observed that four genotypes (*AVPR1B*\_rs33976516 = 1 (one indicate heterozygocity), *KCNN3*\_rs884664 = 2 (two indicate homozygosity for the minor allele), *CACNG2*\_rs2179871 = 2, *KCNQ3*\_rs2469515 = 2) occurred in 46, 45, 49, and 32 3-combinations, respectively. The genotypes were all among the 86 SNPs showing a significantly different distribution between control persons and patients (nominal p-values were 0.013, 0.010, 0.023, and 0.017, respectively). Futhermore; these four observed subgroups of 3-combinations contained all the patients in the material (41, 48, 41, and 37 patients, respectively) having the four genotypes. Such subgroups characterised by relatively many 3-combinations sharing a defining genotype and containing all the patients having this genotype are called clusters. In 1000 permutation tests, at least one cluster of this type (with at least 37 pseudo-patients) were seen 42 times (p = 0.042), two such clusters were seen 3 times (p = 0.003), and at least three or more clusters of this type did not occur once (p\<0.001). Relatively little overlap between the patients in the clusters was observed, as only eleven of the patients were members of two clusters, and no patient was a member of three or four clusters. A total of 156 patients were involved in the four clusters. All patients in a cluster shared the defining genotype, but nested within the four clusters, eleven smaller subgroups were observed. In these subgroups patients sharing the same defining genotype (e.g. *AVPR1B*\_rs33976516 = 1) also shared a second genotype (e.g. *ANK3*\_rs2288358) in combinations with four or more distinct third SNPs (shown in – as the boxes in the third column). Considerable overlap between patients was found within these subgroups. In 1000 permutation tests only 49 subgroups of this type were found (p = 0.049). Additionally, within this type of subgroup, 12 even smaller subgroups were characterized by having the third genotype in the 3-combinations contributed by SNPs from the same gene (shown in – as more SNP numbers in the same box). These 12 small subgroups together comprised 32 3-combinations. Only 19 3-combinations of this type were found in 1000 permutation tests (p = 0.019). ## WTCCC data Of our 803 SNPs 132 were genotyped in the WTCCC bipolar disorder dataset. In order to obtain more SNPs, we performed an imputation using impute2 on the chromosome regions surrounding our selected genes. We could impute 651 SNPs of our 803 SNPs, but after removing SNPs with a low quality we ended up with additional 337 SNPs, leaving 469 SNPs in common between our material and the WTCCC. Chi-square test (or Fishers exact test when appropriate) was performed for each SNP. The genotype distribution was significant different (p\<0.05) between control persons and patients for 51 SNPs, but none remained significant after a Bonferroni correction. Five of the 51 SNPs were among the 86 significant SNPs in our material. The theoretically number of 3-combinations from the 469 SNPs is 461,262,438, while the observed number was 287,931,183. Most, namely 247,477,254 were seen in both controls persons and patients, whereas 16,762,350 3-combinations were found in control persons only, and 23,691,570 3-combinations were found in patients only. Using the above mention criteria for a cluster, three clusters were observed among the patient-specific 3-combinations. The defining genotypes were *NFASC*\_rs12737855 = 2, *NFASC*\_rs7519658 = 2 and *ANK3*\_rs6479700 = 2, containing 124, 135 and 182 combinations, and 159, 150 and 142 patients, respectively (data not shown). All three genotypes were among the 51 significant SNPs. In 1000 permutation tests, at least one cluster of this type (with at least 142 pseudo-patients) were seen 49 times (p = 0.049), two such clusters were seen 3 times (p = 0.003), and at least three clusters of this type did not occur once (p\<0.001). Within the three clusters, we again observed smaller subgroups where patients also shared the second genotype in six or more combinations, and there were a considerable overlap between the patients involved in these subgroups. In 1000 permutation tests only 29 of such subgroups were found (p = 0.029). There was a substantial overlap between the patients in the two clusters with the defining SNPs belonging to the *NFASC* gene (n = 134), while only a limited number of patients belonging to the third cluster (defined by *ANK3*\_rs6479700) were present in the two other clusters (n = 10 and n = 8, respectively, where eight were part of all three clusters). # Discussion Complex diseases may be associated with combinations of SNPs. A number of methodological and theoretical studies have addressed this statistical and data- mining challenge, but clinical investigations using combinations of several SNPs are rare. The problem with combinations is the large numbers created, which is computationally demanding, especially when permutation tests are used as a statistical method. Thus in the present study, with 803 SNPs and combinations of no more than three SNP genotypes at a time, we found 1,985,613,130 3-combinations, close to the theoretical maximum of 2,321,319,627 3-combinations. Due to the relative low number of subjects and genetic factors as allele frequency and non-independence of SNPs located in the same gene region, we had expected to find a smaller number of 3-combinations. In an attempt to identify 3-combinations related to BD, only 3-combinations found exclusively in the patient group were examined in more detail in this study, meaning that only combinations with 100% penetrance were examined. Among the 1,7 billion 3-combinations common for control persons and patients many may be associated with disease; these combinations will be analysed separately. In the 58 million 3-combinations found exclusively in the patients, 45 million were singularities (found in one person only), raising the possibility that they were random. Following this line of reasoning, the 1181 3-combinations, shared by nine or more patients and no control person, may be the most promising in relation to BD, but permutation tests showed that these also might have been random. However, 172 of these 3-combinations were located in four clusters each characterised by a defining genotype, and by inclusion of all patients having this genotype. Occurrence of three or more clusters of this type was not found once in 1000 permutation tests, indicating that at least some 3-combinations in the clusters may be related to BD. In addition, subgroups within the four clusters also shared the second genotypes, and some had the third genotype in the 3-combinations located in the same gene, suggesting that accumulation of several genotypes in a single gene may be important for the disease susceptibility in some cases. We found only 5 nominally significant SNPs in common between the Scandinavian material and the WTCCC material, indicating heterogeneity between the two samples. A cluster with the same defining genotype as in one of the four clusters observed in our material could not be found in the WTCCC material, as none of the four defining SNPs were present among the 132 genotyped SNP or the 337 imputated SNPs in the WTCCC dataset. However, three significant clusters were found. Again three of more clusters were not seen once in 1000 permutation tests. Also nesting 3-combinations sharing the two first genotypes was present in the WTCCC dataset (p = 0.029). An overlap in the clusters observed in the two dataset were not see in the 3-combinations exclusively found in patients, but may be found in the much larger group of combinations common for control persons and patients. An examination of individual patient data in both sample sets shows that most of the patients carry many of the 3-combinations in the clusters although none had exactly identical pattern of 3-combinations, raising the possibility that each patient has a unique genetic background for the disorder. The subgroups of 3-combinations with overlap in patients, sharing two genotypes in more than four 3-combinations, are interesting because different SNP as the third (sometimes even from the same gene (in some cases in close LD)) leads to an accumulation of several genotypes in small group of patients. Such an accumulation may be important for the disease susceptibility. Askland et al. found that although the data from two large independent GWAS, both pointed to ion channel genes as important for BD, only a modest overlap between the two studies was found for the involved genes. The authors suggested that prominent genetic heterogeneity might explain this modest overlap. The present analyses of the 3-combinations of SNP genotypes support the explanation that genetic heterogeneity is prominent in the genetic architecture of BD. This heterogeneity is illustrated by the many patient-specific 3-combinations of SNP genotypes, some of which may be important for BD. The selection of genes in this study is based on their relation to some aspects of signal transmission in the brain, so obviously any combination of genotypes might be related to this function. However, a more narrow relationship may be seen in the cluster defined by *KCNQ3* rs2469515 together with *ANK3* rs12049756 and three different SNPs from *SCN2A* (rs12469667, rs3769949, and rs997508); as the proteins translated from these genes all are located in the node of ranvier. Similarly, in the largest subgroup defined by *CACNG2* rs2179871, and involving 13 combinations with *KCNN3* rs6426998 as the second SNP, many of the nine different genes involved as the third SNP are implicated in the architecture of the sodium channels in the node of ranvier (e.g. *SPTBN4*, *CNTNAP2*, *NFASC*, *SCN2A* and *ANK3*). Our study indicates that BD may show extreme genetic heterogeneity at the population level. At the same time the many 3-combinations in each patient may support gene-gene interactions or epistasis important for BD. However, such interactions probably will involve genes not analysed in the present study. A more profound discussion of functional connections between the genes participating in a combination remain speculative and is preliminary until more genes related to signal transmission are analysed and combinations of more than three genotypes can be carried out. The pronounced genetic heterogeneity and the number of possible interactions on the individual level both suggest that the biology of BD may be very complex; but on the other hand, if the genotypes behind the heterogeneity are associated with a limited number of functions the degree of complexity may be decreased. ## Future direction In this work we have looked at the combinations of three SNP genotypes that were observed in patients only. The next step is to look at the much larger number of combinations seen in both patients and controls. However, this may involve calculations of combinations with more than three SNP genotypes. # Materials and Methods ## Patient sample This study is based on two independent case-control samples from Norway and Denmark, included in the Scandinavian Collaboration of Psychiatric Etiology (SCOPE). The Danish sample consisted of 220 bipolar patients from the Copenhagen area, 162 bipolar patients in Jutland, and 1133 control participants. The sample from Norway included 222 controls and 225 bipolar patients. Thus, a total of 607 unrelated patients and 1355 unrelated healthy control participants were included. The Norwegian patients had been diagnosed according to the DSM-IV and the Danish patients according to ICD-10. The Norwegian and Danish healthy controls and cases have been described in more detail elsewhere. The Norwegian Scientific-Ethical Committees, the Norwegian Data Protection Agency, the Danish Scientific Committees, and the Danish Data Protection Agency approved the study. All patients gave written informed consent prior to inclusion in the project. ## Genes The genes, shown in, were selected based on the following criteria: (1) literature relating corresponding functional proteins to various aspects of the action potential (with focus on the architecture of the sodium channels in the nodes of ranvier and CNS myelination), –; (2) occurrence in loci associated with BD, as indicated by linkage analyses , ; (3) encoding of corresponding proteins that are targets for mood stabilizing drugs, ; and (4) identification in previous studies showing an association with BD,. Additional we used information from different internet sources ([www.ncbi.nlm.nih.gov/Database/](http://www.ncbi.nlm.nih.gov/Database/); [www.ensembl.org/index.html](http://www.ensembl.org/index.html); [www.pubgene.org](http://www.pubgene.org); [www.ensembl.org/index.html](http://www.ensembl.org/index.html); [www.string.embl.de](http://www.string.embl.de); [www5.appliedbiosystems.com/too ls/pathway/all_pathway_list.php](http://www5.appliedbiosystems.com/tools/pathway /all_pathway_list.php); [www.reactome.org](http://www.reactome.org)). We prioritized genes fulfilling more than one criterion, but realize that other genes fulfil the criteria and thus could have been included instead. The genotyping were planned in the autumn of 2007 and preformed in the spring 2008. ## SNP selection and genotyping To cover most of the common variants with tagSNPs, we used a structured gene- wide approach, based on the HapMap CEU population. TagSNP selection was performed at the HapMap website using pair-wise tagging, with r²≥0.8 ([www.hapmap.org](http://www.hapmap.org); HapMap Data Rel 22/phaseII Apr07) and minor allele frequency (MAF)≥0.05. Some SNPs (some with MAF\<0.05) were selected if they resulted in a missense mutation or if they had been linked to BD. Not all selected tagSNP were genotyped, and exclusion was attributable to the following reasons: (1) a design score \<0.4 for the Illumina Platform (n = 109); (2) failure during the genotyping analysis (n = 76); or (3) being discarded during quality control for several reasons \[sample call rate \<90%, more than three clusters seen in the result, and/or SNP not in Hardy-Weinberg equilibrium (HWE; p\<0.001) (n = 94)\]. This resulted in 803 SNP for analyses as described below. Genomic DNA was extracted from whole blood. Most (796) SNPs were genotyped using the GoldenGate 1536plex assay (Illumina Inc.) on Illumina BeadStation 500GX at the SNP Technology Platform, Uppsala University, Sweden ([www.genotyping.se](http://www.genotyping.se)), accredited by the Swedish accreditation agency SWEDAC, and approved according to a quality system based on the international SS-EN ISO/IEC 17025 standard. For the subset of SNPs used in this study, the reproducibility was 99.999% (there was one duplicate error in 70,098 duplicate genotype calls), and the average sample call rate per SNP assay was 99.6%. The four SNPs in *YWHAH* and the three SNPs in *SCN1B* were genotyped using TaqMan genotyping assays according to the manufacturer's instructions. For these, the reproducibility was 99.72%, and the average sample call rate per SNP assay was 98.4%. ## Statistics and data processing The samples were tested for population stratification by calculating the gene- based overall fixation index F_ST using Arlequin Software. The statistical significance of single genotype distribution was assessed using the Chi-square or Fisher's exact test, whereas patterns of SNP combinations were assessed by permutation tests. Combinations of genotypes were calculated using array-based mathematical methods, where data are represented geometrically, hereby facilitating parallel processing, which was performed using programs from Genokey ([www.genokey.com](http://www.genokey.com)) and Dyalog ([www.dyalog.com](http://www.dyalog.com)). Each of the 1000 permutation tests (in our dataset) were performed as follows: 1) A permutation of the entire population (i.e. indicies 1 to 1962) is determined. The result is a new vector with all indices 1 to 1962 in random order. 2) 607 random “patients” are selected, and the remaining 1355 individuals are “controls”. 3) The cluster analysis on the 607 “patients” and 1355 “controls” is determined using exactly the same methods as previously on the biological samples. Likewise, permutation tests were performed in the WTCCC dataset with 1998 random “patients” and 1500 random “controls”. ## Imputation of WTCCC data We obtained genotype data on the bipolar patients (n = 1998) and the UK blood service control group (n = 1500) from the WTCCC data. We searched for SNPs genotyped in both samples and found 132. To get more SNPs in common for the two dataset, we performed imputation using Impute2 (mathgen.stats.ox.ac.uk/impute/impute_v2.html), of SNPs in the chromosome regions around our selected genes (with at minimum of 500 kb surrounding area). We used both the provided hapmap3 dataset and the 1000 genome project samples as reference samples resulting in “Ne” to be 15000. We used default setting with the following exception: we set “iter” to 40 and “k” to 100 to get better genotypes, we selected a call thresh hold to be 0.9, and set the result to include only the SNPs, that we have genotyped. Prior to the imputation we exclude SNPs from the WTCCC dataset, that WTCCC had excluded due to low genotyping quality. We only include imputated SNPs with a certain (more than 80% probability for a given genotype) genotype calling in more than 80% of the subjects. # Supporting Information We thank patients and controls for their participation in the study and the health professionals who facilitated our work. We thank Kristina Larsson, Per Lundmark, Tomas Axelsson, and Ann-Christine Syvänen at the SNP Technology Platform in Uppsala for performing the genotyping. The SNP Technology Platform is supported by Uppsala University, Uppsala University Hospital, and by the Knut and Alice Wallenberg Foundation. This study makes use of data generated by the Wellcome Trust Case-Control Consortium. A full list of the investigators who contributed to the generation of the data is available from [www.wtccc.org.uk](http://www.wtccc.org.uk). Funding for that project was provided by the Wellcome Trust under award 076113. [^1]: Conceived and designed the experiments: PK EM. Performed the experiments: PK BB GLM EM SD. Analyzed the data: PK GLM EM. Contributed reagents/materials/analysis tools: PK EM GLM. Wrote the paper: PK OAA EM. Designed the software used in analysis: GLM. Collection of samples: HD IM LVK MBJ OAA OM TH TW. [^2]: The authors have declared that no competing interests exist.

# Introduction Statistical models for analysis of risk factors for a disease or clinical complications, a main focus of medical research, require that the number of patients is larger than the number of variables (factors) to ensure that the statistical significance of the results can be appropriately established. In practice, most studies assess only the influence of each variable separately rather than the combined importance of a set of variables; the former oversimplistic but yet prevailing approach ignores the possibility of interactions between variables or between groups of variables. The obvious need of developing new statistical tools that take into account the extensive interactions between the very large numbers of variables determining biological processes and hence clinical outcomes is increasingly emphasized in modern medical and bioinformatics research. Medical data sets collected today often have a large number of variables for a relatively low number of patients. This may happen for genetic data sets, where the number of variables (genetic variability, as single nucleotide polymorphism, or gene expression data) can be thousand times greater than the number of patients. Statistical methods are not fully justified in this situation. In such a case, data mining methods can be used instead of, or in addition, to statistical methods. The methods of feature subset selection developed in the scope of data mining play an increasingly important role in the exploratory analysis of multidimensional data sets. Feature selection methods are used to reduce feature space dimensionality by neglecting features (factors, measurements) that are irrelevant or redundant for the considered problem. Feature selection is a basic step in the complex processes of pattern recognition, data mining and decision making. Interesting examples of applications of feature selection procedures can be found, among others, in bioinformatics. A survey of noteworthy methods of feature selection in the field of pattern recognition is provided in. The feature subset resulting from feature selection procedure should allow building a model on the basis of available learning data sets that can be applied for new problems. In the context of designing such prognostic models, the feature subset selection procedures are expected to produce high prediction accuracy. We apply here the *relaxed linear separability* (*RLS*) method of feature selection for the analysis of data on clinical and genetic factors related to *inflammation*. These data were obtained from the so called *malnutrition, inflammation and atherosclerosis* (MIA) cohort of incident dialysis patients with end-stage renal disease in whom extensive and detailed phenotyping and genotyping have been performed,. The cohort was split into two groups: inflamed patients (as defined by blood levels of C-reactive protein, *CRP*, above median) and non-inflamed patients (as defined by a *CRP* below median). Then, genetic and phenotypic (anthropometric, clinical, biochemical) risk factors that may be associated with the plasma *CRP* levels were identified by exploring the linear separability of the high and low *CRP* patient groups. Particular attention was paid in this work to study the complementary role of genetic and phenotypic feature subsets in differentiation between inflamed and non-inflamed patients. Four benchmarking feature selection algorithms were selected for the comparisons with *RLS* method on the given clinical data set: 1) *ReliefF*, based on feature ranking procedure proposed by Kononenko as an extension of the *Relief* algorithm, 2) *Correlation-based Feature Subset Selection - Sequential Forward* algorithm (*CFS-SF*), 3) *Multiple Support Vector Machine Recursive Feature Elimination* (*mSVM-RFE*) and 4) *Minimum Redundancy Maximum Relevance* (*MRMR*) algorithm. The *CPL* method and four other frequently used classification methods (*RF* (*Random Forests*), *KNN* (*K - Nearest Neighbors*, with K = 5), *SVM* (*Support Vector Machines*), *NBC* (*Naive Bayes Classifier*)) were applied for classification of patients on the basis of the selected features. # Methods ## Relaxed Linear Separability Method A detailed description of the *relaxed linear separability* (*RLS*) method as applied in the present study is provided in together with all the definitions. A brief summary of the method is presented below. The *RLS* method of feature subset selection is linked to the basic concept of linear separability. The linear separability means possibility of two learning sets separationby a hyperplane. The linear separability notion originated from the perceptron model linked to the beginning of neural networks. Detection and evaluation of linear separability can be carried out efficiently by minimizing the *perceptron criterion function*. This function belongs to the more general class of the *convex and piecewise-linear (CPL)* criterion functions. The *perceptron* criterion function was modified by adding a regularization component for the purpose of the feature subset selection task. The regularization component has similar structure to those used in the *Lasso regression*. The main difference between the *Lasso* and the *RLS* methods is in the types of the basic criterion functions. The basic criterion function used in the *Lasso* method is that of the *least squared* method, whereas the perceptron criterion function and the modified criterion function are used in the *RLS* method. This difference effects the computational techniques used to minimize the criterion functions. The modified criterion function, similarly to the perceptron criterion function, is convex and piecewise-linear (*CPL*). The basis exchange algorithms allow the identification of the minimum of each of these *CPL* criterion functions. The basis exchange algorithms are similar to linear programming and allow to find the optimal solution efficiently even in the case of large, high dimensional learning sets. The (*RLS*) method of feature subset selection is based on minimization of the modified perceptron criterion function and allows for successive reduction of unnecessary features while preserving the linear separability of the learning sets by increasing the cost parameter in the modified criterion function. The stop criterion for discarding the unnecessary features was based on the cross- validation error (CVE) rate (defined as the average fraction of wrongly classified elements) estimated by the leave-one-out method. The evaluation of the *RLS* approach was previously carried out with good results both when applied on simulated high dimensional and numerous data sets as well as on benchmarking genetic data sets. For example, the *RLS* method were used for processing the *Breast cancer* data set. The number of features (genes) in this set is equal to 24481. The *RLS* method allowed to select from this set the optimal subset of 12 genes and such linear combination of these genes (*linear key*), which allows to correctly distinguish with 100% accuracy two leaning sets composed of 46 cancer and 51 non-cancer patients. ## Alternative Methods for Feature Selection and Classification The *RLS* method of feature subset selection involves generation of the sequence of the reduced feature subspaces (see, equation 7). The sequence is generated in the deterministic manner through a gradual increase of the cost level in the minimized criterion function (see, equation 5). In order to determine the best (final) subspace in the sequence an evaluation of the quality of individual subspaces is needed. Traditionally, the quality of the feature subspaces is evaluated through the quality evaluation of the classifiers built in this subspace. Statistical methods for evaluation and comparison of classifiers can be found in. This section presents a few other previous methods of feature selection and classification that were applied for the analysis of the MIA data sets, for comparison of the results, see Results. Four benchmarking feature selection algorithms were chosen for an experimental comparison with the *RLS* method. One of the selected algorithms, *ReliefF*, is based on feature ranking procedure proposed by Kononenko as an extension of the *Relief* algorithm. The *ReliefF* searches for the nearest objects from different classes and weighs features according to how well they differentiate these objects. The second one is a subset search algorithm denoted as *CFS-SF* (*Correlation-based Feature Subset Selection - Sequential Forward*). The *CFS- SF* algorithm is based on a correlation measure which evaluates the goodness of a given feature subset by assessing the predictive ability of each feature in the subset and a low degree of correlation between features in the subset. These two feature selection algorithms are considered as “the state of the art” tools for feature selection. The third algorithm, *mSVM-RFE*, is a relatively new idea. It is an extension of the *SVM-RFE* algorithm (*Support Vector Machine Recursive Feature Elimination*). The *SVM-RFE* is an iterative procedure that works backward from an initial set of features. At each round it fits a simple linear *SVM*, ranks the features based on their weights in the *SVM* solution, and eliminates the feature with the lowest weight. Multiple *SVM-RFE* (*mSVM- RFE*) extends this idea by using resampling techniques at each iteration to stabilize the feature rankings. The fourth algorithm *MRMR* (*Minimum Redundancy - Maximum Relevance*) is also a relatively new idea. It bases on feature ranking procedure with special ranking criterion. The position of single feature in the list depends both on its correlation with class and dissimilarity to each feature above it in the ranking. To compare feature selection algorithms and to evaluate the selected feature subspaces, four frequently used classification methods, beside the *CPL* method, were applied: The four first classifiers (1) were designed by using *Weka*'s implementation. The *Weka*'s implementation of *ReliefF* and *CFS-SF* was used also for the feature selection and cross validation evaluation of designed classifiers. The *R* implementation of *mSVM-RFE* was used (*SVM-RFE* package). The results of *MRMR* was obtained with the help of the code provided by its author. The *CPL* classifiers based on the search for optimal separating hyperplane through minimization of the *CPL* criterion functions (see, equation 4) was applied using our own implementation. Our own implementation was also used for the *RLS* method of feature selection. ## Clinical Data Sets Two learning sets and were selected from a cohort of patients with chronic kidney disease, the *MIA* cohort. The set contained patients with a high *CRP* levels (above the median value) and the set contained patients with a low plasma *CRP* levels (below the median value). Each patient from the learning sets and was characterized by numerical results of anthropometric or biochemical measurements and by sites of genetic polymorphism (single nucleotide polymorphisms (*SNPs*) or deletions/insertions). The polymorphisms were selected from different candidate genes each harboring one to four of these variations. Each site of the genetic polymorphism was characterized by (usually three) binary features, that described three possible genotypes at this site (for example). The value one of the binary feature represented the appearance of a particular genotype at the polymorphic site. Thus, each patient was represented by the -dimensional feature vector, where is the total number of features and represents the order number (*index*) of a patient in the cohort of patients. The number of genetic features, is lower than the expected value of because several genes appeared in the studied population as only one or two genotype forms, i.e., the polymorphism in these genes was not found or was reduced - such cases were coded with less than three binary features. There was also one gene with three alleles and it was coded with five binary features. These cohort and feature sets were selected from a larger data set and included only those patients for whom at least of features were available and those features that were measured for at least of the patients. In the selected cohort there were still missing data; therefore, for each missing datum, its value for the nearest neighbor in the respective learning set ( or) was assigned. The phenotypic and genetic features were considered separately in the procedure of allocating the missing data. In the case of a missing phenotypic feature value, the nearest neighbour was the patient that had the most similar phenotype, whereas for a missing genetic feature value, the nearest neighbour was the patient that had the most similar genotype. The *ce.impute* procedure of *dprep* package of the *R* programming language was used for the substitution of missing values. During exploration of this database, the computations were performed in feature subspaces divided in two learning sets and. The vectors from the set described patients with high plasma *CRP* levels in the feature subspaces. Similarly, the vectors from the set described the patients with low plasma *CRP* levels. Three basic feature spaces were distinguished as follows: The *RLS* procedure of feature selection was carried out in each of the basic feature spaces (2) separately. # Results The apparent error rate (see, equation 9) and the crossvalidation error rate (see, equation 10) of the optimal linear classifier (see, equation 8) as a function of the dimension of feature subspaces in the sequence (see, equation 7) of the feature spaces, and, definition (2), are presented in. The apparent error rate (*AE*) and the cross-validation error (*CVE*) in feature subspaces of the *phenotypic* space are shown in. The lowest value of (*CVE*) equal to appeared in the feature subspace of the dimension. The features that define this subspace are presented in. The features listed in were ordered according to the absolute values (*factors*) of the components of the optimal weight vector. The features listed in was identified as the one subset of the feature subspace. This subset was not composed from the best single features. It includes the features that are correlated to *CRP* plasma levels as well as those that are not. Most of the *phenotypic features* listed in are in fact expected by medical experts to be related to inflammation but their relative importance is less clear. Whereas the list of phenotypic features in general appears to be biologically plausible, the ranking of the strength of the association as expressed by the value of the factor coefficient provides novel and potentially important insights into the links between the investigated features and the biomarker selected to represent inflammation, i.e. *CRP*. Thus, some of the identified phenotypic features in (i.e., serum fibrinogen, (low) plasma iron, serum ferritin, serum interleukin-6, and white blood cells count) are well established *biomarkers of inflammation*, whereas others are linked to *cardiovascular disease* (plasma troponin T and systolic blood pressure) which is in turn linked to inflammation. However, the negative value for the factor coefficient for systolic blood pressure is an intriguing finding which might reflect that a low blood pressure could be associated with cardiac dysfunction and heart failure, conditions which are known to be associated with inflammation. Other phenotypic features in (height, serum creatinine, plasma insulin, plasma calcium, bone mineral density, hand grip strength, S-triiodothyronine T3, plasma uric acid, plasma fetuin, truncal fat mass, body mass index, glycated hemoglobin) are linked to *nutrition* (height, serum creatinine, bone mineral density, hand grip strength, truncal fat mass and body mass index). It is well established that an abnormal nutritional status with protein-energy wasting in this patient population is strongly linked to inflammation. Several features were linked to *hormonal status* or *metabolism* (plasma insulin, plasma calcium, S-triiodothyronine T3, plasma uric acid, plasma fetuin, glycated hemoglobin); in general, relations between these features and inflammation have been described previously, but the relation with plasma calcium is not expected. Finally, high age and smoking are factors which are associated with inflammation. Feature selection from the *genetic* space is illustrated in. The learning sets and of the space are linearly separable, i.e., the apparent error *AE* is equal to zero. Moreover, the linear separability was preserved during feature reduction from to. In contrast, the lowest value of the average cross-validation error rate appeared for. It should be stressed, that the cross-validation procedure does not separate fully those feature subspaces that are linearly separable. The process of feature selection from the combined *phenotypic* and *genetic* space yielded interesting results shown in. The linear separability in the combined space was found in a large range of subspace dimensions from till. The minimal feature subspace with the linear separability of the learning sets for is composed from both *phenotypic* (i.e., clinical, anthropometric and laboratory) features and *genotypes*. The minimal value of the average cross- validation error rate was low:. This minimum value appeared at the dimension inside the linear separability zone. The optimal feature subspace with was composed from phenotypic features and genotypes. The minimal cross validation error rate in the *phenotypic* space was , and the *genetic* space it was. Combining the *phenotypic* and *genetic* factors (features) resulted in a marked reduction of the CVE error rate to. These results indicate that the *phenotypic* and *genetic* factors are not independent and play complementary roles in describing the inflammatory status of the patients in the MIA cohort. The *confusion matrices* with the mean values obtained by the *leave-one-out* procedure for the *phenotypic* and *genetic* features are presented in for a few selected feature subspaces. The lowest error was found for the subspace with dimension in agreement with the *RLS* method of feature selection. The optimal parameters and may be used to define the linear (affine) transformation of the feature vectors on the one dimensional space : The above transformation described by) was applied in designing the *scatter diagram* (*diagnostic map*) showed in. The horizontal axis (called *phenotypic fraction*) was obtained by transformation (3) applied for *phenotypic* features that constitute the optimal feature subspaces of the *phenotypic* and *genetic* space. Similarly, the vertical axis (called *genetic fraction*) of the diagram was obtained by transformation (3) applied for *genetic* features which constitute the optimal feature subspaces. The *diagnostic map* showed in can be used for diagnosis support. A new patient represented by a feature vector can be situated on the *diagnostic map* as the point determined by). If most of the nearest neighbors of the point (3) on the map belong to the set of the high *CRP* patients, then we infer that the new patient is inflamed. If most of the nearest neighbors of the point (3) on the map belong to the set of the low *CRP* patients, then we infer that the new patient is not inflamed. Similar schemes of decision support are called the *K-nearest neighbours* (*KNN*) in the *pattern recognition* or as the *Case Based Reasoning* (*CBR*) scheme. The transformation of the multidimensional feature vectors from the learning sets and and the feature vector of a currently diagnosed patient on a two- dimensional diagnostic map are aimed at obtaining a *similarity measure*. The measure allows for the determination of the similarity between the vector, representing a newly diagnosed patient, and the *precedents* (*cases*, *verified examples*) from the learning sets (clinical database). Such scheme of the decision support based on the *diagnostic maps* has been used successfully in the medical diagnosis support system *Hepar*. The performance of *RLS* selection method and *CPL* classifier applied in our study was compared to other selection methods and classifiers (see Section “Alternative methods for feature selection and classification”) using the error rate (fraction of misclassified objects from the test set), *CVE*, evaluated in the *cross-validation* (*leave-one-out*) procedure. The results are presented in. The methods *CFS-FS* and *mSVM-RFE* alongside with *RLS* select an optimal subset of features and their prediction power can be assessed using different classifiers. In contrast, *ReliefF* and *MRMR* methods are ranking procedures and od not provide any intrinsic criteria for selection of any optimal subset of features. Such criterion need to be chosen separately. For the purpose of comparison of all these methods, the optimal sets of features for *ReliefF* and *MRMR* were determined for each classifier separately as those with minimal *CVE* for the applied classifier. Thus, the optimal set (and number) of features for these two methods can vary with the choice of classifier. All the applied methods of feature selection were able to reduce the initial number of features. The highest reduction was obtained by *CFS-FS* method, which substantially outperformed in this respect four other methods. The features selected by *RLS* method provided however the lowest average cross validation error *CVE* for all three feature spaces. Especially low errors of (with standard deviation of) obtained for *RLS* method in the combined phenotypic and genotypic feature space demonstrate its good efficiency. The number of features was reduced in this case five times. For the space of genetic features, only *RLS* selection method combined with *CPL* classifier was able to obtain the low average error around, much lower that values of around or higher obtained by other selection methods and classifiers. In the case of phenotypic features, the five selection methods had a similar performance, but *RLS* method yielded slightly lower errors than the four other methods. *MRMR* provided in all three feature spaces lower error values than other methods alternative to *RLS*, especially for *SVM* and *CPL* classifiers; however, the optimal sets of features defined according to the minimal *CVE* value for *MRMR* depended on the selected classifier and this reasult would need further attention and investigation of the scope of these different optimal sets. It is also worth to notice that by allowing for higher errors (similar to those obtained for *CFS- FS* method), one can easily reduce further the number of features selected by *RLS* method as it can be seen in. Among classifiers, *SVM* and/or *CPL* yielded the lowest errors when combined with *RLS* or *CFS-FS* selection methods. *ReliefF* method worked also well with *RF* and *KNN* classifiers. The errors related to the application of *mSVM-RFE* were similar to those related to *ReliefF* and *CFS-FS* methods. The overlap between the features selected by different methods was not high. For example, among the features selected by *CFS-FS* method from the combined phenotypic and genetic features, three were shared among all three methods and seven with only one of the two other methods; five features were specific for the *CFS-FS* method. However, the problem of overlapping between features cannot be easily interpreted because many features are more or less correlated and different methods may select different features fromthose that are mutually correlated. Therefore, an additional analysis would be necessary to investigated this problem; however, this is outside the scope of this study. Among the four applied feature selection methods, *CFS-FS* was the fastest (computation time of the order of 1 sec). *ReliefF* and *MRMR* (together with the selection of optimal set) needed between a few and a few tens of minutes (depending on the applied classifier). The computation time of the *RLS* method was of the order of tens of minutes. The *mSVM-RFE* method had the computation time of about 20 hours. It should be stressed that the relatively long computation time of the *RLS*, *mSVM-RFE*, *ReliefF* and *MRMR* methods was caused mainly by repeated computation in the framework of the cross-validation procedure used by these methods. # Discussion and Conclusions Feature selection is an integral - but often implicit - component in statistical analyses. An explicit systematic feature selection process is of value for identifying features that are important for prediction, and for analysis on how these features are related, and furthermore it provides a framework for selecting a subset of relevant features for use in model construction. The most common approach for feature selection in clinical and epidemiological research is based so far on evaluation of the impact of single features. In this approach, the resulting feature subsets are composed of such features (factors) which have the strongest individual influence on the analyzed outcome (in this case inflammation). Such approach is related to the assumption about the independence of the factors. However, in a complex system, such as the living organism, these factors are more often related than not related. The role of particular factors in a living organism depends among others on (time-dependent) environmental factors and internal conditions, and on (permanent) genetic factors. An advantage of the relaxed linear separability (*RLS*) method is that it may identify directly and efficiently a subset of related features that influences the outcome and that it assesses the *combined* effect of these features as prognostic factors. This characteristic of the approach presented here is clearly visible in the dataset of phenotypic features with minimal cross validation error rate, : this set contains also features that individually do not correlate to the level of *CRP* in plasma, the clinical biomarker used here for discrimination of inflamed and non-inflamed patients. The *RLS* method of feature selection is based on the minimization of the criterion function (see, equation 5) for selected values of the cost level and repeated minimizations of the perceptron criterion function (see, equation 4) in consecutive reduced feature subspaces (see, equation 7). The *CPL* criterion function can be defined for different values of the cost level in the same feature space. Successive increasing of the parameter in the function allows to reduce increasing number of features and, as the result, the obtain the descended sequence of feature subspaces. A feasibility of feature subspaces can be evaluated on the basis of the cross validation experiment with the optimal linear classifier (see, equation 8). The parameters and of the optimal classifier are defined on the basis of repeated minimizations of the perceptron criterion function on elements of the learning sets and in subspace. The application of this method for identifying genetic and phenotypic (anthropometric, clinical and biochemical) risk factors that are associated with inflammation was implemented using a clinical database of patients with chronic kidney disease. A few important properties of the computation results obtained from this cohort can be pointed out. The results show, among others, the scale of the bias of the apparent error (*AE*) estimator (see, equation 9). The bias is illustrated as the difference between the *CVE* curve and the *AE* curve. The optimal feature subspace characterized by the lowest *CVE* error rate (see, equation 10) cannot be identified on the basis of the apparent error *AE* curve because of this bias. The minimum of the *CVE* rate is clear and narrow for the analysis of genetic data, whereas it is less marked for phenotypic and phenotypic-genetic data sets with *CVE* curves fluctuating for a wide range of feature numbers. These two cases may need an analysis of not only the feature space with minimal *CVE* but also the feature spaces with similar, albeit slightly higher *CVE* values. It is also interesting to observe that the lowest values of *CVE* occur for feature subspaces with zero apparent error rate, if genetic and phenotypic-genetic feature spaces are analyzed, whereas for phenotype feature space the minimum is within the range of subspaces with non- zero apparent error rate. Working with large medical data bases one meets often the problem of missing data, which was encountered also in our database. The patients with too many features missing and features that are measured for too low number of patients must be excluded. However, with sufficiently many data one can restore missing values by hypothetical values, and in our study this was done by the value of the nearest neighbour, separately for the phenotypic and genotypic features. Another practical problem is the overfitting of the data that happens when many features are studied for a relatively low number of patients, and this problem occurs also in our database: the two sets of patients with different inflammatory status can be linearly separated as indicated by zero apparent error for all features in the case of genetic, phenotypic and combined sets of features. Therefore, to provide a more reliable method for identifying the most predictive subset of features, the cross validation error was applied together with the *leave-one-out* procedure. These two problems preclude actually any statistical *proof* of the studied associations between features in our patient populations and the study should be considered rather as an example of *exploratory analysis* for associations that should be further investigated. We hope that our approach can supplement the current methods for analyses of such complex data which are difficult to collect, and, at the same time, represent unique and medically promising sets of data. An important characteristic of feature selection methods is the predictive power of the selected feature set, as assessed in the present study by cross validation error. The *RLS* method combined with *CPL* classifier was of similar effectiveness as some other methods if applied for phenotypic features represented mostly by continuous variables, considerably better than all other methods if applied for genetic features represented by discrete (zero - one) variables, and much better than all other methods if applied for combined phenotypic and genetic features represented by mixed type mathematical variables . Therefore, the *RLS*/*CPL* approach may be considered as a viable and promising tool for analysis of the extent by which the genetic pool, and, especially the combination of genetic variability and phenotypic characteristics of the patient, may associate with selected features in patient populations. The computational time for our method depends on two factors: 1) the number of cases and features, and 2) the repetition of calculations for the cross- validation method. The actual computing time for personal computer implementations was in the order of tens of minutes, and was longer than for some alternative methods (see Results), but all the computational times were reasonably short for the current research purpose. However, the computation time may be a limitation of the *RLS* method if applied in the future for data bases with huge amount of data and many patients, or both, and the parallelization of the code or the application of main frame computers may be necessary. Our results suggest that the considerably lower prediction errors obtained for our approach compared to those yielded by faster methods, especially for combined genetic and phenotypic data, make such extensions of the code worthwhile. The comparison between the optimal feature subspaces of the three feature spaces (*phenotypic*, *genetic*, *combined*) showed that the combined *phenotypic* and *genetic* subspace can provide a very low *CVE* error rate of. Such a low error rate opens the possibility for effective computer support of medical diagnosis on the basis of optimal linear combination of selected phenotypic and genetic features. Moreover, an individualization ofdiagnosis and/or therapy can also be considered on the basis of our methods, as, for example, the application of the diagnostic map. Nevertheless, the results of the current study should be considered as hypothesis generating and need to be confirmed in separate evaluations, if possible in another larger group of patients. # Supporting Information [^1]: The authors have the following interests. Baxter Novum - a university division within the Karolinska Institutet - is the result of a grant from Baxter Healthcare Corporation to Karolinska Institutet. Bengt Lindholm is employed by Baxter Healthcare Corporation and serves as well as head of Baxter Novum at the Karolinska Institutet. Jacek Waniewski, Juan Jesus Carrero, Abdul Rashid Qureshi and Malgorzata Debowska are or have been affiliated to Baxter Novum. There are no patents, products in development or marketed products to declare. This does not alter the authors' adherence to all the PLOS ONE policies on sharing data and materials, as detailed online in the guide for authors. [^2]: Conceived and designed the experiments: LB TŁ BL PS OH JA PB JJC ARQ KL LN MS JW. Performed the experiments: LB TŁ PS OH PB ARQ KL LN. Analyzed the data: LB TŁ JA MD JW. Contributed reagents/materials/analysis tools: LB TŁ BL PS OH JA PB JJC ARQ KL LN MS JW. Wrote the paper: LB TŁ BL PS JA JJC MD LN JW.

# Introduction *Mycobacterium tuberculosis* (*Mtb*) is a pathogen causing tuberculosis (TB), which leads to approximately 2 million deaths and 10 million new infections annually. With the emergence of multidrug-resistant (MDR) and extensively drug- resistant (XDR) TB, the current situation is more challenging. In order to search for new antibiotics and more optimized treatment strategy against *Mtb*, a comprehensive understanding of the intracellular lifestyle of this organism is urgently needed. *Mycobacterium* *marinum* (*M. marinum*), a pathogenic mycobacterium that causes disease in fish and amphibians, is genetically close to *Mtb*. Recent studies by Tobin, D.M., et al. and other researchers demonstrated that *M. marinum* was a useful model to study the pathogenesis of TB, and furthermore, to explore potential therapeutic drug target. Transcriptome analysis, which studies all the RNA transcripts during a particular physiological condition, has played a central role for detecting gene expression and transcriptional regulation. It includes investment of transcript structure, operon linkages and absolute abundance of transcripts. However, these information has not been determined on a genome-wide scale for any bacterium until 2009, Passalacqua, K.D., et al. reported the first single-nucleotide resolution view of *Bacillus anthracis* transcriptome. Then a genome-wide map of *Helicobacter pylori* transcriptional start sites (TSSs) and operons were revealed by Sharma, C.M., et al. in 2010. Through transcriptome study, they discovered hundreds TSSs within operons and opposite to genes, which indicates the complexity of gene expression in *Helicobacter pylori* by uncoupling polycistrons and antisense transcription. As for the mycobacterial transcriptome study, Arnvig, K.B., et al. revealed an extensive presence of non-coding RNA in *Mycobacterium tuberculosis*, indicating that post-transcriptional regulations might play an important role in bacterial adaptive responses to changing environments. All these fundamental studies above applied RNA-seq technology, which has led to a fast development of transcriptome study for its relatively lower cost and much more precise measurement of transcripts comparing with hybridization-based microarray assays. RNA-seq is an excellent approach for transcriptome profiling, which uses deep- sequencing technologies to directly determine the cDNA sequence. Besides its relatively lower price and precision mentioned above, RNA-seq is suited to the application of detecting transcripts which correspond to existing genomic sequence. Vera JC, et al have used RNA-seq to reveal the transcriptome of the Glanville fritillary butterfly, which makes RNA-seq especially attractive for non-model organisms with genomic sequences not known. RNA-seq can also reveal the precise location of transcription boundaries to a single-base resolution and have a relatively low background comparing with microarray. Comparison between the two techniques has been reported in several species such as *Candida* *parapsilolis*, *Candida albicans* and *Caenorhabditis elegans*. It has also been applied to discover non-coding RNA at genome wide range. All factors above make RNA-seq useful for studying complex transcriptomes. However, the transcriptome of *M. marinum* has not been revealed experimentally up to now although current tools and database (<http://mycobrowser.epfl.ch/marinolist.html>) have already enabled us to scan its genome and transcriptome from *in silico* prediction. In this research, we took advantage of RNA-seq technology to study the transcriptome of *M. marinum* in exponential and early stationary phase cultures, and investigated the functions for genes that expressed differently in these two phases. To compare the expression levels of different genes, data for CDSs were presented in the form of reads per kilobase per million reads (RPKM). In addition, we predicted potential operons and used qRT-PCR for validation. # Results ## High quality RNA preparation and sequencing profile ### Ribosome RNA (rRNA) removal before cDNA library construction Removal of signals from rRNA is a vital step of RNA-seq technology because these signals may reduce the coverage of mapped results, while decreasing the sequencing meantime. There is only one set of rRNA genes throughout the whole genome of *M. marinum* while they account for 85% of total RNA according to our previous trial (data not shown). In terms of this, we applied an rRNA removal step before cDNA library construction. As a result, 0.3 million reads mapped to rRNA were left in either log phase or early stationary phase culture, which account for 1.38% and 2.64% of total sequenced reads separately. All 5452 genes were detected within two million uniquely mapped reads in log phase sample. While in the early stationary phase sample, one million uniquely mapped reads could cover all genes in the genome (Figure S1 in File S1). A biological replicate of log phase culture was set to evaluate the reproducibility of gene expression profiles using RNA-seq technology. The Spearman correlation coefficient between two samples (r=0.867) indicates the overall pattern of relative gene expression appears quite similar between two biological replicates (Figure S2 in File S1). ### The genome-wide distribution of sequenced reads In order to investigate the transcriptome of *M. marinum* by RNA seq, we harvest bacteria cultures at OD600=0.8 (log phase) and at 4.0 (early stationary phase). RNA was extracted initially, and then followed by the rRNA removal step described above to generate the cDNA library, which was analyzed by Illumina- based sequencing eventually. Two transcriptome files, a log phase of 19.2 million reads and an early stationary phase culture of 12.5 million reads, were generated in total. Adapters of all reads were removed before next step. The generated reads have an average length of 100bp and the sequencing quality of all reads was shown in Figure S3 in File S1. Reads with sequencing error or low quality value (more than 50% of all bases in a single read have quality value \<5) were filtered and the rests were defined as clean data. Respectively, 11.9 million and 8.5 million clean reads were generated from two phases’ transcriptome files, accounting for 62.01% and 67.94% of their total reads. The result of reads mapping was shown in. ## The transcriptome of *M. marinum* ### Gene expression profile analysis To study the gene expression profile of *M. marinum*, we focused on CDSs with RPKM≥5 in either sense or antisense direction. There were 5245 genes with RPKM≥5 during log phase culture, while in early stationary phase culture, the number is 5434. Among these genes whose RPKM≥5 between two samples, 5240 were shared and 5 genes merely expressed in log phase, while 194 genes were specific during early stationary phase. All CDSs were grouped according to the functional classes of their homologs in *Mtb*. 2184 CDSs have no homologs in *Mtb*, which accounts for 40.1% of all annotated CDSs in *M. marinum*. The rest CDSs were classified into 10 categories ([http://genolist.pasteur.fr/TubercuList/help/classif- search.html](http://genolist.pasteur.fr/tuberculist/help/classif-search.html)). To emphasize, in our results there were no “category 4” which stands for “stable RNAs”. shows distribution of RPKM along with the number of genes in each category of two samples. The median RPKM of each category was shown in. Then we analyzed the RPKM distribution of each category between two samples using Wilcoxon test. “Information pathways” (category 2) were significantly up regulated (p\<0.001) in log phase culture as well as “Intermediary metabolism and respiration” (category 7, p\<0.05). In early stationary phase culture, “PE/PPE family” (category 6) and category X (No homologue with H37Rv) were significantly up regulated (p\<0.001) comparing with those in log phase culture. To verify the results, we randomly selected 24 genes (10 from up- regulated genes, 10 from down-regulated genes and 4 from unchanged genes) to perform qRT- PCR experiment. For the 10 genes which were down-regulated in RNA-seq data, eight of them were confirmed by qRT-PCR. However, only half of the up-regulated genes (five out of ten genes) could be verified by qRT-PCR, while the rest of genes belonging to this category did not have significant changes. The qRT-PCR results of unchanged genes were in accordance with the results observed in our RNA-seq data. ### Association between expression level and gene function To test for the association between expression level and gene function, 10% of CDSs with the highest RPKM from both samples were selected and studied according to their function distribution as described above. By comparing the frequency of each category across the whole genome, we found that category 0 (Virulence, detoxification, adaptation), and 2 (Information pathways) were significantly over-represented (P-value\<0.0001, Fisher’s exact test). This shares a same tendency with category 1 (Lipid metabolism), 8 (Unknown), and 10 (Conserved hypotheticals) (P-value\<0.05), which is consistent with what is expected for log phase culture. The actively growing bacteria over-represented mRNA transcripts encoding proteins involved in “lipid metabolism” and “information pathways”. Category X (No homologue with H37Rv) (P-value\<0.0001) and transcripts belonging to “intermediary metabolism and respiration” (P-value\<0.05) were under-represented in log phase culture. The early stationary phase culture showed similar results with category 0, 2 (P-value\<0.0001) and 10 (P-value\<0.05) being over-represented. These results were also reflected in, where we showed 30 CDSs with highest RPKM and the corresponding functional classes. For both log and early stationary phase culture, almost one-third of highest RPKM belong to the category “cell wall and cell processes”. Meanwhile, there were also representatives of “information pathways” and “PE/PPE family”. We also compared our results with previous transciptome study in *Mtb* by Arnvig, K.B., et al. RPKM of *M. marinum* genes and their homologues in H37Rv were compared, and spearman correlation coefficients were 0.495 and 0.326 for log phase and early stationary phase separately (Figure S4 in File S1). Then we further analyzed correlation coefficient of each functional category (Figure S5 in File S1). For the log phase culture, there were four categories with spearman correlation coefficient above 0.5 (0 Virulence, detoxification, adaptation; 2 Information pathways; 3 Cell wall and cell processes; 8 Unknown), while for the early stationary phase culture, spearman correlation coefficient of category 2 (Information pathways) and category 8 (Unknown) were above 0.5. ### Significant gene expression changes To further explore the two gene expression profiles, we looked into those genes with significant changes between two samples. Genes with at least 2-fold change and FDR (false discovery rate) less than 0.001 were defined as differentially expressed genes (See Materials and Methods). Totally, 1446 genes were up- regulated in early stationary phase culture and 847 genes were down-regulated compared with log phase culture. Then these up and down regulated genes were grouped according to their functional classes as described above. Among those genes belonging to “information pathways”, 95 were down regulated, which accounts for 88.0% of all 108 changed genes. In early stationary phase, genes of “PE/PPE family” with significant changes were dominantly up regulated (116/121). Meantime, 262 genes belonging to “intermediary metabolism and respiration” changed at least 2-fold, among which 106 were up regulated and 156 were down regulated. For these genes with no homologues in *Mtb*, 1052 have significant changes between two samples and mostly (82.8%) were up regulated. Furthermore, 8 out of 9 genes were up regulated for the category of “insertion seqs and phage”. In contrast, when we refer to the category of “virulence, detoxification and adaptation”, more genes turn out to be down-regulated. Besides, genes encoding early secreted antigenic target of 6 kDa (ESAT-6) and culture filtrate protein of 10 (CFP-10) stayed unchanged between two samples, together with most of the key components forming ESX-1 secretion system. Three genes (*MMAR_5439, MMAR_5368 and MMAR_1553*) encoding proteins known to be secreted by ESX-1 were down regulated in early stationary culture. In addition, genes constituting ESX-5 secretion system were generally up regulated in early stationary phase culture together with a gene known to encode ESX-5 dependent secreting protein MMAR-3728. ## Operons prediction and validation by qRT-PCR Operons are direct structures to determine whether genes next to each other are transcribed together, so we predicted operons on a genome-wide scale. We considered gene expression level in our sequence profile, coverage of intergenic regions (IGRs) and gene orientation as basic principles for operon prediction (See Materials and Methods). As a result, 898 operons were predicted in total throughout *M. marinum* genome from log phase culture and 978 from early stationary phase culture. In addition, there were 360 predicted operons in common between these two samples. In both samples, most predicted operons were small transcriptional units containing only 2 or 3 genes. For example, in log phase culture, 860 operons have two or three genes, accounting for 96% of the total operons predicted (76% and 20% separately). In addition, 38 operons, almost 4% of all, contain 4 genes or more in our predicted results. The biggest operon predicted in log phase culture, *MMAR_1772*--*MMAR_1777*, includes six genes. These six genes encode five *pps* (polyketide synthase) family (*ppsA- ppsE*) and *fadD26*, the former of which participates in lipid metabolism while the latter has a potential role in activation substrates for the pps polyketide synthase. Then we randomly selected 17 operons from log phase culture for validation using qRT-PCR. Of the 17 operons selected, 14 operons contain two genes and 3 operons contain three genes. Primers were designed to amplify the intergenic region between two genes using cDNA. In the case of operons having three genes, region between the first and the last gene was amplified (Figure S6 in File S1). shows six representative operons and their expression profile. And qRT-PCR results were shown in Figure S7 in File S1, from which we could find that 13 out of 17 selected operons were confirmed positive. Among these 13 operons, 11 operons have two genes and 2 operons have three genes. shows representative qRT-PCR results corresponding to predicted operon maps in. # Discussion Removal of signals from rRNA and high quality RNA preparation are crucial steps of RNA-seq technology. *M. marinum* genome has three rRNA copies, which encodes around 85% of total RNA content (data not shown). This number is close to previously reported 82% by Haas, B.J., et al., when they evaluated the effect of rRNA depletion on RNA-seq transcriptome profiles of *E. coli*. Such high abundance of rRNA not only reduces the coverage of mapped results, but also decreases the sequencing depth when initial number of sequenced reads is set. To overcome this obstacle, a physical removal step has been applied before sequencing. In this study, we used a newly developed Kit designed specifically for Gram-positive Bacteria to get rid of rRNA. As a result, less than 3% of reads were mapped to rRNA. Saturation curves were drawn to evaluate the coverage of CDSs within certain amount of reads using Number of genes detected against Number of uniquely mapped reads. Within two million uniquely mapped reads, all genes were detected in log phase sample. And the number of reads needed was one million to cover all genes for early stationary phase culture. These results benefited from the rRNA removal step before sequencing. Gene expression profile as well as the association between expression level and gene function of *M. marinum* was investigated using RNA-seq in our study. PE/PPE family represents 9.1% of the coding capacity of *M. marinum*, compared with 7.1% for *Mtb*. And it has been proposed that genes belong to PE/PPE family coevolved with the ESX loci, they underwent a specific expansion in the common progenitor of *M. marinum*, *M. tuberculosis*, and *M. ulcerans*. Previous researches showed that different PE/PPE genes are expressed when the bacilli encounter environmental changes such as adaptation to stationary phase, deprivation of oxygen, encountering macrophages. Our result showed 121 genes of PE/PPE family had significant changes and most of them (116) were up regulated in early stationary phase culture. And these significantly changed genes played important roles in the RPKM distribution differences between two samples in. It’s also of interest to look into the ESX-1 and ESX-5 secretion systems, which belong to type-VII secretion systems (T7SSs) in mycobacteria. *MMAR_3728*, a PE_PGRS protein encodes gene that was proved to be ESX-5 dependent, was up- regulated in early stationary phase culture. This could correspond to the fact that genes encoding ESX-5 components were also generally up-regulated. Considering that 116 out of 121 genes with significant changes belong to PE/PPE family were up-regulated, our result strengthens the hypothesis that ESX-5 is a specialized protein secretion system that is devoted to the transport of PE/PPE family proteins. We also found reads mapped in the antisense orientation or at IGRs accounts for 23.29% of total transcriptome in the exponential phase *M. marinum*, similar to the 28% reported for *M. tuberculosis* and 27% reported for *Helicobacter pylori*. While in the early stationary phase culture, these reads represent 40.26% of the total transcriptome, accounting for 1.7 fold (23.29%) of the exponential phase. This is possibly due to a tighter regulation of gene expression at a post-transcriptional level in early stationary phase culture, which may have a significant role in the response to stimulation of stress population heterogeneity. Interestingly, there is an unusual highly expressed gene (*MMAR_5556*) whose RPKM is almost ten times higher than the rest. By blast this gene against reference strain (H37Rv) of *Mycobacterium tuberculosis*. It turns out to be homologue (86% identical sites) with MTS2823, which is small RNA (sRNA) as previously reported by Arnvig, et al. They found that MTS2823 was the most abundant sRNA during exponential growth with a further more than six-fold increase during stationary phase. The expression values of *MMAR_5556* were in the same fashion here. This indicates that *MMAR_5556* may be misannotated as a CDS, and it is more likely to be an sRNA in *Mycobacterium* *marinum*. The expression profiles of log- and early stationary phases have been verified by qRT-PCR. However, not all changes in RNA-seq data could be confirmed by qRT-PCR, and the changes from qRT-PCR experiment were less significant compared with RNA- seq data in general. Besides the different treatments of RNA samples (an additional rRNA removal step for RNA-seq), the differences of gene expression results between these two approaches could be due to the variations between biological repeats, which is one of the limitations of this study. We also compared our RNA-seq results with former transcriptome study in *Mtb* by Arnvig, K.B., et al. The correlation coefficients were found relatively low between these data sets using Spearman Test, this could be due to several reasons. First, the strategies for sample preparations are different. Arnvig, K.B., et al used tobacco acid pyrophosphatase to enrich for small transcripts before sequencing and chose to manually pick out reads mapped to rRNA. While in our study, rRNAs were physically removed before sequencing. This could lead to a general difference on the number of reads mapped to CDs, which varies the expression profiles in turn. Second, the time point for sample collection is different. Here we selected samples of log phase and early stationary phase to analyze transcriptome profiles of *M. marinum*, which could be distinguishable from the “stationary phase” they selected. This may explain the lower correlation coefficient for stationary culture. And again, the single culture analysis used in our study could be responsible for the bias as we discussed above. Operon structure is important in prokaryotic genomes because it determines whether adjacent genes are transcribed together and this further implies whether genes in an operon are co-regulated. Thus, it’s of importance for operon prediction. RNA-seq technology has great advantages in operon prediction over hybridization-based microarray in aspects of supplying information of transcribed intergenic regions. Additionally, it also provides a far more precise measurement of levels of transcripts, as reviewed in. Price, M.N., et al. have established an operon prediction method for prokaryotes based on comparative genomic measures and distance between adjacent genes. The accuracy of their method turns out 85% and 83% for *E. coli* and *B. subtilis* separately. Later, Passalacqua, K.D., et al. have successfully predicted operons in *Bacillus anthracis* using their own sequencing data file and 10 co-operonic gene pairs were validated by RT-PCR. In the field of archaeal transcriptome study, Wurtzel, O, et al. defined more than 1000 operons in *Sulfolobus solfataricus* P2 using whole transcriptome sequencing approach. TB Database offers the “Operon Brower” function to estimate whether two genes are transcribed together in *M. tuberculosis* H37Rv as well as syntenic gene order in related species based on expression correlation information collected from 1260 microarray assays ([<u>http://genome.tbdb.org/annotation/genome/tbdb/Operon Browser.html</u>](http://genome.tbdb.org/annotation/genome/tbdb/OperonBrowser.ht ml)). However, the information of transcribed IGRs is not available there and *M. marinum* is not included in the “Operon Brower” category mentioned above. So we applied a modified method based on transciptome analysis to identify the operons of *M. marinum*. Three hundred and sixty operons were predicted in common from both samples and 13 out of randomly selected 17 operons from log phase culture were confirmed by qRT-PCR. The accuracy is 76.5%, similar to the results in *E. coli* and *B. subtilis* mentioned above. As for the 4 predicted operons that could not be validated here, it could be due to the limited sensitivity of current PCR system and certain genes may only be expressed under specific conditions. There are some known operons reported in *Mycobacterium tuberculosis*. As reviewed by Roback, et al, there are 26 verified operons, of which 22 have homologues in *M. marinum*. We checked these 22 operons in our prediction results and found 18 (82%) of them could be confirmed (including partly matched because these reported operons could be incomplete according to the paper). The known mycobacterial operons missing in our results could either be due to their low expression in our sample or the differences of operon structures between *Mtb* and *M. marinum*. Besides, we also compared our prediction results from log phase sample with 761 predicted operons at MicrobesOnline (<http://www.microbesonline.org/>), and found 472 operons in common. RNA-seq technology is also a powerful tool to search small RNA (sRNA) in bacteria. Till recent, Pellin, D, et al. predicted 1948 candidate sRNAs throughout *Mtb* genome using a bioinformatic pipeline based on the combination of RNA-seq data and comparative genomics. In this study, we also searched for potential sRNA candidates by analyzing the transcripts mapped to IGRs. As a result, 74 and 38 intergenic transcripts were found out in log phase culture and early stationary phase culture separately. All these transcripts were blasted against known sRNAs reported in, and no significant matches were found (E\<10^-5). This is probably attributed to the reason that we manually enriched cDNA with size mainly between 100 and 550bp before sequencing, so those transcripts with small size (\<100nt) could not be detected in our results. Besides, we also blasted these intergenic transcripts in Rfam database ([<u>http://rfam.sanger.ac.uk</u>](http://rfam.sanger.ac.uk)) and found 11 hits, most of which belonged to rRNA. But three hits from log phase intergenic transcripts belong to Cis-regulatory RNA (RF00059_TPP, RF01497_ALIL, and RF01066_6C), which implies a potential regulating function of intergenic transcripts. In summary, we apply RNA-seq technology to explore the transcriptome of *M. marinum*. And through our research, gene expression profiles of two time points were described, with operons being predicted on genome-wide scale meantime. These results may provide insights into the molecular pathogenesis of *M. marinum*, which may shed light on the research of pathogenic *Mtb.* # Materials and Methods Ethics Statement: N/A ## Bacterial strains, media, and growth conditions *M. marinum* strain M (ATCC BAA-535) was used for this study. *M. marinum* strains were grown in Middlebrook 7H9 broth (Difco) supplemented with 10% oleic acid-albumin-dextrose-catalase (OADC), 0.5% glycerol, and 0.05% Tween 80 at 32°C. Cultures were grown to log phase (OD600=0.8) and early stationary phase (12 hours after OD600 stabilized at 4.0). ## RNA isolation and removal of rRNA Bacteria were lysed using 0.1 mm silica beads and RNA was extracted with Trizol. The RNA quality was assessed using a Nanodrop 2000 (Thermo, Fisher) and Agilent 2100 bioanalyzer. Then RNA samples were treated with DNAse and purified using RNeasy MinElute Cleanup Kit (Qiagen) before rRNA removal step. To remove rRNA, Ribo-Zero rRNA Removal Kit (Cat. No. RZPB10106, epicenter) was applied according to manufacturers’ instructions. ## RNA-seq Construction of cDNA libraries was carried out following manufacturers’ instructions of RNA transcriptome discovery Kit (K02421-TS, Gnomegen). cDNA with range between 100bp and 550 was obtained by gel extraction followed by amplification using TruSeq PE Cluster Kit (illumina). Amplified cDNA fragments were sequenced using illumina sequencing technology (Illumina high-seq 2000). The sequencing data were submitted to the National Center for Biotechnology Information Sequence Read Archive under Accession No. SRP026200. ## Analysis of gene expression level After adaptor trimming and quality trimming, the clean reads were mapped to the *M. marinum* transcriptome using Bowtie2. Then, we used samtools and BamIndexStats.jar to calculate the gene expression level, here RPKM value from SAM file. Gene expression difference between log and early stationary phase were obtained by MARS (MA-plot-based method with Random Sampling model), a package from DEGseq. We simply defined genes with at least 2-fold change between two samples and FDR (false discovery rate) less than 0.001 as differential expressed genes. ## Quantitative real-time PCR For quantitative real-time PCR (qRT-PCR) validation experiments, 24 genes were selected (including 10 each from up- and down-regulated genes and 4 unchanged genes according to RNA-seq data). Quantitative real-time PCR was carried out by using a TaKaRa SYBR Premix Ex Taq GC kit in a 7500 real-time PCR system (Life Technologies). ## Function category of *M. marinum* CDSs In order to get the functional category of each *M. marinum* CDS, we started by identifying potential pairs of homologues between *M. marinum* and the well- annotated *M. tuberculosis* H37Rv CDSs. Using BLASTP search, a pair of homologues was defined by protein sequence similarity over 50%. The functional category for each pair of homologue was then referenced from Tuberculist (<http://tuberculist.epfl.ch/>). ## Transcripts assembly First, all clean reads were mapped to the genome sequence using Bowtie2. Then, we did the de-novo transcriptome assembly with Velvet and Oases. All de novo transcripts were then aligned to the reference genome by blat to get their genome location and compared with known transcriptome annotation using cuffcompare. Thus, all de novo transcripts were classified according to their positional relation with known transcripts, labeled by the evidence code provided by cuffcompare. The transcripts referred to those intergenic transcripts (class code: u). ## rRNA content calculation We aligned all clean reads against the rRNA sequence using BLAST with e-value cutoff 1e-10. ## Operon prediction and validate by RT-PCR Operons were predicted using the following set of rules(1). Genes in an operon had the same orientation(2). Coverage of two genes was both ≥5(3). Compare the average coverage of two genes and the coverage of their IGR, the ratio should be ≤1.5(4). When two genes had no IGR, at the same time, they were in the same orientation and both coverages were ≥5, we directly calculated the ratio of their coverage, if it was ≤1.5, these two genes were considered in an operon. RT-PCR was performed using a Takara PrimeScript RT reagent Kit with gDNA eraser. For each pair, primers were designed to amplify across the intergenic region if a contiguous transcript existed or to amplify the overlapped region. Reactions were visualized on 1% agarose gels stained with ethidium bromide. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SW CW HJZ WDT QG. Performed the experiments: SW YQZ. Analyzed the data: SW YQZ XRD GS TL. Contributed reagents/materials/analysis tools: YQZ HJZ. Wrote the manuscript: SW XRD.

# Introduction Information about a person’s income is an asset in several business-relevant domains, such as marketing and recruiting. In the context of marketing, information about a person’s financial situation can be used to target consumers with the right products, e.g. through personalized advertising on platforms, such as Facebook or Google. Which printer—ranging from \$30 to \$5,000—a company like Hewlett-Packard should advertise to a specific consumer depends partly on their financial means. While advertising a printer that is too expensive is likely to prevent consumers from buying it because they cannot afford it, advertising a product that is too cheap might prevent the consumer from buying it because they are accustomed to purchasing higher quality products. In the context of recruiting, information about a person’s current income can function as an anchor to help businesses negotiate (compare). For example, knowing that a candidate’s current salary is in the ballpark of around \$70k allows hiring managers to make a first offer that is ambitious yet plausible. At the same time, people often do not wish to share information about how well off they are financially. Among users of online dating platforms, for example, less than 10% provide information about their income. Similarly, discussing one’s salary with friends and co-workers is highly uncommon. Although the National Labor Relations Act of 1935 granted employees the right to disclose their salaries in the US, the cultural norm still considers it “inappropriate” (; p. 261). In the words of a commentator in *The Atlantic*, “Asking a coworker about pay seems akin to asking about their sex life”. This illustrates that people often consider information about their financial means private. There are good reasons why people might be reluctant to disclose income to others. Consider the two business cases above. In the context of marketing, income targeting can not only be used to help people find the right products but also to exploit their vulnerabilities. For example, expensive pay-day loans could be targeted at individuals who have the biggest need for financial support, yet the least resources to afford an expensive loan. Similarly, the knowledge of a candidate’s current salary might help the employer achieve a better negotiation deal, but it is also likely to perpetuate existing inequalities. For example, all else equal, a job candidate with a current salary of \$80k is likely to be offered a higher salary for a new role than a job candidate with a current salary of \$60k. Given that women–and many minority groups–have traditionally earned less for equal work, asking job candidates about their past salaries may reinforce the existing gender pay gap. In late 2017, several major US cities responded to this problem by passing legislation that bars hiring managers from asking job candidates about their current or past salary. However, barring potential employers from directly inquiring about a candidate’s financial situation in a job interview or hindering Facebook and Google from offering income-based targeting is unlikely to be enough to prevent businesses from obtaining this information. Income can be estimated indirectly from a job candidate’s or customer’s socio-demographic profile, their zip code, or their non-verbal cues in a job interview. In addition, both employers and marketers today have access to a powerful and easy-to-access source of information: The Web and social media. Whether it is their Facebook or Twitter accounts, their browsing logs, or the data captured by their smartphones, the digital profiles people create of themselves–both intentionally and unintentionally–are often extensive. On Facebook alone, there are more than 510,000 comments posted, 293,000 statuses updated, and 136,000 photos uploaded every minute. Research shows that these digital footprints often reveal more about their owners than first meets the eye. Computers have accurately predicted people’s intimate socio-psychological characteristics, such as their personality, political ideology, or sexual orientation, from digital footprints including Facebook profiles, Twitter messages, personal websites, or pictures posted online. A recent study suggests that the same idea holds true for the prediction of people’s financial situations: participants’ Twitter profiles accurately predicted their occupation-based income. However, this study estimated income from participants’ self-reported job title, which considerably limits the conclusions that can be drawn from these results. For example, a person who reported to be an accountant was assigned the average income of an accountant as indicated by the Annual Survey of Hours and Earnings released by the Office for National Statistics of the UK. Consequently, this occupation-based income measure is a rather rough proxy of participants’ actual income that does not capture income variations within professions (e.g. an accountant working for a major strategy consultancy is likely to have a higher income than a self- employed accountant offering advice to small and medium-sized companies). What is missing is an analysis that goes beyond group-level income estimates and tests whether digital footprints can accurately predict a person’s income on the individual level. Can we tell how much a person earns based on the traces they leave on the Internet? The current paper applies machine learning to test whether digital traces on Facebook can be used to predict a person’s self-reported income on the individual level. Although self-report measures are not without drawbacks, this much more nuanced and fine-grained assessment of income allows us to estimate the predictive power of digital footprints more accurately. We further test the generalizability of automatic income predictions by investigating new types of digital footprints: Facebook Likes and Status Updates. Although Facebook profiles merely constitute one of many digital traces of behavior (see), they have been shown to provide highly intimate information about their owners. In addition, they are particularly relevant to business contexts as they are a source that marketers and recruiters are known to consult in their decision- making process. Considering the accuracy of the predictions reported in this paper and the cost required to create them (e.g. with regards to data science capabilities and data access), we return to the two use cases of marketing and recruiting in the discussion to estimate the likelihood that income predictions from digital footprints would be put into practice for each of them. # Method ## Participants We recruited a gender and income representative sample of US Facebook users through a paid Qualtrics panel in early 2015 (data was collected in line with Qualtrics’ and Facebook’s terms of service). A total number of 7,180 participants responded to the survey. We excluded all participants who (i) answered the attention check question incorrectly (N = 2,461), (ii) had used a Facebook ID that appeared in the dataset more than once (N = 516), or (iii) who had fewer than 10 Likes or 500 words in their Status Updates (N = 1,580). This left us with a total of 2,623 participants. Similar to the demographic distribution of Facebook users in the US, the average age of participants in our sample was 35.9 years (SD = 11.0), and 60% of participants were women, suggesting that our sample remained largely representative despite the relatively high attrition (see income distribution). All participants were fully informed of the data they would be sharing when consenting to participate. The study was approved by the University of Cambridge Department of Psychology Ethics Committee. ## Measures and procedure Participants were asked to complete a 5-minute long questionnaire, which included questions about their zip code, age, gender, ethnicity, educational attainment, occupation industry, and personality as measured by the BFI-10, an established short measure of the Five Factor Model of personality. In addition, they were asked about their annual income (“What was your total income in USD, before taxes and other deductions from January 1st 2014 to December 31st 2014? Please include money from jobs or self-employment, net income from business, farm or rent, pensions, dividends, welfare, social security payments, and any other money income that you received during that time”). displays the density distribution of participants’ self-reported income, which is compared to the density distribution of the US population in 2014 reported by the US Census Bureau. Due to the nature of our sample and the fact that we had to exclude a considerable number of participants (see section above), our income distribution deviates from the population distribution in that it is skewed towards low- income participants, even though it follows the same pattern. At the beginning of the questionnaire participants were asked to provide their informed consent and to grant us access to their Facebook Likes (official Page Likes, i.e. excluding any other Likes such as friends’ photos or status updates) and Status Updates. On average, people included in our final dataset had 822 Likes (SD = 1,470) and 18,377 words in their Status Updates (SD = 30,132). Compared to previous publications using Facebook data (e.g. obtained from databases such as myPersonality.org which contains data from 2007 to 2012,), the number of Likes and words in status updates is significantly higher. This is expected, however, as Likes and Status Updates accumulate over time and have seen an exponential growth in recent years. A person, who back in 2012 had 250 Likes, for example, is likely to have liked a substantial number of additional Pages over the following years, while deleting Likes remains rare. ## Prediction models We developed three behavior-based prediction models based on: (i) Likes, (ii) language from Status Updates, and (iii) the combination of Likes and Status Updates. Following the procedure described by for language-based assessment and an adapted version of for Likes-based assessment, we built the prediction models in three steps. In the first step, we turned Likes and Status Updates data into two large matrices—users by Likes and users by words—with users in rows and Likes/words in columns. For the user-Likes matrix, entries were set to 1 if the participant was associated with the Like and to 0 if the participant had no association with the Like. For the user-Status-Update matrix, we first extracted all words and phrases (up to 3 word sequences; e.g. “hit me up”) using a social media tokenizer, a computer program that automatically identifies typical words as well as emoticons, hashtags (e.g. “#election2016”), and slang (e.g. “g’nite”, “smh”–shaking my head). Entries were subsequently set to represent the relative frequency of word use, as well as binary 0/1 indicating whether the word was used at all. Function words, such as articles or pronouns, were retained intentionally as they have been found to significantly relate to individual differences. For example, “the” and other articles have been found to be used more frequently by those with higher intelligence or education while self- references, such as “me,” have been found more frequent among individuals who are depressed or anxious. In the second step, we applied singular value decomposition (SVD;)–a commonly used dimensionality reduction method–to reduce the two matrices to 300 dimensions for Likes and 700 dimensions for Status Updates (dimension based on past work;). Because we have nearly 100,000 dimensions as initial input, SVD along with a Ridge penalty (discussed next), is used to avoid overfitting. To increase the robustness of our SVD solution, we developed the SVD matrices on a much larger sample of Facebook users that was available through the myPersonality dataset. This step allowed us to deal with the sparsity of both Likes and Status Updates data in our sample (i.e. there were more than 855,000 unique Likes, and more than 32.8 million unique words and phrases, many of which were associated only with a handful of participants). In the third step, we ran ridge regression models (LASSO;) to predict logged income from the two low dimensional matrices obtained in step two. Ridge regression is a widely-used machine learning algorithm that applies an L2 penalty to avoid overfitting when one has many independent variables. To avoid overfitting in our model validation, we evaluated the accuracy of our model using separate training samples to fit our models, and hold-out samples to test their predictive power. Such cross-validation procedures including hold-out samples are the gold standard in predictive modeling. They address the problem that traditional in-sample methods (e.g. variance explained as expressed by R²) are highly biased when there is a large number of predictors. In fact, with tens of thousands of predictor variables, one would likely achieve near-perfect prediction accuracy in-sample, but this accuracy is unlikely to generalize to new samples. Specifically, we used a standard 10-fold cross-validation procedure. First, we randomly divided the dataset into ten equal-sized partitions (folds). In 10 iterations, we subsequently trained the LASSO model using data from nine of the folds (training set), and then tested for accuracy over the remaining hold-out fold (testing set). Once the out-of-sample predictions were made for all of the ten folds, we estimated the accuracy of our model by calculating Pearson’s Product-Moment correlation between the actual and predicted income values. We further tested the predictive power of behavior-based predictions by establishing its incremental predictive validity above and beyond the following baseline socio-demographic and psychological comparisons: (i) demographics (gender, age, and ethnicity), (ii) education (8-point scale ranging from “Less than high school” to “Doctorate degree”), (iii) industry (24 industries adapted from the North American Industry Classification System), (iv) zip code income, (v) personality, and (vi) all variables from (i) to (v) combined. We ran significance tests to formally establish whether the added predictive power significantly improved the predictive accuracy of our model. # Results displays the Pearson Product-Moment correlations for all the models we calculated. Likes alone predicted income with an accuracy of r =.27, and Status Updates predicted it with an accuracy of r =.41. Combining Status Updates and Likes further increased the predictive accuracy of all models to an accuracy of r *=*.43 for the model including both Likes and Status Updates. While far from being perfectly accurate, this correlation is considered a strong effect-size in the context of behavioral/psychological factors and is higher than the accuracy of similar approaches used to predict personality factors. Both the language-based and Likes-based models provided incremental predictive accuracy over the socio-demographic and psychological baseline variables. Adding Facebook Likes and Status Updates to the socio-demographic controls increased the variance explained between 10% (when compared to education) and 16% (when compared to personality). Even when taking the most comprehensive baseline of all socio-demographic and personality variables–which together explained 18% of the variance–adding Facebook Likes and Status Updates increased variance explained by 6% to a total of 24% (r = 0.49). Notably, the incremental accuracy was mostly driven by Status Updates, with Facebook Likes adding none or little accuracy to the baseline models. displays Pearson Product-Moment correlations between the predicted and actual income values obtained for the Likes and Status Updates-based prediction models (blue bars). To provide a baseline comparison for evaluating the predictive power of our models, we also included the accuracies obtained from predictions made with psychological and socio-demographic baseline comparisons (and orange and red bars). Finally, we included the accuracy of the model combining all socio-demographic variables, psychological variables, and Facebook data (purple bars). To develop a better understanding of the relationships between Facebook profiles and income, we looked at the individual words (or phrases) and Likes that were most strongly associated with low and high income. shows the 10 Likes that displayed the highest positive correlations with income (right) and the Likes that displayed the highest negative correlations with income (left), when controlling for age and gender. reveals that the Likes of high and low income individuals are markedly different: While the Likes associated with high income refer to concrete expensive travel and retail brands (e.g. “The Cosmopolitan of Las Vegas,” “Paula’s Choice,” or “Janie and Jack”), the Likes associated with low income refer to luxury but in a highly abstract and generalized way (e.g. “Amazing Things”) and often contain entire phrases (e.g. “Dentist Stop Talking to Me, I Cant Talk Your Hand is in My Mouth,” or “if i text a person in the same room as me, i stare at them 'til they get it”). displays two differential word clouds with the words most strongly associated with high income (top) and low income (bottom), controlling for age and gender. All words displayed are significantly related to income. Similar to the Likes, the differences in topics are remarkable and paint a clear and poignant picture. High income individuals talk about vacations (e.g. vacation, flight, beach, Vegas, airport) and pleasant activities that usually require spending a considerable amount of money (e.g. shopping, celebrating), express positive emotions (e.g. excited to, great), and use future-oriented words and phrases (e.g. looking forward to, afterwards). Low income individuals, on the other hand, are highly self-focused (e.g. I need, I can, I got, me), use colloquial language (e.g. idk, cuz), express negative feelings (e.g. hurt, hate, bored), and use more swear words and emoticons. # Discussion Our findings suggest that a person’s income can be predicted from their Facebook profile data. By obtaining access to individuals’ Facebook Likes and Status Updates (with their informed consent) along with a self-report of their person income, we were able to automatically predict personal income with relatively high accuracy (r = 0.43) based on the language and likes. The accuracy level was comparable to what we could achieve using a person’s comprehensive socio- demographic profile including age, gender, ethnicity, zip code level income, education, and industry (r = 0.42). Notably, integrating Facebook Likes and Status Updates with those socio-demographic as well as personality added incremental predictive accuracy to our baseline models (ΔR² between 6% and 16%). As expected, the added value was strongest when considering relatively generic socio-demographic variables, such as age, gender, and ethnicity, but became marginal when considering all socio-demographic and psychological control variables at the same time. Even though around 75% of the variance in income remained unexplained by our models, the prediction accuracy reported in this paper is high enough to be useful in contexts such as targeted marketing that do not require an extremely high degree of accuracy at the individual level. Yet, they do not seem to be accurate enough to establish a person’s income to the extent that it would be useful in the negotiation of starting salaries. Although we could distinguish between a person with a very low income and one with a very high income, our model is not sufficiently accurate to make fine-grained distinctions between a person making \$70k or \$75k a year. However, there is reason to believe that the accuracy of prediction models like ours could be further increased by using more robust and reliable income measures (e.g. actual instead of self-reported salary), as well as a combination of additional external (e.g. their LinkedIn, Facebook and Twitter profiles) and internal (e.g. their CV and application materials) data sources. Importantly, the ability to build predictive models such as the one we introduced in this paper is not limited to a small number of researchers or companies that control huge datasets of personal information (e.g. Facebook or Google). Although it is not trivial to collect the necessary data and generate the predictive models that turn it into insights about a person’s income, our approach could be replicated by any company that receives access to both consumers’ digital footprints and their income information and that has reasonable data science capabilities. This barrier makes it unlikely for small to medium-sized businesses to integrate such predictions into their hiring decisions, as the costs do not outweigh the benefits. For example, employers could receive comparable insights into a candidate’s current income by simply asking them about their “expected salary”–an intervention that comes at no cost. For targeted marketing, however, there is no simple or obvious alternative, making it more likely that larger retail companies will engage in building similar technologies to improve their sales strategy In addition, there is a growing number of companies entering the market of predicting consumer insights from digital footprints (e.g. IBM Watson personality insights). The same way that fintech companies are already using people’s digital footprints to determine their credit scores, other companies might soon offer income predictions from a broad variety of digital sources as part of their portfolio, thereby making it easy for companies to benefit from such predictions without having to train and deploy their own machine learning models. As predictive technologies like the one described in this paper become more prevalent and easily accessible, it becomes paramount to carefully consider and publicly discuss their potential opportunities and challenges. On the one hand, the ability to predict a person’s income unobtrusively from their digital footprint could offer opportunities that help both businesses and consumers if deployed in the right way. As we have outlined in the introduction, businesses might tailor their services and offers to the purchasing power of the individual consumer. Such personalization strategies directly benefit businesses by allowing them to target more accurately, and they could also benefit the consumer by suggesting products that fall into a realistic price (and quality) range, thus alleviating the risk of overspending. In addition, the information about a person’s income–or more broadly, their socio-economic status–could be used to increase the fairness of hiring decisions. As more and more companies transfer to an automatic assessment and screening of job candidates that is driven by computer-based algorithmic choices, it becomes paramount to ensure that these algorithms do not discriminate against specific subpopulations (e.g. women, minorities, low-income individuals,). Because algorithms are trained on historical data, they are prone to formalizing and perpetuating existing biases. For example, if women have been historically hired at a lower rate than men, then the algorithm would conclude that being “male” is a strong predictor of being “successful.” In order to actively overcome such biases, it is necessary to have the socio-demographic information about candidates that one aims to reduce biases against. That is, businesses cannot account for biases against traditionally underprivileged groups if they do not know whether a job candidate is a women or a man, or whether the candidate comes from a low or high socio- economic background. While managers can no longer legally–and should not ethically–ask their applicants about their previous income or socio-economic standing directly, estimating this information and using it in the context of auditing and de-biasing prediction models and decisions indirectly (without revealing it to the manager), could increase the fairness of hiring processes as long as the data processing was strictly controlled so that it does not influence the manager. On the other hand, together with the larger body of literature on predicting highly intimate characteristics from digital footprints (e.g.), our findings demonstrate the need for ethical guidelines for predictive technologies, as well as regulations on a policy level. Such guidelines and regulations are needed to protect people’s privacy and to ensure that these new technologies are used in the best interest of individuals and society at large. In most parts of the world–including the US–our approach to predicting a person’s income from his or her Facebook profile could be implemented without the knowledge of users. Although some of the more progressive data protection regulations, such as the European General Data Protection Regulation (GDPR), consider the implications of “profiling” and regulate its application (e.g. by requiring transparency and consent), they do not prohibit them entirely. There remains a substantial degree of freedom in how businesses implement predictive technologies like the one described in this paper; this makes an ongoing public and political discussion on the ethical challenges of those technologies paramount. Taken together, the current research showed that Facebook Likes and Status updates not only predict self-reported income with the same degree of accuracy as standard socio-economic variables, but they also added incremental predictive power. The accuracy of these predictions is high enough for low stakes applications such as targeted marketing where outcomes are usually measured at the group level. However, but they are–in their current form–high enough for applications such as recruiting and salary negotiations, where outcomes are measured at the individual level and where there is a simple alternative to making a prediction. While providing opportunities for businesses to improve their services and aim for algorithmic fairness in their decisions, our findings also raise several critical questions regarding privacy and data protection. Progressive data protection regulations such as the GDPR alleviate those concerns to some extent, but they may not fully prevent the technology presented in this paper from being used in a way that is potentially harmful to people. For example, job candidates might feel the pressure to check a box saying they agree to their data being processed, even though they consider it invasive. It is therefore critical that businesses adhere to ethical principles and carefully consider both the potential costs and benefits before implementing income predictions from digital footprints. In addition, consumers need to be made aware of the possibilities that predictive technologies hold, and the fact that social media data (and other digital footprints) can reveal a lot more about them than they might think. Together with giving them the power to decide who uses their data, and for what purposes, this will allow them to decide for themselves whether they are willing to forsake their privacy in order to benefit from better services, or whether they consider the risks of potential abuse as being too high. [^1]: The authors have declared that no competing interests exist.

# 1. Introduction MicroRNAs (miRNAs) are a class of short (approximately 21-nucleotide), non- coding RNAs which can regulate gene expression at the post-transcriptional level. Upon loading into the Argonaute (Ago) proteins, which are the catalytic components of the RNA-induced silencing complex (RISC), miRNAs interact with the target mRNAs. These interactions result in mRNA repression, destabilization and thus prevent the target genes from producing functional peptides and proteins. There are more than 2,000 annotated humans miRNAs deposited in miRBase (<http://www.mirbase.org/>). Because of the extensive targets, miRNAs are involved in a variety of cellular pathways, from development to pluripotency to oncogenesis and so on. The induction of gene silencing and repression via miRNAs typically requires complementary base pairing between specific regions of the miRNA and its target mRNA. Canonical miRNA-mRNA interactions require the target site complementarity to the seed sequence, nucleotides 2–8, of the miRNA. However, there are many examples of functional miRNA-mRNA interactions that occur without perfect seed pairing, indicating the non-canonical interactions. There are multiple experimental techniques available to identify targets of miRNAs. Through miRNA-overexpression studies combined by gene-expression analysis, a large number of miRNA:target-gene interactions have been identified. Due to the complexity of regulation network, many affected genes may not be the direct targets of miRNAs. Recently, other methods are developed to investigate the direct interaction of miRNA and target sites, such as CLIP-seq (cross- linking immunoprecipitation with sequencing), PAR-CLIP (photoactivatable- ribonucleoside-enhanced CLIP), iCLIP (individual-nucleotide resolution CLIP) etc. All these methods can catch the miRNA:target hybrids in the RISC complex and identify lots of miRNA:target-site interactions through high-throughput sequencing. In addition to experiment techniques, there are lots of computational prediction algorithms developed to predict miRNA targets such as TargetScan, miRTarget3, and miRWalk etc. Most methods are mainly based on attributes of the mRNA sequence itself, thermodynamic stability of the miRNA-mRNA duplex, evolutionary conservation, or statistical inference based machine learning. Recently, some deep learning (DL)-based approaches were developed to predict miRNA target sites and/or transcripts. But, these methods also depended heavily on the hand-crafted features of the miRNA-target duplex. Besides the deep learning methods to predict the mirna-targets, there are other method to be used for the similar task. Also, it has been shown that the finding such regulatory relationship can be used for cancer subtyping. Because of not fully understood rules that govern miRNAs targeting process and different training datasets for different algorithms, there is limited overlap between the targets that are predicted by various programs. So, it is still a challenge to develop more reliable computation methods based on more accurate miRNA:target datasets. Deep learning (DL), which can autonomously learn and identify patterns from raw data, has been shown to be an effective method for classification tasks in domains with complex feature representation. Convolutional neural networks (CNNs) are characteristic of convolution layers which can automatically extract features from input datasets and have showed great success in image recognition. The convolution layer, consisting of a combination of linear convolution operation and nonlinear activation function, is usually followed by a pooling layer which provides a typical down-sampling operation such as max-pooling. Through using multiple convolution and pooling layers, CNN models can learn the patterns from low to high levels in the training dataset. Lots of CNN architectures have been developed to address biological problems and showed to be successful. Here, we designed a multilayer convolutional neural network to predict the target sites of miRNAs without need of feature extraction in advance. To the best of our knowledge, we first applied CNN to extract complex features from raw sequences of miRNA:target-site duplex, which were used for prediction of miRNA targets. Our method can also be used to predict the target gene of miRNAs through scanning the full length of gene transcripts. # 2. Materials and methods ## 2.1 Datasets preparation ### MiRNA:Target-site datasets To get a more reliable dataset, only the experiment-validated direct interactions of miRNAs and targets were collected. The positive miRNA:target data were downloaded from three sources: 1. Study of miRNA interactome by CLASH (crosslinking, ligation, and sequencing of hybrids) in HEK293 cells. 2. Study of miRNAs to their target sites in *C*. elegans using modified CLIP methodology and re-analysis of published mammalian AGO-CLIP data for miRNA-chimeras yielded \~17,000 miRNA:target-site interactions. 3. miRNA:target-site interaction data in MirTarBase with strong experimental evidence (immunoblot, luciferase reporter assay, qRT-PCR). The miRNA sequences were retrieved from miRBase. All the data were merged followed by removing the duplicates of miRNA:target-site sequences and the concatenated miRNA:target chimeras longer than 110 nt (nucleotides). Finally, we got the positive dataset with 42,085 positive interactions with the labels of “1”. There are more than 557 thousands of miRNA:target-gene interactions deposited in MirTarBase (Release 7.0) with strong or weak evidence. To generate the negative dataset, we made pseudo combinations of miRNA and gene avoiding the miRNA: gene pairs in MirTarBase. The cDNA sequences of genes in the pseudo combinations were retrieved from human genome (GRCh38) and mouse genome (GRCm38) using SAMtools. The pseudo target sites were taken randomly from 3’UTR (untranslated region), 5’UTR or ORF (open reading frame) of genes with a proportion of 7:2:1 and the overall length of each pseudo miRNA:target chimera was set to be 110 nt (nucleotides). Altogether, we generated the negative dataset containing 94,764 human and 22,531 mouse miRNA:pseudo-target site interactions with the labels of “0”. The negative and positive datasets were merged together and separated randomly into train (149,439), validation (4,941) and test (5,000) datasets. In the 10-fold cross validation (CV) experiments, the merged dataset was divided into 10 segments with about the same number of miRNA:target chimeras (15 938). In each experiment, nine segments were used for training while the remaining one was used for evaluating the performance of the model. ### MiRNA:Target-gene datasets To predict the target genes of miRNAs, the positive and negative experiment- validated miRNA:target-gene pairs were downloaded from MirTarBase and Diana TarBase respectively. Next, we strictly selected the most convinced data. The positive dataset was composed of all the interactions of miRNAs and genes with strong evidence while the negative dataset only contained those with direct evidence of no interaction. The data appearing both in the positive and negative dataset were removed. The mature miRNA sequences were downloaded from miRBase database and the transcripts sequences of genes were retrieved from human cDNA (complementary DNA) annotation file (Homo_sapiens.GRCh38.cdna.all.fa). Since there are many transcripts for one gene, we only use the longest transcript and the shortest transcripts to represent positive and negative interaction genes, respectively. All the interactions of miRNAs and genes which we failed to retrieve the sequences were removed from the datasets. The final experiment- validated positive dataset contains 7815 items and negative dataset contains 281 items. ## 2.2 Sequence padding and vectorization Since the contextual sequence around the target site in the mRNA has great impact on the interaction, we considered both the direct interaction sequences and the contextual sequences in mRNAs for our mode. So, the miRNA:target-site chimeras contain the contextual sequences around the target sites, which keeps more information for the deep learning. Upon inspecting the positive miRNA:target-site chimeras, we selected the positive data with the length of no more than 110. Since different positive miRNA: target-site chimeras had different lengths, we padded each positive sequence into the length of 110 for batch learning in the next model training process. The padded sequences were randomly generated avoiding 4 continuous pairing bases with corresponding miRNAs. Next, we generated pseudo miRNA: target-site chimeras with each length of 110 nt to be the negative samples. To encode RNA sequences, the “one-hot” encoding was used for each base ("A":\[1,0,0,0\],"U":\[0,1,0,0\],"G":\[0,0,1,0\],"C":\[0,0,0,1\]). After encoding, each miRNA:target-site chimera can be represented by a 110 x 4 tensor, which was used in our supervised deep learning. ## 2.3 Architecture of the proposed multi-layer convolutional neural network The designed architecture and parameters of the deep convolutional neural network (CNN) were showed in. In the model, the input sequences were first convolved by sixteen kernels with the size of 2 over a single spatial dimension (filters: 16, kernel size: 2) followed by max pooling. Second, the output tensors flowed through the second convolution layer (filters: 32, kernel size: 3) and max-pooling layer. And then there were the third convolution layer (filters: 64, kernel size: 4) and max-pooling layer, followed by the last convolution (filters: 128, kernel size: 5) and max-pooling layer. All the max- pooling layers took the maximum value with the size of 2. After multi-layer convolution and max-pooling operations, all the extracted features were passed to a fully-connected layer (units: 128). The last layer is only one unit obtained by the Sigmoid activation function on the probability of the miRNA:target-site chimeras. The activation functions for other layers, if needed, are chosen to be “relu”. For generalization of our model, we added two dropout layers before fully-connected layers as showed in as well as L2 regularization on the parameters of the fully-connected layer with 128 units. The total number of parameters was 166,001, mostly due to the fully-connected layers. ## 2.4 Model training and evaluation The loss function we employed was the cross entropy between the predicted values and the actual labels (“1” or “0”). The Adam optimizer was applied to learn the network weights in a back-propagation fashion. During the training process, the generalization error was also monitored using validation dataset. The training process was not stopped until the loss on evaluation dataset did not decrease any more. After training, the learned parameters as well as the model structure were stored. Using the trained model, we calculated the classifier performance on the test dataset in terms of sensitivity, specificity, F1-Score, Matthews Correlation Coefficient and accuracy. (TP: true positive, TN: true negative, FP: false positive, FN: false negative) Sensitivity: $$Sen. = \frac{TP}{TP + FN}$$ Specificity: $$Spe. = \frac{TN}{TN + FP}$$ F1-Score: $$F1 = \frac{2*TP}{2*TP + FP + FN}$$ Matthews Correlation Coefficient (MCC): $$MCC = \frac{TP*TN - FP*FN}{\sqrt{\left( {TP + FN} \right)*\left( {TN + FP} \right)*\left( {TN + FN} \right)*\left( {TP + FP} \right)}}$$ Accuracy: $$Acc. = \frac{TP + TN}{TP + TN + FP + FN}$$ Also, we plotted the receiver-operating characteristic curve (ROC) with the calculated area under the ROC curve (AUC). With decreasing thresholds on the decision function used, corresponding false positive rates (fpr) and true positive rates (tpr) were computed. ROC curve was drawn based on a series of fpr and tpr. ## 2.5 Code implementation and availability Because only a single candidate site to be classified as positive is required for long target sequences, the model is particularly sensitive to false positive. Hence, we filtered out the candidate sites of mRNA using RNAhybrid with the minimum free energy (MFE) of miRNA:target duplexes \< = -20 kcal/mol. The implemented cnnMirTarget was based on the well-trained CNN model using the training dataset and can be used to predict whether miRNAs can target candidate sequences or even full length of mRNAs. The cnnMirTarget takes the inputs of miRNAs and the target sequences as parameters. If the target sequence provided is less than 110, it will be padded into 110 with the same strategy described in the methods 2.2. For long mRNA sequences, all the candidate sites are filtered out using RNAhybrid followed by prediction using the trained CNN model. The final prediction result is based on the maximum prediction value of all the candidate sites. If the output value is greater than the threshold of 0.5, the prediction result is “True”, and otherwise “False”. The cnnMirTarget source code, which is written in python with keras library, is freely available through GitHub (<https://github.com/zhengxueming/cnnMirTarget>). ## 2.6 Prediction performance comparison with other methods We next tested the performance of the cnnMirTarget and compared with that of other state-of-the-art target prediction software tools on the experiment- validated positive and negative miRNA:target-gene datasets (see section 2.1). All the predicted datasets of miRTarget3 (miRDB v5.0), metaMIR and TargetScan (v7.0) were downloaded from each website and searched for each miRNA:target-gene pair in the positive and negative interactions. Since the miRTarget3 and TargetScan had no negative datasets available, the miRNA:target-gene pairs not found in their interaction datasets were considered as negative. Based on the searched results, the performance was evaluated by sensitivity, specificity, F1-Score, MCC and accuracy. # 3 Results ## 3.1 Performance of our model on the test miRNA:Target-site dataset For the training/evaluation/test dataset splitting, the model was trained on the training dataset with enough epochs, evaluated on the evaluation dataset and finally the performance was tested on the test dataset. In the 10-fold CV, we trained our model with the nine folds while the remaining one fold was used for testing the performance in each time. For conciseness, we showed the average performance along with standard error (SE) for the 10-fold CV experiments. As shown, we got similar values of sensitivity (column 2), specificity (column 3), F1-score (column 4), MCC (column 5) and accuracy (column 6) for different dataset splitting strategies in our model. In the training/evaluation/test dataset splitting, the overall accuracy of prediction is 97.36% and all the other values are more than 93%, indicating high generalization performance of our trained model. To further evaluate our model, we plotted the ROC curve of prediction on the test dataset. As shown in, the AUC of ROC curve is 99.50%, indicating high performance for recognizing the target sequences of the miRNAs. ## 3.2 Comparison with other methods Since our CNN model showed high performance to predict the target sites of miRNAs, we wanted to test the performance and compare with other methods on predicting target genes of miRNAs. Due to the long length of mRNAs, there was great possibility of false positive as described in the methods 2.5. So, we introduced a filter step to find the candidate sites in the cnnMirTarget. To get the most reliable interaction dataset of miRNAs and genes, we carefully selected the experiment-validated positive and negative datasets as described in methods 2.1. Because of the small number of negative data available, the merged dataset is highly imbalanced. To classify targets or non-targets of miRNAs, the threshold we used here is 0.5. The prediction performance for miRTarget3, metaMIR, TargetScan7 as well as our cnnMirTarget was evaluated on the experiment-validated positive and negative miRNA:target-gene datasets (described in section 2.1). The prediction results for miRTarget3, metaMIR, TargetScan7 as well as our cnnMirTarget were showed in the. The true positive and true negative interactions predicted by our cnnMirTarget are 5,179 and 162 respectively. Compared with miRTarget3 and TargetScan7, cnnMirTarget has better performance on the positive dataset (Column 2). On the other hand, cnnMirTarget has better performance on the negative dataset than metaMIR (Column 3). Although miRTarget3 and TargetScan7 have high specificity, they missed many true miRNA:gene interactions (low sensitivity: 12.80% and 18.62%, respectively). And metaMIR has comparably higher sensitivity, but with many false positives (low specificity: 25.62%). Overall, our algorithm outperforms other methods for predicting miRNA-mRNA interactions indicated by F1-Score, MCC as well as accuracy. # 4. Discussion Different from traditional machine learning algorithms, deep learning can automatically extract patterns from canonical and non-canonical pairing between the miRNAs and its targets. In this study, we used four layers of convolution followed by max-pooling operations in our model, which extract features hierarchically from miRNA:target-site chimeras. By using two dropout layers and the L2 regularization in the first dense layer, our model showed little generalization error on the evaluation dataset during the training process. The high performance of trained model on test dataset showed the learned features can be used to predict the interaction sites of miRNAs with high accuracy. In summary, our trained CNN-based model can predict the interaction of miRNAs:target-sites with high performance. In machine learning especially deep learning, it is vital important to collect large amounts of reliable data. For miRNA targets prediction, different methods used different datasets to train their models. So, there is little overlap among them, which makes the prediction of miRNA targets difficult and challenging. In this study, we only chose the experiment-validated data mainly from high- throughput sequencing. Different from other rule-based machine learning methods, the convolutional neural network need equal amount of positive and negative datasets to train the model. Since it is harder to collect the data of negative miRNA:target-site interactions, we generated a large negative dataset as described in methods. Although our methods showed great performance on the test dataset to predict target sites, the accuracy to predict target genes of miRNAs was dramatically decreased. Since one gene can express different transcripts under different conditions, an experiment-validated miRNA:target-gene interaction does not mean that the miRNA can target any transcript of the gene. In fact, we found much controversial data appearing both in positive and negative experiment-validated miRNA:target-gene interaction dataset. Different from predicting the target sites of miRNAs, there are harder to predict the interactions of miRNAs and genes because of many other factors involved the process. For an example, the secondary structure of mRNA may affect the accessibility of miRNA. Moreover, the stability of miRNA:target hybrids has great importance on the interactions. Although we filtered out the candidate sites using RNAhybrid, the minimum free energy (MFE) should be carefully selected. Also, there exist many weak binding sites in some mRNA. So, the synergistic effects should be considered. Furthermore, there are complicated molecular interaction networks in the cell, which affect the interactions of miRNAs and target mRNAs. The binding sites on mRNA can be occupied by other proteins or RNAs and the miRNA:mRNA interactions may be eliminated by circRNAs. Also, Ago protein binding sites in mRNA may bring mRNAs to miRNAs, which leads to interactions. In brief, there are many factors that can affect the interaction of miRNAs and mRNAs, which should be taken into consideration to improve our model in the future. So far, there are tens of computational prediction algorithms to predict miRNAs targets. But, owing to the great difference of prediction results for different predictors, the prediction of miRNA target is still a challenge. Here, we designed the CNN model to learn both canonical and non-canonical interactions automatically from experiment validated miRNA:target-sites chimeras. The results showed that our model is very successful to predict the target sites of miRNA, but there are great room to improve the performance on predicting miRNA:target- gene interaction. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Canine hip dysplasia (CHD) is a common trait in most dog breeds. This orthopedic condition causes instability and subluxation of the hip with secondary signs of osteoarthritis and clinical signs of lameness. Breed prevalences vary widely from 1% to 75%. In German Shepherd Dogs, prevalence of CHD is estimated at 35%. There is strong evidence in support of a genetic predisposition to CHD in German Shepherd Dogs and many other dog breeds. Heritability estimates for German Shepherd Dogs from different European countries were at h²  =  0.20 to 0.35 –. Results from radiographic screenings for CHD of 48,367 German Shepherd Dogs born in 2001–2007 were used for estimation of heritabilities in threshold and mixed linear-threshold models. In this sample of German Shepherd Dogs, heritability was 0.25 for CHD in the linear model and 0.19–0.27 for binary trait definitions regarding only borderline as CHD-affected and mild to severe CHD cases as CHD-affected versus CHD-free dogs. In a study of 13,124 Australian German Shepherd Dogs born between 1976 and 2005, the heritability of the summed phenotype constructed from nine ordinally-scored British Veterinary Association Hip Traits was 0.30. Linear models are commonly applied for genetic analyses of CHD as this methodology most correctly reflects the underlying nature of the data. Complex segregation analyses demonstrated involvement of a major gene for the German Shepherd Dog. Genome-wide linkage studies showed nine genome-wide significant quantitative trait loci (QTL) for CHD in German Shepherd Dogs. A linkage study in a Labrador Retriever-Greyhound crossbred family revealed twelve dog chromosomes (CFA) with chromosome-wide significant markers for CHD. In Portuguese Water Dogs, QTL for signs of CHD were demonstrated on CFA1 and 3. A genome-wide association study (GWAS) for CHD and osteoarthritis (OA) across several dog breeds including Labrador Retriever-Greyhound crosses identified four CHD-associated and two OA-associated SNPs. The CHD-associated SNPs were located on CFA3, 11 and 30, but not within QTL of the Labrador Retriever- Greyhound crossbred linkage study. In 174 Bernese Mountain Dogs, two different CHD-regions were identified on CFA14. A third CHD-associated region was located on CFA37. A Dutch study on 48 CHD-affected and 30 CHD-free Labrador Retrievers revealed significant SNPs on CFA8 within a previously reported quantitative trait locus (QTL) in German Shepherd Dogs. A 10-bp intronic deletion haplotype within *FBN2* on CFA11 was shown to be associated with CHD. Dogs homozygous for this haplotype had significantly less *FBN2* mRNA in their femoral head articular cartilage. The mutant *FBN2* haplotype was identified in 49 different breeds, but homozygous mutant haplotypes were only prevalent in Labrador and Golden Retrievers. The objective of the present study was to perform a GWAS on 192 dogs to identify SNPs associated with CHD, followed by a validation of CHD-associated SNPs in a stratified sample of 843 dogs. We have chosen the German population of German Shepherd Dogs as this population is well suited for an association study, since this breed represents one of the largest purebred dog populations in Europe with a large phenotypic and genetic variance for CHD and a consistent recording system of CHD including the collection of EDTA-blood samples. # Results The genome-wide scan using the canine 127K Affymetrix SNP chip (Affymetrix, Santa Clara, CA, USA) revealed five SNPs with –log₁₀P-values\>4.3 suggestive for association with CHD. We detected a genome-wide significant CHD- associated SNP on CFA24 in a region which was not significant in a previous linkage study for German Shepherd Dogs. We determined the genome-wide threshold for significance at a -log₁₀P-value \>5.98 which corresponds to a P-value \<0.05 after applying the Bonferroni correction for multiple testing. The quantile-quantile (Q-Q) plots illustrated that inflation due to stratification effects had been removed by the mixed linear model employed for CHD. The validation study included the five most significant SNPs on CFA19, 24, 26 and 34. Two SNPs were located on CFA34 and each one SNP on CFA19, 24 and 26. All these five SNPs were evaluated in 843 German Shepherd Dogs. The minor allele frequencies (MAF) ranged from 0.07 to 0.48. Each one SNP on CFA24, 26 and 34 was significantly associated with CHD in the validation set, whereas the SNP on CFA19 and one SNP on CFA34 did not pass validation. The CHD-QTL on CFA24, 26 and 34 were confirmed in the analyses when dogs with the CHD grades C and D were separately compared with all CHD-free dogs. The CHD-QTL on CFA24 was also significant for the separate analysis for CHD-E. A combined analysis for dogs free from CHD and dogs with CHD-D and CHD-E gave slightly higher –log₁₀P-values compared to the analysis with CHD-A versus CHD-D dogs (data not shown). The strongest association was found for the SNP on CFA24. In all analyses considering CHD-free German Shepherd Dogs versus mildly to severely CHD-affected dogs (C+D+E or C or D or E or D+E), BICF2S2367279 reached -log₁₀P-values at 10–40. Odds ratios (ORs) were highest for BICF2S2367279 with values at 4.3–5.8. The CHD-associated genotype G/G of BICF2S2367279 on CFA24 occurred in 312/843 dogs and 27/312 were free from CHD and all other 285/312 were CHD-affected. Corresponding figures for BICF2S2367279 on CFA26 were for the CHD-associated genotype A/A were 369/843, 66/369 (CHD- free) and 303/369 (CHD-affected), respectively. The joint genotypes G/G and A/A had a frequency of 222/843 and only 10/222 were free from CHD, whereas 212/222 were CHD-affected. The proportion of phenotypic variance of CHD explained by the cumulative effects of all five SNPs was 19.6% (all CHD-grades), 21.6% (CHD-A vs C), 31.6% (CHD-A vs CHD-D), 20.7% (CHD-A vs. CHD-E) and 28.0% (CHD-A vs. CHD-D and CHD-E). # Discussion The present GWAS identified a novel genome-wide significant CHD-QTL on CFA24, which was validated in a large sample of German Shepherd Dogs stratified by sex, age, birth year, CHD-status and ancestry. A slight overestimation of the –log₁₀P-values may be assumed in the detection sample because we used a linear model and not a cumulative logit model. However, the –log₁₀P-value for the SNP on CFA24 was still below the Bonferroni significance threshold of 0.05. Further two QTL from a previous linkage study could be confirmed through a GWAS and a validation study. In close proximity to the SNP BICF2S2367279 on CFA24, the gene *SRC* (*v-src sarcoma (Schmidt-Ruppin A-2) viral oncogene homolog*) is located. This gene is involved in bone formation mediating cytoskeletal reorganization as well as osteoclast survival in response to RANKL in vitro. Targeted disruption of *src* function in mice caused osteopetrosis, which is characterized by increased bone density due to impairment of osteoclast bone resorbing activity. Osteoclasts from src-null mice failed to form ruffled borders impairing their ability to resorb bone matrix. On CFA26, the SNP BICF2P281364 is located within *KSR2* (*kinase suppressor of ras 2*) which encodes a scaffold protein interacting with members of the RAS/MAPK signalling pathway. KSR2 has been suggested to be involved in activation of molecules like p38 MAPK, cell cycle control proteins and proteins of the ubiquitin proteasome. The p38α and p38β MAPK proteins are important regulators of bone homeostasis controlling expression and activation of transcription factors implicated in osteoblastogenesis like RUNX2. In addition, KSR2 is predicted to mediate PI3K activation. Via PI3K, a RANKL mediated signalling pathway is activated which negatively regulates osteoclast survival. On CFA34, the SNP BICF2P1086886 is located approximately 0.7 Mb downstream of the *triple functional domain (PTPRF interacting)* (*TRIO*) gene. The protein encoded by *TRIO* contains three functional domains, a serine/threonine kinase domain and two guanine nucleotide exchange factor (GEF) domains for the family of Rho-like GTPases, specific for Rac1 and RhoA. Rac1 and RhoA act as antagonists, both playing an important role in chondrogenic proliferation and differentiation. Chondrocyte-specific deletion of Rac1 in mice leads to dwarfism due to reduced chondrocyte proliferation. Inhibition of Rac1 expression in micromass culture resulted in reduced mRNA levels of the chondrogenic markers collagen II and aggrecan, and decreased accumulation of glycosaminoglycans indicating that Rac1 promotes chondrogenesis. Rac1-deficient chondrocytes had severely reduced levels of inducible nitric oxide synthase protein (iNOS) and nitric oxide production. Mice deficient for iNOS had reduced chondrocyte proliferation and resembled the phenotype of Rac1-deficient growth plates. RhoA overexpression in chondrogenic ATDC5 cells resulted in increased proliferation and a marked delay of hypertrophic differentiation, whereas inhibition of Rho/ROCK signaling inhibited chondrocyte proliferation and accelerates hypertrophic differentiation. Therefore, changing the balance between the GTPases RhoA and Rac1 due to mutations in *trio* will lead to disturbances in cartilage development. In agreement with previous reports, the CHD-associated regions and the candidate genes identified here, support the key role of enchondral bone formation in the pathogenesis of CHD. A candidate region identified in German Shepherd Dogs and Labrador Retrievers on CFA8 at 29 Mb harbours the potential candidate gene *LRR1*. The encoded protein is involved in proteoglycan synthesis through NF-κB signalling transduction and cartilage integrity like iNOS and MAPK proteins. In Bernese Mountain Dogs, *FN1* was identified as a candidate gene for CHD. FN1 had been proposed to have a function in matrix organization of cartilage. In an across-breed GWAS, *EVC* and *EVC2* on CFA3, *PTPRD* on CFA11 and *MFAP1* on CFA30 were suggested as CHD candidate genes. All these genes are highly expressed in cartilage and mutations in these genes can cause chondrodysplasia (*EVC*, *EVC2*), Marfan syndrome (*MFAP1*) or restless legs syndrome (*PTPRD*). For the candidate genes on CFA24, 26 and 34, we could find a joint network involved in bone formation using GENEMANIA (<http://www.genemania.org/>). Joint co-expression and physical interactions had been indicated for *TRIO, SRC* and *KSR2*. This may indicate the QTL identified harbour genes contributing to CHD on the same pathways. Using the candidate genes *LRR1*, *FN1*, *PTPRD* and *MFAP1*, we could extend the network for *TRIO, SRC* and *KSR2*, primarily through *RAC1*. In conclusion, this study identified and corroborated CHD-loci on three different chromosomes for German Shepherd Dogs and is a further step towards elucidation of the genes underlying CHD. The three validated and significantly CHD-associated SNPs are located within or in close proximity to genes which are involved in a joint network regulating bone formation, osteoclast activity, chondrocyte proliferation and differentiation. # Material and Methods ## Ethics Statement All animal work has been conducted according to the national and international guidelines for animal welfare. The dogs in this study were included with the consent of their owners. The blood samples were collected by veterinarians during the routine examination for CHD regarding principles of good veterinary practice. These diagnostic procedures had to be carried out anyway. All blood- sampling of dogs was done in veterinary practices for small animals by trained staff. CHD examinations are compulsory for all dogs intended for breeding and these examinations were performed in veterinary clinics and practices according to the rules for hip screening of the Fédération Cynologique Internationale (FCI). Approval from the ethics committee was not obtained because blood sampling was done during a diagnostic veterinary procedure according to the German Animal Welfare Law (released on 05/18/2006, last changes on 08/07/2013). ## Animals A total of 1035 German Shepherd Dogs had been included in the present association study for CHD. The sample for the GWAS to identify significantly associated SNPs contained 192 dogs and the validation set 834 dogs. First, we selected the dogs for the GWAS and then, the validation sample was collected from available databases and bio-banks. The study population was a stratified sample from all registered and X-rayed dogs of the German population of German Shepherd Dogs born between 2000 and 2005. Dogs had an age at examination of 12–14 months in each cohort. Coancestry among cases and controls as well as within cases and within controls was minimized. In addition, the samples used were stratified by sex, age at radiographic examination, birth year, coancestry and CHD-affection. The pedigree records from eleven generations were employed to calculate relationship coefficients among all dogs using PEDIG software. The study population had mean relationship coefficients among all individual dogs \<0.1%. Cases and controls were chosen with equal proportions of sexes and birth cohorts from the years 2000–2005. Dogs were 12–14 months old at radiographic examination. Radiographs and CHD-scores according to the official guidelines of the FCI as well as EDTA-blood samples for parentage testing were collected by the Association for German Shepherd Dogs (SV, Augsburg). Parentage testing is mandatory for all German Shepherd Dogs intended for breeding. For all dogs included in the present study, parentage has been confirmed using an approved set of microsatellites for parentage control. The radiographs were made by veterinarians officially approved by the association of radiographic diagnosis for genetically influenced skeletal diseases of small animals (GRSK) and the German Association for Dog Breeding and Husbandry (VDH) strictly following the requirements for radiographs of hip joints of the FCI. All X-rays were evaluated by a recognized veterinary expert and a subsample of 200 X-rays was re-evaluated by further two experts. Consistency of diagnoses was \>99%. These officially recorded CHD-grades were supplied by the SV and used in all subsequent analyses. CHD-A includes dogs with normal hips, CHD-C dogs with slight signs of CHD, CHD-D dogs with moderate signs of CHD and CHD-E dogs with severe signs of CHD using X-rays from both hip joints. Dogs with mild, moderate or severe signs of CHD were defined as CHD-affected. Controls were dogs free from any signs of CHD. ## Genome-wide Association Study A genome-wide screening for polymorphic and CHD-associated SNPs was performed using the canine 127K Affymetrix SNP chip (Affymetrix). This sample included 192 German Shepherd Dogs whereof 96 were free from CHD, 65 mildly, 22 moderately and 9 severely CHD-affected. Cases and controls were balanced by CHD score, sex, birth year, age at veterinary radiographic examination and coancestry. The animals were unrelated at least at the grandparent level. Relationship coefficients with the validation sample were as low as possible and did not exceed 0.1%, on average. The dogs from both data sets, the detection and validation samples, were unrelated at least on the grandparent level. Quality criteria were minor allele frequency (MAF) \>0.05, genotyping rate per SNP and animal \>0.90 and HWE (p\<0.00001). After filtering for quality criteria, 47,729 SNPs remained for the final analysis. Data analysis was done for the CHD-score as quasi-continuous target trait. We used a mixed linear model with sex and genotype as fixed effects and a random animal effect through an identity-by- state-kinship (IBS) matrix. A Q-matrix to detect population structure was estimated using principal components (PCAs). We also tested several extended models employing up to five PCAs to show the robustness of the outcome of the GWAS. All these models yielded the same highly significant associated SNP as the final model. Thus, adding principal components for cryptic data structure did not change the results of the final model. Thus, the IBS matrix reflected the genomic relationship matrix among all individuals genotyped and captured the relatedness among animals as well as the cryptic family structure. The analysis was run using TASSEL, version 3.0.164. ## Validation Study Cases and controls were matched by sex, birth year, age at radiographic examination, CHD score and coancestry. We edited 843 German Shepherd Dogs out of a group of 12,096 dogs which fitted our study design and had mean relationship coefficients \<0.1% with any other individual dog in the sample. The animals were born in 2000–2005 and purebred following the rules of the SV. Distribution of CHD-scores in the 405 males and the 438 females was as follows. In the males, 136 (16.1%) CHD-A, 176 (20.9%) CHD-C, 72 (8.5%) CHD-D and 21 (2.5%) CHD-E; in the females, 141 (16.7%) CHD-A, 173 (20.5%) CHD-C, 80 (9.5%) CHD-D and 44 (5.2%) CHD-E. Dogs scored as CHD-A were treated as unaffected (n = 277), dogs scored as CHD-C (n = 349), CHD-D (n = 152) and CHD-E (n = 65) as affected. Association analyses were performed using odds ratios with their 95% confidence limits and χ²-tests for genotypic and allelic associations and the allelic trend with the CHD status (CHD-A vs. CHD-C/D/E) and with CHD-A versus CHD-C, CHD-D or CHD-E. We tested for the different CHD grades separately to see whether the association is consistent across the different CHD-grades or may be caused by a specific CHD-grade. The CASECONTROL procedure of SAS/Genetics, version 9.3 (Statistical Analysis System, Cary, NC, USA, 2013), was used for these calculations. ## DNA Preparation and Genotyping DNA was isolated from EDTA-blood samples using the NucleoSpin Kit 96 Blood Quick Pure Kit (Macherey-Nagel, Düren, Germany). Five SNPs on CFA19, 24, 26 and 34 were chosen for validation from the GWAS. Genotyping of the validation SNP set consisting of five SNPs was performed using the ABI Prism SNaPshot™ Multiplex System (Life Technologies by Applied Biosystems, Darmstadt, Germany). Single base extension (SBE) primers were designed using online primer design software BatchPrimer3 (<http://probes.pw.usda.gov/cgi- bin/batchprimer3/batchprimer3.cgi>). According to the manufacturer's instructions, the primer sequences should differ in length by at least four to six nucleotides and should not contain possible extendable hairpin structures. All primers used here had undergone HPLC purification. The primer design was in this way that a difference of seven nucleotides between the SBE primers was achieved by adding non-homologous polynucleotides (poly (dNTP)) at the 5′end. The SBE detection was performed using an ABI Genetic Analyzer 3500 (Life Technologies by Applied Biosystems). Data evaluation was done using GeneMapper, version 4.2 (Life Technologies by Applied Biosystems). ## Statistical Analysis Allele and genotype frequencies of the SNPs were calculated for the different CHD-grades as well as the observed heterozygosity (H₀), the polymorphism information content (PIC) and Hardy-Weinberg equilibrium were estimated using the ALLELE procedure of SAS/Genetics. Association analyses were performed using odds ratios with their 95% confidence limits and χ²-tests of the CASECONTROL procedure of SAS/Genetics for genotypic and allelic associations and the allelic trend with the CHD status (CHD-A vs. CHD-C/D/E) and with CHD-A versus CHD-C, CHD-D or CHD-E. These four definitions of phenotypes were used for all analyses. A SNP was regarded as genome-wide significantly associated for –log₁₀P-values\>5.98 and suggestive for association with –log₁₀P-values\>4.3. The same cutoffs were applied for the validation study. A general linear model including all CHD-associated SNPs was used to estimate the proportion of the phenotypic variance explained by these SNPs. Calculations were performed using SAS, version 9.3. # Supporting Information The authors thank the Associaton of German Shepherd Dogs (Verein für Deutsche Schäferhunde e.V., SV), Augsburg, Germany, for providing the pedigree data, X-rays, the official CHD scores and EDTA-blood samples collected for paternity testing. The authors are grateful to U. Philipp and H. Klippert-Hasberg for expert technical assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: OD. Performed the experiments: LF OD. Analyzed the data: LF OD. Contributed reagents/materials/analysis tools: OD. Wrote the paper: LF OD.

# Introduction Advances in rehabilitation technology have led to a growing class of computer- controlled devices that provide powered assistance for human motion. These include powered protheses, orthoses, and exoskeletons. The behavior of these devices can often be tuned via a set of control parameters. For instance, the BiOM powered ankle prosthesis has 11 parameters governing its stiffness, the power and timing of the actuated push-off, and other functionality. The powered prosthesis for trans-femoral amputees developed at Vanderbilt had 42 virtual- stiffness parameters at one point in its development. Although high numbers of parameters make devices highly adaptable, they also make the selection of proper parameter values more difficult. One might hope to select appropriate parameters settings using models and simulations, but parameter settings selected in this way are not always optimal when evaluated experimentally. Offline models cannot fully capture the variability of end-users nor the ways that users will interact with different parameter settings. This makes an *a priori* selection of parameters for a specific end-user a challenge. For this reason, final parameter selections are almost always experimentally. That is, the process is driven by experimental observations of the user interacting with the device at different parameter settings. The methods for tuning parameters to a particular end-user are often based on heuristics. The tuning may involve a trial-and-error process that is driven by subjective evaluations. This is similar to the process used by a prosthetist when aligning a conventional passive prosthesis. The process requires expert knowledge, is time consuming, and research has shown that the subjectivity of these methods makes them unreliable. It would be highly desirable to automate the tuning process to remove subjectivity and enable assistive devices to continue to adapt even after the initial fitting. The development of such an automated tuning process begins with the definition of a suitable physiological objective. An optimization algorithm that is able to quantify this objective can automatically compare parameter choices and identify those that are better. Possible examples of physiological objectives include the reduction of energy spent by the user, improved user comfort, or the enforcement of a desired user training effort. Generally, none of these objectives can be measured directly. Instead, they must be estimated from related indicators, such as from oxygen consumption, muscle activation, heart rate, or joint motion. The relationship between these indicators and the physiological objective can be complex and is often concealed by unrelated effects, such as noise or other disturbances. To reconstruct the objective, we might have to combine information from multiple sensors and take into account delays and other dynamic effects. In this paper we focus our work on using metabolic energetic cost as the physiological objective. We estimate this cost by using oxygen consumption measured at the mouth as the related indicator. In our methods, we compensate for the dynamics of energy conversion in the body and we employ algorithms that are robust against the large amounts of noise that are typical for respiratory measurements. Naturally, these methods are very specific to energetic cost. However, similar ideas can be applied to a wide range of physiological objectives as well as to different indicators and sensor modalities. Metabolic energetic cost is commonly used to characterize and improve device performance. It is a measure of the energy being consumed by a user to sustain an activity. As the user of an assistive device is able to rely more on the device to perform the necessary work, the energy required of a user goes down. In this way, energetic cost provides a clear way to characterize the quality of assistance being provided. For example, Malcolm et al. demonstrated that measurements of energetic cost can be used to identify the energetically optimal timing setting of a powered ankle-foot orthosis. Deviations away from this optimal setting resulted in increased user effort. There is also evidence that, in many situations, humans seek to minimize their energetic cost in unassisted walking. Research has shown that a subject’s choice of walking speed, step- frequency and step width will minimize the energy required to walk a given distance. In addition, using metabolic effort as the objective might allow us to capture some other important characteristics of providing beneficial assistance. For example, if the wrong parameter setting of an assistive device causes discomfort or pain, the user might start fighting the device or engage in other compensatory actions. Through this mechanism, discomfort and other issues will likely lead to an increase in energy consumption and are thus accounted for in measures of energetic cost. The use of measurements of energetic cost is challenging because they are sparsely sampled, noisy, and dynamically delayed. The most practical measures during locomotion come from measuring the volumes of oxygen and carbon-dioxide in inspired and expired air. These measures are used to approximate the consumption rates of chemical fuels in the body. The sampling rate of these measures corresponds to the breath-rate of the subject. During light exercise, a typical breath rate of 20 breaths-per-minute would result in a sampling rate of 0.3Hz. Variability in breath volumes results in noisy measurements with a Gaussian distribution and a signal to noise ratio of approximately four. Additionally, the measures do not always reflect the user’s current level of effort. Rather, the effect of changes in effort appears gradually over time. For example, if a sitting subject were asked to stand and begin walking, it would take over a minute for the energetic cost measures to rise and plateau at the value corresponding to walking. When looking to identify optimal parameter settings, the difficulties of respiratory measurements are typically addressed by averaging the effects of many steady-state breaths. To characterize the energetic cost at a given parameter setting, researchers will often wait three minutes to allow the measures to plateau at steady-state values and then average measurements over an additional three minutes while the subject continues the activity. This process is repeated with several parameter settings to approximate the relationship between the energetic cost and the parameter through a curve-fitting process. The minimum of this curve is taken as a proxy for the optimal parameter settings. We refer to this method as *Steady-State Cost Mapping*. *Mapping*, refers to the fact that this method seeks to fit a curve or surface over a large number of parameter settings to estimate the shape of the overall energy- parameter relationship. *Steady-State Cost* is the energetic cost averaged at a single parameter setting after some waiting time to avoid any effects of the dynamic delay and noise. With only a single parameter to tune, Steady-State Cost Mapping can be quite effective. The time required to collect each data point, however, greatly limits the broader applicability of this method. The number of data points required to map the interactions of multiple parameters increases by the power of the number of parameters. If *N* settings were evaluated to map the effect of a single parameter, *N*² would be required for a similar mapping of two parameters. For example, evaluating a single parameter at five settings takes about half an hour. Two parameters, with five settings each would require two and a half hours. Experiments by Bertram used eight different values for the speed and step frequency of subjects. Though he only tested 49 of the 64 possible combinations, the experiments required subjects to walk for over four hours. Evaluating all the possible combinations for the 11 parameters of the BiOM or the 42 parameters of the Vanderbilt leg would be infeasible. Nor is such an exhaustive mapping necessarily useful. When the goal is to identify the energetically optimal parameter settings, parameter settings evaluated far from the minimum do not greatly affect the result. The surfaces that are fit to measurements are designed to match the *local* relationship between the parameters and energetic cost. For this reason, evaluating parameter settings close to the minimum would provide a more accurate estimate of the minimum location than the evaluation of settings far from the minimum. Yet, this is only possible if we have some prior knowledge about the approximate location of the minimum, or if we can estimate the location in an iterative process. The purpose of our work is to improve the methods used to identify device parameter settings. Specifically, we propose methods that enable the faster selection of energetically optimal parameter settings. These methods rely on the the estimation of *Instantaneous Energetic Cost*. *Instantaneous* Cost refers to power being used by the body at any particular instant in time. Estimates of Instantaneous Cost provide a way of characterizing the energetic requirements of various conditions without necessarily waiting for respiratory measurements to reach steady-state at each condition. This enables a real-time optimization that relies on continuously varying parameters that are updated with an actual human body in the loop. # Methods The optimization seeks to identify optimal parameter values ***p***^⋆ that minimize a physiological value function *x*. This value function is estimated indirectly through measurements $\hat{\mathbf{y}}$ of indicators ***y***. For example, we might seek to minimize the energetic effort of a person walking with an assistive device. In this example, the value function *x* is the subject’s instantaneous energetic cost, and the indicator *y* is oxygen consumption measured at the mouth ($\overset{.}{V}O_{2}$), obtained via respiratory measurements $\hat{y}$. When we assume that a reduction of effort can be achieved by the proper tuning of controller parameters, the cost function can be expressed as *x*(***p***) in which the controller parameters constitute the free parameter vector ***p***. The current work demonstrates a faster method for approximating the optimal setting of a single parameter *p* (*Instantaneous Cost Mapping*) and a method that can search for the minimum when multiple parameters ***p*** are involved (*Instantaneous Cost Gradient Search*). These methods are compared with the traditional *Steady-State Cost Mapping* approach. The estimation of Instantaneous Cost relies on a function that serves as a surrogate for the unknown relationship between energetic cost and the parameter settings. The surrogate function is adapted to match experimental measurements of energetic cost. This surrogate function can be interpreted as a *response surface* that characterizes the energy-parameter relationship and can be used to drive an optimization. The *Instantaneous Cost Mapping* uses a cubic polynomial fit to the entire feasible range of a parameter. The measurements for the polynomial are taken while linearly varying the parameter value from one extreme to the other over a fixed time interval. The estimate of the optimal parameter setting is found by evaluating the minimum of the polynomial. The *Instantaneous Cost Gradient Search* uses gradient descent techniques adapted from Finite Difference Stochastic Approximation. This method approximates a gradient using measures taken at points near the current parameter settings. The gradient (i.e., a linear response surface) is used to iteratively drive the parameter settings to a local minimum. This process takes place with a “body-in-the- loop”–it is driven by measurements of energetic cost collected in real-time. The three methods are illustrated in. In the study presented here, the device parameter *p* that we optimize is simply the step frequency at which a subject is commanded to walk. That is, rather than using a robotic assistive device unique to our lab, our “computer-controlled assistive device” is a metronome that subjects are instructed to follow. Our use of step frequency allows the methods of this work to be easily replicated and improved by other groups. Because step frequency is easy to prescribe and measure and has a clear metabolic minimum, it has been used in experiments studying the way that humans optimize their own walking. The energetically optimal step frequency can be approximated by the step frequency that a subject selects on their own (without a metronome). This self-selected step frequency is referred to as a subject’s “preferred”. Our experiments with eight subjects used the preferred step frequency of each subject as the ground truth for the energetic minimum. ## Parameter Exploration Strategies By exploring the parameter settings, ***p***, in an intelligent way, we can learn about the shape of the cost function, *x*(***p***), and find energetically optimal values of the parameters, ***p***^⋆. The algorithms we test explore the parameters in different ways: the *Steady-State Cost Mapping* relies on evenly-spaced parameter settings evaluated in separate trials, the *Instantaneous Cost Mapping* evaluates a continuous sequence of parameter settings in a single trial, and the *Instantaneous Cost Gradient Search* tests nearby parameter settings to intelligently update the parameter choice. ## Response Surface Identification The instantaneous energetic cost, *x*, cannot be measured directly, but only estimated through the *respiratory response*, *y*. These measurements are intermittent, very noisy and exhibit dynamics similar to a low-pass filter. Without replicating the complex biological processes that govern respiration and metabolism, we describe the dynamic relationship of *instantaneous energetic cost* *x* and the *respiratory response* *y* by a simple lumped-parameter model with experimentally identified coefficients. In particular, our previous work shows that a first-order, linear dynamic model with a single time constant, *τ*, can describe the dynamics of respiratory metabolic measurements during walking. That is, the dynamic response of *y* is akin to a low-pass filter with a time constant of *τ*. In, this grey-box model was able to account for 82–99% of the measured variability. Mathematically, the relationship between the respiratory response of energetic cost, *y*^*i* (corresponding to breath, *i*), and the instantaneous energetic cost, *x*^*i*(***p***^***i***), can be expressed as: $$\begin{matrix} {y^{i} = \left( {1 - \frac{h_{i}}{\tau}} \right)y^{({i - 1})} + \frac{h_{i}}{\tau}x^{i}\left( \mathbf{p}^{i} \right).} \\ \end{matrix}$$ In this time-discrete dynamic formulation, *h*_*i* is the time since the previous breath and *τ* is the time constant of the subject’s metabolic response. If *x* and *y* were at the same value (i.e., steady-state conditions) and the value of *x* was suddenly changed, the subsequent values of *y* would begin to asymptotically approach the new value of *x*. *τ* characterizes the rate of this asymptotic approach. After a time equal to 3*τ* the error between *x* and *y* will be approximately 95% smaller than before. Previous experiments characterizing instantaneous energetic cost during walking found an average value of *τ*, in their subject pool, to be 42 seconds with a standard deviation of 12 seconds. We approximate the instantaneous cost function *x*(***p***) by finding a *response surface* that best predicts the respiratory measurements while taking into account the dynamics of metabolic energy conversion. This response surface is defined by a vector of *m* constants, ***λ***. The constants alter the shape of the response surface and are identified by the algorithm to match the experimental measurements of energetic cost. We denote this approximation of the underlying relationship between the parameters, ***p***, and the instantaneous energetic cost, *x*(***p***) with an overbar: $\overline{x}\left( {\mathbf{p},\mathbf{\lambda}} \right)$. The optimization is conducted with respect to this response surface, similar to work by Lawrence et al.. While there are many functions that could be used to represent the relationship between the parameters and the required energy, the response surfaces used in this work are polynomial series with coefficients ***λ***. Polynomials are useful for approximating unknown functions because they are linear with respect to ***λ***. Recursively evaluating using $\overline{x}\left( {\mathbf{p},\mathbf{\lambda}} \right)$ and an initial breath estimate ${\bar{y}}^{1}$ returns a predicted respiratory response $\bar{y}$. For a given set of *n* noisy breath measurements, $\hat{y}$ (where $\;\hat{}\;$ refers to actual measurements), the best fit response surface minimizes the sum of squared error, *e*, between the predicted respiratory response, $\bar{y}$, and the actual measurements, $\hat{y}$: $$\underset{\mathbf{\mathbf{\lambda}},{\overline{y}}^{1}}{\text{min}} \text{      }e = \sum\limits_{i = 1}^{n}\left( {{\hat{y}}^{i} - {\overline{y}}^{i}} \right)^{2} = \left( {{\hat{y}}^{1} - {\overline{y}}^{1}} \right)^{2} + \sum\limits_{i = 2}^{n}\left( {{\hat{y}}^{\, i} - {\overline{y}}^{i}\left( {\overline{x}\left( {\mathbf{p}^{i},\mathbf{\lambda}} \right),{\overline{y}}^{i - 1},h^{i},\tau} \right)} \right)^{2}$$ The solution to includes a set of optimal shape-defining constants, ***λ***^⋆, that result in a response surface, $\overline{x}\left( {\mathbf{p},\mathbf{\lambda}^{\star}} \right)$, with the best fit between the predicted respiratory response, $\bar{y}$, and the actual breath measurements, $\hat{y}$. This method of response surface identification is similar to Least- Squares techniques of dynamic system parameter estimation. A well-fit response surface provides a way for us to characterize the effect of the parameters on the energetic cost in an analytic form. Because our polynomial response surfaces are linear with respect to ***λ***, can be expressed as a system of linear equations $$\begin{array}{r} {\mathbf{A}\begin{pmatrix} {\overline{y}}^{1} \\ \lambda_{1} \\ \vdots \\ \lambda_{m} \\ \end{pmatrix} = \begin{pmatrix} {\overline{y}}^{1} \\ \vdots \\ {\overline{y}}^{n} \\ \end{pmatrix}.} \\ \end{array}$$ This linear form allows us to efficiently find the optimal initial breath estimate, ${\bar{y}}^{1 \star}$, and shape parameters, ***λ***^⋆ that satisfy. This is done using the pseudoinverse of **A**, **A**⁺ = (**A**^*T* **A**)⁻¹ **A**^*T*. The properties of the pseudo-inverse allow us to find the best-fit response surface by multiplying **A**⁺ with the respiratory measurements $\hat{y}$. The respiratory response, $\bar{y}$, predicted by ${\bar{y}}^{1 \star}$ and ***λ***^⋆ will have the least squared-error possible. <img src="info:doi/10.1371/journal.pone.0135342.e026" id="pone.0135342.e026g" /> ( y ¯ 1 ⋆ λ 1 ⋆ ⋮ λ m ⋆ ) = A \+ ( y ^ 1 ⋮ y ^ n ) **A** is a (*n*) × (*m* + 1) matrix whose elements are defined recursively by the following pattern. <img src="info:doi/10.1371/journal.pone.0135342.e027" id="pone.0135342.e027g" /> A i j = { 1 i = 1 , j = 1 0 i = 1 , j ≠ 1 A ( i \- 1 ) j · ( 1 \- h i τ ) i \> 1 , j = 1 A ( i \- 1 ) j · ( 1 \- h i τ ) \+ h i τ ∂ x ( λ , p i ) ∂ λ j \- 1 i \> 1 , j \> 1 When using a polynomial response surface, $\bar{x}\left( \mathbf{\lambda},p \right)$ for a single parameter, *p* of the form $\bar{x}\left( \mathbf{\lambda},p \right) = \lambda_{1} + \lambda_{2}p + \lambda_{3}p^{2} + \ldots + \lambda_{m}p^{m - 1}$, the matrix **A** is defined as follows $$\begin{matrix} {\mathbf{A} =} & \left( \begin{matrix} 1 & 0 & 0 & \cdots \\ \left( {1 - \frac{h_{i}}{\tau}} \right) & \frac{h_{i}}{\tau} & {\frac{h_{i}}{\tau}p} & \cdots \\ \left( {1 - \frac{h_{i}}{\tau}} \right)^{2} & {\frac{h_{i}}{\tau}\left( {1 - \frac{h_{i}}{\tau}} \right) + \frac{h_{i}}{\tau}} & {\frac{h_{i}}{\tau}p\left( {1 - \frac{h_{i}}{\tau}} \right) + \frac{h_{i}}{\tau}p} & \cdots \\ \vdots & \vdots & \vdots & \vdots \\ \end{matrix} \right. \\ & {\left. \begin{matrix} \cdots & 0 \\ \cdots & {\frac{h_{i}}{\tau}p^{m - 1}} \\ \cdots & {\frac{h_{i}}{\tau}p^{m - 1}\left( {1 - \frac{h_{i}}{\tau}} \right) + \frac{h_{i}}{\tau}p^{m - 1}} \\ \vdots & \vdots \\ \end{matrix} \right).} \\ \end{matrix}$$ In the Instantaneous Cost Mapping, the response surface identification process is done only once, after the experiment is completed. In the case of the Instantaneous Cost Gradient Search, this identification process is done repeatedly during the experiment to iteratively drive the parameters towards energetically optimal values. We implemented this method in MATLAB, and have made it available on the MATLAB File Exchange. ## Instantaneous Cost Mapping For the Instantaneous Cost Mapping method, we fit a cubic polynomial response surface to the parameter space by continuously changing parameter values during the experiment. The analytical minimum of the surface is returned as the optimal parameter choice. ### Parameter Exploration in the Instantaneous Cost Mapping The parameter space is explored by linearly changing the parameter value from one extreme to the other (e.g. *p*_*min* to *p*_*max*) over the course of time *T*. <img src="info:doi/10.1371/journal.pone.0135342.e031" id="pone.0135342.e031g" /> p ( t ) = p m i n \+ ( p m a x \- p m i n T ) t Larger values of the interval, *T*, result in an increased number of samples which reduces the uncertainty of the shape parameters. Our experiments use *T* = 5 min. The algorithm begins by commanding a step frequency at either 25% above or 25% below the subjects preferred step frequency. This step frequency gradually changes over the course of five minutes to the other value (e.g., from -25% to 25%). ### Response Surface Type for the Instantaneous Cost Mapping We assume that the underlying relationship between the parameters and energetic cost can be approximated by a cubic polynomial function. <img src="info:doi/10.1371/journal.pone.0135342.e032" id="pone.0135342.e032g" /> x ¯ ( λ , p ) = λ 1 \+ λ 2 p \+ λ 3 p 2 \+ λ 4 p 3 The minimum of the polynomial is the unique point where the first derivative of the polynomial (with respect to *p*) is zero and the second derivative is positive. ## Instantaneous Cost Gradient Search This method iteratively approaches the local minimum of the value function by using the gradient (i.e., slope) of the value function to update its guess of the best parameter settings. ### Parameter Exploration in the Instantaneous Cost Gradient Search Gradient searches work by evaluating the gradient of the energy-parameter relationship near the current parameter settings. This gradient can be thought of an arrow pointing in the direction of improvements in energetic cost. The algorithm works by following this arrow towards the minimum. For example, if one imagines an energy-parameter relationship that looks like a parabola, the gradient would always be pointing in the direction of the minimum. The general form of a gradient descent is $$\begin{matrix} {{\overline{\mathbf{p}}}_{k + 1} = {\overline{\mathbf{p}}}_{k} - \alpha_{k}\mathbf{J}_{k}\:\:\:\:\alpha_{k} > 0} \\ \end{matrix}$$ where ${\overline{\mathbf{p}}}_{k}$ is the vector of parameter values at the current iteration, ${\overline{\mathbf{p}}}_{k + 1}$ contains the parameter values at the next iteration, **J**_*k* is the local gradient and *α*_*k* is a small gain that determines the size of the change in parameters. Larger values of *α*_*k* can be used to improve the progress of the algorithm far from a minimum and smaller values can be used to improve the convergence of the algorithm near a minimum. This method assumes that the cost function is differentiable with respect to the parameters. **J**_*k* at the current parameter settings ${\overline{\mathbf{p}}}_{k}$ is a vector containing the local slopes of the value function with respect to each parameter (*p*₁, *p*₂, *p*₃ etc.). <img src="info:doi/10.1371/journal.pone.0135342.e037" id="pone.0135342.e037g" /> J k = ∇ ( x ) p = ( ∂ x ∂ p 1 ∂ x ∂ p 2 ∂ x ∂ p 3 ⋮ ) \| p ¯ k For a value function evaluated through measurements, one cannot evaluate the gradient directly. An estimate of the gradient can be found by fitting a linear response surface to parameter values near the current iteration. Because of the uncertainty propagation from measurement noise, this estimate will be a stochastic approximation of the true gradient. Finite-Difference Stochastic Approximation (Kiefer-Wolfowitz) optimization techniques provide a template for identifying and utilizing these uncertain gradients in the search for a minimum. These methods estimate gradients by perturbing parameters with a “finite difference” and observing the response. The estimates of the gradients are used with a “scheduled” gain, *α*_*k* that decreases with as the iteration number, k, increases according to the following equation. <img src="info:doi/10.1371/journal.pone.0135342.e038" id="pone.0135342.e038g" /> α k = A 0 α 0 A 0 \+ k γ 0 \< γ ≤ 1 In this method, measurements are taken at a perturbation distance, *c*_*k*, from the current iteration point, ${\overline{\mathbf{p}}}_{k}$. In our experiments, *c*_*k* was kept at the constant at 5% of the subjects preferred step frequency. The parameters of the gain schedule were *A*₀ = 3, α₀ =.0004^*hz*²/_ml/min, γ = 1. To estimate the gradient in this one-dimensional case with *n* measurements of energetic cost, the parameter value was first perturbed upwards for $\frac{n}{2}$ breaths and then downwards for another $\frac{n}{2}$ breaths. *i* is the number of breaths since the gradient fitting began. In our experiments, we evaluated the gradient with 30 breaths at each perturbation point (i.e., *n* = 60). <img src="info:doi/10.1371/journal.pone.0135342.e042" id="pone.0135342.e042g" /> p i = { p ¯ k \+ c k , if i ≤ n 2 p ¯ k \- c k , if n 2 \< i ≤ n ### Response Surface Type for the Instantaneous Cost Gradient Search A linear response surface is fit to the set of *n* measurements of energetic cost taken near the current iteration point, ${\bar{p}}_{k}$. This surface allows us to extract an estimate of the gradient, $J_{k} = \lambda_{2}^{\star}$. <img src="info:doi/10.1371/journal.pone.0135342.e045" id="pone.0135342.e045g" /> x ¯ ( λ , p ) = λ 1 \+ λ 2 p ## Experimental Methods We implemented the three algorithms and evaluated their performance with eight healthy subjects walking to the beat of a metronome. provides subject metrics. Subjects provided, written informed consent according to procedures approved by the University of Michigan Institutional Review Board. The methods of the study were approved prior to the commencement of the study by the IRB listed below. Study eResearch ID: HUM00020554 Health Sciences and Behavioral Sciences Institutional Review Board (IRB-HSBS) 2800 Plymouth Rd., Building 520, Room 1170, Ann Arbor, MI 48109-2800 (734) 936-0933 <irbhsbs@umich.edu.> We evaluated each subject in two sessions with no more than 10 days between sessions. We asked subjects to fast for at least two hours before each session to ensure accurate metabolic measurements. All walking was at 1.25 m/s on a Woodway PRO 27 treadmill (WOODWAY USA, Inc., Waukesha, WI, USA). We conducted the Steady-State Cost Mapping in the first session. The Instantaneous Cost Mapping and Instantaneous Cost Gradient Search were conducted in this order in the second session. The time constant of each subject’s metabolic response was identified during the Instantaneous Cost Mapping method. We measured oxygen consumption with a portable respirometer (*K*4*B*², COSMED, Rome, Italy). Oxygen consumption, $\overset{.}{V}O_{2}$, was taken as a proxy for metabolic energetic cost. Energetic cost values calculated with oxygen consumption alone will not differ by more than 3% from energetic cost values calculated with both oxygen consumption and *CO*₂ production. The step-frequencies were commanded through an Arduino-controlled metronome and measured with a force-sensitive- resistor (FSR) under the first metatarsal of the right foot. The FSR sensor was attached to a wireless transmission system (Trigno Wireless EMG, DELSYS, Natick, MA, USA). The measured step-frequency was filtered with a four-foot-fall (of the right foot) moving average filter. The filter was designed to roughly associate each breath with the contemporaneous step frequency measurements. Given a typical breath-rate and cadence of 20 ^{breaths/minute} and 1.8 Hz, respectively, we would expect three foot-falls of the right foot between breaths. From these filtered values, step frequencies concurrent to the recorded breaths were extracted through linear interpolation and used for processing. All algorithms were implemented in MATLAB (The MathWorks, Inc., Natick, MA, United States). We developed custom scripts to extract real-time $\overset{.}{V}O_{2}$ data from the COSMED software and real-time step frequency from the EMG software into MATLAB. In the Steady-State Cost Mapping, subjects first stood at rest for six minutes so we could evaluate their resting metabolic rate. We then asked the subjects to “walk normally” for six minutes. We averaged the last three minutes of measured step frequencies to establish the subject’s preferred step frequency. We then evaluated subjects at step frequencies commanded by the metronome between 25% above and 25% below their preferred step frequency at 5% intervals in a random sequence (including an evaluation at 0%, i.e., with the subject’s preferred step frequency commanded by the metronome). In each condition, we allowed three minutes to reach steady-state before averaging the measurements of oxygen consumption over an additional three minutes at the condition. Waking bouts were preceded by rests and did not last more than 18 minutes. We again evaluated subjects standing at rest for six minutes. A third-order polynomial was fit to the means of the steady-state recorded step frequencies and $\overset{.}{V}O_{2}$. The minimum of this polynomial was taken as an estimate of the true energetic minimum. For the Instantaneous Cost Mapping, subjects stood at rest for three minutes, then walked for three minutes at a starting cadence randomly selected to be either 25% above or below their preferred cadence. This data was used to identify the time-constant of the subject’s metabolic dynamics. Since our previous work found no significant difference between time constants for increases and decreases in activity level, we did not evaluate an additional, decreasing step. In order to identify the time constant, we needed to approximate the true instantaneous cost underlying the measures. We approximated the instantaneous cost as a step input that moves from a standing rest value up to a starting cadence value. For the standing rest value, we averaged the respiratory measurements from the last minute at that condition. For the starting cadence, we also used an average of the last minute of data. With this definition of instantaneous cost, *x*, MATLAB’s optimization toolbox was used to search for the value of *τ* and ${\bar{y}}^{1}$ that minimized the sum of the squared error between the predicted dynamic response, $\bar{y}$, and the measurements,$\hat{y}$, taken during the entire first six minutes of the test. This process was adapted from our prior work. These six minutes were followed by a metronome beat which moved from the starting cadence to the other extreme over the course of five minutes (i.e., moving from -25% to +25% or vice versa). Using the time constant we identified from the first part of the test, we used the data from the last five minutes of the test to identify the best-fit cubic polynomial response surface. The minimum of this polynomial was taken as an estimate of the true minimum. In the Instantaneous Cost Gradient Search, the starting condition was randomly chosen to be either 20% above or below the subjects preferred cadence. Subjects began by walking for three minutes at the starting condition to reach steady- state before the optimization commenced. The data acquired during this interval was not used in the optimization. The gradient was evaluated using 60 breaths (approximately 3 minutes). The perturbation width was 5%. For example, if the starting condition was 20% above the subject’s preferred step frequency, after warming up at the starting condition, the algorithm first commanded a metronome beat 25% above the subject’s preferred value for 30 breaths and then commanded a beat at 15% for 30 breaths. The algorithm then iterated along the gradient according to and repeated the gradient evaluation process. At any point or after five iterations, the optimization was stopped and the subject was allowed to rest as desired. When the test resumed, the algorithm again waited three minutes with the subject walking at the parameters defined by the last iteration before evaluating the gradient again. The algorithm was terminated after 15 iterations. The choice of the number of breaths, *n*, and the gain schedule were evaluated in simulation to verify their performance under the noise conditions. This algorithm assumed the subject’s time constant to be forty seconds. The estimated minimum was taken to be the termination point of the algorithm. # Results The minima predicted by the three algorithms are listed in and illustrated in. The Steady-State Cost Mapping required 66 minutes of walking. The minima of the resulting curves had an average relative error of 0.975% ± 4.08% (mean ± SD) with respect to the subjects’ preferred step frequencies. The Instantaneous Cost Mapping required only 8 minutes of walking. The time constants of our subjects had an average value of 28 sec ± 11 sec (mean ± SD). The minima of the of the resulting cubic polynomials had an average relative error of −0.59% ± 4.57% (mean ± SD) with respect the subjects’ preferred step frequencies. After fifteen iterations, the final conditions of the Instantaneous Cost Gradient Search had an average relative error of −0.29% ± 4.77% (mean ± SD) with respect to the subjects’ preferred step frequencies. The average walking-time for this method was 56.5 minutes. illustrates the progress of this algorithm for each subject. compares the results of the three methods for two representative subjects. The absolute error for each method was 3.41% ± 2.11% (mean ± SD) for the Steady- State Cost Mapping, 3.92% ± 2.89% (mean ± SD) for the Instantaneous Cost Mapping and 3.56% ± 1.92% (mean ± SD) for the Instantaneous Cost Gradient Descent. # Discussion These methods represent a valuable new tool for the selection of parameters in devices that interact with the human body. Our experiments demonstrate that the optimization algorithms we propose are robust to variations between subjects. The means of the error of the minima predicted by the three methods are each within a standard error (SE) from 0%. This means that none of the methods exhibited a significant bias towards selecting minima above or below the subjects’ preferred step frequencies. The accuracy of a particular method can be characterized by the variance of the estimated minima. A smaller variance would suggest that the method is more likely to select minima close to the true energetic minimum. In this sample, the Steady-State Cost Mapping has a slightly smaller variance than the other two methods. However, a Levene test for homogeneity of the variances does not show the variances to be significantly different (*p* = 0.88). Moreover, an F-test on the absolute error of the methods does not reveal any significant difference (*p* \> 0.49). This suggests that methods proposed in this work can identify energetically optimal parameters settings with similar accuracy to traditional methods. ## Faster Minimum Identification For the experimenter, Steady-State Cost Mapping is by far the simplest method to implement. In this method, all of the data is post-processed and discrete parameter settings can be evaluated in completely separate trials (which permits parameters to be changed offline). This method, however, is time consuming and fatiguing for subjects. The Instantaneous Cost Mapping, on the other hand, can be much faster than Steady-State Cost Mapping. The implementation of this method, however, requires the parameters to be varied during the test. Like the Steady-State Cost Mapping, this method assumes a function shape for the underlying energy-parameter relationship with the assumption that the minimum of this function will be near the true energetic minimum. For clinical populations that fatigue quickly, the Instantaneous Cost Mapping provides the means to estimate energetic minima without excessive walking-time. ## Global vs Local Methods When there are many interacting parameters to be tuned, gradient based optimization methods, such as the Instantaneous Cost Gradient Search are more efficient than mapping methods. Rather than map the entire parameter space, gradient searches iteratively improve parameter values relative to the current setting. If one imagines a parameter-cost relationship as a terrain with peaks and valleys, a gradient search algorithm can be thought to “flow” downwards towards the lowest point. Instead of blindly sampling the entire parameter space, the algorithm only needs to evaluate parameter settings along its path, increasing algorithmic efficiency. As a downside, gradient searches are “local” methods, meaning they can only find parameter combinations that reduce metabolic cost local to the starting conditions. If two valleys are separated by a ridge, the algorithm is unable to move between them. When the starting point is in the basin of the lowest valley (i.e., somewhat close to the global minimum), the algorithm will be able to reach the best-possible parameter setting. It is difficult to state whether the local nature of gradient searches poses a problem. It has been shown, for example, that the energetic cost relationship to power and timing parameters in an assistive ankle orthosis is well-characterized by a single valley. For new devices, however, the shape of the cost landscape is always unknown a priori. The importance of the starting point choice depends on whether the cost landscape contains a single or multiple valleys. ## Algorithm Parameter Selection The performance of the algorithms presented in this work can be affected by a poor choice of the parameters governing their operation. When the noise in the measurements of energetic cost is large compared to the slopes of the response surface, algorithm performance will be negatively impacted. In general, adjusting the parameters to lengthen the duration of the testing time should improve performance. For instance, the Steady-Sate Cost Mapping works best when the data is averaged over a sufficient time. The more samples that are collected at each steady-state condition, the greater the confidence in the resulting mean. If we assume the underlying measurements come from a normal distribution, we would expect the standard error of the calculated means to have an inversely proportional relationship to the square-root of the number of samples, *n*. The Instantaneous Cost Mapping method can be improved by increasing the duration of the parameter sweep. A slow parameter sweep reduces the attenuation in the metabolic dynamics. A quick parameter sweep may result in a rapidly changing instantaneous energetic cost. These rapid changes in cost will be will be attenuated by the low-pass-filter-like dynamic response of the breath-to-breath measures. Additionally, an increased number of samples helps to average-out the noise. If the parameter sweep is slow, the shape of the estimated energy- parameter relationship will be driven by a more accurate mean of the noisy signal. The order of the Instantaneous Cost Mapping polynomial can also be adjusted to improve algorithm performance. The order of the polynomial can be decreased to “smooth” the effect of the noise. Doing so, however, may also hide important features of the underlying energy-parameter relationship that could be captured by higher-order approximations. For example, quadratic approximations are symmetric and only have one maximum or minimum. Higher-order approximations could be used to capture asymmetry or additional minima or maxima. These higher- order fits, however, are more-likely to be affected by noise. The negative impact of the noise would be most pronounced when using a relatively quick parameter sweep. The Instantaneous Cost Gradient Search has many parameters than can be tuned to improve performance or adapt the algorithm. Increasing the perturbation width, *c*, and number of breaths measured at each evaluation perturbation point can improve the estimate of the gradient. Large perturbation widths, however, hurt the “local” nature of the gradient estimate. The gain schedule and number of iterations have a more complex effect on the algorithm performance. The gain schedule can be shaped to implicitly assume that the algorithm will start far from the minimum and will approach the minimum with some approximate rate. At early iterations, the evaluation point is assumed to be far from the minimum and the algorithm has the authority to take large steps. At later iterations, the algorithm may be closer to the minimum where the gradient is difficult to estimate and where large steps may cause the algorithm to overshoot. The decrease in gain with increasing iterations helps limit the step size as the algorithm approaches the minimum. ## Scaling to Multiple Parameters Neither the Steady-State Cost Mapping nor the Instantaneous Cost Mapping readily scales to the optimization of multiple parameters simultaneously. A Steady-State Cost Mapping method that uses *N* evaluation points with a single parameter would require *N*^*d* evaluation points with multiple parameters (where *d* is the number of parameters). For example, our Steady-State Cost Mapping was performed over 11 different step frequencies. This required a little over an hour of walking. If we were to attempt to identify the optimal combination of these 11 step frequencies with 11 different values of a related parameter, such as walking velocity, subjects would need to walk for over 12 hours. If it would take 12 hours of walking to map two parameters, one can imagine the difficulty of mapping the combinations of the tens of parameters used in powered prostheses. Using the techniques of the Instantaneous Cost Mapping the parameter values could be varied continuously, but even this method relies on a exhaustive sampling of the parameter space. The response surface will be most accurate near parameter combinations that have been tested. Without a priori knowledge of the minimum location it would be necessary to spread the experimental evaluations over all the possible combinations of parameters. Of our three methods, the Instantaneous Cost Gradient Search is the best suited method for use with high numbers of parameters. The number of required perturbations to fit the local response surface scales only linearly with the number of parameters. In our experiments, each of the 15 evaluations the gradient for the single parameter required about three minutes. If we were to repeat our experiments with multiple parameters, evaluating the gradient with our current methods would require an additional three minutes of walking for each additional parameter. ## Time Constant Sensitivity The Instantaneous Cost methods are somewhat sensitive to errors in the estimate of the subject’s metabolic time constant. For example, when parameters are increased from one extreme to the other in the Instantaneous Cost Mapping, errors in the time-constant will result in an error in the predicted “phase lag” between the respiratory measurements and the underlying change in cost. To compensate for this effect, one could vary the parameter from one extreme to the other and then back again in a single trial. In contrast, the sensitivity of the Instantaneous Cost Gradient Search to errors in the estimation of the time constant is only moderate because errors in the time constant typically only result in errors of the magnitude of the gradient (not the sign of the gradient). When observing respiratory measurements immediately after a step change in the underlying cost, the gradient is essentially estimated by observing the *rate* of change in the breath measurements. This rate is determined by both the difference between the currently measured cost and the steady-state cost (*λ*₁−*y*^*i*−1) and the ratio of the gradient to the time constant ($\frac{\lambda_{2}}{\tau}$). This is apparent when we rewrite in terms of the linear response surface defined in $$\begin{array}{r} {\left( y^{i} - y^{i - 1} \right) = \frac{h_{i}}{\tau}\left( \lambda_{1} + \lambda_{2}p - y^{i - 1} \right) = \frac{h_{i}\lambda_{2}}{\tau}p + \frac{h_{i}}{\tau}\left( \lambda_{1} - y^{i - 1} \right).} \\ \end{array}$$ If there is an error in the magnitude of the time constant, it will result in an error in the magnitude of the estimated gradient, *λ*₂. The effect of an error in the magnitude of the gradient is something akin to an error in the magnitude of the gain, *α*_*k*. This gain does not need to have a precise magnitude for convergence to occur. For this reason, the algorithm did not rely on an individually characterized time constant for each subject. As it turned out, the average of our individually-identified time constants (28 seconds) was lower than the average-time constant of previous studies on which we based the time constant used the Instantaneous Cost Gradient Search (40 seconds). This did not impede convergence. ## Subject Adaptation Compared to the other methods, the Instantaneous Cost Gradient Search is the most robust to subject adaptation. The method relies only on recent measurements. This allows the algorithm to be stopped and restarted at any point. Thus allowing its implementation to be spread out over multiple sessions or days as needed. This property could also be beneficial if subjects are unfamiliar with a particular set of device parameters. This unfamiliarity may cause a subject to use their muscles to work against the assistance provided by the device. Such a suboptimal walking strategy will result in higher measures of energetic cost. In the Steady-State Cost Mapping or Instantaneous Cost Mapping, these higher measures would result in a response surface that is higher in that region. Given enough time, however, subjects might adapt to unfamiliar conditions and walk in an energetically improved way. With the Instantaneous Cost Gradient Search, one can imagine a situation where the algorithm approaches a set of parameter settings that are unfamiliar to the user. The measured energetic cost will be skewed upward by the suboptimal walking strategy employed by the subject. The gradient, however, will not necessarily be affected by this unfamiliarity. Improved parameter setting choices will still result in a *relative* improvement in energetic cost (even when the *absolute* cost is skewed by the unfamiliarity). This relative improvement in energetic cost can still be used to direct the search. # Conclusions In this work, we have demonstrated novel methods to automatically identify parameters of assistive devices via the optimization of a desired physiological objective. To prove the validity of our ideas, we tested the performance of our our algorithms by manipulating the step frequency of eight subjects. These experiments confirmed that our algorithms were able to identify energetic minima (which correspond to subjects’ preferred step frequency) with similar accuracy to the traditional method of Steady-State Cost Mapping. Our methods enable considerable time savings and readily extend to multi-dimensional parameter spaces. To the best of our knowledge, the presented Instantaneous Cost Gradient Search method is the first time that an algorithmic, online optimization of energetic cost was performed with the human body in the loop. summarizes the qualitative differences between our algorithms and traditional methods. The presented step-frequency study provides an interesting proxy for assistive devices. Commanding step-frequency is inherently safe and its simplicity enables an easy implementation. Furthermore, the existing literature on the relation of energetics and step-frequency allows a rigorous assessment of our results. The final goal of our work, however, is the application of the same methods in the tuning of powered prostheses, orthoses, and exoskeletons. When using energetic cost as the physiological objective, the necessary algorithmic adaptions will be minor. Although assistive devices may introduce questions about adaptation and local minima, they have a distinct advantage over optimizing step frequency. Humans are poor “computer-controlled devices”. The algorithms in this work cannot guarantee that the “commanded” step frequency was followed precisely by the subject. In fact, reveals large differences between the commanded and measured frequency. As a consequence, the algorithms were forced to rely on noisy measurements of step frequency. In contrast, with an assistive device the parameters would be firmly under control. In particular, we observed a bias towards walking at the preferred step frequency. It is conceivable that this improves the performance of the mapping methods because the data set becomes denser closer to the optimum (and thus in the range where it is important). We do not believe that this effect improves the Instantaneous Cost Gradient Descent, as this method is trying to evaluate a local gradient rather than estimating the global minimum. The primary challenge in the implementation would be to extend the optimization to multiple dimensions and to guarantee patient safety. As these algorithms develop, they will find increasing use for devices where traditional tuning methods are infeasible. In the long run, we do not see our approach being limited to energetic cost as the objective. Using similar methods, one can, for example, optimize gait symmetry by evaluating kinematic sensor information or enforce a desired training effort by evaluating inverse dynamics or EMG data. While such objectives will require different methods for sensor data processing, the underlying algorithmic principles will be very similar. # Supporting Information We thank the lab of Dr. Deanna Gates for graciously allowing us to use their equipment. Funding for the work of the RAMlab came through the MCubed program at the University of Michigan (CDR). Funding for the work done by the Locomotion Lab was supported by U.S. Army Research Office grant \#W911NF-13-1-0268 (JMD), NSERC Discovery Grant (JMD), and Vanier Canadian Graduate Scholarship (JCS). This material is based upon work supported by the National Science Foundation Graduate Research Fellowship (WF) under Grant No. DGE 1256260. Any opinion, findings, and conclusions or recommendations expressed in this material are those of the authors(s) and do not necessarily reflect the views of the National Science Foundation. [^1]: The authors of this manuscript have read the journal’s policy and have the following competing interests: J. Maxwell Donelan is the founder and scientific advisor to Bionic Power. This does not alter the authors’ adherence to PLOS ONE policies on sharing data and materials. [^2]: Conceived and designed the experiments: JMD WF CDR JCS. Performed the experiments: WF. Analyzed the data: WF. Contributed reagents/materials/analysis tools: JCS WF. Wrote the paper: WF CDR. Edited the manuscript: WF CDR JCS JMD. Figure design: WF CDR JCS JMD.

# Introduction Bone metastases are the main cause of cancer-associated death in patients with cancer and represent one of the most common sites of metastasis.Breast, prostate, and lung cancers collectively represent the primary tumors that more frequently metastasize to bone. Furthermore, considering that a reduced bone matrix is found in patients suffering from breast and prostate cancers treated with hormonal therapies, bone metastasis grows into a compartment bone that is itself already weakened. Among the potential complications related to bone metastasis, pain is the symptom most frequently referred by patients resulting in a drastic deterioration of their quality of life (QoL).For patients with overt bone metastases, current treatment objectives are designed to decrease tumor burden, reduce skeletal-related events and maximize pain control. Current management of skeletal metastasis includes pain management/analgesia, systemic therapy, radiation therapy (RT), surgery, and ablative techniques. However, the long- term-results of many of these treatment modalities may be not fully satisfying. Thermoablation is a family of non-surgical approaches used to treat otherwise unrespectable tumors and using different power sources. Various delivery methods of ablation exist. They include radiofrequency ablation (RFA), microwave ablation, high intensity focused ultrasound (HIFU), laser ablation, and cryoablation (CA). These techniques, initially developed for patients with malignant primary or metastatic liver tumors, have proven to be useful therapeutic options for the management of bone tumors. However, these ablative approaches have rarely been used in combination in with RT. Recently has been demonstrated that the association of RFA with RT was well tolerated, with a satisfactory profile of adverse events.. In this study, for the first time, we investigated whether the sequential addition of RT to CA could favorably affect clinical management of painful bone metastatic lesions compared with CA and RT delivered as individual treatments. # Materials and Methods ## Patient selection In this study, patients older than 18 years with radiological and histological confirmed painful solitary bone metastases were prospectively selected to undergo CA procedures followed by RT. Magnetic resonance imaging (MRI) examination, computed tomography (CT), and nuclear isotope (technetium 99) bone scan were planned within the 4 weeks prior to the procedures. Other eligibility criteria included: (i) a pain score of 5 or more on the validated visual analog scale (VAS) over the prior 24 hours (or a score of less than 5 with the use of narcotic medications); (ii) pain localized to the site of the bone metastases; (iii) life expectancy of greater than 3 months; (iv) and Karnofsky performance status (KPS) score of greater than 70. Exclusion criteria were: (i) a painful area previously treated with RT or palliative surgery; (ii) radiographic evidence of spinal cord or cauda-equina compression; (iii) lesions positioned within 0.5 cm from a critical structure such as the spinal cord, brain, aorta, inferior vena cava, bowel, or bladder; (iv) abnormal fracture of the treatment site. The combined treatment was as arranged as follows: CA followed by RT 15 days later when technically feasible. Between September 2011, and April 2014, 41 consecutive patients receiving CA followed by RT were treated. Sixteen out of 41 subjects (39%) were excluded since (i) refused to participate in the study (3/41; 7%), (ii) they did not satisfy the inclusion criteria (5/41; 12.2%) and (ii) they did not match with any subject included in the groups treated by CA or RT after propensity analysis (8/41; 19.5%). The group of patients undergoing the CA-RT was retrospectively compared and matched by propensity analysis with a group of subjects treated with RT or CA. The subjects included in RT or CA groups were retrospectively selected from all who were treated in the Radiotherapy and Interventional Radiology units, respectively. Analgesic consumption of all patients was recorded. Written informed consent was obtained from the subject and the study was approved by the San Salvatore Hospital IRB (Deliberation n. 89/2012) ## Treatment procedures RT was delivered by three-dimensional conformal technique with a dose of 20 Gy in five fractions of 4 Gy over 1 week. Percutaneous cryotherapy ablation was carried out under conscious sedation with an argon-based cryotherapy system (SEEDNET GOLD Cryoablation System, Galil Medical LTD, Israel). After sterile preparation, one or more cryoprobes (ICEROD, Galil Medical LTD, Israel) were introduced into the target lesion by CT guidance by experienced radiologists. The cryoprobes were introduced in a parallel arrangement approximately 2 cm apart. For larger lesions, a cluster of cryoprobes was placed within 1 cm of the tumor margin to provide adequate coverage of the outer border of the target lesion. Cryoprobe positioning was confirmed by CT imaging. Rapid freezing of the target lesion (-100°C within a few seconds) was obtained and two 15-minute freezes separated by a 10-minute thaw were used at each cryoprobe position. At the end of each procedure, CT was performed to ensure that the extent of ablation was confined to the target tissue and that there was no substantial damage in the tissue surrounding the target. After all combined procedures, patients were observed for 2 hours in the recovery room and then were admitted to the hospital for a minimum of 24 hours. ## Patient assessment A full physical examination was performed and data on direct and indirect changes in pain levels were assessed by VAS and medication level questionnaire. QoL was assessed with a single question from the McGill Quality of Life Questionnaire (MQOL).The strongest evidence of validity comes from comparison with the single-item quality of life measure. ## Study endpoints and response criteria The primary endpoints were the percentage of patients with a (1) complete response (CR) at 12 weeks after treatments as previously described. The secondary endpoints were (1) the rate of subjects requiring analgesics at 12 weeks after treatments and (2) the changes in self-experienced QoL. Complete response was defined as a pain score of 0 at treated site with no concomitant increase in analgesic intake (stable or reducing analgesics in daily oral morphine equivalent \[OMED\]). Partial response was defined as pain reduction of 2 or more at the treated site on a scale of 0 to 10 scale without analgesic increase, or analgesic reduction of 25% or more from baseline without an increase in pain. Pain progression was defined as increase in pain score of 2 or more above baseline at the treated site with stable OMED, or an increase of 25% or more in OMED compared with baseline with the pain score stable or 1 point above baseline. Stable-intermediate response was defined as any response that is not captured by the complete response, partial response, or pain progression definitions. ## Statistical analysis The data analyzed in this report were derived from a population-based observational study. In order to reduce treatment selection bias and determine realistically the treatment effects, a case control-matched propensity analysis was performed. Pairwise nearest neighbor matching with a caliper was used to minimized distance within matched sets applying a caliper of 0.1 on the propensity score scale. Increasing the number of controls included in each matched set may increase the precision of the estimated treatment effect. The optimum number of controls needed to match to each treated subject may range from 1 to 5. Increasing the number of controls matched to each treated subject will increase the size of the matched sample, resulting in estimates of treatment effect with increased precision. Multivariate logistic regression was used to calculate the predicted probability of the dependent variables as well as the propensity score for all observations in the data set. The dependent variables included in the multivariate analysis were age, KPS, primary tumors, anatomic site, VAS scale, and QoL before procedures. A 1:5 matched analysis was performed where one case (subject treated by CA or CA followed by RT) was matched to five controls (subjects treated by RT). The primary endpoint of this feasibility study was that, for patients with painful bone metastasis, pain relief achieved following CA-RT in terms of CR should be higher than that achieved following RT alone. The current study was powered to determine an increase of 21% or greater in the CR at 12 weeks after CA-RT with respect to RT alone. The literature indicates that from 11 to 21% of intention-to-treat (ITT) patients achieved CR after RT. Thus, we set the rate of CR after RT at 11% (P0 = 11%). Using a two-sided test and a 5% type I error adjusted for Bonferroni correction (p\<0.0166), with the matched control to case ratio of 1:5, 25 subjects in the experimental groups (CA group and CA-RT group) and 125 in the control group (RT) would provide greater than 80% power to detect an increase of 21% (P1 = 32%). Continuous variables not normally distributed (Shapiro-Wilk test) were presented as medians and 95% confidence intervals (95%CI). The Mann-Whitney U test was used to evaluate the difference between two groups and Kruskal-Wallis test was used to evaluate the difference among more than two tests. If the Kruskal-Wallis test was statistically significant, apairwise comparison of subgroups was performed according to Conover. Dichotomous variables were summarized by absolute and/or relative frequencies. The chi-squared test or Fisher’s exact test was used to evaluate the difference between two groups. For multiple comparisons, the alpha value threshold was adjusted by using Bonferroni correction. All tests were two-sided except where specified and were determined by Monte Carlo significance. An alpha value threshold of 0.05 was used. All statistical analyses were performed using the SPSS statistical analysis software package, version 10.0 (IBM Corporation 1 New Orchard Road, Armonk, New York 10504–1722 United States). # Results ## Local pain control and self-perceived QoL lists the clinical and demographic characteristics of treated patients stratified according to propensity score and treatment received. All subjects were valuable at the end point of 12 weeks. This was mainly due to the selection criteria (solitary bone lesion) which configure this group of patients as long survivors. A higher proportion of subjects treated by CA (32% \[8/25\]; p = 0.018) or by CA followed by RT (72% \[18/25\]; p\<0.0001) experienced a CR at 12 weeks with respect to patients treated by RT alone (11.2% \[14/125\];). Interestingly, the addition of RT to CA significantly improved the rate of CR with respect to CA used as individual treatment (p = 0.011). A PR was more frequently observed in patients treated by RT (42.4% \[53/125\]) or CA (36% \[9/25\]) with respect to subjects treated by CA followed by RT (12% \[3/25\]). The higher rate of CR observed in patients treated with CA followed RT with respect to those treated with CA or RT paralleled an improved self-rated QoL. At 12 weeks, patients treated by CA followed by RT reported a higher improvement in self-rated QoL (MQOL score: 7, CI95% 5.4 to 9) with respect to subjects treated by RT alone (MQOL score: 5, CI95% 4 to 5). At the same time, patients treated by CA alone experienced a significant improvement of self-rated QoL (MQOL score: 6, CI95% 5 to 8) with respect to subjects treated by RT alone (MQOL score: 5, CI95% 4 to 5). Finally, although an improvement in the self-rated QoL between CA and CA-RT was observed, the difference did not reach statistical significance. All patients received oral narcotic analgesic over the month before treatment with CA, RT, or CA followed by RT.In this regard, at 12 weeks, 13.6% of patients (17/125) in the RT group, 36% of patients (9/25) in the CA group, and 76% of patients (19/25) in the CA-RT group did not require narcotic medications. This figures paralleled with the reduction in the 24-hours median morphine-equivalent dose. Before treatments the 24-hours median morphine-equivalent dose was 80 mg (95%CI 75–80) in the RT Group, 70 mg (95%CI 65–80) in the CA Group and 80 mg (CI95% 65.7–80) in the CA+RT Group with no significant difference among groups (p = 0.67). A significant decrease in the 24-hours median morphine-equivalent dose at week 12 was documented in the CA (p = 0.008) and CA-RT (p = 0.004) groups with respect to the RT group. Overall, patients tolerated the CA and combined treatments well. Forty (80%) out of the 50 patients performed CA without complications. The complications were listed in. # Discussion Radiation treatment is one of the most important non-surgical approaches for the management of osteolytic or osteosclerotic bone lesions with sub-optimal overall response rate. Interventional radiologists developed different ablative approaches, which demonstrated outstanding effects in terms of clinical efficacy in a wide range of clinical scenarios including bone metastatic disease. However, a number of patients may respond unsatisfactorily to these treatments in terms of pain control. Recently, the management of painful bone metastases was effectively achieved by combining ionizing radiation with RFA. These encouraging preliminary results induced us to study if the addition of RT to CA would result in improved clinical efficacy. To our knowledge, no existing empirical study has addressed the question of whether percutaneous CT-based CA followed by RT achieves better pain control and improved self-rated QoL than CA or RT delivered as individual treatments in the management of osteolytic and/or mixed painful bone lesions. It is well-known that the localization of the bony metastases may affect the way subjects perceived their symptoms. This is an important methodological point since an heterogeneity the localization of the bony metastases may configure as a study bias. This bias was overcome by using propensity analysis and a balance among groups for this variable was achieved. Cryotherapy is a well-known ablative technique which, thanks to new miniaturized argon-gas devices, creates an “ice-ball” deep in bone, with less peri- and post- procedural pain. Callstrom and co-workers described the results of clinical trials in which percutaneous CA was used for the palliation of painful metastatic lesion. Interestingly, they found that CA was an effective and safe way for palliation of pain related to metastatic lesions. Although authors did not specifically investigate this issue, they found that 62% of studied subjects received RT course prior to CA. In these subjects no improvement in pain response with respect to subjects who did not perform RT was found. In contrast, we found that when RT was delivered after CA, it favorably impacted on pain control in subjects with painful bone lesions. The main methodological difference with Callstrom’s study is that in ours, the radiation course was added sequentially to CA. The temporal sequence of two treatments was based on the hypothesis that RT may improve CA efficacy. Recent studies have reported achieving a synergistically increased rates of survival and tumor local control in both animal models and in patients with stage I lung cancer, when RFA was followed by RT. We know that RT efficacy is largely dependent on tissue oxygenation, whereas this biological parameter does not influence CA efficacy. Bone lesions have a hypoxic microenvironment in their inner cores and CA, whose effectiveness is largely independent of tumor oxygen content, may effectively kill hypoxic centrally located tumor cells. In contrast, the outer ring contains fewer hypoxic tumor cells on which RT can effectively exert its cytotoxic effect. Additionally, it is possible that perturbation of the tumor microenvironment after ablation may improve the efficacy of subsequent RT on the malignant tissue. However, the exact mechanisms through which CA followed by RT may relieve symptomatic pain remains largely unknown. The pain experienced by subjects suffering from bone metastases may have several pathophysiologal mechanisms, including: (i) microfractures by direct metastatic bone invasion; (ii) periosteum distortion due to increased pressure on the endosteum; (iii) nerve-root compression or muscle spasm; and (iv) release of chemical mediators involved in the conduction of nociceptive impulses to the central nervous system. The decrease of cancerous cell burden within bone tissue associated with a lowering of endosteum pressures and modification of the release of pain-related chemical mediators related with the combinational treatment may substantially influence the perception of pain intensity. The improvement of CR rate we found, in terms of pain control, showed that 11.2% of subjects treated by RT achieved a CR. The rate of CR was significantly improved with respect to RT in subjects treated by CA (32%) and interestingly, when RT was sequentially added to CA, the percentage of patients achieving a CR was 72%. These data paralleled the improvement of self-rated QoL observed in subjects treated by CA and CA followed by RT with respect to subjects treated by RT alone. When RT was sequentially added to CA the subjects experienced improved QoL, although the difference did not reach statistical significance. In terms of medical adjunctive treatment, all subjects took narcotic medications before treatments. At 12 weeks post treatment, 13.6% of patients in the RT group, 36% in the CA group, and 76% in the CA-RT group did not require narcotic medications and the 24-hour median morphine-equivalent dose significantly decreased with respect to baseline, with the major decrease found after CA or CA followed by RT. Finally, although the rate of complications we reported seems to be higher than that reported by Calstrom, the combinational treatment was well tolerated since 84% of patients were ablated without complications. The recorded complications were all transient and none of treaded subject presented irreversible injures at 12 weeks after treatments. Several limitations affect our study. The main ones are the sample size and the use of a non-randomized study design. To date, large randomized controlled trials (RCT) have provided the strongest evidence for the efficacy of therapeutic procedures or treatments in the clinical setting. However, this bias has been mitigated by the use of a strategy based on propensity score analysis, which helped us to obtain groups of patients randomized post-hoc for important clinical characteristics. Thus, comparative analysis by propensity-matched pairs contributed to the results being less prone to methodological biases than other usual statistical methods. Finally, the positive results obtained in terms of both QoL and pain control may be partially imputable to the adjunctive oncological medical treatments that the subjects included in this study performed concomitantly with RT, CA and CA+RT. If these medical treatments have contributed to the effects measured in this study is unclear. In this regard, a strategy of assessment based on the measurement of outcome measures before the start of adjunctive oncological medical treatments may not ensure that the pain or the QoL perceived at that moment may be similar to that experienced when RT, CA or CA+RT was then effectively delivered. This situation is imputable to the fact that these medical treatments may go on for several weeks and during this time period the symptoms may profoundly change. However, we believe of having minimized this issue by adopting the propensity score analysis. In this regard the percentage of subjects using oncological systemic adjuvant treatments did not significantly differ across the groups making the confounding effect of this variable on outcome measure balanced in the groups. So, if we reasonably assume that at the time of local treatment with RT, CA or CA+RT the rate and the type of oncological medical treatments were comparable across the groups any improvement measured in pain or QoL was imputable to the aforementioned local treatments. Different may be the question concerning the fact if one or more of these oncological medical treatments may have potentiated the effects of RT, CA or CA+RT. It is very difficult to answer to this question since we cannot be sure which treatment is responsible for the enhancement of local treatments effects. All these reasons make our conclusions not equally generalizable to the population of patients suffering from solitary painful bone metastases who are both treated or not treated with multimodal oncological medical treatments. It is our opinion that the results here described are better suitable for the larger population of patients who is actively treated with multimodal oncological medical treatments. Despite these methodological limitations our data seem to suggest, for the first time, that the addition of sequential RT to CA favorably impacts on the pain of painful bone metastases. However our results have to be interpreted with caution and to serve as a framework around which to design future large-scale RCT. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MDS GLG LZ PB LG PF FM FV RM VT EDC CM. Performed the experiments: MDS GLG LZ PB LG PF FM FV RM VT EDC CM. Analyzed the data: MDS GLG LZ PB LG PF FM FV RM VT EDC CM. Contributed reagents/materials/analysis tools: CM EDC VT GLG. Wrote the paper: MDS GLG LZ PB LG PF FM FV RM VT EDC CM.

# 1. Introduction In 2019, it was estimated that 75 million people in the world require a manual wheelchair (MWC). MWC users daily face physical environmental barriers such as slopes, cross-slopes, curbs, and uneven terrain that affect their access to buildings and urban areas. Yet, accessibility for people with disabilities is crucial for their social and professional integration. Standards and regulations have been established to impose some architectural rules to make public buildings and squares accessible to everyone. However, the regulations are mainly based on the aspects of required space and maximum slope inclination. Despite the improvement of the overall accessibility of public areas, these regulations remain unsatisfactory for a large proportion of MWC users. The limitations imposed by environmental barriers in MWC locomotion can be described using the International Classification of Functioning, Disability, and Health (ICF). The ICF is a framework for describing “dynamic interactions between a person’s health condition, environmental factors, and personal factors”. The ICF can therefore be used to identify key elements that need to be addressed in rehabilitation, to guide the classification of assistive technology, or even to determine the relationship linking wheelchair skills and capabilities with participation frequency and mobility. From that, previous studies have, in particular, revealed the need for better training in overcoming environmental barriers. In addition, the ICF framework could be used by clinicians to adapt MWC training programs according to their patients’ capabilities and life projects. To that end, it appears necessary to be able to associate a barrier’s difficulty with the user’s capabilities. This could be achieved by the quantification and comparison of the physical demands associated with the various environmental barriers encountered. Biomechanical analysis of locomotion is a reference method to investigate physical demands associated with MWC locomotion. Such biomechanical analysis classically includes the quantification of joint motion and intersegmental loads (forces and torques). Thus, several studies have investigated the physical demands of MWC propulsion when crossing various environmental barriers from a biomechanical point of view. Illustrations of environmental barriers that were recreated in a laboratory to that end can be found. However, in general, only one type of barrier was investigated in each study, and it appears that no study investigated more than two types of obstacles, hindering the comparison of results between barriers. Moreover, studies seem to use a variety of experimental protocols and investigated different biomechanical parameters. For these reasons, researchers may encounter difficulties when looking for concise data on the influence of environmental barriers on a biomechanical evaluation of MWC locomotion. To address this gap, the purpose of this study was to identify and synthesize data and experimental methods from the literature on the biomechanics of MWC propulsion for various and frequent environmental barriers that are encountered daily by MWC users. # 2. Methods The present study conducted a systematic review to identify and analyze existing studies that reported biomechanical parameters of MWC propulsion while overcoming environmental barriers. Because handrim propulsion is the most frequent system of manual propulsion adopted by MWC users due to its higher compliance with the constraints of activities of daily living indoors, the review focuses on the biomechanics of manual handrim propulsion. ## 2.1 Systematic literature review To answer the question: “What are the biomechanics involved to overcome specific environmental barriers?”, a systematic search was performed based on the methodology of Harris et al. and Moher et al. to identify relevant articles published until May 2021 within the Pubmed and Scopus database. The request, launched on May 3, 2021, focused on biomechanical parameters and especially on spatio-temporal parameters, kinematics, kinetics, and muscle activations during MWC propulsion to overcome environmental obstacles, as well as on the experimental methods used to obtain the aforementioned parameters. More precisely, the request was: *(bioengineering OR biomechanic*\* *OR kinematic*\* *OR velocity OR velocities OR (joint angle*\**) OR kinetic\* OR force*\* *OR torque*\* *OR moment*\* *OR (motion capture) OR electromyography) AND wheelchair AND (propulsion OR slope OR kerb OR curb OR ground OR floor OR rolling resistance OR activities OR activity OR ambulation OR locomotion OR situation)* The keywords used for this search were determined after reviewing the results of a preliminary search, which had identified the four most studied in the literature: slope, cross-slope, curb, and ground type. ## 2.2 Article selection Articles were selected following the flow diagram recommended by the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA). After eliminating duplicates, all titles were screened for inclusion by three of the authors. The inclusion criteria were: original study or systematic review; study written in English; and features experimental results on slopes, cross-slopes, curbs, and ground types during MWC locomotion. Exclusion criteria were articles about electric wheelchairs, power-assisted wheelchairs, sports wheelchairs, other propulsion systems than manual handrim, and hemiplegia-pattern propulsion. All other abstracts and articles were screened by the same authors. The articles on subject-based studies dealing with an environmental barrier were selected and then sorted according to the barrier type: slope, cross-slope, curb, and ground type. For the analysis, biomechanical parameters were divided into four *a priori* defined categories: spatio-temporal parameters (push time, recovery time, cycle time, speed, etc); kinematics (joint angles); kinetics (handrim forces and torques, rate of rise, fraction of effective force, net joint moments, mechanical work and power, etc); and muscle activity. A more detailed definition of these biomechanical parameters can be found in. # 3. Results The first search resulted in a total of 1429 references, and 1093 articles remained after removing duplicates. The screening through the title filter resulted in 266 references. After reading the abstracts, 59 articles were selected, and finally, 34 papers were included in this review after the full text read. This selection process is summarized in. The 34 retrieved articles included populations between 7 and 128 participants (Total: 756, Mean \[M\]: 22, standard-deviation \[SD\]: 25). Cohorts included able-bodied (AB) subjects and MWC users (MWU), among whom spinal cord injured (SCI) subjects, subjects with lower limb amputation, cerebral palsy, neuropathy, or Friedreich’s Ataxia. AB and SCI subjects were studied in 10 and 22 studies, respectively. Experimental design, acquisition methods, and measurement tools were also found to differ between studies. The MWC was propelled overground, on a treadmill, or over a stationary ergometer. Kinematics were recorded either with motion capture systems, inertial measurement units, video cameras or optical encoders. Kinetics were systematically recorded with instrumented wheels. An overview of the retrieved studies is provided in. A subsection dedicated to each investigated environmental barrier (slope, cross-slope, curb, ground type) summarizes the experimental methods used in these studies (also reported in Tables –) as well as the obtained biomechanical results. A compilation of the detailed numerical results of the studies is appended as supplementary material. ## 3.1 Slope ### 3.1.1 Methods on slopes Twenty-five articles investigated MWC propulsion on a slope, all during slope ascent. The number of participants ranged between 7 and 128 (M: 23, SD: 29) and the studied populations were mostly MWU (SCI or other motor disabilities). Experimental design differed across studies, both in terms of the propulsion experimental environment (overground, treadmill, or on a stationary ergometer) and the slope (grade mostly between 2° and 5° but could reach up to 15°). Similarly, the acquisition methods and measurement tools were not consistent between studies. Kinematics recording was most often based on opto-electronic motion capture systems, but also on systems based on inertial measurement units, simple 2D cameras, or optical encoders. Kinetics were always measured through instrumented wheels (six-component dynamometers), generally mounted on one side only. One study investigated the kinetics of both wheels using only one instrumented wheel mounted separately on the right and left sides in different trials. Two of the ten studies which used only one instrumented wheel reported having mounted a matching "dummy" wheel on the opposite side to ensure inertial symmetry. Outcome measurements included spatio-temporal parameters (*e*.*g*. MWC mean velocity, cycle frequency, push and recovery phases durations, etc.), kinematics (glenohumeral, elbow, neck, and trunk angles), handrim kinetics (tangential, radial, and total forces; fraction of effective force, mechanical work and power), joint kinetics (shoulder net joint moments and glenohumeral joint contact force), and muscle activity (percentage of maximal voluntary isometric contraction). ### 3.1.2 Results on slopes *3*.*1*.*2*.*1 Spatio-temporal parameters*. Under uncontrolled conditions (*i*.*e*. overground), MWC speed was found to decrease with increasing slope. Contradictory results were obtained on cycle frequency: MWU tended to increase their cycle frequency with slope on a long ramp, whereas AB decreased cycle frequency with slope on a short ramp. Moreover, when the MWC speed was constant across the different slope inclinations (speed imposed by the treadmill belt), cycle frequency tended to increase with increasing slope in SCI subjects, but was not affected with AB subjects. Push phase duration at the reference level (*i*.*e*. grade = 0°) was similar in all studies that reported this information. When the speed was imposed (*i*.*e*. on a motorized treadmill), the push phase duration was not modified by slope. On the opposite, in overground and stationary ergometer studies, where speed was self-selected, push phase duration increased with the grade. All studies reported a decrease in recovery phase duration with the increase of slope inclination. Seven studies reported data on contact angles. Four of these studies used treadmills but investigated different populations, namely AB and MWU, and highlighted significant differences between those populations in contact angle even on a zero grade slope: contact angle was higher on the same slope when experimenting on AB subjects, and seemed to remain constant with different grades of slope in AB subjects, whilst contact angle tended to decrease with increasing slope in MWU *3*.*1*.*2*.*2 Joint kinematics*. Important differences can be noted between studies in all degrees of freedom (DoF) of the glenohumeral joint. In particular, the evolution of the glenohumeral flexion-extension range of motion (RoM) with the grade differed with either an increase, no observed change, or even a decrease for AB users in one study. On the contrary, results on trunk inclination are in agreement between studies with an increase of trunk flexion- extension RoM with the grade. An increase of the neck extension with the grade, consistent with the increase of the trunk extension to keep the gaze orientation, was also observed. Wrist flexion-extension and radio-ulnar deviation RoM also tended to increase, as well as elbow flexion-extension and pronation-supination RoM. Finally, one study reported maximal scapular angles (down-up, antero-posterior, and internal-external rotations), showing a decrease in maximal downward and anterior rotations, and an increase in internal rotation with increasing slope. *3*.*1*.*2*.*3 Handrim and joint kinetics*. Results on handrim kinetics show noticeable differences between studies when compared to similar or close grades. However, evolution with grade was consistent between studies with an increase of both mean and peak total force, as well as of its tangential and radial components. The handrim mechanical work and power also increased with the grade. Results on the fraction of effective force were however less clear with a mean value that tended to slightly decrease, be maintained, or increase. Few and disparate outcome data were provided on joint kinetics during slope ascent. An increase of the mean and peak glenohumeral net joint moment and the peak elbow net joint moment with slope was however reported. Two studies reported data on glenohumeral joint contact forces, which require the assessment of muscle forces through a musculoskeletal model, and found a significant increase of the three components of this force with the slope grade. *3*.*1*.*2*.*4 Muscle activity*. Most studies reported peak EMG value, but five studies reported mean EMG activity during propulsion. Although most studies reported normalized muscle activity using maximum voluntary contraction testing, one article reported un-normalized EMG activity as voltage measured by the sensor. The muscles investigated in the studies were often different, although most studies measured the muscle activity of the anterior deltoid and pectoralis major. On equivalent slopes, the different studies gave different values of normalized muscle activity for these two muscles. However, it was observed that muscle activity of all of the studied muscles was found to consistently increase with the grade. Some studies reported muscle activity during locomotion higher than the one observed during maximum voluntary contraction testing for some subjects. ## 3.2 Cross-slope ### 3.2.1 Methods on cross-slopes Four articles studied cross-slope propulsion. Seven to twenty-five (M: 14, SD: 8) MWU—mainly SCI subjects—took part in these experiments. Trials were performed overground, or on a treadmill, always at self-selected speeds. Cross-slope inclination ranged between 1.4 and 6°. Cross-slope length was only reported in one study (7.2 m). Kinematics recording was based on an opto-electronic motion capture system or on an inertial measurement unit-based system. Kinetics were systematically measured through a six-component instrumented wheel. The downhill side was systematically measured, with only one study reporting using a dummy wheel, and only one study equipping both wheels. EMG activity of the downhill side was recorded in two studies and focused on three muscles: the pectoralis major, the anterior deltoid, and the infraspinatus. Outcome data were spatio-temporal parameters (MWC speed, cycle frequency, push and recovery phase duration, contact angle), handrim kinetics (tangential and total handrim forces, fraction of effective force, propelling torque, mechanical work, and mechanical power), shoulder joint kinetics (glenohumeral joint contact force) and muscle activity (peak and/or mean of the percentage of maximal voluntary isometric contraction). One study compared the kinetics at the dominant and non-dominant hand sides, while the MWC’s right wheel was downside, without investigating the effect of the side of the dominant hand (two participants left-handed). ### 3.2.2 Results on cross-slopes *3*.*2*.*2*.*1 Spatio-temporal parameters*. The only study reporting data across different grades of cross-slopes showed a decrease of the speed, an increase of the cycle frequency (i.e. decrease of the cycle duration), an increase of the push phase duration, and a decrease of the recovery phase duration with increasing slope. Contact angles on the downhill side did not appear to be affected by the grade of the cross-slope. *3*.*2*.*2*.*2 Joint kinematics*. The only study investigating body kinematics during cross-slope propulsion found an increase in downhill glenohumeral flexion/extension and internal/external rotation RoM compared to level-ground propulsion. On the contrary, downhill glenohumeral abduction/adduction RoM decreased on the cross-slope and trunk flexion/extension RoM tended to increase only in SCI subjects (and not in AB subjects). *3*.*2*.*2*.*3 Joint and handrim kinetics*. Peak and mean total forces were shown to increase with increasing grade of the cross-slope or compared to level- ground. The propelling torque on the downhill wheel as well as the mechanical power of this torque were also increased with the grade of the cross-slope. The downhill glenohumeral joint contact force, assessed through a musculoskeletal model, was increased by the cross-slope with respect to level ground in every direction (posterior, superior, medial, and total). *3*.*2*.*2*.*4 Muscle activity*. Finally, results on downhill side muscle activity based on EMG data showed an increase of mean muscle activity for all investigated muscles during propulsion on a cross-slope compared to level ground for AB and SCI populations; with an increase of peak muscle activity for the anterior deltoid and pectoralis majors, and a decrease of peak activity for the infraspinatus muscle in SCI participants. ## 3.3 Curb ### 3.3.1 Methods on curbs Two studies investigated curb ascent with a MWC, involving five and fifteen SCI participants. Curb height ranged from four to twelve centimeters and curbs were negotiated overground with momentum. Initial instantaneous MWC speed at the beginning of the curb ascent was not reported in any publication. Kinematics measurements were performed through an opto-electronic motion capture system in both articles but with a small number of cameras for both (less than four). Handrim kinetics were measured using a six-component instrumented wheel, either on one or on both sides. It was not reported whether a dummy wheel was used to equilibrate the MWC when only one instrumented wheel was mounted. EMG data were recorded in one study and focused on four muscles: biceps, triceps, pectoralis major, and anterior deltoid muscles. Outcome data were trunk inclination and upper-limb joint angles (shoulder, elbow, and wrist joints), upper-limb net joint moments (shoulder, elbow, and wrist joints), and muscle activity. ### 3.3.2 Results on curbs *3*.*3*.*2*.*1 Joint kinematics*. Reported results on kinematics showed an increase in the RoM of the shoulder and elbow joints with increasing curb height. In general, this increase was related to an increase of the maximal angle value or a decrease of the minimal value of the angle only. The shoulder internal-external rotation RoM was noticeably increased in both the internal and external rotation ranges. Changes in the wrist RoM remained limited in spite of a slight increase of the peak flexion angle. Finally, the trunk inclination was also modified by the curb height with an increase of the RoM and a noticeable increase of the trunk flexion. *3*.*3*.*2*.*2 Joint and handrim kinetics*. Regarding results on net joint moments, both studies found consistent results for peak total shoulder and elbow moments at high curb level (*i*.*e*. 10 and 12 cm). Furthermore, peak and mean net shoulder moments were increased for all three moment components, but more especially for the flexion and internal rotation moments. At the elbow, there was also an increase in the total net joint moment, lower than that of the shoulder. The flexion component was the most affected. At the wrist, the increase with curb height was also more limited than at the shoulder and the elbow. The extension and radial deviation components were the most affected. Comparison between joints showed that the higher the initial moment value (*i*.*e*. at a curb height of four centimeters), the higher the increase. It can also be noticed that extremely high variability (*i*.*e*. standard deviation) was found in upper-limb joint kinetics. *3*.*3*.*2*.*3 Muscle activity*. Finally, regarding muscle activity, all four muscles were found to increase their activity with curb height. The biceps brachii and the anterior deltoid muscles appeared to be the most involved between the four studied muscles. Very high variability was also found on these outcome variables. ## 3.4 Ground type ### 3.4.1 Methods on ground types Twelve studies investigated the influence of various ground types on MWC propulsion. The experiments were conducted on MWU populations ranging from eight to 128 participants (M: 31, SD: 36), among which most were SCI participants. Indoor ground types were mostly studied and one study investigated grass and pavers. Kinematics were recorded using an opto-electronic motion capture system or inertial measurement units. Kinetics were recorded using instrumented wheels mounted on both sides of the MWC or one side only. It was not reported if a dummy wheel was also mounted when only one instrumented wheel was used. Muscle activity was recorded using EMG. Outcome parameters included the spatio-temporal parameters of propulsion (speed, stroke frequency, push phase duration, contact angle), handrim kinetics (tangential, radial, and total handrim forces, fraction of effective force, propelling torque, mechanical work, and power), and EMG data expressed in percentage of maximal voluntary contraction for normalization purposes, or directly as measured in voltage. ### 3.4.2 Results on ground types *3*.*4*.*2*.*1 Spatio-temporal parameters*. Results showed that self-selected speed was the highest on smooth concrete, tile, and paved grounds, whereas it was the lowest on high-pile carpet, polyfoam mat, grass, and wood grounds. Stroke frequency was the highest on concrete, grass, and paving. High-pile carpets seemed to induce a decrease in speed compared to low-pile carpets, and so did aggregate concrete compared to smooth concrete. In one of two studies, a decrease of stroke frequency was also reported between high-pile and low-pile carpets, while in general, similar stroke frequencies were reported for carpet and tile. *3*.*4*.*2*.*2 Joint kinematics*. Regarding the kinematics of upper limbs, results indicated an increase in the RoM of the shoulder, elbow, neck, and trunk during locomotion on a polyfoam mat compared to locomotion on tiles. *3*.*4*.*2*.*3 Joint and handrim kinetics*. The reported results on handrim kinetics showed that propulsion on smooth concrete, tile, and linoleum resulted in the lowest values in peak and mean handrim forces, propelling torque, as well as output work and power. Propulsion on low-pile carpet also presented low values in handrim forces, propelling torque, and output work and power. High- pile carpet, aggregate concrete, polyfoam mat, pavers, and grass were the most constraining ground types with high values in peak, mean, and rate of rise handrim forces, propelling torque, and output work and power, with grass propulsion having the highest of these values. Fraction of effective force was reported in two articles only, and showed propulsion asymmetry between the subjects’ dominant and non-dominant sides and presented a high variance among subjects; it was the lowest on smooth concrete, and the highest on grass, as well as generally high on ground types that present higher values in handrim forces and propelling torque. *3*.*4*.*2*.*4 Muscle activity*. Lastly, regarding muscle activity, an increase of the mean activity was found for the anterior deltoid and the triceps brachii from abrasive tile to padded carpet, while similar to decreased voltage values were found from linoleum to carpet for these muscles in. Muscle work was also found to double for the anterior deltoid from tile to padded carpet. # 4. Discussion ## 4.1 Investigated environmental barriers Four different barrier types representing obstacles encountered daily by MWC users were considered and investigated in the literature: slopes; cross-slopes; curbs; and ground types. Among these four barrier types, the slope has been studied the most, always during the ascent, while cross-slopes and curbs (ascent only) were scarcely studied. Yet, the study of curbs and cross-slopes appears particularly relevant since they require specific propulsion strategies. It should be noted that differences in the biomechanics of the uphill and downhill sides during cross-slopes were not investigated. Out of the thirty-four retrieved studies, nine investigated multiple barriers at once—albeit not more than two. The scarcity of studies on cross-slopes and curbs diminishes the strength of the conclusions drawn by these studies. Indeed, a larger number of studies may have demonstrated contradictory results, as is the case for the retrieved studies on slopes (due to different experimental setups, processing, or populations). The discrepancy of focus between slopes/ground types and curbs/cross-slopes cannot possibly be explained by the lack of cross- slopes or curbs encountered during MWC locomotion in urban areas, since the uneven ground usually encountered may present such environmental barriers, albeit of low grades. Similarly, descending slopes and curbs, or technically challenging situations such as crossing a door threshold with or without a ramp deserve to be studied. For some of these environmental situations, a task analysis could also be considered by separating start-up, propulsion, braking, and turning. Future studies should therefore be conducted on several different environmental barriers simultaneously, with a special focus on the reproduction of the environments and tasks that are encountered daily by MWUs. Indeed, measuring spatio-temporal parameters, kinematics, kinetics, and muscle activity using the same methods for all barriers would allow the identification of a set of parameters reflecting the difficulty of any environmental barrier encountered in daily MWC locomotion. Furthermore, to allow for comparison of results between studies, the experimental methods and protocols must be clearly defined and explained. Indeed, the speed of the MWC when approaching a curb strongly influences curb negotiation. Similarly, muscle fatigue may impact how the different barriers are approached, and especially curbs and cross-slopes. Consequently, future research should focus on the standardization of protocols and experimental methods regarding MWC locomotion. ## 4.2 Experimental design ### 4.2.1 Studied populations Significant variations were observed in the recruited populations, composed mainly of SCI and AB subjects (twenty-two and ten articles, respectively), but also of lower-limb amputees or subjects affected by cerebral palsy, neuropathy, or Friedreich’s Ataxia. Although the level of experience in MWC locomotion has been shown to significantly affect user biomechanics, the MWC locomotion skills of the AB subjects were not specified and therefore this may have influenced the results obtained on each environmental barrier. Even when discarding AB subjects, the included MWC users were characterized by various physical conditions, anthropometries, and abilities. While such differences can lead to different propulsion strategies over the same locomotion conditions, it is interesting to have this variety represented in the studied cohorts, to have a representative population of real-world MWC users. ### 4.2.2 Reproduction of environmental barriers The difference in the number of studies investigating each barrier may not only be due to a heterogeneous distribution of interest amongst researchers, but also due to practical reasons regarding the methods available to study each barrier. Indeed, researchers can use inclined motorized treadmills or stationary ergometers to simulate slopes and potentially cross-slopes, whereas experiments with curbs and ground types all need to be conducted overground. Propulsion strategies implemented on a motorized treadmill or a stationary ergometer replicating a slope or cross-slope may differ from those typically used overground. When motorized treadmills were used, the subjects were sometimes secured using safety belts, which were reported to have some looseness in order to limit their influence on the subject’s propulsion. Yet, even when secured, the subject may unconsciously fear to fail to sustain the speed of the treadmill and therefore fall, leading to safer propulsion strategies than those that they would have adopted overground. When a stationary ergometer is used, slope simulation is achieved by adding a rolling resistance equivalent to the work needed to ascend the desired slope, sometimes coupled with an incline of the MWC. Yet, the stationary ergometer fails to reproduce the increased risk of wheelchair tipping during slope ascension, as well as the risk of backtracking when an insufficient moment of propulsion is applied to the handrim by the user. It should also be noted that when using a treadmill, propulsion biomechanics may be impacted by the surface of the treadmill belt which differs from everyday overground surfaces, leading to different strategies over a similar slope. This remark is also valid for different surfaces during overground propulsion on slope and cross-slope. ### 4.2.3 MWC configuration MWC configuration is one of the main determining factors when optimizing locomotion for a given user, as it affects propulsion biomechanics as well as other locomotion factors, such as stability. MWC stability, for example, is strongly affected by environmental barriers such as curbs or slopes. Yet, most of the reviewed studied did not report the configuration of the investigated MWC, and those that did provided only a brief description of the MWC dimensions. The issue lies in the lack of consensus on methodology to characterize and report MWC characteristics/configuration, leading to a major bias limiting the comparison across studies and subjects. ## 4.3 Joint kinematics and kinetics estimation Upper-limb kinematics and subject kinetics were reported for ascending slope propulsion as well as, to a lesser extent, for curb and cross-slope, but not for ground types. Yet, when reported, methodological differences in kinetic and kinematic acquisition (opto-electronic motion capture system, system based on inertial measurement units) and in data processing (musculoskeletal model used for computation of joint angles and moments, point and basis of expression of net joint moments) hinder rigorous comparisons of studies on the same barrier, and prevent the formulation of a reliable evidence-based synthesis of the propulsion biomechanics for each barrier. Lastly, poor data acquisition accuracy may lead to improper conclusions, especially for kinematics and kinetics quantities. This observation may explain some of the contradictory results reported in the studies such as those involving slopes. When investigating handrim kinetics, all studies used instrumented wheels, but most of them only mounted such wheels on one side of the MWC, whereas mounting them on both sides would enable the comparison of kinetics on each side of the MWC user and the evaluation of possible asymmetries in propulsion strategies. Moreover, only four studies reported the use of a dummy wheel to balance the MWC equipped with one instrumented wheel, which is crucial to ensure natural propulsion strategies. During level-ground propulsion over concrete, which is a situation expected to stress the user symmetrically, a relative difference of 20% between dominant and non-dominant sides of the user was found. The only study that investigated cross-slope locomotion using instrumented wheels on both sides of the subjects’ MWC also reported results indicating an asymmetry in handrim forces, propelling torques, mechanical works, and powers when comparing dominant and non-dominant sides of the user. However, they did not report which side was uphill or downhill, which is the most interesting paradigm for interpretation of the results on cross-slopes. Reported studies also tended to use different musculoskeletal models, yet the definition of joint coordinate systems linked to musculoskeletal models influences both kinematic and kinetics results. Although there is consensus on upper-extremity joint coordinate system definition for kinematics since 2005, the ISB has made recommendation on the reporting of kinetics only as of. Only two studies reported joint contact forces estimations. The reason could be that such a parameter requires a deeper dive into musculoskeletal modeling and simulation because it requires, as a prerequisite, to assess muscle forces. Furthermore, the definition of such a model influences the accuracy with which joint contact forces are estimated. Further studies should take better advantage of musculoskeletal models specifically developed and tailored to study MWC locomotion, and the sharing of these models would favor the standardization of the results. It should be noted that none of the studies presented in this review reported the uncertainties in the determination of the parameters of interest, while the different choices of models or measurement devices might have resulted in significant uncertainties. For instance, multibody kinematics optimization was found to generally carry reconstruction residual errors on markers ranging from four to forty millimeters, and between three and ten degrees of error against true bone kinematics for shoulder rotations. Moreover, the measurement uncertainty of kinetic measurement devices given by manufacturers has to be applied and propagated with those kinematic uncertainties to rigorously compare results on body kinetics. Therefore, future studies should provide recommendations on how to assess and propagate modeling and measurement uncertainties in order to allow a more rigorous comparison of results across different studies. ## 4.4 Muscle activity estimation Fourteen studies reported results acquired using EMG, ten of which focused on slope propulsion. All studies but one normalized EMG data acquired during locomotion by EMG data of maximum voluntary contraction, hence reported muscle activity highly depends on the physical capacity of each participant. It is therefore difficult to give an estimate of activity for a specific muscle and barrier, as these results are highly dependent on both the subject’s physiology and propulsion strategy. Moreover, maximum voluntary contraction normalization is subjected to uncertainty under the risk of incorrectly testing for maximum voluntary contraction. In particular, when normalization is done improperly, there may be trials where recorded muscle activity is higher than its maximal value, characterized by results above 100% of maximum voluntary contraction. For instance, Requejo et al. reported mean muscle activity higher than 100% for eight subjects, but it might also be the case for some subjects in other studies in which the mean muscle activity was averaged over all the participants. One study reported un-normalized EMG data, which is therefore presented in Volts, preventing the comparison of muscle activity with other studies. # 5. Conclusion This review highlighted discrepancies in focus given to each environmental situation in the literature. Slope ascent and ground types were studied much more than cross-slope or curb ascent. Furthermore, the review evidences a lack of consensus on the parameters of interest to report and on the methods used to conduct experiments. These variations and lack of consensus make it impossible to cross-reference studies to compare situations. Nevertheless, for each environmental barrier, this review provides an unprecedented overview of its current biomechanical assessment through the report of numerical values of all biomechanical parameters retrieved from the relevant literature (in tables provided in supplementary material). At the end of this review process, we recommend a more systematic approach when reporting materials, methods, and results for the reflection of the difficulty of any environmental barrier encountered in MWC locomotion: (i) effectively reporting barriers’ lengths, grades, or heights; (ii) striving for standardization or a report of the approach conditions of the barrier, such as velocity, especially on curbs; (iii) reporting the configuration of the used MWC, and if it was fitted to the subject’s morphology; (iv) reporting rotation sequences for the expression of moments and kinematics, and when used, the definition of the musculoskeletal model; (v) when possible, reporting measurement uncertainties and model reconstruction errors. # Supporting information The authors would like to thank Pr. Philip Fink for his generous help checking this manuscript’s grammar and spelling. 10.1371/journal.pone.0269657.r001 Decision Letter 0 van der Woude Lucas Academic Editor 2022 Lucas van der Woude This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 26 Dec 2021 PONE-D-21-13523Manual wheelchair biomechanics while overcoming various environmental barriers: a systematic reviewPLOS ONE Dear Dr. Rouvier, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Apologies for the delayed return of your manuscript and the accompanying reviews, which is fully on my account. The reviewers appreciate the manuscript and the work it has entailed. They are positive, yet have a number of major suggestions to improve the manuscript and its readability. Other than that they have provided extensive smaller remarks. Major points are the overlap between the results text section and the table info. A more complementary text to the central role of the tables is suggested. Use SI-system units throughout the manuscript. What is the role for Power output (W), external or user-related power production in your analyses? I miss this info in the tables. Add definitions The conclusion should be short and concise, not repeating or adding to the discussion. Did the authors conduct a quality assessment of the different papers, which is often typical for the systematic review? Be consistent in terminology, abbreviations between the different tables and sections. Please submit your revised manuscript by Feb 09 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Lucas van der Woude Academic Editor PLOS ONE Additional Editor Comments (if provided): Apologies for the delayed return of your manuscript and the accompanying reviews. The reviewers appreciate the manuscript and the work it has entailed. They are positive, yet have a number of major suggestions to improve the manuscript and its readability. Other than that they have provided extensive smaller remarks. Major points are the overlap between the results text section and the table info. A more complementary text to the central role of the tables is suggested. Use SI-system units throughout the manuscript. What is the role for Power output (W), external or user-related power production in your analyses? I miss this info in the tables. Add definitions The conclusion should be short and concise, not repeating or adding to the discussion. Did the authors conduct a quality assessment of the different papers, which is often typical for the systematic review? Be consistent in terminology, abbreviations between the different tables and sections. Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: N/A Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: This review gives a very good and extensive overview on the existing literature of manual wheelchair propulsion while overcoming different barriers in daily life of wheelchair users. The results are presented in detail in the tables which is very convenient for the interested reader to extract the information of interest. However, the presentation of the results in the text is sometimes lengthy and a repetition of what is presented in the table. This makes the review difficult to read. In my view, the review would benefit from shortening the text in the results section and referring to the tables instead. Detailed comments to the manuscript are written below. Introduction: Line 55: delete (MWU) Line 55-57: This sentence is not clear. What is the link between ICF and physical demands associated with barriers? Please rephrase. Line 67-68: Sentence is not complete. Line 69: replace "in the same study" with "in each study". Line 78-80: these two sentences are not well placed, they should be placed earlier in the introduction not just at the end where it's out of context. Line 80: to be specific, add "manual" to handrim propulsion Methods: Line 83: replace "consisted of" with "consisted in". There are other small language flaws in the manuscript, please let it check by a native English speaker. Line 95-98: did you also try to include "inclination" as an addition to "slope"? This might have resulted in more studies. Line 118: S1 Appendix: The appendix does not includ all parameters mentioned here, and not all parameters from the appendix are mentioned here. Please make it consistent. Regarding the S1 Appendix: \- Contact angle: you define it as "angle distance travelled by the non-dominant hand on the handrim during the push phase". Why is it limited to the non- dominant hand? \- Rate of rise: your definition is not correct, not detailed. Is it the mean resultant force divided by the contact time of the whole cycle? Besides that, in literature there are different definitions: • taking the derivative of FR with respect to time and then determining the maximum value during the first third of the stroke • (first) peak of the resultant force, divided by the time to reach the peak. Therefore, write it more precisely in your definition. \- Fraction of effective force: There are also different definitions used, either FEF = (Mwheel·r−1)·Ftot−1 or Ftan2/Fres2 Therefore, be more specific in your definition Results Line 133: include a reference to table 1. Table 2: van Drongelen 2005: information on kinematics is missing. I did not check the whole table on missing information, but just spotted this one. So please check again whether everything is included here. Table 3: - what does \* mean? \- what does sEMG mean? \- Since it is a new table, explain the abbreviations below. Same accounts for the following tables. Table 5: why is in this table height and weight of the participants included, but not in the others? Keep it consistant over the tables. Line 156 and further: As already indicated in the general comments, all the information you give in here is already displayed in the table. The first part of the methods, where you summarize the range of participant number, the studied population and the range of slopes measured, is ok (although also not really needed since it is already indicated in the table). The second part on how the measurements were performed is obsolete, since you list again what is mentioned in the table (i.e. 19 articles measured kinematics, 15 article measured kinetics..). Please shorten this part a lot or even delete it and refer to the table. The same accounts for cross-slope, curb and ground types. Line 157: it should be referenced to table 2 instead of table 1. Line 187 and further: In general, I like the overview on the results in the appendix, it gives a good overview. If possible it would be good to take the tables out of the appendix and place it in the manuscript Then you could also shorten the text of the results part. Line 187: Please refer to appendix Table S2, also further on in the results section. Line 188: replace "speed decreases with the grade of the slope" with "speed decreases with increasing slope" Line 193: replace "increase with the grade" with "increase with increasing slope" Line 215: replace "increase in internal rotation with the slope" with "increase in internal rotation during propulsion on a slope" Line 241: should be referred to table 3 instead of 2. Line 250: before you always called it instrumented wheel instead of handrim dynanometer. Please use one of the labels conistently throughout the manuscript Line 270: "on speed, cycle frequency and duration of push and recovery phase" can be deleted. In general, write more concisely and omit the information that is not really necessary or is a repetition. This makes it easier and more convenient for the reader. Line 274-275: give reference to the study and Appendix Line 296: should be referred to table 4 instead of 3. Line 296: "real" MWC sounds odd, please delete. Line 313: add reference to appendix. Line 344: should be referred to table 5 instead of 4. Line 365: this is another example of sentences that are not needed, a reference to the appendix is more meaningful. Discussion Line 408: "investigated in literature" instead of "from literature" Line 409-414: This information belongs to the results, there you already listed how many studies investigated which barrier. It does not have to be mentioned again, especially not with numbers of studies. You might want to say in one sentence what has been studied most. Line 431: safety belts were not used in all of the studies conducted on treadmills Line 443: Number of studies reporting kinematics and kinetics is for the results section, not for the discussion. Line 467: include reference to the study. References Reference 1: what kind of a reference is this? if it is a website, please indicate the url. Reference 11: Title is written twice in the reference Figure 1: indicate in the figure why the records (n=1098) they are excluded S2 Table: Check the table, units are not always indicated. For example in Slope: MWC speed, recovery phase duration, contact angle, sometime it's indicated after the numbers instead of in the the title line (i.e. body kinetics), or both in the title line and after numbers (Peak un-normalized EMG). Please make it consistent. Reviewer \#2: Comments to editor: The authors presented a nicely written systematic review on environmental barriers and manual wheelchair biomechanics. The introduction, rationale and methods of this study are presented in a clear way. The results still have quite some redundant information, which is already present in the tables, it is recommended to shorten parts of the results to improve the readability. The discussion section raises very interesting points, only the structure needs some finetuning. The conclusion should be shorter and to the point. A recommendation of Acceptance with Major Revision has been given. General comments: A nicely written systematic review, which presents the work in a very detailed and extensive way. In more detail: Introduction: the introduction is written well, only the ending could use some finetuning. Methods: Methods section is well presented and only lacks a clear inclusion/exclusion overview. Results: The results were separated for the four areas (slopes, cross-slopes, curbs and ground type), as well as methods and results. Regarding the methods on slopes, cross-slopes, curbs and ground type: It is unnecessary to present all results on methods this extensively. Most information can also be found in the presented tables, which already cover a lot of information. Removing parts in these sections might help with the readability of the whole paper. On the Results on slopes, cross-slopes, curbs and ground type part: Adding subheadings for the subcategories: spatio-temporal, kinematics, kinetics and muscle activity are recommended. Discussion: In the discussion section a lot of important information is discussed. It feels like a lot of separate interesting discussion points are raised but the structure is a bit lacking. The transitions from one paragraph to another, as well as the general structure should use some finetuning. Conclusion: In this section a lot of information is discussed, which can either be removed or combined with some of the discussion points. Try to get the conclusion short and to the point. Smaller general points: The consistency in the use of references, sometimes all references are listed extensively, but there are cases in which some the references are missing, examples are Lines 380-382 and Lines 426-442. Maybe a figure with the 4 environmental barriers might be a good addition to have a clear overview and explanation of the 4 barriers (cross-slope might be unclear to some naïve readers). Specific comments: Abstract: Page 2, Lines 33-35: Make the conclusion at the end a bit stronger, something similar to: “It is recommended to standardize the procedure, studying a wide set of…” Introduction: Page 3, Lines 45-46: The sentence seems confusing, maybe something similar to: “Despite the improvement of overall accessibility of public areas…” Page 3, Lines 46: Recommended to start a new paragraph when ICF is introduced Page 3, Lines 61-62: Last sentence of this paragraph feels a bit redundant Page 4, Line 75: Remove the word “therefore” in this sentence Page 4, Lines 78-80: This paragraph is unnecessary, the end of the previous paragraph is a more proper ending of the introduction Methods: Page 4, Line 83: Change the sentence a bit: “The present study conducted a systematic review to identify…” Page 5, Lines 105-106: Remove “Appended to this paper…” and add “(S1 table)” add the end of “…(PRISMA) Checklist \[24\]” Page 5, Lines 106-110: Separate into inclusion/exclusion criteria, inclusion: biomechanical data, experimental work, English languages, exclusion: electric and power assisted wheelchairs etc. Page 5, Line 111: Remove the word “three” from the sentence Page 5, Lines 114-116: Very nice and detailed description of the parameters Page 6, Line 117: Change towards: “A more detailed definition of the biomechanical parameters can be found in S1 Appendix” Page 6, Line 118-119: Sentence feels a bit redundant Results: Page 6, Lines 129-132: Split the sentence into two Page 6, Lines 134-137: Cut the sentences a bit, similar to: “Experimental design, acquisition methods and measurement tools differed between studies. The MWC was propelled overground… …ergometer. Kinematics were either recorded…”. Page 6, Lines 137-138: What is the use of “on the contrary”? Page 6, Lines 140-142: Try to rephrase the sentence a bit, it is unclear what is part of the tables and what of the supplementary material. Maybe similar to: “Table 1 represents the synthesis of all studies, the remaining tables (2-5) summarize the study review for the four categories (slope, cross-slope, curb, ground type). … are appended as supporting information (S2 Table)”. Page 18-19, Methods on slopes: See general comment Page 19, Line 188-189: Sentence can be removed (“Results on cycle frequency…) and combined with the next sentence. Page 19, Lines 191-194: Try to shorten the sentence: “…increase with the grade in SCI subjects, but unaffected with AB subjects.” Page 19: Lines 201-203: What kind of differences were found? Page 20: Line 212: Remove the part “in the study where they were reported”. Page 20: Lines 219-220: Remove the part: “…in all studies reporting these data”. Page 21: Line 233: Change to past sense Page 21-22, Methods on cross-slopes: See general comments Page 22: Lines 266-267: Is an interpretation sentence, better be part of the discussion Page 22, Lines 270-274: Try to shorten this sentence, parts at the beginning of the sentence can be removed. Page 23-24, Methods on curbs: See general comments Page 23, Line 296: Remove the word “real” Page 25-26, Methods on ground types: See general comments Page 27, Lines 380-382: The references are missing Discussion Page 27-28, Lines 401-406: This paragraph only repeats the aims, without actually giving any relevant information, combining some of this information with the next paragraph might be a solution. Page 28, Lines 417-425: This part can be shortened a bit, there is some repetition, especially with the last sentence. Page 29, Lines 428-430: Future and past sense are both used in this sentence Page 29-30, Lines 443-454: The paragraph starts with summarizing the number of articles investigating kinematics/kinetics. Despite main focus of this paragraph seems on the differences in acquisition methods. Try to get the most relevant information at the start/end of the paragraph. Page 30, Lines 469-476: This feels like a separate paragraph, since the main focus of this paragraph was on the use of instrumented measurement wheels on both side of the MWC. Page 31, Lines 486-490: It is unclear why this sentence is presented in between brackets, would make the argument stronger to present it as an example. Page 32, Lines 505-508: Try to combine these sentences into one strong sentence. Page 32, Lines 509-516: This paragraph does not add too much to the whole review and could, feels like lose paragraph at the end of the discussion, see also general comments Conclusion: Page 32-34: A conclusion is preferably short and to the point. The conclusion now consists of around five paragraphs in which also a lot of recommendations are given. Some of these recommendations are already discussed previously and don’t necessarily strengthen the paper. Page 32, Lines 519-524: The first paragraph summarizes/repeats what has already been repeated multiple times Page 33, Line 526: This is one of the main lines to be used for the conclusion. Page 33, Lines 530-536: This all can be said in one sentence in the discussion part. Page 33, Lines 537-540: Interesting point, also combine with some of the paragraphs in the discussion part (Future recommendation) Page 33, Lines 541-543: Repeating the fact that cross-slopes and curbs should be studies, but now focused on asymmetrical propulsion. Page 33-34, Lines 544-551: This seems like one of the main parts of the conclusion of the paper \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Ursina Arnet Reviewer \#2: **Yes: **Riemer Vegter & Thomas Rietveld \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0269657.r002 Author response to Decision Letter 0 10 Feb 2022 Dear reviewers, Editor, Thank you for your encouraging comments while reviewing the manuscript and for allowing us to improve its content. Please find below how we have addressed each of the reviewers’ comment. The reviewers’ comments are reported in black, whereas the answers are reported in blue. A version of the revised manuscript has also been uploaded. Changes are highlighted in yellow in the revised version of the manuscript. Kindest regards, Théo Rouvier, on behalf of all the authors of the manuscript entitled "Manual wheelchair biomechanics while overcoming various environmental barriers: a systematic review” Reviewer 1: General comments: This review gives a very good and extensive overview on the existing literature of manual wheelchair propulsion while overcoming different barriers in daily life of wheelchair users. The results are presented in detail in the tables which is very convenient for the interested reader to extract the information of interest. However, the presentation of the results in the text is sometimes lengthy and a repetition of what is presented in the table. This makes the review difficult to read. In my view, the review would benefit from shortening the text in the results section and referring to the tables instead. Detailed comments to the manuscript are written below. Thank you for your appreciation and for your detailed review of our work. You will find hereafter responses to your detailed comments. Following your advice, the text in the “results” section was shortened and references to the tables were added. Introduction: Line 55: delete (MWU) “(MWU)” was deleted from line 47 but the abbreviation was redefined at line 130 of the revised manuscript since it is used in the “results” and “discussion” sections. Line 55-57: This sentence is not clear. What is the link between ICF and physical demands associated with barriers? Please rephrase. We believe that ICF could be better implemented for MWC users if clinicians were able to adapt training programs in accordance with the difficulty of a barrier with regards to a user’s capabilities. The quantification of the physical demands associated with barriers could be used as a quantification of a barrier’s difficulty. The paragraph was rephrased. (l. 55-60) Line 67-68: Sentence is not complete. The sentence was rephrased. Line 69: replace "in the same study" with "in each study". This was corrected (line 67 of the revised manuscript). Line 78-80: these two sentences are not well placed, they should be placed earlier in the introduction not just at the end where it's out of context. The sentences have been moved up in the text at the beginning of the “Methods” section (lines 82-86 of the revised manuscript). Line 80: to be specific, add "manual" to handrim propulsion The adjective “manual” was added to be more specific, following the reviewer’s comment. (line 86 of the revised manuscript). Methods: Line 83: replace "consisted of" with "consisted in". There are other small language flaws in the manuscript, please let it check by a native English speaker. The sentence was rephrased. The manuscript was checked by an English speaker and language flaws have been corrected throughout the manuscript. Line 95-98: did you also try to include "inclination" as an addition to "slope"? This might have resulted in more studies. Following the reviewer’s suggestion, we performed a new search on PubMed with the addition of the term “inclination”, which resulted in 21 new results compared to the initial PubMed search, among which some had already been retrieved with the initial Scopus search. Based on the screening on the title, most are related to seat or backrest inclinations, and not to the ground inclination (slope grade). Finally, only 1 reference appeared to be relevant, but it did not meet the inclusion criteria based on the screening on the abstract. We agree with the reviewer that it remains possible that we missed some papers relative to very specific situations due to the absence of very specific keywords, but we believe that the global conclusions drawn on methodologies will not be challenged. Line 118: S1 Appendix: The appendix does not include all parameters mentioned here, and not all parameters from the appendix are mentioned here. Please make it consistent. Thank you for your comment, the parameters defined in the appendix and those mentioned in the “methods” section are now consistently the same. Results: Line 133: include a reference to table 1. A reference to table 1 was added. Table 2: van Drongelen 2005: information on kinematics is missing. I did not check the whole table on missing information, but just spotted this one. So please check again whether everything is included here. Thank you for noticing this mistake. Information on kinematics recording in van Drongelen 2005 was added to Table 2. All the tables were double checked following your comment. Table 3: - what does \* mean? \* Represents data that was extrapolated by the authors using the data given by the study. The signification of “\*” was added to the tables, in the caption. \- what does sEMG mean? sEMG means Surface EMG, this was detailed in the tables’ captions. \- Since it is a new table, explain the abbreviations below. Same accounts for the following tables. All abbreviations are now defined in the caption of each table. Table 5: why is in this table height and weight of the participants included, but not in the others? Keep it consistant over the tables. It is true that data reported in tables was not consistent. We initially chose to report height and weight of subjects for the studies focusing on ground types because some of these studies normalized kinetic data on subject height and weight. But it is indeed confusing to report such data only for those studies and we don’t consider it crucial to the comprehension of our systematic review. For these reasons we normalized the tables and removed data on subject height and weight. Line 156 and further: As already indicated in the general comments, all the information you give in here is already displayed in the table. The first part of the methods, where you summarize the range of participant number, the studied population and the range of slopes measured, is ok (although also not really needed since it is already indicated in the table). The second part on how the measurements were performed is obsolete, since you list again what is mentioned in the table (i.e. 19 articles measured kinematics, 15 article measured kinetics..). Please shorten this part a lot or even delete it and refer to the table. The same accounts for cross-slope, curb and ground types. Thank you for your comment. Following advice from both reviewers, the presentation of the experimental methods of the reviewed studies was shortened in the corpus of the manuscript. Full detail of the methods can be found in tables 2-5, and only key information was retained in the text. Line 157: it should be referenced to table 2 instead of table 1. This was corrected. Line 187 and further: In general, I like the overview on the results in the appendix, it gives a good overview. If possible it would be good to take the tables out of the appendix and place it in the manuscript Then you could also shorten the text of the results part. We agree with the reviewer’s that the tables provide a nice overview of the results, and, initially, it was our aim to place them in the manuscript. However, given the lack of consensus in the literature on the outcome parameters of interest, it is difficult to provide tables that would fit over a page. The tables have indeed a large number of columns and few rows, which compromise their insertion in the text. Therefore, we chose to summarize the results in the paper and to refer to appendix table S2. Line 187: Please refer to appendix Table S2, also further on in the results section. We added a reference to the S2 Table in the slope sections, as well as in other sections (l.180-181 of the revised manuscript). Line 188: replace "speed decreases with the grade of the slope" with "speed decreases with increasing slope" The sentence was rephrased following the reviewer’s suggestion (l.184). Line 193: replace "increase with the grade" with "increase with increasing slope" The sentence was rephrased following the reviewer’s suggestion (l. 188). Line 215: replace "increase in internal rotation with the slope" with "increase in internal rotation during propulsion on a slope" This was changed to “increase in internal rotation with increasing slope”. Line 241: should be referred to table 3 instead of 2. The reference to the table was corrected. Thank you for spotting this error. Line 250: before you always called it instrumented wheel instead of handrim dynanometer. Please use one of the labels conistently throughout the manuscript Thank you for noticing this inconstancy. We replaced three usages (all of them) of “handrim dynamometer” by “instrumented wheel”. Line 270: "on speed, cycle frequency and duration of push and recovery phase" can be deleted. In general, write more concisely and omit the information that is not really necessary or is a repetition. This makes it easier and more convenient for the reader. Thank you for your comment which allowed us to increase the readability of our manuscript. Line 274-275: give reference to the study and Appendix Reference to the study and to the appendix were added (appendix reference l. 262-263 and study reference l. 268). Line 296: should be referred to table 4 instead of 3. This was corrected (l.293). Line 296: "real" MWC sounds odd, please delete. This was deleted. Line 313: add reference to appendix. A reference to the appendix table was given (l. 306-307). Line 344: should be referred to table 5 instead of 4. This was corrected. Line 365: this is another example of sentences that are not needed, a reference to the appendix is more meaningful. These sentences were deleted (reference to the appendix l. 352-353). Discussion Line 408: "investigated in literature" instead of "from literature" The sentence was rephrased following the reviewer’s suggestion (l. 392). Line 409-414: This information belongs to the results, there you already listed how many studies investigated which barrier. It does not have to be mentioned again, especially not with numbers of studies. You might want to say in one sentence what has been studied most. The sentence was shortened following the reviewer’s suggestion (l. 393-394). Line 431: safety belts were not used in all of the studies conducted on treadmills Thank you for your comment. The sentence was changed, and now states that safety belts were sometimes used in studies conducted in treadmills (l. 444-447). Line 443: Number of studies reporting kinematics and kinetics is for the results section, not for the discussion. The sentence was shortened following the reviewer’s suggestion. Line 467: include reference to the study. The reference to the study was added (l. 451). References Reference 1: what kind of a reference is this? if it is a website, please indicate the url. The reference was corrected to display authors, organisation, and website url. Reference 11: Title is written twice in the reference This was corrected. Figure 1: indicate in the figure why the records (n=1098) they are excluded The 1098 records were excluded after either a screening on title or abstract. These records were excluded based on our inclusion/exclusion criteria. These criteria are detailed in section (2.2 – Article selection) of the manuscript. “The inclusion criteria were original study or systematic review, study written in English and experimental work. Exclusion criteria were articles about electric wheelchairs, powered-assisted wheelchairs, sports wheelchairs, other propulsion systems than manual handrim, hemiplegia-pattern propulsion”. It was added to the PRISMA flow-chart that these records were excluded after title or abstract screening. Reasons detailing why studies were excluded in the full-text screening were also added to the flow-chart. S2 Table: Check the table, units are not always indicated. For example, in Slope: MWC speed, recovery phase duration, contact angle, sometime it's indicated after the numbers instead of in the the title line (i.e. body kinetics), or both in the title line and after numbers (Peak un-normalized EMG). Please make it consistent. Thank you for this comment. It is true the reporting of units was inconsistent. All units are now consistently reported. Units for all values except those that are normalized are reported in the title line. Units for values that are normalized are always reported in each cell, because studies did not always use the same normalization criterion, even on the same biomechanical parameter, and therefore had different units for the same parameters. Appendix \- Contact angle: you define it as "angle distance travelled by the non-dominant hand on the handrim during the push phase". Why is it limited to the non- dominant hand?- Rate of rise: your definition is not correct, not detailed. Is it the mean resultant force divided by the contact time of the whole cycle? Besides that, in literature there are different definitions: • taking the derivative of FR with respect to time and then determining the maximum value during the first third of the stroke • (first) peak of the resultant force, divided by the time to reach the peak. Therefore, write it more precisely in your definition. \- Fraction of effective force: There are also different definitions used, either FEF = (Mwheel·r−1)·Ftot−1 or Ftan2/Fres2 Thank you for your comment, the definition of these biomechanical parameters was detailed. We presented both definitions for the two latter parameters, and indicated which one was more present in the reviewed articles. Reviewer 2: A nicely written systematic review, which presents the work in a very detailed and extensive way. In more detail: Thank you for the appreciation of our work and your thorough review of the manuscript. Introduction: the introduction is written well, only the ending could use some finetuning. The last sentences of the introduction were moved up in the introduction so as to provide a better ending of the introduction, following the reviewer’s comment. Methods: Methods section is well presented and only lacks a clear inclusion/exclusion overview. Inclusion and exclusion criteria were separated in the Methods section following the reviewer’s suggestion. Results: The results were separated for the four areas (slopes, cross-slopes, curbs and ground type), as well as methods and results. Regarding the methods on slopes, cross-slopes, curbs and ground type: It is unnecessary to present all results on methods this extensively. Most information can also be found in the presented tables, which already cover a lot of information. Removing parts in these sections might help with the readability of the whole paper. On the Results on slopes, cross-slopes, curbs and ground type part: Adding subheadings for the subcategories: spatio-temporal, kinematics, kinetics and muscle activity are recommended. Thank you for your comment. Following your recommendations, the presentation of the reviewed methods in the “Results” section was shortened to only contain information we deemed key for the reader. Also, as suggested, subheadings were added in the “Results” section for the different biomechanical parameters (spatio-temporal, kinematics, kinetics, muscle activity). Discussion: In the discussion section a lot of important information is discussed. It feels like a lot of separate interesting discussion points are raised but the structure is a bit lacking. The transitions from one paragraph to another, as well as the general structure should use some finetuning. Thank you for your comment, we have worked on improving the flow of the “Discussion” section. Paragraphs were re-arranged and subheadings were added to place a greater emphasis on the different key points addressed in the section. We think that this new general structure of the discussion improves its readability compare to the previous draft. Conclusion: In this section a lot of information is discussed, which can either be removed or combined with some of the discussion points. Try to get the conclusion short and to the point. Thank you for your comment and suggestion, the conclusion indeed felt too long. Points addressed in the discussion were either shortened or moved up into the “Discussion” section. The conclusion is now only two paragraphs short. Smaller general points: The consistency in the use of references, sometimes all references are listed extensively, but there are cases in which some the references are missing, examples are Lines 380-382 and Lines 426-442. Thank you for noticing these missing references. References were added in both paragraphs as suggested (l. 365, 443-45 of the revised manuscript). Maybe a figure with the 4 environmental barriers might be a good addition to have a clear overview and explanation of the 4 barriers (cross-slope might be unclear to some naïve readers). Thank you for your suggestion. Thank you for your comment. Pictures of the recreated environmental barriers were moved from the supplementary materials (S1-3 Figs) to the corpus of the manuscript, and fused in one figure Abstract: Page 2, Lines 33-35: Make the conclusion at the end a bit stronger, something similar to: “It is recommended to standardize the procedure, studying a wide set of…” Following the reviewer’s comment, the concluding sentence of the abstract was rephrased to “It is recommended to standardize the procedure when studying various physical environmental situations and to systematically report MWC configuration(s). Furthermore, a wider set of situations should be studied” (l. 33-36). Introduction: Page 3, Lines 45-46: The sentence seems confusing, maybe something similar to: “Despite the improvement of overall accessibility of public areas…” The sentence was rephrased following the reviewer’s suggestion (l. 45-47). Page 3, Lines 46: Recommended to start a new paragraph when ICF is introduced A new paragraph dedicated to the ICF was started (l. 48-60). Page 3, Lines 61-62: Last sentence of this paragraph feels a bit redundant Indeed, the last sentence was removed following the reviewer’s comment. Page 4, Line 75: Remove the word “therefore” in this sentence The word “therefore” was removed from the sentence. Page 4, Lines 78-80: This paragraph is unnecessary, the end of the previous paragraph is a more proper ending of the introduction The sentences of the last paragraph were not deleted as some studies focusing on other means of propulsion were retrieved from the literature and are not discussed in the present review. However, following both reviewer’s comments, they were moved in the Methods section. The introduction now ends on: ” To fill this gap, the purpose of this study was to identify and synthesize data and experimental methods from the literature on the biomechanics of MWC propulsion for various and frequent environmental barriers that are daily encountered by MWC users.” (paragraph moved to l. 82-86). Methods: Page 4, Line 83: Change the sentence a bit: “The present study conducted a systematic review to identify…” The sentence was changed following the reviewer’s comment (l. 82). Page 5, Lines 105-106: Remove “Appended to this paper…” and add “(S1 table)” add the end of “…(PRISMA) Checklist \[24\]” This was changed (l. 108). Page 5, Lines 106-110: Separate into inclusion/exclusion criteria, inclusion: biomechanical data, experimental work, English languages, exclusion: electric and power assisted wheelchairs etc. The criteria for article selection into the present systematic review were separated into inclusion and exclusion criteria following the reviewer’s comment (see l. 109-112 of the revised manuscript) Page 5, Line 111: Remove the word “three” from the sentence Following the reviewer’s suggestion, “Three” was removed from the sentence as it was already cited earlier in the manuscript that three authors participated in the article selection process. Page 5, Lines 114-116: Very nice and detailed description of the parameters Thank you for your comment Page 6, Line 117: Change towards: “A more detailed definition of the biomechanical parameters can be found in S1 Appendix” The sentence was changed (l. 119). Page 6, Line 118-119: Sentence feels a bit redundant The sentence was moved to the section 2.1 of the manuscript. Indeed, it was a bit out of context in section 2.2 “Article selection” but it seemed relevant to the authors to insist of the methodological aspects of the studies as well as on the obtained results (l. 95-96). Results: Page 6, Lines 129-132: Split the sentence into two The sentence was split into two sentences and a reference to Table 1 was added following both reviewers’ comments (l. 129-133). Page 6, Lines 134-137: Cut the sentences a bit, similar to: “Experimental design, acquisition methods and measurement tools differed between studies. The MWC was propelled overground… …ergometer. Kinematics were either recorded…”. The sentence was cut into three (l. 134-138). Page 6, Lines 137-138: What is the use of “on the contrary”? “On the contrary” was removed from the sentence. Page 6, Lines 140-142: Try to rephrase the sentence a bit, it is unclear what is part of the tables and what of the supplementary material. Maybe similar to: “Table 1 represents the synthesis of all studies, the remaining tables (2-5) summarize the study review for the four categories (slope, cross-slope, curb, ground type). … are appended as supporting information (S2 Table)”. Following the reviewer’s suggestion, the sentence was rephrased to clarify the organization of the review: “An overview of the retrieved studies is provided in Table 1. A dedicated subsection to each investigated environmental barrier (slope, cross-slope, curb, ground type) synthesizes the experimental methods used in these studies (also reported in Tables (2-5)) as well as the obtained biomechanical results with summarizing the experimental method. A compilation of the studies' numerical results is appended to this document as supplementary material (S2 Table).” (l. 139-143). Page 18-19, Methods on slopes: See general comment As answered before in general comments, the presentation of the experimental methods of the reviewed studies was shortened. Full detail of the methods can be found in table 2, and only key information was retained in the text. Page 19, Line 188-189: Sentence can be removed (“Results on cycle frequency…) and combined with the next sentence. This sentence was removed. Page 19, Lines 191-194: Try to shorten the sentence: “…increase with the grade in SCI subjects, but unaffected with AB subjects.” The sentence was shortened as suggested. Page 19: Lines 201-203: What kind of differences were found? Thank you for this comment, we indeed did not detail which differences were found. A sentence detailing these differences was added to the manuscript: ”Contact angle was higher on the same slope when experimenting on AB subjects, and seemed to remain constant with different grades of slope in AB subjects, whilst contact angle tended to decrease with increasing slope in MWU.” (l. 198-200). Page 20: Line 212: Remove the part “in the study where they were reported”. This part was removed. Page 20: Lines 219-220: Remove the part: “…in all studies reporting these data”. This part was removed. Page 21: Line 233: Change to past sense The verb “give” was changed to the past tense, thank you for spotting that error. Page 21-22, Methods on cross-slopes: See general comments See general comments for response: the review of the studies’ methods was shortened in the text. While tables still give full detail on the methods of the reviewed studies, the text now holds only key information. Page 22: Lines 266-267: Is an interpretation sentence, better be part of the discussion Thank you for your suggestion. The sentence was moved into the discussion section (4.1) (l. 395-397 of the revised manuscript). Page 22, Lines 270-274: Try to shorten this sentence, parts at the beginning of the sentence can be removed. The sentence was rephrased (l. 265-268). Page 23-24, Methods on curbs: See general comments See general comments for response: the review of the studies’ methods was shortened in the text. While tables still give full detail on the methods of the reviewed studies, the text now holds only vital information. Page 23, Line 296: Remove the word “real” This word was removed. Page 25-26, Methods on ground types: See general comments See general comments for response: the review of the studies’ methods was shortened in the text. While tables still give full detail on the methods of the reviewed studies, the text now holds only vital information. Page 27, Lines 380-382: The references are missing References were added (l. 365). Discussion: Page 27-28, Lines 401-406: This paragraph only repeats the aims, without actually giving any relevant information, combining some of this information with the next paragraph might be a solution. Thank you for your comment. The paragraph was shortened to avoid repetition and allows for a better introduction of the review’s focus on experimental design issues (l. 386-388). Page 28, Lines 417-425: This part can be shortened a bit, there is some repetition, especially with the last sentence. This part of the paragraph was split into two, creating one more paragraph to present our point better. The penultimate sentence of the part was changed to get rid of repetition (l. 406-415). Page 29, Lines 428-430: Future and past sense are both used in this sentence The sentence was changed so the verbs are conjugated accordingly (l. 417-418). Page 29-30, Lines 443-454: The paragraph starts with summarizing the number of articles investigating kinematics/kinetics. Despite main focus of this paragraph seems on the differences in acquisition methods. Try to get the most relevant information at the start/end of the paragraph. Thank you for this comment. Along with rewiever 1’s suggestions, the first sentence of the paragraph was shortened to skip the summary of articles investigating kinematics/kinetics (l. 471-472). Page 30, Lines 469-476: This feels like a separate paragraph, since the main focus of this paragraph was on the use of instrumented measurement wheels on both side of the MWC. Indeed, not starting a new paragraph felt odd. This was corrected and a new paragraph was created discussing joint coordinate systems usage (new paragraph l. 497-507). Page 31, Lines 486-490: It is unclear why this sentence is presented in between brackets, would make the argument stronger to present it as an example. Thank you for your remark, the part in the brackets was moved out of brackets to be used as an example (l. 529-531). Page 32, Lines 505-508: Try to combine these sentences into one strong sentence. These two sentences were combined into one (l.466-468 of the revised manuscript). Page 32, Lines 509-516: This paragraph does not add too much to the whole review and could, feels like lose paragraph at the end of the discussion, see also general comments The paragraph was moved up in the discussion, where it was more relevant (section 4.3) and was slightly reformulated (l. 508-517). Conclusion: Page 32-34: A conclusion is preferably short and to the point. The conclusion now consists of around five paragraphs in which also a lot of recommendations are given. Some of these recommendations are already discussed previously and don’t necessarily strengthen the paper. Thank you for your comment. Following your recommendations, the conclusion was shortened and is now straighter to the point. Page 32, Lines 519-524: The first paragraph summarizes/repeats what has already been repeated multiple times The first paragraph was deleted. Page 33, Line 526: This is one of the main lines to be used for the conclusion. Page 33, Lines 530-536: This all can be said in one sentence in the discussion part. The paragraph was moved to the discussion section and split into different sentences adding weight to their respective new paragraphs. Page 33, Lines 537-540: Interesting point, also combine with some of the paragraphs in the discussion part (Future recommendation) The paragraph was moved in the discussion, in the section 4.1 discussing the environmental barriers studied in the retrieved literature (l. 406-409). Page 33, Lines 541-543: Repeating the fact that cross-slopes and curbs should be studies, but now focused on asymmetrical propulsion. This paragraph indeed felt like a repetition of what was already discussed previously, for this reason, it was deleted. Page 33-34, Lines 544-551: This seems like one of the main parts of the conclusion of the paper This paragraph is now the conclusion’s and our manuscript’s end (l. 545-552 of the revised manuscript). 10.1371/journal.pone.0269657.r003 Decision Letter 1 van der Woude Lucas Academic Editor 2022 Lucas van der Woude This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 7 Apr 2022 PONE-D-21-13523R1Manual wheelchair biomechanics while overcoming various environmental barriers: a systematic reviewPLOS ONE Dear Dr. Rouvier, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by May 22 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Lucas van der Woude Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: (No Response) Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: N/A Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The authors have addressed my comments well and the manuscript is now much clearer and easier to read. I only have some small comments to this revised version: Abstract: In my view the conclusion stated in the abstract does not reflect the conclusion of your whole paper. Please include all recommendatios given in the conclusion section (i-v), or state it more general. Furthermore, do you suggest to standardize the measurements or the reporting of it? Methods: Line 107: You refer to the PRISMA checklist when describing article selection. The prisma Checklist does not specify how to select articles but specifies what should be reported in a systematic review. So better refer to figure 2 or rewrite the first sentence. Line 110: You mention systematic review as a inclusion criteria, but another inclusion criteria is experimental work. If experimental work is a inclusion criteria, systematic reviews (as mentioned in the line above) will not be included. Results: Figure 2: In the description of figure 2 you indicate classification, but this is not addressed in figure 2. Line 286: Please indicate whether EMG was measured on both sides or just on up-/ or downhill side. Line 344: include "wheel" after instrumented Discussion: Line 440: include "due" in front of to. Conclusion: Line 548-553: here you state what you recommend for further studies. Is this what you miss in the current studies, or is this all you would like to have reported in studies? And what for example about the participant characteristics? Reviewer \#2: Comments to editor: The authors addressed all essential comments previously made. Only some minor revisions are necessary, after which the paper is recommended to be accepted. General comments: Thank you for addressing all comments and the changes made to the manuscript. The changes have improved the readability of the manuscript. Only some minor adjustments are recommended. A small note: Sometimes a tab is used to begin a new paragraph, sometimes not, there is no real consistency throughout the whole paper. Specific comments: Abstract: Page 2, Lines 33-36: It would be recommended to make 1 strong final sentence: “It is recommended to standardize the procedure when studying various physical environmental situations, systematically report MWC configuration(s) and study a wider set of situations.” Introduction: The introduction now indeed ends with a logical flow towards the purpose of the study, well done. Methods: Page 5, Lines 109-112: The exclusion criteria are now nicely stated, only the inclusion criteria are still a bit too general (English language, original study, experimental work). It would be recommended to also add something similar to: “Studies investigating slope, cross-slope, curb and ground type in manual wheelchair users.” Results: Removing parts on the ‘methods’ section for all barriers improved the readability of the results part of the manuscript. Adding subheading really helped for the structure as well, great work. Page 19, Line 180-182: These two sentences can be removed or moved a bit up, it is already previously stated on Page 7, Line 144 that the numerical results can be found in S2 Table. Now the same sentence is stated for all barriers on Page 22, Lines 263-264, Page 24, Lines 307-308 and Page 26, Lines 353-354. Stating that information once around line 144 will probably be enough, similar to: ‘Supplementary material on the detailed results on all environmental barriers (slopes, curbs etc.) can be found in S2 Table.’. Page 22, Line 265: The subheading ‘3.2.2.1. Spatio-temporal parameters’ is too big, should be similar to the ‘3.2.2.2. Joint Kinematics’. Discussion Page 27, Lines 387-389: This introductory statement might not necessarily be needed, in the next paragraph it is also summarized that environmental barriers were investigated. Page 28, Lines 411-422: One nice combined paragraph can be made out of these three separate paragraphs with the last one only being one sentence. Page 31, Lines 483-497: This has become a rather long paragraph to only state one thing: ‘Instrumented wheels should be used on both sides’. Especially lines 490-495 are repetitions of the first sentences of the paragraph. Page 32, Lines 509-511: Which studies are you referring to? Page 32, Line 511: ‘… might have resulted in significant uncertainties’. Conclusion: The conclusion is a lot more shortened and to the point, well done. Page 33, Line 538: The first sentence is a bit unclear, seems like there is missing some information. Page 33, Line 545: ‘… parameters retrieved from the relevant literature.’ \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Ursina Arnet Reviewer \#2: **Yes: **Riemer Vegter & Thomas Rietveld \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0269657.r004 Author response to Decision Letter 1 25 Apr 2022 Dear reviewers, Thank you for your encouraging comments while reviewing the revised manuscript and for allowing us to keep improving its content. Please find below how we have addressed each of the reviewers’ comment. The reviewers’ comments are reported in black, whereas the answers are reported in blue. A version of the revised manuscript has also been uploaded. Changes are highlighted in yellow in the revised version of the manuscript. Kindest regards, Théo Rouvier, on behalf of all the authors of the manuscript entitled "Manual wheelchair biomechanics while overcoming various environmental barriers: a systematic review” Reviewer 1: The authors have addressed my comments well and the manuscript is now much clearer and easier to read. I only have some small comments to this revised version: Thank you for your endearing comments. We will try to take the latter comments into account as best as we can. Abstract: In my view the conclusion stated in the abstract does not reflect the conclusion of your whole paper. Please include all recommendations given in the conclusion section (i-v), or state it more general. Furthermore, do you suggest to standardize the measurements or the reporting of it? Thank you for your comment. We added the five recommendations made in the conclusion to the abstract. These recommendations include suggestions to both standardize the measurements (ii) and the reporting of it (i, iii, iv, v). Methods: Line 107: You refer to the PRISMA checklist when describing article selection. The prisma Checklist does not specify how to select articles but specifies what should be reported in a systematic review. So better refer to figure 2 or rewrite the first sentence. Indeed, we should not have referred to the prisma checklist. The sentence now refers to figure 2, which was moved up in the manuscript to appear after this paragraph. Line 110: You mention systematic review as a inclusion criteria, but another inclusion criteria is experimental work. If experimental work is a inclusion criteria, systematic reviews (as mentioned in the line above) will not be included. Thank you for your comment. This error was fixed by modifying the phrasing of the inclusion criteria, which now state “features experimental results” instead of “experimental work” Results: Figure 2: In the description of figure 2 you indicate classification, but this is not addressed in figure 2. Thank you for your comment. This was from an earlier draft of the figure. Classification was removed from the figure’s title. Line 286: Please indicate whether EMG was measured on both sides or just on up-/ or downhill side. EMG was measured only on the left side. This precision was added in the presentation of the methods and results of muscle activity in cross-slopes. Line 344: include "wheel" after instrumented This was corrected. Discussion: Line 440: include "due" in front of to. This was corrected. Conclusion: Line 548-553: here you state what you recommend for further studies. Is this what you miss in the current studies, or is this all you would like to have reported in studies? And what for example about the participant characteristics? Thank you for this question. These recommendations are based on what we miss in most studies of the literature, but also what we would like see reported in future studies. For example, some studies -but not all- reported barrier’s lengths and other characteristics, and we deem it is important that this practice be generalized. On the other hand, no reviewed study reported measurement uncertainties or model reconstruction errors, and we think it would be a truly interesting metric for researchers trying to compare their results with the state of the art. Participant characteristics are also a very interesting metric, but we feel that the state of the art has reached closer to a consensus, hence why we did not include them in our recommendations. Reviewer 2: Thank you for addressing all comments and the changes made to the manuscript. The changes have improved the readability of the manuscript. Only some minor adjustments are recommended. A small note: Sometimes a tab is used to begin a new paragraph, sometimes not, there is no real consistency throughout the whole paper. Thank you for your encouraging comments. Tabulations are now systematically used at the start of the first paragraph of a section. Abstract: Page 2, Lines 33-36: It would be recommended to make 1 strong final sentence: “It is recommended to standardize the procedure when studying various physical environmental situations, systematically report MWC configuration(s) and study a wider set of situations.” Thank you for your comment. Alongside reviewer 1’s comments, we decided to end the abstract by the five recommendations listed in the conclusion. The sentence “Furthermore a wider set of situations should be studied” that was used to close the abstract was moved up to let our recommendations close the abstract. Introduction: The introduction now indeed ends with a logical flow towards the purpose of the study, well done. Thank you very much for your comment. Methods: Page 5, Lines 109-112: The exclusion criteria are now nicely stated, only the inclusion criteria are still a bit too general (English language, original study, experimental work). It would be recommended to also add something similar to: “Studies investigating slope, cross-slope, curb and ground type in manual wheelchair users.” Thank you for your comment. We changed Inclusion criteria and replaced “experimental work” by “features experimental results on slopes, cross-slopes, curbs, and ground types during MWC locomotion”. Results: Removing parts on the ‘methods’ section for all barriers improved the readability of the results part of the manuscript. Adding subheading really helped for the structure as well, great work. Thank you for your comment. Page 19, Line 180-182: These two sentences can be removed or moved a bit up, it is already previously stated on Page 7, Line 144 that the numerical results can be found in S2 Table. Now the same sentence is stated for all barriers on Page 22, Lines 263-264, Page 24, Lines 307-308 and Page 26, Lines 353-354. Stating that information once around line 144 will probably be enough, similar to: ‘Supplementary material on the detailed results on all environmental barriers (slopes, curbs etc.) can be found in S2 Table.’. Thank you for your comment. Duplicates were removed, and the reference to Table S2 is now only listed in the overview of general results. Page 22, Line 265: The subheading ‘3.2.2.1. Spatio-temporal parameters’ is too big, should be similar to the ‘3.2.2.2. Joint Kinematics’. This was corrected. Discussion Page 27, Lines 387-389: This introductory statement might not necessarily be needed, in the next paragraph it is also summarized that environmental barriers were investigated. The introductory statement was deleted. Page 28, Lines 411-422: One nice combined paragraph can be made out of these three separate paragraphs with the last one only being one sentence. Thank you for your comment. The three paragraphs were merged into one Page 31, Lines 483-497: This has become a rather long paragraph to only state one thing: ‘Instrumented wheels should be used on both sides’. Especially lines 490-495 are repetitions of the first sentences of the paragraph. Thank you for your comment. The paragraph indeed felt long. We shortened the paragraph to restrain from repeating ourselves, and focused the latter half of the paragraph on the fact that the only study that investigated cross-slopes using two instrumented wheels, reported results on the dominant and non-dominant sides of the user without expressing which one was downhill and uphill, which is also an important factor in the biomechanics of cross-slope locomotion Page 32, Lines 509-511: Which studies are you referring to? We added the precision as to which studies we are referring to: all studies presented in this review. Page 32, Line 511: ‘… might have resulted in significant uncertainties’. This was corrected Conclusion: The conclusion is a lot more shortened and to the point, well done. Thank you for your comment. Page 33, Line 538: The first sentence is a bit unclear, seems like there is missing some information. The first sentence was rephrased: “This review highlighted discrepancies in focus given to each environmental situation in the literature”. Page 33, Line 545: ‘… parameters retrieved from the relevant literature.’ This was corrected 10.1371/journal.pone.0269657.r005 Decision Letter 2 van der Woude Lucas Academic Editor 2022 Lucas van der Woude This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 26 May 2022 Manual wheelchair biomechanics while overcoming various environmental barriers: a systematic review PONE-D-21-13523R2 Dear Dr. Rouvier, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements.  Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Lucas van der Woude Academic Editor PLOS ONE 10.1371/journal.pone.0269657.r006 Acceptance letter van der Woude Lucas Academic Editor 2022 Lucas van der Woude This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 31 May 2022 PONE-D-21-13523R2 Manual wheelchair biomechanics while overcoming various environmental barriers: a systematic review Dear Dr. Rouvier: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Professor Lucas van der Woude Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction The US National Institutes of Health (NIH) funded five multi-institutional research and surveillance centers for influenza (Centers of Excellent for Influenza Research and Surveillance \[CEIRS\]) starting in 2006. One of the two primary focus areas of the CEIRS program is to conduct influenza surveillance in domestic and wild animals with the aim of identifying novel and emergent influenza A strains which could transmit to humans. The structure of the CEIRS surveillance labs is somewhat unique. Unlike most networks of government or academically affiliated veterinary diagnostic or public health labs, there are no specific recommendations or standards provided for influenza A testing or for influenza A antibody detection. Individual labs select the most appropriate test for their workflow and specimen type. Furthermore, the diversity of influenza these labs could encounter is unusually broad. First, the labs may test specimens from potentially any animal species although wild waterfowl, domestic poultry, swine, sea mammals and horses are among the most common target species. Secondly, the surveillance labs are located world-wide in 23 countries and include labs on each of the 6 inhabited continents; therefore influenza A from any geographic lineage may be present in samples. In 2012 an external quality assurance (EQA) program was implemented for the CEIRS animal surveillance labs based on the framework described by Wiegers. Similar to other EQA programs the goal was to ensure that all participating labs were utilizing tests with adequate sensitivity and specificity and to provide a way for labs to evaluate and train their personnel for adequate performance. Here we report the results of 5 years of testing and discuss the implementation of an international EQA testing for an infectious disease. # Methods ## General overview of testing and logistics Two sample types were distributed for molecular testing for influenza A and optional subtype identification: 1) avian origin influenza virus, and 2) mammalian origin influenza virus. A third sample set consisted of animal origin serum to evaluate detection of antibodies to influenza A and alternated between swine and chicken sera. Each lab selected which sample sets they would complete. Testing was conducted eight times at approximately six month intervals from June 2012 through February 2016. Labs were only required to participate once per year, but could participate more often. The materials were distributed in coded vials labeled with sample set date, sample number and lot number. Labs requiring more than one set of any type were provided with different lots (each lot had a unique composition and sample order) to minimize the influence from one individual’s results on another individual’s interpretation. Samples remained blinded until all results had been returned. Labs were instructed to process the material exactly as they would process their routine surveillance samples. For reporting purposes 80% or greater correlation with the expected results was considered “passing”. A cut-off of 80% for a passing score was selected because this is the value used by the Clinical Laboratories Improvement Amendments for proficiency testing in clinical laboratories for virological tests of moderate and high complexity. Although Ct values for rRT-PCR, or OD readings for serology, were requested on score sheets, scoring was based on how the technician interpreted the result, i.e. as positive or negative. This approach was implemented because the goal of testing was to imitate a clinical sample and the practical question was whether a sample would be treated as positive or negative for follow-up and eventual reporting by the individual conducting the test. Importantly, due to variations in testing methods and training of individuals there was no standard for cut-off values by Ct value and OD readings (and not all labs used tests which had numerical result). Labs with technicians that scored below 80% on any test were contacted to identify the possible cause. Subtype identification was optional and was only scored for samples and subtypes for which the lab tested. For example, if a lab tested an H3N2 for H3, but not N2, they would only be scored for whether they identified H3 correctly. A result was treated as a false negative if the actual subtype was tested for, but not identified and if no other subtype was identified in the sample (i.e. it was negative by all attempted tests). A result was considered a false positive when another subtype was incorrectly identified, regardless of whether a test for the actual subtype was conducted. ## Molecular sample sets The molecular detection sample sets were comprised of inactivated whole virus. Only isolates not classified as “select agents” by the US Department of Agriculture (USDA) were utilized. A limited number of strains that had been modified to be non-pathogenic by reverse genetics (and which had removed from the select agent list by USDA) were included because of the importance of these strains for subtype identification. Influenza A was propagated in embryonated chickens eggs (ECE) or cell culture by standard methods as appropriate for each isolate. Virus was inactivated in 1% Environ (Steris Inc. Mentor, OH) in distilled water, which has been shown to stabilize the RNA for the first four rounds of testing or 0.1% beta-propiolactone in brain heart infusion broth for the last four rounds of testing. Material for each sample was safety tested by completing two replicates of two blind passages each, in ECE using standard methods for influenza to ensure that the virus had been completely inactivated. Each isolate was then diluted to a pre-determined target concentration and the concentration was confirmed by testing in triplicate to ensure that the target concentration was reliably achieved. Each sample set included a total of 10 samples which represented a range of concentrations: one to three negative (no template) samples and the remaining samples were a mixture of weak positive, positive and strong positive concentrations. The Ct value ranges were based on the ranges of mean egg infectious dose (EID₅₀) equivalents observed for virus shed by infected animals using primers and probe that are a100% match to the virus sequence (e.g. the strong positive was approximately equivalent to 10⁵−10⁷ EID₅₀ per ml with a Ct value of \<26.9; a positive was approximately equivalent to 10³−10⁵ EID₅₀ per ml with a Ct value range of 27.0–32.9; and weak positives were approximately equivalent to 10¹−10³ EID₅₀ per ml for Ct values ≥33). Negative samples contained no template and had an expected Ct value of 0.0. Isolates were selected to represent numerous genetic lineages of the influenza M gene from around the world. Because the M gene was the only gene targeted by the type A influenza RT-PCR tests utilized by participating labs diversity in this gene was important to demonstrate adequate sensitivity with variants. Additional criteria for isolate selection were inclusion of the HA and NA subtypes and lineages that would be most relevant for the surveillance target populations. ## Internal quality control and assurance of molecular samples Initial preparation of the sample sets included testing all materials at their final target concentration in triplicate to ensure that the target concentration was achieved and that reproducibility was adequate. Avian molecular samples were internally calibrated with two different real-time PCR instruments each with RT- PCR reagents from a different manufacturer (7500 FAST \[Applied Biosystems, Foster City, CA\] with the Ambion AgPath kit \[Life Technologies, Waltham, MA\] and the Smart Cycler II \[Cepheid, Inc., Carlsbad, CA\] with the OneStep RT-PCR kit \[Qiagen, Inc, Valencia, CA\]). Both variations used the USDA M gene primers and probes. After the second round of testing it was determined that use of a second test did not provide better information and was discontinued; only the 7500 FAST based test was used subsequently. Mammalian molecular samples were only tested with the 7500 FAST and Ambion AgPath kit, using USDA M gene primers and probes. The test materials were monitored for deterioration by running material that was retained at 1–2 week intervals until all results were returned. Molecular sample sets were stored at both the recommended 4°C and at -20°C, ambient temperature and 37°C to evaluate the effect of different temperatures on sample stability. ## Serology sample material Each serology sample set consisted of 15 samples for the first two rounds, then 10 samples in subsequent rounds of testing. One or two negative samples were included in each set and the remaining samples were adjusted to represent a range of antibody concentrations (weak positive through strong positive) with sterile phosphate buffered saline. All sera were treated at 56°C for one hour in a water bath prior to internal characterization. Diluted sera were tested every two weeks by the methods listed below to ensure sample stability. Mammalian and avian sample sets were initially offered simultaneously, by the third round of testing the species of origin alternated between avian and swine to simplify logistics. To produce sera for the avian antibody sample set, isolates were selected for diversity in the NP protein sequence because this protein is targeted by most in-house and commercial type A influenza antibody tests. The isolates, in allantoic fluid, were inactivated with 0.1% beta-propiolactone and were used to prepare a vaccine with a commercial oil adjuvant, Montanide ISA 70VG (Seppic, Inc. Fairfield., NJ), in accordance with the manufacturer’s instructions. Five week-old specific pathogen free white leghorn chickens were then vaccinated with 0.5ml per bird by the subcutaneous route. Serum was collected 3-weeks later and tested by commercial avian influenza ELISA kits (IDEXX, Westbrook, ME; and Biochek, Scarborough, ME). This specific serum generation in chickens was approved by the Southeast Poultry Research Lab, Institutional Animal Care and Use committee. Serum was only tested by agar gel immunodiffusion (AGID) by standard methods if results were returned from labs which used AGID. Convalescent positive and negative swine sera were obtained from pigs experimentally infected with swine influenza virus (National Center for Animal Diseases, USDA; using IACUC approved procedures) or from commercial pigs in the US. Each serum sample was tested by commercial blocking-ELISA (bELISA) (MultiS- Screen, IDEXX, Westbrook, ME) to characterize the antibody concentration. # Results ## Logistics Attempts were made to minimize transit time; the goal was less than 48hrs in transit and to ensure that materials would pass through customs. Therefore it was critical to contact the recipients prior to shipment and acquire all permits in advance. The most common requirement were letters stating methods of testing, certification that safety testing was completed, and that all material was non- infectious. One country required isolate-by-isolate safety testing data, which is only available a short time prior to shipment, and because of the length of the permitting process, the permits were not issued before the sample material had expired and the material could not be shipped. Two other countries which did not require the same in depth data also had lengthy permitting processes which prevented shipment. Most countries had no permit requirements since the material was non-infectious, although certification letter were required for customs clearance. In most cases there were no issues with clearing customs. However, one country to which six shipments were completed, had very inconsistent requirements and the length of time the material would spend waiting to clear was variable and, in one shipment expired as it spent more than six weeks in customs. Further complicating shipment was that in some cases the recipient could obtain permits for the serology samples, but not molecular samples and vice versa. ## Participation During the eight rounds of testing, 41 labs (a lab is defined as groups under a single principal investigator) from 23 countries ordered at total of 132 avian influenza molecular sample sets, 121 mammalian influenza molecular sample sets and 90 serology sample sets. Results were returned from: 119 (90.1%) avian molecular, 109 (90.1%) mammalian molecular, and 69 (76.6%) serology sample sets. Of the total tests ordered, results were not returned or the EQA testing could not be completed for 13.5% (46/343) of all sets of samples ordered. Reasons included: 1) import permits could not be obtained, so the material could not be shipped; 2) shipments were delayed in transit or customs and the samples degraded over time or from high temperatures so could not be used; 3) there were technical or logistical problems at the labs, e.g. moving, new personnel, equipment or reagent problems, the lab was tasked to work on an emergency outbreak situation; 4) the participating lab never provided results and never responded to inquiries; and 5) labs returned results for fewer sample sets than they received. A total of 111 individual technicians completed at least one type of EQA tests in at least one round of testing. Of the 111 technicians, 31 completed two types of samples and 15 completed sample sets for all the types (avian, mammalian and serology). Participants also held diverse roles in the lab: principal investigators, lab managers, professional technicians, post-docs, graduate students and temporary technicians. ## Avian influenza molecular samples A total of 119 avian influenza molecular sample sets were completed for influenza A detection by 71 technicians from 31 labs. Technicians passed by scoring 80% or greater agreement with expected results on 103 (86.6%) of the samples sets. The discrepant results consisted of 18.2% false positives and 81.8% false negatives. Twenty-eight labs reported methods data. The RNA extraction procedures, RT-PCR reagents, primers and probe sequences and instruments varied widely among the participants. All labs reported using commercial RNA extraction kits, however even labs that used the same kits, sometimes used different procedures based on differing starting volumes and final RNA volumes. Real-time RT-PCR (rRT-PCR) was the most common platform and was used by 26 of the 28 labs that reported methods; the remaining two labs used conventional RT-PCR. All rRT-PCR and RT-PCR based tests targeted the M gene and the most common primer/probe sets were the USDA avian influenza test (8 labs used as reported or with modifications), the US CDC/WHO universal fluA primers (used by 5 labs) or the Fouchier primers (used by 3 labs). The remaining labs reported using other published or non-published primers and probes. Three labs did not provide information on the specific primers sets utilized. The RT-PCR reagents were highly variable among labs regardless of the primers and probe sets utilized and were dictated by regional availability and lab preference. Subtype testing was optional and was completed by technicians from 16 labs. Individual labs tested for different subtypes including: H1 including both 2009 pandemic (pdm(H1N1)09) specific tests and non-pdm(H1N1)09 H1 tests, and, H3, H5, H6, H7, H9, N1 and N2 in samples. Subtype identification was correct for 100 (54.1%) of the subtype tests run (n = 185). False negatives (the test for the correct subtype was negative) accounted for 46.3% of the discrepant results and false positives (an assay was positive for the wrong subtype) accounted for 53.7% of the discrepant results. Based on poor results during the first two rounds of testing two labs changed their M gene primers and probes from an in-house set to a published set and two labs added or modified their H5 or H7 subtype tests to expand testing specificity. The changes resulted in improvements in all subsequent rounds of testing in which they participated. ## Mammalian influenza molecular samples A total of 109 mammalian influenza molecular sample sets were completed for influenza A detection by 56 technicians from 21 labs. Passing scores were achieved on a total of 94 (86.2%) sample sets. Discrepant results were comprised of 29.0% false positives and 71.0% false negatives. Methods data were reported by 19 of 22 labs. All labs used tests which targeted the M gene, however one technician reported using a test targeting the NP gene, although the other technicians in the same lab used an M gene based test. One lab used conventional RT-PCR; all others used real-time RT-PCR tests. Similar to the avian molecular testing, most labs used the USDA type A influenza test or modifications of it (8 labs), the US CDC/WHO primers (5 labs) or the Fouchier test (2 labs). The remaining labs used other published or in-house primers and probes. Similar to the avian samples, RT-PCR reagents were highly variable among labs participating in mammalian influenza testing. Subtype identification was attempted by 12 labs with either real-time RT-PCR or conventional RT-PCR. The actual subtypes assayed varied among labs but included: H1 (both non-pdm(H1N1)09 and pdm(H1N1)09), H3, H5, N1 and N2. Overall the correct subtype was identified by 155 of 204 tests (76.0% correct). Incorrect results included 20.4% false positives and 79.6% false negatives. Two labs modified their type A influenza testing procedures to improve sensitivity with a more diverse set of isolates. One participated in subsequent rounds of testing and improved their score. ## General molecular testing results Fifteen labs participated in testing with both the avian and mammalian molecular sample sets, of which 10 used the same procedure for both sets of samples (seven labs used the USDA test or modified USDA test and three labs used CDC/WHO M gene primers for both avian and mammalian testing, one used an in-house test). Four labs used different primers for avian and mammalian samples. No labs reported using an internal positive control. Twenty-five technicians from 10 of the 35 labs that conducted testing with molecular methods for type A influenza (avian or mammalian) received a failing score (less than 80% agreement with expected results) at some time. Thirty-three percent of the failing scores were from one lab. Six of those labs completed avian and mammalian samples sets and poor scores were observed with both; the remaining four labs only participated in the avian EQA. Nine of these technicians participated in at least one more round of testing and six improved their scores. Reasons for failure were determined by following-up with the lab and included: the participant was inexperienced with the technique, a reagent failure occurred, sample degradation from spending an extended time (greater than six weeks) in customs, or the RT-PCR procedure was insensitive and required modification. ## Serology A total of 69 serology samples sets for influenza A antibody were completed by 45 individual technicians from 23 labs. Passing scores were achieved with 60 (87.0%) of the sample sets. Commercial ELISA kits were utilized by 19 labs (13 of which were from one manufacturer), one lab used AGID (three total tests) and three labs did not report serological methods. Eleven labs used ELISA kits that were not species specific (i.e. blocking ELISA format). Discrepant ELISA results consisted of 98% false negatives and 2% false positives and 100% of the AGID discrepant results were false negatives. Nine technicians from four labs did not pass a serology test at some time during the EQA program. Reasons for failure were traced back to inexperienced technicians, use of AGID, or use of a new ELISA kit. ## General results and outcomes Data integrity errors were noted with 16 sets of results, all from research labs. Typical problems were that the wrong lot or no lot number was reported, the wrong result reporting sheet was used, the results sheet was altered, or data were missing. Testing also helped to identify an equipment problem in one lab. Three labs used the EQA testing material to train new personnel or to help verify new or updated procedures. Another positive outcome was that four labs that did not have access to a reliable source of positive control material retained samples for future use as control material. # Discussion An EQA program was implemented for surveillance labs in year five of the CEIRS program with the goals of: ensuring accurate, consistent results; detecting problems with procedures or operator errors; and to provide material which could be used for assay verification or technician training. Surveillance labs in the CEIRS network are located worldwide and test specimens from diverse domestic and wild avian and mammalian species with the goal of identifying and characterizing novel influenza A viruses in animals. Therefore, the full diversity of influenza A could be present in samples processed by these labs; it is critical for labs to detect unusual isolates (either novel lineages or lineages which have been disseminated from their normal host or geographic region). The labs themselves are also functionally diverse, and included: research labs, veterinary diagnostic labs and even two labs that processed human samples in addition to the animal surveillance samples. Participating labs were government, academic and private, which accounts for the diversity in roles among the participants (students through principal investigators). Interestingly, data integrity issues were exclusively with research labs and may reflect a difference in training, where diagnostic labs place more emphasis on data handling. Participation at the individual level also indicated that most technicians are specialized between the molecular and serological tests; only 25.5% (28/111) participated in both serological and molecular testing. Detection of influenza A (avian or mammalian origin) by molecular methods was very reliable despite a lack of harmonization of the testing procedures. Adequate personnel training and utilization of properly validated methods were crucial. Type A influenza detection discrepancies were mostly false negatives, which suggest sensitivity was the problem more often than cross-contamination (or sample mix-ups). Importantly, the discrepant samples were not always the weakest samples based on analyte concentration. The sensitivity of individual tests correlated to the match between primer/probe sequences and the sample sequence. Similarly, the performance characteristics of the tests varied substantially among the isolates as evidenced by large variations in Ct values for individual samples, for example strong positives could have high Ct values when the primers/probe were a poor sequence match for the isolate (data not shown). Wide variations in Ct values were also reported by Popowich *et al*.. In general it would be advisable for labs to use influenza detection assays that are well validated and which are continually monitored for performance with emerging lineages (e.g. tests which are used by national and international diagnostic laboratory networks). Subtype identification was poor for both avian (54.1% accuracy) and mammalian (75.9% accuracy) isolates. The slightly better accuracy with mammalian isolates is likely due to there being fewer subtypes and less variability among the mammalian lineages, as the isolates are predominantly H1, H3, N1 and N2. Importantly both false negatives and false positives accounted for the errors. False negatives were due to mismatches between the primers and probes. False positives were due to cross reactions, since primers and probes often targeted conserved regions, which are sometimes adequately conserved enough between subtypes for binding to occur. Importantly, subtype identification was attempted by only a few labs. The consequence of misidentified subtypes in this context is unclear since most of the labs will confirm the subtype by sequencing and/or serology. Because of the poor accuracy of identifying random subtypes in avian surveillance samples by rRT-PCR, laboratories should consider using gene sequencing instead. The accuracy of the molecular testing was overall slightly lower than results reported from other influenza molecular proficiency testing of US public health labs, international veterinary diagnostic labs and Chinese public health labs. The difference is probably due to the wider diversity of influenza A isolates used in this testing versus the other testing programs which focused on few or even a single lineage. Testing with a variety of isolates is critical for animal influenza surveillance labs since they can encounter the broadest diversity of isolates relative to public health or regional veterinary diagnostic labs. In contrast Stelzer-Braid reported poorer results with H5N1 testing, but attributed that to the inclusion of very weak samples as results improved when labs modified testing to optimize their limits of detection. Antibody detection was highly reliable with commercial ELISA kits. The few false positives could be attributed to variations in instruments used to read the plates, how well samples were mixed, how the plates were washed, and how long the plates were left before reading (extended times could result in a negative control with an optical density that is too high and a reduced ability to detect true positives). Agar gel immuno-diffusion was utilized by only a few labs, but was prone to false negatives. Compared to commercial ELISA, interpreting AGID requires more skill and the reference reagents must be carefully calibrated to achieve adequate sensitivity. Commercial ELISA kits may be more expensive than AGID on a per sample basis, but assuming they undergo some form of licensure, they offer consistent quality control of the kit materials from the manufacturer, robustness, and a simple format where the results can usually be interpreted by software. One feature of real samples that the molecular EQA samples could not reproduce was the sample matrices that each lab could encounter. These were clean samples, without inhibitors, so the robustness of RNA extraction procedures could not be evaluated. Utilization of an internal positive control or extraction positive control would help ensure that negative results are reliable. Similarly, the serological samples were from only a few species and variations in test sensitivity based on species variations could not be determined. Variation in the sensitivity of blocking ELISAs for antibody from different species has been documented. Overall, the EQA testing had an appreciable impact on the methods and approaches of participating labs. Six labs provided feedback that they modified their molecular detection procedure to improve detection or subtype identification based on the results of the EQA testing. Five of those labs participated in subsequent rounds of testing and improved their scores. Another lab identified an equipment problem when poor results were observed. They were able to fix the problem and complete a repeat of the test within the same testing cycle and achieve 100% agreement with expected results. Several labs were also able to use the material to train technicians and verify their proficiency or to verify new test methods. Importantly, the poor results for influenza A detection or antibody detection were accounted for by only a few labs, indicating that training may be needed since the reported test procedures were validated and should have been adequate. Finally, the results observed here demonstrate a lack of reliability for influenza A subtype identification by RT-PCR methods. The authors gratefully acknowledge: Scott Lee, Kira Moresco, Cristian Flores and Janet Anderson for technical assistance with this work and Dr. Amy Vincent and Dr. Marie Culhane for supplying pig serum. Support was provided by US Department of Agriculture, ARS CRIS Project 6040-32000-063-00D and by federal funds from the National Institute of Allergy and Infectious Diseases, National Institutes of Health, under IAA No. AAI12004-001-00001 and the Minnesota Center for Excellence in Influenza Research and Surveillance HHSN266200700007. Its contents are solely the responsibility of the authors and do not necessarily represent the official views of the NIH. Mention of trade names or commercial products in this publication is solely for the purpose of providing specific information and does not imply recommendation or endorsement by the U.S. Department of Agriculture. USDA is an equal opportunity provider and employer. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** ES CC. **Data curation:** ES SF JM. **Investigation:** ES CC SF JM. **Methodology:** ES CC SF JM. **Resources:** ES CC SF JM. **Writing – original draft:** ES. **Writing – review & editing:** ES CC SF JM.

# Introduction Mast cells play a critical role in the development of asthma and related allergic diseases. Mast cell activation leads to the release of a battery of mediators including preformed granule-derived mediators such as histamine and proteases and newly synthesized prostaglandins, leukotrienes and cytokines. Excess release of these mediators as a result of aberrant activation contributes to allergic disease states. FcεRI-dependent activation of mast cells is characterised by an influx of extracellular Ca²⁺ that is essential for mediator release. A major pathway through which this influx occurs is through Ca²⁺ release activated Ca²⁺ (CRAC) channels, also known as store-operated channels. These channels are activated by the inositol 1,4,5-triphosphate (IP₃)-mediated depletion of the endoplasmic reticulum (ER) Ca²⁺ stores that occurs following cell surface receptor-dependent activation of phospholipase C. CRAC channels were first characterised in rodent mast cells twenty years ago, but the molecular components of the CRAC channel were only recently identified. STIM1 acts as the sensor of the ER Ca²⁺ concentration and transmits this information to the CRAC channel pore. Orai1 (also known as CRACM1) was subsequently identified as the Ca²⁺-selective pore forming protein in the plasma membrane. Two further homologues are expressed in mammalian cells, Orai2 and Orai3. These show a high degree of sequence homology with Orai1 but have distinct functional properties. Heterodimerisation between Orai channel subunits has been reported in heterologous expression systems. It is not yet known whether this also occurs in mast cells. Orai channels are essential for both human and rodent mast cell mediator release. Ca²⁺ influx, degranulation, leukotriene (LT)C₄ release and TNFα production are all greatly reduced in foetal liver-derived mast cells from a Orai1 knockout mouse. Similarly we demonstrated that block of Orai channels in human lung mast cells (HLMCs) using the specific blockers Synta-66 and GSK-7975A reduced Ca²⁺ influx, degranulation, LTC₄ release and cytokine secretion. Human and rodent mast cells express all three Orai subunits at the mRNA level. However the relative contribution of these channels to Ca²⁺ influx in human mast cells is not currently known as current Orai blockers inhibit all family members. In mast cells derived from the mouse Orai1 knockout, Ca²⁺ influx was reduced by 70% with the remaining Ca²⁺ influx blocked by Orai channel inhibitors suggesting that Orai2 and/or Orai3 also contribute to Ca²⁺ influx in rodent mast cells. However, the substantial differences evident between rodent and human mast cells mean that it cannot be assumed that this is the case in human mast cells. Understanding the role of individual Orai family members in HLMCs is important because the development of pharmacological strategies that target individual family members will reduce off-target inhibition of other Orai family members in other cells and body systems. In this study we therefore investigated the relative contributions of Orai channels to Ca²⁺ influx and mediator release in HLMCs. # Methods ## Ethics statement All patients donating tissue gave written informed consent and the study was approved by the National Research Ethics Service (reference 07/MRE08/42). ## Human mast cell purification and cell culture HLMCs were purified from macroscopically normal human lung obtained within one hour of resection for lung cancer and cultured as previously described. Final mast cell purity was \>99% and viability \>97%. Cells were activated with an anti-FcεR1α antibody (Fisher Scientific) as required. ## Adenoviral Transduction of HLMC Adenoviruses (Ad5C20Att01) encoding shRNA directed against human Orai1, Orai2, Orai3 and luciferase, or directing the expression of GFP, were purchased from Biofocus Ltd (Leiden, The Netherlands). Optimal conditions for knockdown of Orai transcripts in HLMC were determined empirically for each virus 2-7 days post infection, at a multiplicity of infection of 10-250. Knockdown efficiency was determined by quantitative RT-PCR. Satisfactory knockdown of transcripts was observed 4 days post infection using a multiplicity of infection of 100 (Orai1 and -2) or 250 (Orai3). This time period was used for all subsequent experiments. Matched experiments using a virus expressing shRNA targeted against luciferase or expressing GFP were used to control for nonspecific effects of adenoviral infection on Orai channel expression and cell viability. Adenoviruses (Ad5C20Att01) expressing dominant-negative pore mutations of Orai1 (E106Q), Orai2 (E80Q) and Orai3 (E81Q) were prepared by Biofocus Ltd using cDNAs supplied by our laboratory. Mutants were generated by the Quik Change method (Stratagene) using Orai1-3 cDNAs cloned into vector pcDNA3 as template. Mutants were verified by sequencing of the entire mutant cDNAs. ## Quantitative RT-PCR Total RNA was isolated from cells using an RNAqueous-Micro kit (Applied Biosystems, Warrington, United Kingdom) according to the manufacturer’s instructions. Approximately 100 ng of total RNA was used to generate cDNA using random decamer primers and a Retroscript kit (Applied Biosystems). TaqMan probes for human Orai1 (Hs00385627_m1), Orai2 (Hs00259863_m1), Orai3 (Hs00752190_s1) and beta-actin (Hs99999903_m1) together with TaqMan Gene Expression Master Mix (all Applied Biosystems) were used for quantification of Orai transcripts. Reactions were run on a Stratagene Mx3000P (Agilent Technologies, Stockport, United Kingdom) real time thermocycler. ## Cell survival assay HLMCs were plated at 5 x 10⁴ cells/well of a 24 well plate in 1 ml culture medium. Recombinant adenovirus was added as appropriate and cells incubated for 96 hours. Cells were pelleted and then resuspended in 20 µl DMEM. An equal volume of trypan blue was added and cells counted in a haemocytometer. Exclusion of trypan blue was used to assess cell viability. ## Patch clamp electrophysiology The whole-cell variant of the patch-clamp technique was used. Patch pipettes were made from borosilicate fiber-containing glass (Clark Electromedical Instruments, Reading, UK), and their tips were heat polished, typically resulting in resistances of 4–6 MΩ. The standard pipette solution contained (in mM): CsCl (140), MgCl₂ (2), HEPES (10), NaATP (2) GTP (0.1) EGTA (5); pH 7.3 with KOH. Prior to use 30 µM IP₃ was added to the pipette solution. The standard external solution contained (in mM), NaCl (140), KCl (5), CaCl₂ (2), MgCl₂ (1), HEPES (10) and glucose (5); pH 7.3 with NaOH. For recording, mast cells were placed in 35 mm dishes containing standard external solution. Whole cell currents were recorded using an Axoclamp 200A amplifier (Axon Instruments, Foster City, CA, USA), and currents usually evoked by applying voltage commands to a range of potentials in 10 mV steps from a holding potential of -20 mV. The currents were digitised (sampled at a frequency of 10 kHz), stored on computer and subsequently analysed using pClamp software (Axon Instruments). Capacitance transients were minimised using the capacitance neutralisation circuits on the amplifier. Correction for series resistance was not routinely applied. Experiments were performed at 27°C, temperature being controlled by a Peltier device. Experiments were performed with a perfusion system (Automate Scientific Inc, San Francisco, CA, USA) to allow solution changes, although drugs were added directly to the recording chamber. Currents in some experiments were also evoked using a ramp protocol consisting of a continuous voltage ramp from -120 mV to +120 mV. ## Mediator assays β-hexosaminidase release was used as a measure of degranulation. 2.5x10⁴ HLMCs in a volume of 90 µl were added to a 96 well V bottom plate in duplicate. Plates were pre-incubated at 37°C for 10 min before activation of cells by the addition of 10 µl 10x anti-FcεRIα antibody (final dilution of antibody 1:300). Cells were incubated at 37°C for 30 min and then either centrifuged, and the supernatant decanted for measurement of mediator content or lysed in 1% Triton X-100 (Sigma) for the determination of total β-hexosaminidase content. 40µl of supernatant or cell lysate were incubated in triplicate with 80µl 4-nitrophenyl N-acetyl-β-D-glucosaminide (Sigma) in 0.05M citrate buffer pH4.5 in a 96 well plate at 37°C for 1-2 hours. The reaction was terminated by the addition of 200µl carbonate buffer pH10.0 and absorbances determined at 405 nm in a microplate reader (Wallac). Net β-hexosaminidase release was expressed as a percentage of the total β-hexosaminidase content. LTC₄ release was measured by ELISA (Cambridge Bioscience Ltd, Cambridge, United Kingdom). ## Statistical analysis Data were compared using paired or unpaired t test and ANOVA or repeated measures ANOVA as appropriate, using GraphPad Prism 5 software. P\<0.05 was considered to be statistically significant. N numbers represent number of HLMCs (from at least 3 independent donors) used for patch clamp experiments, or the number of independent experiments from different donors for mRNA expression, mediator release and viability experiments. # Results ## Orai1 but not Orai2 plays a major role in Ca²⁺ influx in HLMC We used an adenoviral delivery system to transduce HLMCs with shRNAs targeted against Orai1, -2 and -3, or against luciferase as a control. The shRNAs were effective in downregulation of the mRNAs for Orai1 and -2 in HLMCs 4 days following transduction with virus, as shown by quantitative RT-PCR. Knockdown of Orai3 transcripts however required the use of a multiplicity of infection (moi) of adenovirus (moi 250), significantly higher than that required to achieve knockdown of Orai1 and Orai2 transcripts (moi 100). However, at 250 MOI Orai3 and luciferase control transfected cells appeared unhealthy in that they could not be patch clamped and exhibited reduced mediator release (see below). Interpreting the functional effects of Orai3 knockdown was therefore considered unreliable. The level of Orai3 transcripts was also reduced in cells transduced with shRNA targeted against Orai1, although this did not reach statistical significance and was markedly less than knockdown of Orai1 mRNA. The expression of shRNAs against Orai1 and -2 did not affect cell viability monitored by Trypan blue exclusion over the time course of these experiments, compared to cells expressing the luciferase shRNA control. To elicit Orai currents in HLMCs IP₃ was included in the patch pipette to deplete intracellular Ca²⁺ stores. HLMCs transduced with the luciferase shRNA developed an inwardly rectifying current characteristic of that carried by Orai channels. These currents were both qualitatively and quantitatively similar to those we have previously observed from untransduced HLMCs. In experiments to examine the effect of Orai1 knockdown, a mean baseline current of -6.2 ± 1.2 pA (n=19 cells) at -120 mV, with a reversal potential of -4.6 ± 2.7 mV, was recorded from control cells transduced with luciferase shRNA. After 4 min in the presence of IP₃, these currents increased to -33.1 ± 2.1 pA, with a shift in reversal potential to 28.0 ± 4.4 mV (p\<0.0001 compared to baseline for both current and reversal potential). We have previously observed in patch clamp experiments that addition of the Orai channel blocker GSK-7975A at a concentration of 1 µM to HLMCs resulted in the complete inhibition of Orai current in these cells. Addition of 1 µM GSK-7975A similarly inhibited Ca²⁺ influx into IP₃-activated HLMCs transduced with the control shRNA. Mean current at -120 mV was reduced to -6.1 ± 1.0 pA with a shift in reversal potential to -2.6 ± 1.9 mV (p\<0.0001 for both current and reversal potential compared to control). We observed a substantial reduction in the size of Ca²⁺ currents recorded from IP₃-activated HLMCs transduced with shRNA directed against Orai1 compared to luciferase control. Baseline currents in these cells at -120 mV were -4.9 ± 0.8 pA, with a reversal potential of -9.3 ± 1.8 mV (n=13). Following 4 min in the presence of IP₃, these currents increased to -18.0 ± 1.3 pA, with a shift in reversal potential to 16.7 ± 2.9 mV (p\<0.0001 for current, p=0.00018 for reversal potential). This represents a reduction in current to 48.6 ± 8.6% (p=0.0292) of the level of that recorded in control cells as a result of Orai1 knockdown (baseline currents subtracted). Currents generated by IP₃ were again completely blocked following the addition of 1 µM GSK-7975A, being reduced to -5.2 ± 1.3 pA, with a reversal potential of -8.5 ± 2.1 mV (p\<0.0001 for current, p = 0.00019 for reversal potential compared to post IP₃ data). In contrast, we observed only a relatively small reduction in Orai currents in HLMCs transduced with the shRNA targeted against Orai2 compared to control which did not reach statistical significance. Mean baseline current at -120 mV recorded from HLMCs transduced with the control shRNA was -5.9 ± 1.3 pA, with a reversal potential of -7.0 ± 2.1 mV (n=11), while that recorded from cells transduced with the Orai2 shRNA was -5.2 ± 1.2 pA, with a reversal potential of -8.9 ± 2.4 mV (n=14). Following activation with IP₃, currents at -120 mV increased to -30.3 ± 3.8 pA, with a shift in reversal potential to 21.4 ± 4.5 mV in control cells (p\<0.0001 for both current and reversal potential) and to -23.4 ± 2.9 pA, with a reversal potential of 10.7 ± 3.2 mV in cells transduced with the Orai2 shRNA (p\<0.0001 for current, p=0.00059 for reversal potential). This represents a reduction in current to 74.5 ± 12.1% (p=0.3475) of the level of that recorded in control cells as a result of Orai2 knockdown (baseline currents subtracted). In both cases, currents generated by IP₃ were completely blocked by the addition of 1 µM GSK-7975A. HLMC transduced with shRNA targeted against Orai3 or luciferase control using virus at the higher moi of 250 were not possible to record in our patch clamp experiments due to the inability to obtain tight seals. Consequently we were unable to satisfactorily record Orai currents from these cells. ## Knockdown of Orai1 but not Orai2 attenuates HLMC mediator release We next determined the effect of Orai channel knockdown on mast cell mediator release following FcεRIα-dependent activation by measuring the release of β-hexosaminidase as a marker of degranulation and LTC₄ secretion as a marker of arachidonic acid metabolism. In experiments to assess the effect of Orai1 or Orai2 knockdown, net FcεRIα-dependent release of β-hexosaminidase from cells transduced with the luciferase shRNA control virus was 23.4 ± 4.2%, not significantly different to that observed from untreated cells (25.1 ± 4.2%, p=0.371, n=10 independent experiments with HLMC from different donors). FcεRIα- dependent release of LTC₄ from control virus transduced cells was 741.6 ± 226.8 ng/10⁶ cells, again not significantly different from untreated cells (884.7 ± 446.9 ng/10⁶ cells, p=0.9599, n=5). In contrast, the higher moi used in experiments to assess the effects of Orai3 knockdown appeared to result in some non-specific effects, with both β-hexosaminidase and LTC₄ secretion reduced in cells transduced with the luciferase control virus (β-hexosaminidase release 14.7 ± 0.7% with luciferase control, versus 31.4 ± 4.0%, untreated, p=0.0084, n=6; LTC₄ secretion with luciferase control 651.6 ± 242.8 ng/10⁶ cells versus 1199.0 ± 450.2 ng/10⁶ cells untreated, p=0.0146, n=5). This again suggests that Orai3-transfected cells were not healthy, and it was therefore not possible to assess the contribution of Orai3 to mediator release. Knockdown of Orai1 markedly reduced FcεRIα-dependent β-hexosaminidase release to 53.1 ± 4.2% (n=9, p\<0.0001) of that from control cells. In contrast only a slight reduction of β-hexosaminidase release was observed on knockdown of Orai2 (89.0 ± 6.7% of control cells, n=7, p=0.116). A similar pattern of results was obtained on examining the effect of Orai channel knockdown on the release of LTC₄ from activated HLMC. Knockdown of Orai1 resulted in a significant reduction of LTC₄ release to 44.1 ± 5.7% (n=5, p=0.016) respectively of control cells. A much smaller but significant reduction was observed on knockdown of Orai2 (86.1 ± 3.5% of control cells, n=5, p=0.037). ## Dominant-negative pore mutants of Orai channels abolish mast cell degranulation and greatly reduce Ca²⁺ influx To further characterise the composition of Orai channels in HLMCs, recombinant adenoviruses expressing a dominant negative pore mutant of Orai1 (E106Q) or the equivalent mutation in Orai2 (E80Q) and Orai3 (E81Q) were assembled and their effect on activated HLMCs assessed. A recombinant virus directing the expression of GFP was used as a control. After 4 days of culture, there was no significant difference in HLMC viability between cells transduced with the GFP control and those transduced with the Orai1 E106Q expressing virus (n=5, p=0.312), the Orai2 E80Q virus (n=5, p=0.503), or the Orai3 E81Q virus (n=5, p=0.116). This indicates that expression of the dominant negative Orai mutants had no effect on cell viability compared to the GFP control virus during this time period. β-hexosaminidase release from cells transduced with the GFP control virus was 25.5 ± 4.5%, (n=7), not significantly different from untreated cells (27.4 ± 4.8%, n=7, p=0.595). However β-hexosaminidase release from cells transduced with viruses expressing Orai1 E106Q (0.50 ± 0.47%, n=4), Orai2 E80Q (0.92 ± 0.22%, n=4), or Orai3 E81Q (0.22 ± 0.30%, n=4) was essentially abolished. The expression of Orai dominant-negative channels also had a dramatic effect on Ca²⁺ influx into HLMCs measured using patch clamp electrophysiology. Currents at -120 mV elicited by IP₃ in HLMC transduced with the Orai1 E106Q mutant were only 16.8 ± 6.2% (p=0.00016) of those recorded in GFP control cells (baseline currents subtracted). Similarly, currents at -120 mV were reduced to 18.3 ± 7.8% (p=0.0049) of the level in control cells as a result of expression of the Orai2 E80Q mutant; while currents at -120 mV in Orai3 E81Q expressing cells were 15.2 ± 6.1% (p=0.0049) of the level of control. The addition of 1 µM GSK-7975A to IP₃-activated HLMC expressing either GFP, Orai1 E106Q, Orai2 E80Q or Orai3 E81Q resulted in a reduction of current back to baseline levels (not shown). ## Levels of Orai transcripts are altered by mast cell activation Given the apparent roles of Orai1 but not Orai2 in mediator release from HLMCs, we next examined whether levels of Orai transcripts were altered following HLMC activation. Quantitative RT-PCR was performed on RNA purified from HLMCs following activation with FcεRIα antibody for 1 hour and 4 hours or from control untreated cells. Levels of Orai1 and -3 transcripts were little altered 1 hour post activation, while levels of Orai2 transcripts were slightly up-regulated though by an amount that did not reach statistical significance. However, changes in the levels of Orai transcripts were more marked 4 hours following activation. Orai3 transcripts were significantly down-regulated by -3.50 ± 0.60 fold, p=0.0028, n=5. In contrast there was a small but significant up- regulation of Orai1 (1.57 ± 0.16 fold, p=0.0324, n=5) and Orai2 transcripts (1.90 ± 0.32 fold, p = 0.0271, n=5). # Discussion Ca²⁺ influx into human mast cells is essential for the release of mediators such as histamine, tryptase, leukotrienes and cytokines following their activation through the high affinity IgE-receptor (FcεRI). We have recently shown that members of the Orai ion channel family play a major role in FcεRI-dependent Ca²⁺ influx in human lung mast cells. However the individual role if any, of each of the three members of the Orai channel family has not been defined. In this study we present evidence that Orai1 plays important roles in Ca²⁺ influx and mediator release from HLMCs. In addition, we find that levels of Orai transcripts are not static following activation – rather they appear to be actively regulated. Our results are consistent with Orai1 but not Orai2 playing a major role in the influx of extracellular Ca²⁺ into and mediator release from HLMCs following their activation. shRNA knockdown of Orai1 significantly reduced the Ca²⁺ influx that occurs following depletion of intracellular Ca²⁺ stores by IP₃. In addition there was a reduction of approximately 50% in both HLMC degranulation, as measured by the release of β-hexosaminidase, and of newly synthesized LTC₄ release. In contrast shRNA knockdown of Orai2 resulted in only small reductions in Ca²⁺ influx and a marginal although significant reduction in LTC₄ release but not degranulation. Orai2 may therefore play a minor role in Ca²⁺ influx in activated HLMCs, in keeping with its relatively lower levels of expression at the mRNA level. It was not possible to interpret the results of our knockdown of Orai3. While HLMCs transduced with shRNA against Orai3 or luciferase appeared viable morphologically and by Trypan blue exclusion, the high multiplicity of infection required reduced FcεRI-dependent mediator release in the luciferase shRNA control. In addition cells became impossible to study using patch clamp electrophysiology. It is noteworthy that knockdown of Orai1 also reduced Orai3 mRNA expression although this did not reach statistical significance, and this was significantly less than the knockdown of Orai1 expression. It is therefore possible that Orai3 also contributes to HLMC mediator release following FcεRI- dependent activation. However, CRACM3 exhibits rapid inactivation during voltage steps, which is not evident in HLMC CRAC currents, suggesting that Orai1 is the dominant conducting channel. Furthermore, in spite of clear knockdown of Orai1 here, residual Orai3 channels were not evident from the electrophysiological characteristics of the residual raw currents (data not shown). A contribution of Orai3 cannot therefore be confidently excluded, but Orai1 appears to be the dominant channel mediating FcεRI-dependent mediator release from HLMCs. This is in keeping with the marked inhibition of mediator release observed from Orai1 knockout mouse mast cells. We also investigated possible heteromultimerisation amongst Orai subunits by over-expression of dominant-negative pore mutations of Orai1, -2 and -3. Each Orai channel consists of a hexamer of Orai subunits, with each subunit contributing to the central Ca²⁺ conducting pore. We reasoned that if for example Orai1 and Orai3 were to form heteromultimeric channels then expression of a Orai1 dominant-negative construct (E106Q) or a Orai3 dominant negative construct (E81Q) would have a greater effect on Ca²⁺ influx into and mediator release from HLMCs than if Orai1 and -3 form homomeric channels. However we found that expression of Orai1, -2 or -3 channels carrying dominant-negative pore mutations all greatly reduced Ca²⁺ influx and completely abolished mast cell degranulation. These results suggest that Orai channels potentially exist as heteromultimers in HLMCs. Since Orai2 appears to be expressed at lower levels and is less important functionally than Orai1, any heteromulitmers are likely to consist predominantly of Orai1, and possibly Orai3. Ion channels are also implicated in cell survival. In HLMCs and the human mast cell line HMC-1, TRPM7 plays a critical role in maintaining cell survival. Here, we found knocking down Orai1 or Orai2, or inhibiting Orai currents through the transduction of dominant-negative Orai constructs, had no effect on HLMC survival. This is again in keeping with data from the Orai1 knockout mouse. In our previous study, pharmacological blockers of Orai channels reduced degranulation by approximately 50-70% in spite of completely blocking FcεRI- dependent Ca²⁺ currents in individual cells. Knockdown of Orai1 or Orai3 in this study similarly reduced degranulation by approximately 50%, although knockdown was not complete. However, our finding that over-expression of dominant-negative Orai pore mutants completely inhibited HLMC degranulation suggests that the role of Orai channels in HLMC degranulation may be underestimated by pharmacological tools and shRNA knockdown. Furthermore, these results suggest that Orai channels may be the sole Ca²⁺ influx pathway contributing to FcεRI-dependent HLMC degranulation, and that other channels postulated to contribute to this in mast cells such as TRP family members are not operative in HLMC. Our results suggest that under conditions of sustained activation, Orai isoforms may be differentially expressed, allowing the fine-tuning of Ca²⁺ influx into HLMCs. We observed a significant down-regulation of Orai3 transcripts in HLMCs following a four hour period of activation with anti-FcεRIα antibody. Orai1 transcripts were up-regulated over the same time period, though by a smaller amount. There was a similar up-regulation of Orai2 transcripts, further supporting at least some role for Orai2 in Ca²⁺ influx in HLMCs. Differential expression of Orai3 transcripts has also been observed in mammary epithelial tissue and cell lines and also in human T helper (T_H) cells. However in these cases, the development of cancer (mammary tissue/cell lines) or subjection of cells to oxidative stress (effector T_H cells) resulted in upregulation of Orai3 expression. Because of the critical requirement for Ca²⁺ influx for mast cell mediator release, Orai channels are of great interest as potential therapeutic targets for the treatment of asthma and related allergic conditions. Because of the contribution of Orai1 to the HLMC secretory response, a greater understanding of the specific functions of Orai1 in mast cells may permit the fine-tuning of anti-Orai strategies, with the development of drugs which target these family members selectively. This in turn would prevent the off-target inhibition of Orai2 in other cells and body systems, therefore potentially reducing unwanted treatment side-effects. [^1]: The authors have declared that no competing interests exist [^2]: Conceived and designed the experiments: PB IA MLL. Performed the experiments: IA SMD. Analyzed the data: IA SMD MLL PB. Wrote the manuscript: IA, PB.

# Introduction The transcription coactivator p300, which encodes a 300 kDa protein, is a member of the lysine acetyltransferase 3 (KAT3) family of histone acetyltransferase (HAT) transcriptional coactivators and was originally identified as a protein that was bound to the adenoviral E1A. p300 is a large nuclear protein found in most mammalian cells, and it plays an important role in the regulation of genes controlling cell proliferation, apoptosis, differentiation, the cell cycle and DNA repair. Numerous transcription factors can interact with p300 either directly or indirectly through coactivators to stimulate the transcription of specific genes. p300 has been shown to possess acetyltransferase activity, which enables it to influence chromatin structure by modifying nucleosomal histones. The acetylation of amino-terminal lysine residues on histone tails loosens chromatin, thereby increasing the accessibility of the DNA to transcriptional factors. This modification of histones opens the chromatin and facilitates the direct recruitment of specific transcriptional regulators that regulate gene expression. OCT4, NANOG and SOX2 have been identified as core factors that cooperatively form a positive regulatory circuitry for regulating and maintaining the self- renewal and pluripotency of embryonic stem (ES) cells. OCT4, NANOG and SOX2 are also expressed in somatic stem cells, and the overexpression of these pluripotent factors in somatic stem cells results in increased proliferation and differentiation potential in these cells. Several reports have provided strong evidence for the critical role of histone acetyltransferases such as p300, Tip60, Mof and Gcn5 in regulating the expression of genes of the core transcriptional network in mouse ES cells. p300 is involved in regulating the transcription of *OCT4, NANOG* and *SOX2* target genes and thus regulates the transcriptional network in ES cells. p300 can be recruited to genomic sequences bound by OCT4, NANOG or SOX2, and the depletion of any of these factors reduces the binding of p300 at such sites. p300 acetylates histones at the distal regulatory region of NANOG and is therefore directly involved in modulating NANOG expression in differentiating mouse ES cells. p300 may also interact with the transactivation domain of SOX2 and thereby synergistically coactivate SOX2 and OCT3/4. p300 can acetylate the DNA-binding domain of SOX2 and thus enhance global acetylation in mouse ES cells. Dental pulp cells (DPCs) are composed of ectodermic and mesenchymal components containing neural crest-derived mesenchymal progenitors endowed with plasticity and multipotency. The ability of these cells to form colonies is greater than that of MSCs from the bone marrow. DPCs are easily obtained from extracted teeth, possess high proliferative ability, and can be reprogrammed into induced pluripotent stem (iPS) cells at relatively high rates. OCT4, NANOG and SOX2 have been detected in the early passages of cells derived from the dental pulp and might be markers of differentiation of DPCs. p300 can be detected during murine tooth development. However, the expression profile of p300 in HDPCs has not been described, and it is not known whether p300 is involved in regulating the expression of these pluripotency factors or promoting odontogenic differentiation in human dental pulp cells. In this study, we explored the expression pattern of p300 in HDPCs. We then developed HDPC/p300 and HDPC/p300-ΔHAT cell lines. We show that p300 upregulates the expression of NANOG and SOX2 but not OCT4; we also show that the inherent HAT activity of p300 is required for this regulation. The overexpression of p300 can improve the odontogenic differentiation potentiality of HDPCs that have been induced to undergo odontogenic differentiation; however, it has no impact on the proliferation of HDPCs. # Results ## p300 expression levels in wild-type HDPCs First, we examined the levels of p300 mRNA and protein in wild-type primary HDPCs and in HDPCs serially passaged one to seven times. As shown in, p300 mRNA and protein could be detected in the HDPCs. p300 levels were highest in the primary HDPCs and decreased in subsequent passages. p300 levels were almost undetectable in HDPCs passaged seven times. Thus, our data show that p300 mRNA and proteins are present in wild-type HDPCs. We next examined the expression of p300 in HDPCS undergoing odontogenic differentiation. Odontogenic differentiation was induced in HDPCs for 7 and 14 days; real-time qPCR and western blotting showed that p300 mRNA and protein levels decreased during this period and that p300 levels were lower after 14 days than after 7 days. These findings suggest that the expression of p300 is down-regulated during the induction of odontogenic differentiation. ## Stable overexpression of p300 and p300-ΔHAT in HDPCs To investigate whether HDPCs respond to stimulation with p300 or p300-ΔHAT, we stably transduced HDPCs with lentiviral vectors overexpressing p300 or p300-ΔHAT or with a control vector. GFP expression was measured 3 days after the transduction using fluorescence microscopy. We then examined the expression levels of p300 or p300-ΔHAT with real-time qPCR analysis and western blotting. We found that the levels of p300 and p300-ΔHAT mRNA and protein were higher in p300-overexpressing and p300-ΔHAT-overexpressing HDPCs (HDPC/p300, HDPC/p300-ΔHAT) than in the negative control cells (HDPC/V). The expression of GAPDH did not vary significantly among the groups, suggesting that p300 and p300-ΔHAT were stably expressed in the HDPCs. ## Effects of p300 on the proliferation of HDPCs The growth rates of HDPC/p300, HDPC/p300-ΔHAT and HDPC/V cells cultured in normal culture media were compared. Cell proliferation was measured at 1, 2, 3 and 4 days with the CCK8 assay. As shown in, the three types of HDPCs grew at similar rates over the 4 days of observation, and no significant differences in growth rates were observed among the three groups. To confirm these results, we tested cell proliferation with a BrdU assay. As shown in, no significant differences in DNA synthesis were observed among the three groups. These results suggest that p300 may not play an important role in enabling the proliferation of HDPCs. ## p300 regulates the expression of key pluripotency factors in HDPCs To evaluate the effects of the overexpression of p300 on the endogenous expression of the key pluripotency genes *OCT4, NANOG* and *SOX2* in HDPCs, we performed real-time qPCR and western blotting analyses to determine whether mRNA and protein levels of these genes differed in the HDPC/p300 and HDPC/V groups. Real-time qPCR demonstrated that *OCT4, NANOG* and *SOX2* mRNA levels were increased in the p300-transfected HDPCs. As shown in, the patterns of change in the protein and mRNA levels of these pluripotency factors differed. Whereas the protein levels of NANOG and SOX2 were upregulated in cells with p300 overexpression, the protein level of OCT4 was not significantly altered. We then evaluated whether the intrinsic HAT activity of p300 was necessary for the activation of NANOG and SOX2 expression in HDPCs. As shown in, real-time qPCR revealed that the deletion of the p300 HAT domain, which abrogated the HAT activity of p300, partially suppressed its ability to promote the transcription of the *NANOG* and *SOX2* genes. However, significantly higher levels of protein and gene expression were observed in the p300-ΔHAT mutant than in the control. We next confirmed the effects of the wild-type p300 and p300-HAT mutants by western blotting analysis; as was the case with mRNA levels, we found that the levels of NANOG and SOX2 proteins were decreased in cells lacking the HAT domain. These results suggest that functional HAT activity is partially required for the modulation of NANOG and SOX2 mRNA and protein expression by p300. ## p300 enhances the activities of the NANOG and SOX2 promoters Co-transfection studies were performed to determine whether p300 is involved in regulating the activities of the *NANOG* and *SOX2* promoters. First, the constructed promoter reporter plasmid and the exogenous p300 or p300-ΔHAT expression plasmid were transiently co-transfected into HDPCs. The cells were harvested and the relative luciferase activity was measured. As shown in, *NANOG* and *SOX2* promoter activities were increased by p300 overexpression; in contrast, stimulation with the p300-ΔHAT plasmid resulted in much lower levels of *NANOG* and *SOX2* promoter activities. However, significantly higher levels of *NANOG* and *SOX2* promoter activities were observed in the p300-ΔHAT mutant than in the control. The HDPCs were next transiently transfected with varying amounts of the p300 expression plasmid and the reporter plasmid. shows that p300 was able to promote the transcriptional activities of the *NANOG* and *SOX2* promoters in a dose-dependent manner. ## Effects of p300 on odontoblastic differentiation and the mineralization potential of HDPCs following the induction of odontoblastic differentiation Our results showed that the overexpression of p300 in HDPCs upregulated the expression of NANOG and SOX2, the markers of cell “stemness”. We next determined whether p300 induced the HDPCs to be more stem-cell-like (and therefore less differentiated) and whether p300 played a role in regulating the odontoblastic differentiation potential of HDPCs. To test whether p300 can induce the HDPCs to become less differentiated, we measured the mRNA levels of critical marker genes of odontoblastic differentiation (*DMP-1, DSPP, DSP, OPN* and *OCN*) in p300-transfected HDPCs cultured under normal culture conditions using an empty vector as a control. The results indicated that mRNA levels of *DMP-1, DSPP, DSP, OPN* and *OCN* were significantly reduced and demonstrate that the overexpression of p300 can suppress the expression of odontoblastic marker genes and induce HDPCs to be less differentiated. It has been reported that the odontoblastic differentiation ability of HDPCs can be improved by culturing the HDPCs in odontoblastic induction medium for several days. To investigate whether p300 plays a role in regulating the odontoblastic differentiation potential of HDPCs, we examined the expression of odontoblastic differentiation markers, the activity of ALP, and the formation of mineralized nodules in HDPC/p300 and HDPC/p300-ΔHAT cells induced to undergo odontoblastic differentiation. The cells were cultured in a differentiation induction medium for 7 and 14 days. The mRNA levels of *DMP-1, DSPP, DSP, OPN* and *OCN* genes were measured by real-time qPCR. The alterations in the expression levels of *DMP-1, DSPP, DSP, OPN* and *OCN* were similar at all time points in HDPC/p300 and HDPC/p300-ΔHAT cells. Significantly higher levels of mRNA of these genes were observed in HDPC/p300 and HDPC/p300-ΔHAT cells than in HDPC/V cells on days 7 and 14 of odontoblastic induction. These results indicate that p300 can upregulate the expression of odontoblastic marker genes when HDPCs are induced to differentiate, even in the absence of HAT activity. We next measured ALP activity (an early marker of osteoblasts and odontoblasts) in cells not exposed to odontoblastic induction and in cells cultured in odontoblastic induction medium for 3 days. As shown in , HDPC/V cells had higher levels of ALP activity under normal culture conditions. However, when the cells were induced to differentiate for 3 days, HDPC/p300 and HDPCs/p300-ΔHAT cells had greater levels of ALP activity than the control cells. These results indicate that the overexpression of p300 results in increased ALP activity in HDPCs induced to differentiate. The formation of nodules is considered to be an important feature of mineralization in HDPCs. To identify the effect of p300 on the formation of mineralized nodules, cells were cultured in differentiation medium for 21 days and alizarin red S staining was used to assess the formation of calcified nodules. On day 21, mineralized nodules were detected by microscopy; the overexpression of p300 and p300-ΔHAT accelerated the formation of nodules to a similar extent. The number of mineralized nodules was significantly higher in HDPC/p300 and HDPC/p300-ΔHAT cells than in HDPC/V cells. ## p300 is recruited to the promoter of OCN and DSPP and increases the acetylation of lysine 9 of histone H3 of OCN and DSPP promoters To verify whether p300 can be recruited to the promoters of *OCN* and *DSPP* in odontoblastic induction medium, we performed ChIP assays of HDPC/p300 cells. Antibodies against IgG were used as negative controls. The results of this assay showed that anti-p300 but not IgG antibodies immunoprecipitated the *OCN* and *DSPP* promoter in HDPC/p300 cells. Therefore, p300 was recruited to the promoter region of *OCN* and *DSPP*. We then examined whether p300 can affect the histone acetylation of *DSPP* and *OCN* promoters by ChIP assay using an anti-H3K9Ac antibody. demonstrates that the overexpression of p300 enhanced the level of H3K9Ac in the promoter region of the *OCN* and *DSPP* genes. # Discussion p300 is one of several transcriptional coactivators that has been shown to possess intrinsic acetyltransferase activity. p300 acts as a transcriptional adapter for many DNA-binding activators, and its expression has been detected in multiple cell lines. Chen et al. previously reported that p300 expression can be detected during murine tooth development ; however, no data on the expression pattern of p300 in human dental pulp cells are available. In this study, we demonstrate that p300 mRNA and protein can be detected in wild-type HDPCs and that p300 levels gradually decrease when the cells are serially passaged. p300 expression decreased when HDPCs were induced to undergo odontogenic differentiation, which might be related to the reduced FBS content of the odontoblastic induction media. It has been demonstrated that the key pluripotency molecules OCT4, NANOG and SOX2 can cooperatively maintain the regulatory network responsible for self- renewal and pluripotency in ES cells. These pluripotency markers are also expressed in somatic stem cells that have superior expansion and differentiation potential. Previous reports have confirmed that the expression of OCT4, NANOG and SOX2 genes can be detected in dental pulp-derived cells. Although they are expressed at low levels, these genes regulate the “stemness” of dental pulp- derived cells. p300 activity in ES cells is closely linked to the expression of OCT4, NANOG and SOX2, and p300 plays a critical role in the ES cell transcription network. According to a previous study, the expression level of NANOG but not OCT4 was markedly lower in embryoid bodies (EBs) from p300^−/− mouse ES cells than in EBs from wild-type ES cells. This finding suggested that p300 is involved in regulating NANOG expression during the differentiation of mouse ES cells and that this regulation most likely occurs through the epigenetic modification of histones near the NANOG-encoding sequence. The interaction between p300 and SOX2 has also been corroborated by *in vitro* studies. p300 can acetylate the DNA-binding domain of SOX2, thus enhancing global acetylation in mouse ES cells. Mimicking acetylation at a key lysine residue also enhances the interaction of SOX2 with the nuclear export machinery. In our study, we transfected HDPCs with expression vectors for the wild-type p300 and its HAT- deletion mutant to determine whether p300 regulates the expression of OCT4, NANOG and SOX2. We found that the overexpression of p300 upregulated the production of *OCT4, NANOG* and *SOX2* mRNA and of NANOG and SOX2 proteins. The protein levels of OCT4 were not significantly altered, perhaps due to post- transcriptional modifications of OCT4. The overexpression of p300 increased the activities of promoter reporter genes associated with *NANOG* and *SOX2*. We found that the overexpression of a mutant form of p300 lacking the HAT domain in HDPCs resulted in the decreased expression of NANOG and SOX2 mRNA and proteins. The downregulation of NANOG expression by p300-ΔHAT was particularly significant. Li et al. reported that the overexpression of p300 upregulated the promoter activity and the mRNA and protein expression of the human pituitary tumor transforming gene (*hPTTG*). Moreover, the HAT activity of p300 was critical for the regulatory function of this protein. p300 interacts with a series of cardiac-specific genes; however, when the HAT of p300 is inhibited by curcumin, histone acetylation in the promoter regions of these gene does not occur, and the expression of these genes is down-regulated. Our results also showed that the HAT activity of p300 is important for the transcriptional regulation of NANOG and SOX2 in HDPCs. However, the p300-ΔHAT mutant, which lacks the HAT domain, is still able to produce significantly higher levels of these proteins than the control, which has normal endogenous levels of p300. We also showed that p300-ΔHAT was able to stimulate the promoter activities of these genes, albeit at a much lower level. This finding implies that the HAT domain of p300 may partially mediate the regulation of these factors. The upregulation of the expression of these genes and proteins in the p300-ΔHAT mutant cells may be a result of the ability of p300 to indirectly acetylate the relevant histones; however, this upregulation may also be attributable to other functions of p300. The pluripotency factors NANOG and SOX2 are involved in maintaining the regulatory network responsible for the self-renewal ability and differentiation potential of ES cells and somatic stem cells. Here, we show that the overexpression of p300 in HDPCs upregulates the expression of NANOG and SOX2. We next evaluated whether p300 could induce HDPCs to be less differentiated and whether p300 overexpression is involved in regulating the odontoblastic differentiation potential of HDPCs induced to undergo odontoblastic differentiation. We measured the expression levels of the genes *DMP-1, DSPP, DSP, OCN* and *OPN*, which have been used as mineralization markers for the odontoblast-like differentiation of HDPCs, in HDPC/p300 and HDPC/V cells cultured under normal culture conditions and in odontoblastic differentiation induction medium. DMP-1, a candidate protein for dentinogenesis imperfecta, is present in the extracellular matrix of dentine and bone. DSPP is a major non- collagenous protein found in dentin that is essential for odontoblast differentiation and dentin mineralization. DSP and OCN are considered late markers for odontoblast and osteoblast differentiation. OPN is also mainly expressed in the bone and in dentine. We found that mRNA levels of the odontoblastic differentiation markers *DMP-1, DSPP, DSP, OPN* and *OCN* were significantly decreased in HDPCs overexpressing p300 cultured under normal conditions; this finding suggests that p300 mainly interacted with the “stemness” markers *NANOG* and *SOX2* in non-inductive conditions. Thus, the cells were induced to be more stem-cell-like and therefore less differentiated. Paino et al reported that valproic acid (VPA), a histone deacetylase (HDAC) inhibitor, did not reduce cell viability or cell proliferation or impact the cell cycle profile of human dental pulp cells; however, VPA did significantly enhance matrix mineralization by increasing OPN and BSP expression. Jin et al demonstrated that Trichostatin A (TSA), another HDAC inhibitor, promoted the proliferation and odontoblast differentiation of HDPCs induced to undergo mineralization differentiation *in vitro* and enhanced dentin formation and odontoblast differentiation *in vivo* during tooth development. The overexpression of p300 did not overtly affect the ability of HDPCs to proliferate in our study. When grown in odontoblastic induction medium, HDPC/p300 cells produced increased levels of *DMP-1, DSPP, DSP, OPN* and *OCN* mRNA, indicating that p300 increased the odontoblast differentiation ability of HDPCs. The enhanced expression of the odontogenic marker genes under inductive medium was different from that under non-inductive medium, which may be related to weakened interactions of p300 with “stemness” genes. These results were further confirmed by the increased ALP activity (an effective early marker of osteoblasts/odontoblasts) observed in HDPC/p300 cells after differentiation induction. Consistent with this finding, alizarin red S staining showed that more mineralized nodules formed at the late stage of odontogenic differentiation in HDPC/p300 cells. However, the p300 mutants lacking the HAT domain induced to differentiate had similarly increased levels of *DMP-1, DSPP, DSP, OPN* and *OCN* mRNA, similarly increased ALP activity and similarly enhanced mineralized nodule formation as the cells with overexpressed p300. Thus, our findings indicate that the overexpression of p300 improves the odontogenic differentiation potential of HDPCs induced to undergo odontoblastic differentiation, and that HAT may not be required for this regulation. We next focused on the mechanism by which p300 regulates the expression of odontogenic differentiation markers. ChIP assays showed that p300 was recruited to the promoter region of *OCN* and *DSPP*, which implies that p300 regulates the expression of odontogenic markers by directly binding to their promoters. The overexpression of p300 also enhanced the level of H3K9Ac in the promoter regions of the *OCN* and *DSPP* genes, indicating that p300 enhanced the histone acetylation of the *OCN* and *DSPP* promoters and thus promoted the transcription of the genes. However, we also showed that the HAT activity of p300 has no significant effect on odontogenic differentiation. p300 has been shown to not only possess acetyltransferase activity but also to interact with numerous transcription factors directly or indirectly through coactivators to stimulate the transcription of specific genes. When p300 is recruited to target genes, other acetyltransferases might also be recruited as coactivators to the same sites to affect the histone acetylation of the target genes. The acetylation of the promoter regions of the *OCN and DSPP* genes may occur due to compensation by other acetyltransferases for the effect of p300. Hence, we inferred that p300 might be acting as a coactivator to regulate the expression of odontoblastic marker genes. In conclusion, we have shown that p300 is expressed in wild-type HDPCs, that p300 is involved in the upregulation of the key pluripotency molecules NANOG and SOX2 and that the p300 HAT domain is partially required for this upregulation. The mRNA expression of odontoblastic markers, the ALP activity and the formation of mineralized nodules were enhanced when HDPCs overexpressing p300 were induced to differentiate. Thus, our study provides valuable insights into the molecular mechanisms by which p300 regulates the expression of NANOG and SOX2 and the odontogenic differentiation potential of HDPCs. Our findings may have important implications for tooth regeneration studies. # Materials and Methods ## Ethics Statement All patients and guardians on behalf of the minors/children who enrolled in our study gave written informed consent for this investigation; the study was approved by the Ethical Review Board of the Guanghua School of Stomatology of Sun Yat-sen University. The investigation also conforms to the principles outlined in the Declaration of Helsinki. ## Cell culture HDPCs were isolated and cultivated as previously described by Gronthos et al. HDPCs were isolated from impacted third molars extracted for orthodontic reasons from healthy donors aged 17 to 25 years. Only healthy teeth without carious disease or hyperemic pulp tissue were selected. Immediately after extraction, the teeth were washed with 70% ethanol and with phosphate-buffered saline (PBS, pH 7.4). Subsequently, the teeth were cut around the cemento-enamel junction to expose the pulp chamber; the pulp tissue was gently separated from the crown and root using sterile forceps and minced into small pieces. The small fragments were digested for 30 minutes at 37°C in a solution with 3 mg/mL type I collagenase and 4 mg/mL dispase (Gibco, CA, USA). Following digestion, the pulp tissues were placed in 25 cm² culture flasks containing DMEM supplemented with 10% fetal bovine serum (FBS), 100 units/ml penicillin and 100 mg/ml streptomycin (Gibco, CA, USA) and incubated at 37°C in an atmosphere of 95% O2 and 5% CO2. The media were changed every 3 days, and outgrowths from the minced pulp tissue explants were observed after 10 to 14 days of culture. When the cells reached confluence, they were harvested by treatment with 0.25% trypsin with 0.25 mM EDTA (Gibco, CA, USA) and serially passaged for further experiments. Cells from the second or third passage were used for further study. ## Construction and production of plasmids and recombinant lentiviral vectors The expression vectors containing human wild-type (wt) p300 (pCI-p300) or its HAT-deletion mutant (pCI-p300, HATΔ1472–1522) were generously provided by Dr. Joan Boyes (Institute of Cancer Research, London, UK). The recombinant lentivirus vectors pCDH-CMV-MCS-EF1-copGFP containing either human p300 or its HAT-deletion mutant sequences were constructed. The empty lentivirus vector pCDH-CMV-MCS-EF1-copGFP was used as a negative control. The identities of all recombinant vectors were confirmed by polymerase chain reaction (PCR) and DNA sequencing. The recombinant lentiviral vectors were produced by transfecting 293T cells. After the 293T cells reached 40% confluence, they were cotransfected with 20 µg of lentivirus vector DNA and packaging vectors composed of 15 µg of psPAX and 5 µg of pMD2.G in serum-free Opti-MEM using Lipofectamine 2000 reagent (Invitrogen Co., Carlsbad, CA, USA) according to the manufacturer's instructions. Six hours after the transfection, the transfection mix was replaced with fresh DMEM supplemented with 10% FBS. The lentivirus-containing culture medium was harvested and filtered through a 0.45 µm protein binding filter 24 and 36 hours post-transfection. The collected viruses were either used immediately or stored at −80°C in 10% FBS for later use. The *NANOG* and *SOX2* promoter fragments were amplified by PCR using the following primers: *NANOG*: 5′-GTAGAAGGAATGAGAAGACT-3′ (sense) and 5′-TCTTTTAACCACGC TGCACT-3′ (antisense); *SOX2*: 5′-GCTCACTTCCTCTGACTCT- 3′ (sense) and 5′-TGGCAA CAACCAAAACAGT-3′ (antisense). The fragments were cloned into pGL3-Basic vectors (Promega, Madison, WI, USA) between HindIII and BamH1 sites, creating the *NANOG* and *SOX2* promoter/luciferase plasmid. ## Establishment of stable p300/p300-ΔHAT overexpression in HDPCs Cells were maintained in the culture medium and incubated at 37°C in 5% CO₂ until they reached approximately 40% confluence. Subsequently, the cells were infected in DMEM with 10% FBS containing the recombinant lentiviral vectors for 4 hours at a multiplicity of infection (MOI) of 50; the supernatant was then replaced with fresh culture medium. Polybrene (8 µg/ml) was added to increase the infection efficiency. Cells were infected with the lentivirus three times over the course of 48 hours. The expression of green fluorescent protein (GFP) detected using fluorescence microscopy (Zeiss, Jena, Germany) showed that approximately 80% of the cells were infected within three days of the transfection. Puromycin (Sigma-Aldrich, St Louis, MI, USA) was added at a concentration of 1.25 µg/ml to the medium for 2 to 4 days to select stably transfected cells. ## Odontoblastic induction of HDPCs For odontoblastic differentiation, cells were seeded in 24-well and 12-well plates at a density of 2×104 and 5×104 cells per well, respectively, and cultured in DMEM containing 10% FBS until they reached 80% confluence. The media were then replaced with odontoblastic differentiation media containing DMEM supplemented with 5% FBS, 50 µg/ml ascorbic acid, 10 mM b-glycerophosphate and 10⁻⁷ M dexamethasone (Sigma-Aldrich, St Louis, MI, USA). The culture medium was changed every 3 days. ## Cell proliferation assay The effect of p300 on the proliferation of HDPCs was examined with the CCK-8 (Cell Counting Kit-8) assay (Dojindo, Kumamoto, Japan) and BrdU assay (Sigma- Aldrich, St Louis, MI, USA) according to the manufacturer's protocol. For the CCK-8 assay, a total of 3×10³ cells per well were cultured in 96-well plates at 37°C in 100 µl of DMEM with 10% FBS. After 1, 2, 3 or 4 days of incubation, the supernatant was removed, and 110 µl of DMEM medium containing 10 µl of CCK-8 were added to each well for another 2 hours at 37°C. The optical density (OD) value for each well was read at 450 nm using an automated microplate reader (Sunrise, Tecan, Switzerland). For the Brdu assay, the HDPSC/v, HDPSC/p300-ΔHAT and HDPSC/p300 cells were labeled with 10 µmol/L BrdU for 4 hours and fixed with 70% ethanol at −20°C overnight. The cells were washed three times in PBS containing 0.5% IFS (Gibco, CA, USA) and treated with 2 mol/L HCl with 0.5% IFS for 20 minutes at room temperature. To stain the nuclei, cells were incubated in serum-free medium with an anti-BrdU antibody (ab152095-Abcam, Cambridge, UK) for 20 minutes at 4°C. A secondary FITC-conjugated rabbit anti-mouse antibody (EarthOx, CA, USA) was applied for another 20 minutes. After extensive washing, the stained cells were counted under a fluorescence microscope (Zeiss, Jena, Germany). ## Real-time quantitative PCR Total RNA was extracted from dental pulp cells using TRIzol reagent (Invitrogen, Carlsbad, CA, USA) according to the manufacturer's instructions. The extracted RNA was pretreated with RNase-free DNase (Promega, Madison, WI, USA), and 2 µg of RNA from each sample were reverse-transcribed for cDNA synthesis using the RevertAid First Strand cDNA Synthesis Kit (Fermentas, Ontario, Canada) and random primers. Subsequently, the cDNA was used as a template for the PCR. Real- time quantitative RT-PCR was performed using the LightCycler 480 SYBR Green I Master (Roche, Basel, Switzerland) with specific primers according to the manufacturer's instructions. GAPDH was used as a house-keeping gene to normalize gene expression levels. The following reagents were mixed in a final reaction volume of 20 µl: 10 µl of Master Mix, 1 µl of forward and reverse primers, 3 µl of H2O, and 5 µl of cDNA. Cycling conditions were as follows: 1 cycle at 95°C for 3 minutes, followed by 40 cycles at 95°C for 10 seconds, 55°C for 10 seconds, and 72°C for 30 seconds. Relative differences in PCR results were calculated using the comparative ΔΔCt method. Primers were designed using Primer Express v 3.0 software (Applied Biosystems, Inc. Foster City, CA, USA) as follows: *p300*: 5′-GCGGCCTAAACTCTCATCTC-3′ (sense) and 5′-TCTGGTAAGTCGTGCTCCAA-3′ (antisense); *NANOG* 5′-GATTTGTGGGCCTGAAGAAA-3′ (sense) and 5′-ATGGAGG AGGGAAGAGGAA-3′ (antisense); *OCT4*: 5′-GTGGAGGAAGCTGACAACAA-3′ (sense) and 5′-GGTTCTCGATACTGGTTCGC-3′ (antisense); *SOX2*: 5′-GCTTAGCCTCGTCGATG AAC-3′ (sense) and 5′-GCTTAGCCTCGTCGATGAAC-3′ (antisense); *DMP-1*: 5′-TGGGCA TAGATTTCCTCTTTG-3′ (sense) and 5′-TGAGCAGGATGCTGATCTTC-3′ (antisense); *DSPP*: 5′-GCCACTTTCAGTCTTCAAAGAGA-3′ (sense) and 5′-GCCCAAATGCAAAAATATG TA-3′ (antisense); *DSP*: 5′-CAGGATGTACTATTCTCGGCG-3′ (sense) and 5′-GTGTTCT GGTTCTGGTGCCT-3′ (antisense); *OCN*: 5′-CTCACACTCCTCGCCCTATT-3′ (sense) and 5′-TTGGACACAAAGGCTGCAC-3′ (antisense); *OPN*: 5′-GTGATTTGCTTTTGCCTCCT-3′ (sense) and 5′-GCCACAGCATCTGGGTATTT-3′ (antisense); *GAPDH*: 5′-AAGGTGAA GGTCGGAGTCAA-3′ (sense) and 5′-AATGAAGGGGTCATTGATGG-3′ (antisense). ## Western blotting Cells were harvested in protein lysis buffer (50 mM Tris, 150 mM NaCl, 1% Triton X-100, 1% sodium deoxycholate, 0.1% SDS, sodium orthovanadate, sodium fluoride, EDTA) containing protease inhibitors (Beyotime, Haimen, China) and incubated on ice for 30 minutes. Protein concentrations were measured with a bicinchoninic acid (BCA) protein assay (Beyotime, Haimen, China). Thirty micrograms of protein were separated by 10% sodium dodecyl sulfate- polyacrylamide gel electrophoresis (SDS-PAGE) and electrophoretically transferred onto a polyvinylidene fluoride (PVDF) membrane (Millipore, Billerica, MA, USA). The membrane was incubated for 1 hour at room temperature in TBST containing 5% skim milk to block nonspecific protein binding and incubated at 4°C overnight with the primary antibodies anti-p300 (ab54984-Abcam, Cambridge, UK; 1∶200), anti-NANOG (ab62734-Abcam, Cambridge, UK; 1∶1000), anti-OCT4 (ab19857-Abcam, Cambridge, UK; 1∶500), anti- SOX2 (ab97959-Abcam, Cambridge, UK; 1∶500) and anti-GAPDH (ab8245-Abcam, Cambridge, UK; 1∶3000). After washing, the membrane was incubated for 1 hour with an HRP-conjugated secondary antibody (ab6728-Abcam, Cambridge, UK; 1∶1000) at room temperature. Antibody binding was visualized with an enhanced chemiluminescence system (Millipore ECL Western blotting Detection System, Millipore, Billerica, MA, USA), and band densities were obtained and normalized to GAPDH and the background using ImageJ. ## Luciferase reporter assay The HDPCs were seeded in 24-well plates and cultured for 24 hours. The HDPCs were then co-transfected with the promoter/luciferase plasmid and the pCI-p300 plasmid or the HAT-deletion mutant (pCI-p300, HATΔ1472–1522) plasmid using Lipofectamine 2000 reagent (Invitrogen); the empty plasmid pCI was used as a negative control. Cells were incubated at 37°C for 5 to 8 hours, after which the culture medium was changed. Cells were harvested after another 24 hours. Cell lysates were collected and the luciferase activity was measured with a Turner Designs TD-20/20 Luminometer in the Dual-Luciferase Assay System (Promega). A second luciferase gene from Renilla reniformis provided constitutive activity as an internal control. ## Alkaline phosphatase activity analysis Cells were seeded at a density of 2×10⁴ cells per well in 24-well plates and cultured in DMEM with 10% FBS until they reached 80% confluence; the cells were then cultured in odontoblastic differentiation medium for 0 and 3 days. The medium was replaced every 3 days. Cells were washed three times with PBS; 100 µl of 1% Triton X-100 were then added to each well and the plates were incubated at 4°C overnight. An ALP assay kit (Nanjing Jiancheng Bioengineering Institute, Nanjing, China) was used to analyze ALPase activity according to the manufacturer's instructions. The absorbance was measured at 520 nm using an automated microplate reader (Sunrise, Tecan, Salzburg, Switzerland). The protein content was quantified using a BCA protein assay (Beyotime, Haimen, China). The amount of ALP in the cells was normalized against the total protein content. ## Alizarin red S staining To assess *in vitro* mineralization, cells were seeded in 12-well plates and cultured in DMEM with 10% FBS until they reached 80% confluence; the cells were then cultured in the differentiation medium for 21 days. After culture, the cells were washed three times with PBS (pH 7.4), fixed in a 4% paraformaldehyde solution for 15 minutes, and washed three times with deionized water. Cells were then incubated in 1% Alizarin red S for 30 minutes, washed 5 times with distilled water, and dried at room temperature. The mineralized nodules were photographed using an inverted microscope (Zeiss, Jena, Germany). ## Chromatin immunoprecipitation assays (ChIP) HDPC/p300 and HDPC/V cells were cultured in odontoblastic differentiation medium for 7 days. The cells were then fixed by adding formaldehyde at a final concentration of 1% and incubated for 10 minutes at 37°C. Glycine was added to the samples to terminate the cross-linking reaction. The samples were incubated for 5 minutes at room temperature, lysed with SDS lysis buffer and protease inhibitor, and sonicated to generate 500–1000 bp DNA fragments. Agarose gel electrophoresis was used to confirm that the sonification had produced the appropriate degree of shearing of chromatin. Cross-linked proteins were immunoprecipitated using antibodies against p300 (ab14984-Abcam, Cambridge, UK), H3K9Ac (ab10812-Abcam, Cambridge, UK) and IgG (ab2410-Abcam, Cambridge, UK). Complexes were washed in low-salt, high-salt and LiCl buffers once and in a TE buffer twice. The DNA was then extracted and precipitated. To remove the protein-DNA cross-links, NaCl was added to the samples and the samples were incubated overnight at 65°C. The eluates were diluted with ChIP dilution buffer and underwent a secondary immunoprecipitation reaction. To digest the isolated protein, 10 µl of 500 mM EDTA, 20 µ l of Tris HCl (pH 7.2), and 1 µl of proteinase K were added to each sample; the samples were then incubated for 1 hour at 45°C. The enrichment of the DNA template was amplified by PCR using primers designed to specifically amplify the *OCN* and *DSPP* promoter region containing p300 binding sites. The following primer sequences were used: *OCN*: 5′-CTTTGGCTGGCAGTCCCT-3′ (sense) and 5′-GCTTGCTCTTCCC TCCTCT-3′ (antisense); *DSPP*: 5′-GGCCACTTTCAGTCTTCA′ (sense) and 5′-ATCAAACTGG CTTCATCT-3′ (antisense). ## Statistical analysis Each experiment was performed in triplicate and repeated at least three times. Experimental groups were compared using one-way analysis of variance (ANOVA) or repeated measures ANOVA with SPSS13.0 software (SPSS, Chicago, IL, USA). Values are expressed as the means ± SD. All P-values are two-tailed and *P*\<0.05 was considered statistically significant. We thank Dr. Joan Boyes (Institute of Cancer Research, London, UK) for kindly providing pCI-p300 and pCI-p300 (HATΔ1472–1522) plasmids. We also thank Guang Shi, PhD, from College of Life Science, Sun Yat-sen University, for her advice. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: QX. Performed the experiments: TW HL. Analyzed the data: YN. Wrote the paper: TW HL QX. Revised the manuscript: QX.

# Introduction The calpain system is composed of a family of Ca²⁺-dependent cysteine proteases, their activity being regulated by cytosolic calcium. Calpains originally comprised two proteases, calpain 1 (Capn1) or µ-calpain (micromolar) and calpain 2 (Capn2) or m-calpain (millimolar). Also referred to as the “conventional” or “ubiquitous” calpains, they are specifically inhibited by calpastatin. Calpains appear to be involved in cell motility, signal transduction, cell cycle progression, gene expression regulation and apoptosis. Calpains are known to play an important role in macroauthophagy, being regulators of autophagosome formation, and in some viral oncogenetic mechanisms. Furthermore, calpains represent a promising target for cancer therapy, since they appear to play a key role in metastatic cell migration and angiogenesis. Calpain 3 (Capn3), also known as calpain p94, is a member of the so-called tissue-specific subfamily, being predominantly expressed in skeletal muscle. Its structure is similar to that of the other members of this family but it presents three specific sequences not found in any other calpain: a novel sequence (NS) at the N-terminus, insertion sequence 1 (IS1) within the catalytic domain, and insertion sequence 2 (IS2) upstream of the Ca²⁺-binding domain. It has been suggested that Capn3 may play a role in sarcomere remodelling and in mitochondrial protein turnover. Point mutations in the Capn3 gene are responsible for limb girdle muscular dystrophy type 2A (LGMD2A), an autosomal recessive disease characterized by progressive atrophy and weakness of the proximal limb muscle. Furthermore, several variants of Capn3 have been reported in many tissues including the eye, peripheral blood mononuclear cells, and astrocytes, thus suggesting that Capn3 is important for tissues other than skeletal muscle. Recently, novel Capn3 isoforms in white blood cells have been sequenced and are being investigated to develop a new approach for performing the molecular diagnosis of LGMD2A at mRNA level. To our knowledge, no investigation has so far been carried out on Capn3 expression in veterinary and comparative spontaneous carcinogenesis. The aim of the present paper is to report, for the first time in medical literature, the activation of the Capn3 protease in urothelial tumors of the urinary bladder in cattle. Such tumors are very commonly found in adult cattle grazing on lands rich in bracken fern. Furthermore, a strong relationship between bracken fern and bovine papillomavirus type 2 (BPV-2) has been established, so that BPV-2 infection appears to be a pivotal event in the bladder carcinogenesis of cattle. # Methods ## Ethics Statement In our cases we didn't perform any experimentation as we collected tissue samples directly in public slaughterhouses. All the animals we studied were slaughtered after a mandatory clinical *ante-mortem* examination, as required by European Union legislation. ## Bladder samples Samples of bladder neoplastic urothelium were collected at public slaughterhouses from twelve 4- to 24-year-old cows that had suffered from chronic enzootic hematuria for several years. Samples of bladder mucosa without any apparent gross lesions due to tumor proliferations were also obtained from the same animals. All animals had been raised in hilly/mountain cattle households in the South of Italy and were known to have grazed on pastures rich in bracken fern. Normal bladder mucosa was obtained from five 4- to 14-year-old healthy cows which had grazed on pastures in which no bracken was present. Bladder samples were routinely divided into four parts. One part was fixed in 10% buffered formalin. Two parts were immediately frozen in liquid nitrogen, stored at −80°C and utilized for molecular procedures. The remaining part was frozen in isopentane pre-cooled in liquid nitrogen and stored at −80°C until further processed for immunohistochemistry. ## Histopathology The tissues fixed in 10% buffered formalin were routinely processed for paraffin embedding. Histologic diagnosis was assessed on 5-µm-thick haematoxylin-eosin (HE)–stained sections using morphological criteria suggested in a recent report on the new histological classification of urothelial tumors of the urinary bladder of cattle. ## E2F3 immunoprecipitation for proteomic analysis E2F3 overexpression is known to play a crucial role in bladder carcinogenesis. To investigate its potential molecular partners, a co-immunoprecipitation experiment with E2F3 was performed. Tissues were lysed in ice-cold buffer containing 50 mM Tris-HCl (pH 7.5), 1% (v/v) Triton X-100, 150 mM NaCl, 2 mM PMSF, 1.7 mg/ml Aprotinin, 50 mM NaF, and 1 mM sodium orthovanadate. Lysates were clarified by centrifugation at 11,000 g for 30 minutes. Supernatants were collected, and protein concentration was determined by a modified Bradford assay (Bio-Rad). One mg per sample of proteins was immunoprecipitated using, in a first step, IgG mouse (Sigma) with 30 µl of G-sepharose (Ge Healthcare) for preclearing, and, in a second step, 2 µg of anti-E2F3 antibody (Upstate) with the same amount of G-sepharose. Immunoprecipitates were washed four times in a complete lysis buffer and finally heated in Laemmli buffer composed of glycerol 40% (Sigma), β-mercaptoethanol 0,35 M (Sigma), SDS 5% (Sigma), and Blue Bromophenol (Roche). Immunoprecipitates were separated on 4–12% polyacrylamide gels and then submitted to in-gel digestion. ## In-gel digestion of IP protein bands Gel bands for mass spectrometric analysis were basically processed according to Shevchenko *et al.*. Sliced gel pieces were washed with 100 mM NH₄HCO₃ and acetonitrile (1∶1, v/v) (buffer A). HPLC-grade acetonitrile was obtained from Sigma-Aldrich (St. Louis, MO). Proteins were in- gel reduced by 10 mM DTT, and subsequently alkylated with 20 mM iodoacetamide. After a washing step with buffer A, the gel pieces were dried in a vacuum centrifuge, and rehydrated at 4°C in a digestion buffer (50 mM NH₄HCO₃, 5 mM CaCl₂) containing 25 ng/µl trypsin. After overnight incubation, peptides were extracted from the gel using three separate washings with a mixture of acetonitrile/water/formic acid 70/25/5 (v/v/v). Extracts were combined and dried down in a vacuum centrifuge. The lyophilized digests were reconstituted in 30 µl of loading pump solvent (see Nanoscale LC-MS/MS Section). An aliquot of the solution (10 µl) was then injected for nanoscale LC-MS/MS analysis. ## Nanoscale Liquid Chromatography coupled with tandem Mass Spectometry (nano LC-MS/MS) Chromatography was performed on an Ultimate nano LC system from Dionex (Sunnyvale, CA). The analytical nano LC column used was an in-house packed 75 µm i.d., 40 cm long Integra Frit™, column from New Objective (Cambridge, MA), filled with 4 µm C₁₂ silica particles Jupiter Proteo from Phenomenex (Torrence, CA); a mixture of the peptide (10 µL) was loaded onto an in-house packed 150 µm i.d., 3 cm long Integra Frit™ (New Objective) trapping column (packing bed length 1 cm) at 12 µL/min of loading pump solvent, consisting of H₂O/acetonitrile/trifluoroacetic acid (TFA) 97.95:2:0.05 (v/v/v). After a 2-minute washing, the trapping column was switched on-line to the analytical column, and gradient separation started at 200 nL/min. A binary gradient was used for peptide elution. Mobile phase A was H₂O/acetonitrile/formic acid/TFA 97.9:2:0.09:0.01 (v/v/v/v); mobile phase B was H₂O/acetonitrile/formic acid/TFA 29.9:70:0.09:0.01 (v/v/v/v). Gradient was from 5 to 45% B in 60 minutes at 200 nL/min flow rate. After 10 minutes at 95% B, the column was re-equilibrated at 5% B for 30 minutes before injection. MS detection was performed on a QSTAR XL hybrid LC-MS/MS from Applied Biosystems (Foster City, CA) operating in positive ion mode, with nESI potential at 1800 V, curtain gas at 15 units, CAD gas at 3 units. Nanoelectrospray ionization was achieved via distal coated Pico Tips™ 20 µm ID, 10 µm tip ID (New Objective). Information-dependent acquisition (IDA) was performed by selecting the two highest peaks for MS/MS analysis after a full TOF-MS scan from 400 to 1600 m/z lasting 4 seconds. Both MS/MS analyses were performed in enhanced mode (3 second/scan). Threshold value for peak selection for MS/MS was 20 counts. ## RNA isolation, cDNA synthesis, and PCR Total RNA was isolated both from normal and neoplastic mucosa. Furthermore, RNA was also isolated from the same neoplastic bladders in an apparently unaffected mucosa region (since it did not show any gross neoplastic lesions), according to manufacturer's instructions (Paris kit, Ambion). cDNAs were synthesized from bovine bladders or human PBMC starting from total RNA (4 µg), using Thermoscript RT-PCR system (Invitrogen). Calpain 3 transcript was amplified from cDNAs using PCR_x Taq Polymerase Enhancer System (Invitrogen). A pair of primers flanking a region of the protease sequence known to contain calpain 3 (Bos Taurus calpain 3 p94, CAPN3 GenBank Accession No NM_174260.2) specific insertion sequence 1 (IS1) were used: forward primer (Sn 831) 5′-CTGCTGGAGAAGGCTTATGC-3′, reverse primer (Asn 1845) 5′-CCCGCATGTTGATGTAGGTT-3′. Another pair of primers flanking a region of the protease sequence known to contain calpain 3 specific insertion sequence 2 (IS2) was also used: forward primer (Sn 1884) 5′-GTCATCGTGCCCTCCACC-3′, reverse primer (Asn 2669) 5′-TCAGGCATACATGGTGAGCTGCAG-3′. PCR was performed using the following parameters: a denaturation step for 2 min at 98°C; then 95°C for 40 s, 55°C for 40 s and 72°C for 2 min, for 40 cycles. To determine the relative levels of calpain 3 transcript in different animals, equal amounts of each cDNA sample were amplified in the presence of primers (forward primer 5′-ACGACCCCTTCATTGACC-3′, reverse primer 5′-TGCTTCACCACCTTCTTG) specific for glyceraldehyde-3-phosphate dehydrogenase (GAPDH). PCR conditions were the following: a denaturation step for 2 min at 98°C; then 95°C for 30 s, 55°C for 30 s and 72°C for 1 min, for 22 cycles. PCR products were separated by electrophoresis on 1.2% agarose gel (Biorad). ## Detection of calpain 3 activity in bovine bladders Bovine bladder samples collected as above described were stored at −80°C until further processed. Bladders from healthy or pathological animals were minced, homogenized and lysed by sonication (six bursts, 30 sec each) in five volumes of ice-cold 50 mM Na Acetate buffer pH 6.7, containing 1 mM EDTA, 0.5 mM 2-mercaptoethanol. The particulate material was discarded by centrifugation (100,000 g for 15 min, 4°C) and the soluble fraction (35 mg protein) was loaded onto a ion-exchange DE52 column (4 ml), previously equilibrated in 50 mM Na Acetate buffer pH 6.7 containing 0.1 mM EDTA, 0.5 mM 2-mercaptoethanol. The absorbed proteins were eluted with a linear gradient (70 ml) 0–0.4 M NaCl in 1 ml fractions. Calpain activity was assayed on aliquots (150 µl) of each eluted fraction as previously described. ## Immunohistochemical examination Six µm-sectioned frozen samples were fixed in acetone at 4°C for 5 min, then blocked for endogenous peroxidase in 0.3% H₂O₂ in methanol for 20 min. Tissue sections were then incubated overnight at room temperature with an anti-rabbit calpain p94 mouse monoclonal antibody (Chemicon International, Billerica, MA, USA), diluted 1∶100 and 1∶200. Slides were washed three times with PBS, then labelled with streptavidin biotin (LSAB kit; DakoCytomation, Denmark) for 30 min, followed by incubation with streptavidin conjugated to horseradish peroxidase (LSAB Kit; DakoCytomation, Denmark). Color development was obtained following 5–20 min of diaminobenzidine (DakoCytomation`,` Denmark) treatment. Sections were finally counterstained with Mayer's hematoxylin. ## Viral DNA analysis A small fragment of frozen tissue of the urothelial tumors was digested by Proteinase-K in a lysis buffer (50 mM KCl, 10 mM Tris HCl, pH 8.3, 2.5 mM MgCl2, 100 µg/ml gelatin, 0.45% NP-40, and 0.45% Tween-20), in order to recover the genomic DNA. Ten microliters of each sample were amplified by PCR utilizing one unit of Taq polymerase (Platinum*Taq* Invitrogen, Milan, Italy) in 50 µl of the buffer provided by the manufacturer with 3 mM MgCl₂, and 2.5 mM of each dNTP. The reaction was carried out in an iCycler (Bio-Rad Laboratories, Milan, Italy) using the 5′-TACTGTTTCTGCTGCTATTT-3′ forward primer and the 5′-ACAAATCAAATCCACATAATAGTA-3′ reverse primer that amplify a small fragment of BPV-2 (125 bp). PCR conditions were as follows: denaturation for 2 min at 95°C, followed by 35 cycles of denaturation at 95°C for 30 s, annealing at 50°C for 30 s, and extension at 72°C for 1 min. The final PCR products were electrophoresed in 2% agarose gel and visualized by ethidium bromide stain. Each experiment included a blank sample consisting of reaction mixture without DNA and a positive sample consisting of a recombinant plasmid carrying the genomic sequence of BPV-2 (kindly provided by Dr. M. S. Campo, Glasgow University, Scotland). A band corresponding to the size of the amplified sequences of BPV-2 was detected in the examined cancer samples. To confirm the PCR data, the amplified products were purified through silicagel membranes by the QIAquick PCR quantification kit according to the manufacturer's instructions (QIAgen, Milan, Italy) and then subjected to direct sequencing in an automated apparatus (Biogen, Rome, Italy). ## Immunoprecipitation and immunofluorescence for BPV-2 E5 protein Tissues were lysed in ice-cold buffer containing 50 mM Tris-HCl (pH 7.5), 1% (v/v) Triton X-100, 150 mM NaCl, 2 mM PMSF, 1.7 mg/ml Aprotinin 50 mM NaF, and 1 mM sodium orthovanadate. Lysates were clarified by centrifugation at 10,000 g for 30 minutes. Supernatants were collected, and protein concentration was determined by a modified Bradford assay (Bio-Rad). One mg per sample of proteins was immunoprecipitated using 2 µg of anti-E5 antibody (kindly provided by Dr. Campo, Glasgow University, Scotland) and 30 µl of G-sepharose (Ge Healthcare). Immunoprecipitates were washed four times in complete lysis buffer and finally heated in LDS loading buffer 4X (Invitrogen) at 70°C for 10 minutes according to the manufacturer's protocol. Immunoprecipitates were separated on 4–12% polyacrylamide gels and transferred to nitrocellulose filter membranes (Biorad). Membranes were blocked in 5% nonfat dry milk, incubated with primary antibodies, detected by the appropriate secondary antibodies, and revealed with an enhanced chemiluminescence system (Amersham Biosciences). For immunofluorescence, paraffin sections were deparaffinized, rehydrated and heated in a microwave oven (twice, for 5 min each at 750 W) to allow antigen to be unmasked. Slides were then incubated with a 1∶50 dilution of rabbit anti-E5 antiserum (kindly provided by Dr Schlegel, Georgetown University, USA) and, thereafter, with a FITC-conjugated secondary antibody (Chemicon). Immunofluorescence was analyzed with a Zeiss LSM 510 laser scanning confocal microscope (Carl Zeiss GmbH, Jena, Germany). ## Western blot analysis In order to validate protein identification based on a single protein as detected by proteomic approach and also investigate the expression levels of E2F3 and retinoblastoma tumor suppressor protein (pRb), the latter known to regulate E2F activities, western blot analysis was performed. Briefly, tissues were lysed in a buffer containing 50 mM Tris-HCl (pH 7.5), 1% (v/v) Triton X-100, 150 mM NaCl, 2 mM PMSF, 1.7 mg/ml Aprotinin, 50 mM NaF, and 1 mM sodium orthovanadate, using an Ultra Turrax (Ika-Werke). LDS loading buffer 4X (Invitrogen) was added to the protein samples (40 µg) and they were then heated at 70°C for 10 min as indicated in manufacturer's instructions. Electrophoresis of the proteins was carried out in a MOPS 4–20% gradient gel (Invitrogen). Proteins were blotted on PVDF membranes and subsequently incubated with the appropriate antibodies. Protein bands were detected using ECL (Amersham). The following antibodies (Abs) were used: anti-E2F3 mouse monoclonal Ab (Upstate), anti-pRB goat polyclonal Ab (Santa Cruz), anti-serpin3 mouse monoclonal Ab (Sigma), and anti-actin and anti-tubulin mouse monoclonal Abs (Sigma) as controls. Appropriate secondary antibodies were also utilized (Amersham Biosciences). # Results ## Microscopical morphology of the tumors The histological patterns of examined tumors were diagnosed as low grade papillary carcinoma (three cases), high grade papillary carcinoma (one case), low grade invasive carcinoma (two cases), high grade invasive carcinoma (two cases), primary carcinoma *in situ* (CIS) (two cases), papillary urothelial neoplasm of low malignant potential (PUNLMP) (one case), and papilloma (one case). ## Proteomic analysis Nano LC-MS/MS analysis and database search of tryptic peptides generated by in- gel digestion of protein bands immunoprecipitated with E2F3 allowed the identification of the proteins reported in. Data were searched on the Mascot search engine ([www.matrixscience.com](http://www.matrixscience.com)) against the MSDB database (updated in April 2008) using the following parameters: MS tolerance 10 ppm; MS/MS tolerance 0.3 Da; fixed modifications carbamidomethyl cysteine; enzyme trypsin; max. missed cleavages 1; taxonomy *other mammalia*. Protein hits based on two successful peptide identifications were considered valid. Protein hits based on a single peptide identification with Mascot score higher than the significance level (\>15) were retained after manual validation. ## Detection of calpain 3 transcript in cDNAs from bovine bladders The presence of calpain 3 transcript was determined in bovine neoplastic urothelium, on bladder cDNAs obtained from different animals. As shown in, all neoplastic bladder samples contained calpain 3 transcripts although at different levels. Theses transcripts were not present in bladders from healthy animals; only in one sample trace amounts of calpain 3 mRNA were detected. We have also observed that calpain 3 transcripts were detectable in apparently unaffected bladder portions, belonging to the same affected animals. Although the amounts of cDNA used for PCR were normalized with respect to GAPDH, quantification of calpain 3 transcripts cannot be precisely determined due to the high number of PCR cycles performed. In fact, the PCR fragment detected following 40 PCR cycles is much less intense in pathological bovine bladders than in human PBMC cDNA, known to express a Capn3-like protease. The data reported in indicate also that pathological bovine bladders express a calpain 3 transcript lacking both IS1 and IS2 inserts typical of the CAPN3 gene. In conclusion, the calpain 3 expressed in pathological bladders shows peculiar properties due to its low level of transcription and to the absence of IS1 and IS2 structures. The absence of IS1 internal structure prevents the auto-inactivation step catalyzed by the protease following its exposure to Ca²⁺ , whereas the lack of IS2 containing a putative nuclear localization signal, may alter the cellular protease trafficking. ## Detection of calpain 3 activity in bovine bladders To determine the presence of active or activable calpain 3 in the pathological bladder we submitted crude extracts of bovine bladders isolated from both normal and pathological animals to ion-exchange chromatography in order to separate calpain 3 from the ubiquitous µ- and m-calpains, and from calpastatin. As shown in, a peak of Ca²⁺-dependent activity eluted from the chromatographic column of the pathological bladder at a ionic strength different from that of µ-, m-calpain and calpastatin. The elution of this calpain 3 protease in this ion-exchange chromatography is slightly different from that previously observed with PBMC-calpain 3, probably because of the difference in the protein structure and in the pH employed in the analysis. In crude extracts from healthy bladders no calpain 3 activity was detectable both in normal and pathological bladders the level of m-calpain and calpastatin (data not shown) were comparable. Thus, the presence of calpain 3 mRNA and protein is a characteristic feature of these urinary bladder tumors. Immunohistochemically, Capn3 was not manifest in normal urothelial cells from healthy cattle. Capn3 was weakly detected in urothelial cells from apparently unaffected mucosa of the neoplastic bladders. In urothelial cancer cells, Capn3 was observed mostly in the cytoplasm. Nuclear positivity was also manifest. Basal cells were the predominant ones showing a strong immunoreactivity for Capn3. ## Virus analysis BPV-2 DNA was amplified in the tumor samples examined. PCR analysis, validated by direct sequencing of the amplified product, demonstrated the presence of true BPV-2 sequences. ## Immunoprecipitation and immunofluorescence for BPV-2 E5 protein Western blot analysis showed a marked presence of E5 protein in tumor tissue cells. The protein was also present, if only less markedly, in apparently unaffected tissue cells from neoplastic bladders. A cytoplasmic immunofluorescence typical of E5 protein expression was clearly shown in tumor urothelial cells. ## Western blot analysis To establish the expression levels of E2F3 protein, we performed a western blot analysis and we observed that E2F3 was markedly overexpressed in tumor samples. Since E2F3 interacts with the retinoblastoma tumor suppressor protein (pRb), we also investigated pRb expression and detected a severe downregulation, to actual inactivation of pRb protein in the same cancer samples. Finally, we performed an investigation to validate the effective presence of SerpinA3-7 protein identified by a single peptide by means of proteomic analysis. We showed that the molecular complex obtained using an anti-E2F3 antibody was also composed of SerpinA3-7. # Discussion Although various calpain substrate proteins are associated with carcinogenesis, the molecular identities of these substrates are largely unknown. It has been suggested that Capn3 and the conventional calpains have common substrate specificities, as several proteins known to be potential targets of Capn3 appear the same substrates of conventional calpains. Recently, calpains have been shown to be involved in proteolysis of NORE1A, a potential Ras effector, and of Ras-association domain family 1 (RASSF1A) proteins. Downregulation of NORE1A and RASSF1A proteins might be involved in some carcinogenesis mechanisms. Calpain 4 (Capn4) has been found to be associated with metastasis and recurrence of hepatocellular carcinoma (HCC). Therefore, it has been proposed that Capn4 might be a target for cancer therapy. In addition, Calpain 6 (Capn6) expression was found increased in some uterine tumors. High levels of Capn3 without any protease activity were detected in human melanoma cell lines. Therefore, it has been suggested that Capn3 may play a pro- apoptotic role in melanoma cells, and it could be a useful diagnostic marker for monitoring melanoma development and progression. In our cases, Capn3 expression was detected in twelve papillomavirus-positive neoplastic urothelial lesions. Capn3 protein was initially identified with nanoscale liquid chromatography coupled with tandem mass spectrometry (nano LC- MS/MS) in a co-immunoprecipitation experiment on E2F3. Capn3 expression was confirmed by reverse transcription polymerase chain reaction (RT-PCR) and its Ca²⁺-dependent proteolytic activity was assayed following ion exchange chromatography. Morphologically, Capn3 expression was documented by immunohistochemical examination. To our knowledge, this is the first study showing Capn3 activation in spontaneous oncogenesis in medical literature. We did not investigate whether any oncogenic proteins of BPV-2 could be involved in Capn3 activation. It has been shown that human papillomavirus (HPV) E7 oncoprotein is able to down-regulate pRb expression, since it binds to µ-calpain and activates its proteolytic activity, resulting in cleavage of pRb. Other mechanisms are likely to exist in animal cancers in which BPV-2 infection may play a central role. We have also shown that papillomavirus-associated urothelial cancers of cattle are characterized by an overexpression of sigma 2 receptors. These receptors are known to play a crucial role in modulating cytoplasmic \[Ca²⁺\] in cancer cells. It is reasonable to suggest that sigma 2 receptors may be involved in Capn3 activation, as it has been shown that even small increases of resting cytoplasmic \[Ca²⁺\] are responsible for Capn3 proteolytic activity. The role of the Capn3 protein in urothelial bladder cancer still remains to be elucidated. We have shown also E2F3 protein overexpression and a dramatic decrease of Rb protein in bladder cancers. Taken together our data allow us to suggest that pRb may be a calpain substrate also in bovine urothelial tumors. E2F3 levels are known to be regulated by pRb and have a crucial role in activating cell proliferation in human bladder cancers,. As Capn 3 was only detected in molecular complexes immunoprecipitated with E2F3, it can be argued that the proteolitically active form of Capn3 protein might be directly involved in molecular pathways leading to the overexpression of E2F3 transcription factors, very likely via Rb protein degradation. Furthermore, Capn3 is known to be a regulator of the conventional calpain system. Therefore, it cannot be excluded that Capn3 may play a role in regulating E2F3 protein expression in an indirect manner, as conventional µ-calpain is known to promote the degradation of Rb protein. It has been suggested that inactivation of Rb pathway and overexpression of E2F3 are obligate events in some human bladder tumors. Our suggestion appears to be strengthened by the proteomic profiles of the other complexes from the same samples, immunoprecipitated with antibodies, vs proteins known to play a role in BPV-2 infection, such as PDGF receptor beta. No Capn3 was detected in them. Indeed, extracellular matrix components such as lumican and decorin were predominantly detected in these complexes (data not shown). Our observations appear to be consistent with the very interesting emerging evidence of a broader role of PDGFRs in tumor stromogenesis, and with the increasing interest in the role of decorin and lumican via the tyrosine kinase receptor family in cancer biology. Further studies are needed to better understand the role of the calpain system in bladder carcinogenesis. If our results are validated by other studies, Capn3 will definitely appear to play an important role in the molecular pathway of bovine urinary bladder tumors. As a consequence, it may prove useful as a diagnostic biomarker for monitoring urothelial tumor development and progression and as a potential target for cancer therapy. Cattle suffering from urothelial tumors, whose incidence may be ∼90% in adult animals, may serve as an animal model useful for gaining insight into new molecular pathways involved in naturally occurring bladder carcinogenesis and for evaluating *in vivo* potential new drugs against specific targets, or for proposing novel therapeutic strategies that are urgently needed nowadays. It is also worth to remember that the establishment of reliable and reproducible animal models for bladder cancer remains an ongoing challenge, since developing therapeutic agents requires *in vivo* models. Furthermore, the UICC Study Group suggested that biological models may still be trail blazing for the natural history of cancer, although molecular models have fostered an impressive progress over the last decades. Finally, it is worthwhile noting that cattle has already been investigated and, it has been found to be a good animal model for several other human diseases,. [^1]: Conceived and designed the experiments: SR FR. Performed the experiments: SR. Analyzed the data: SR RDT CR RS VR MG GB MA OP GC FR. Wrote the paper: SR FR. [^2]: The authors have declared that no competing interests exist.

# 1 Introduction Images are generally contaminated by noise during acquisition, transmission and compression and real-life images are often degraded with mixed noise and it is hard to identify the type and model the noise. Images with high resolutions are desirable in many applications, e.g., object recognition, image classification, and image segmentation in medical and biological science. As an essential low- level image processing procedure, image denoising has been studied extensively and belong to a special type of classical inverse problems. The general observation with additive noise can be modeled as **Y** = **X** + **N**, where **Y** is the noisy observation, and **X** and **N** present the original image and white Gaussian noise, respectively. Though a plethora of noise removal techniques have appeared in recent years, for example, Convolutional Neural Network (CNN) have proved very promising on denoising tasks for which large training sets are available, but when the training data are scarce, their performance suffers from overfitting. Therefore image denoising for real-life noise still remains an important challenge in order to recover the images with high quality. Image denoising problem is in general ill-posed and it requires appropriate regularization. Over the past few decades, numerous image denoising methods have been developed. This is usually achieved by minimizing a suitable energy functional that characterizes a trade-off between data-fidelity and regularity. Frobenius norm is often employed to measure the data fitting loss for additive Gaussian noise. Sparse signal representation describes a signal that can be approximated as a linear combination of as few as possible atoms from a given dictionary. Recently, Elad showed that sparse overcomplete representation approach is quite effective in denoising images, supported by recent study that better denoising performance can be achieved by using a variant of sparse coding methods. In order to promote sparsity more extensively than convex regularization, it is also standard practice to employ non-convex optimization. In image denoising, following, each noise patch **y**_*i* is extracted from the noisy image **Y**. In order to better exploit group sparsity, we group a set of similar patches $\mathbf{Y} = \left\lbrack \mathbf{y}_{1},\mathbf{y}_{2},\ldots,\mathbf{y}_{n} \right\rbrack \in \mathbb{R}^{m \times n}$. Thus, denoising problem becomes the recovery problem of **x**_*i* from **y**_*i*. Now let us consider the group sparsity defined by a group norm \|\|**A**\|\|_*p*,2: $$\begin{array}{r} {\left( \mathbf{D},\mathbf{A} \right) = \arg\min\limits_{\mathbf{D},\mathbf{A}}\frac{1}{2}\left| \middle| \mathbf{Y} \right. - {\textbf{DA}{||}}_{F}^{2} + \left. \lambda \middle| \middle| \mathbf{A} \middle| \right|_{p,2}^{p},\mspace{720mu} 0 < p \leq 1,} \\ \end{array}$$ where $\mathbf{A} = \left\lbrack \alpha^{1},\alpha^{2},\ldots,\alpha^{n} \right\rbrack \in \mathbb{R}^{m \times n}$ is related to image patches by **X** = **DA**. We note that the group norm (quadratic symmetric gauge function, see 2.4.2 of)\|\|⋅\|\|_*p*,2 is defined by $$\begin{array}{r} {\left| \middle| \mathbf{A} \middle| \right|_{p,2} \triangleq {\parallel (||}\alpha^{1}{||}_{2}\left.,\ldots, \middle| \right|\alpha^{n}{||}_{2}{) \parallel}_{p},} \\ \end{array}$$ where *α*^*i* = \[*α*_*i*,1, …, *α*_*i*,*m*\]^*T* denotes the *i*^*th* column of matrix **A** in $\mathbb{R}^{m \times n}$. In recent years, many research is devoted to address the group sparse optimization problem, aiming at the improvement of efficiency and accuracy (e.g., see survey paper and references therein). Once all group sparse codes **A** are achieved, the latent clean image **X** can be reconstructed as **X** = **DA** by standard approach(see Theorem 1 and 2 in). The main contributions of this paper are illustrated as follows: 1. We unify the group-based sparse coding in and the Schatten-*p* norm minimization problem in by proving their mathematical equivalence. 2. A fixed-point iteration scheme is developed for sparse optimization in *ℓ*_*p* space with *p* ∈ (0, 1\] by using proximal operator and we a new solution to Schatten-*p* norm minimization problem is obtained, which appears to be more accurate and rigorous than. 3. Regarding to image denoising, we find that the optimal value of *p* is inversely proportional to the noise level except for high level noise, where the best values of *p* are 1 and 0.95, when the noise levels are 75 and 100, respectively. 4. The proposed Spatially Adaptive Fixed Point Iteration (SAFPI) algorithm attains the best denoising performance on the value of PSNR and SSIM, being able to retain the image structure information, which outperforms many state-of-the-art denoising methods such as BM3D, WNNM and WSNM. The rest of the paper is organized as follows. In Section 2.1, we prove the equivalence of group-based sparse coding and the Schatten-*p* norm minimization problem and propose a new solution to Schatten-*p* norm minimization problem. A fixed point iteration for solving sparse optimization in *ℓ*_*p* space with *p* ∈ (0, 1\] is formulated and discussed. In Section 2.2, we establish an image denoising scheme using nonlocal self-similarity and Schatten-*p* norm minimization. In Section 3, based on the new developed Spatially Adaptive Fixed Point Iteration (SAFPI) algorithm, we present the experimental results using a set of standard benchmark images. And the comparison with several existing methods are also provided to demonstrate our improvement. Finally, the paper ends with concluding remarks. # 2 Materials and methods ## 2.1 Proximal operator for Schatten-p norm minimization ### 2.1.1 Background Consider a matrix $\mathbf{Y} \in \mathbb{R}^{m \times n}$, then **Y**^*T* **Y** is a positive semidefinite matrix. The eigenvalues of **Y**^*T* **Y** are called the singular values of **Y**, denoted by *σ*₁(**Y**), …, *σ*_min{*m*,*n*}(**Y**) in decreasing order (see page 246 of). Let *r* = **rank**(**Y**), it is clear that $$\begin{array}{r} {\sigma_{r + 1}\left( \mathbf{Y} \right) = 0,\ldots,\sigma_{\min\{ m,n\}}\left( \mathbf{Y} \right) = 0.} \\ \end{array}$$ The matrix **Y** also has the following Singular Value Decomposition (SVD) **Y** = **U**Σ**V**^*T*, where $\mathbf{U} \in \mathbb{O}(m),\mathbf{V} \in \mathbb{O}(n)$ ($\mathbb{O}$ is the set of orthogonal matrices) and *σ*_*n* is an *m*×*n* diagonal matrix with diagonal entries *σ*₁(**Y**), …, *σ*_min{*m*,*n*}(**Y**). We introduce the Schatten-*p* norm (0 \< *p* \< ∞) of **Y**, which is defined as $$\begin{matrix} {\left| \left| \mathbf{Y} \right| \right|_{S_{p}} = \left( {\sum\limits_{i = 1}^{\text{min}{\{{m,n}\}}}\sigma_{i}^{p}\left( \mathbf{Y} \right)} \right)^{1/p}.} \\ \end{matrix}$$ Special cases of the Schatten-*p* norm include the nuclear norm (*p* = 1) and the Frobenius norm (*p* = 2). Next we analyze the relationship between group-based sparse coding and the Schatten-*p* norm minimization problem, which improves Theorem 2 in. But our approach is based on the “symmetry” technique (similar to for other purpose), which is essentially different from. **Theorem 1** *The group-based sparse coding problem* *is equivalent to a Schatten-p norm minimization problem*. Eqs, and imply that any operation designated for sparse coefficient vector *α*’s can be conveniently implemented with singular values of **X** (only differs by a constant scalar). The Schatten-*p* norm (0 \< *p* ≤ 1) has been widely used to replace the nuclear norm for better approximating the rank function. There are extensive study for the Schatten-*p* norm optimization problem in literature. Note that the main difference between group sparse coding and the Schatten-*p* norm minimization problem is that group sparse coding has a dictionary learning operator while the Schatten-*p* norm minimization problem does not involve such operation. ### 2.1.2 Computation of proximal mapping using fixed point iterative method Now let us recall the definition of proximal mapping. **Definition 2** *The proximal mapping of a mapping* $\left. \Theta:\mathbb{R}\mapsto\mathbb{R} \right.$ is $$\begin{array}{r} {\textbf{Prox}_{\Theta}(x) = \arg\min\limits_{w}\left\{ (w - x)^{2} + 2\lambda\Theta(w) \right\}.} \\ \end{array}$$ *The proximal mapping of* $\left| \middle| \mathbf{Y} \middle| \right|_{S_{p}}^{p}$ *is defined as*: $$\begin{array}{r} {\textbf{Prox}_{\lambda} \parallel \cdot \parallel_{p}\left( \mathbf{Y} \right) = \arg\min\limits_{X}\frac{1}{2}\left| \middle| \mathbf{Y} \right. - \left. \mathbf{X} \middle| \right|_{F}^{2} + \left. \lambda \middle| \middle| \mathbf{X} \middle| \right|_{S_{p}}^{p}.} \\ \end{array}$$ And we have the following celebrated theorem: **Theorem 3** *\[Theorem 1 of**\] If matrix* $\mathbf{Y} \in \mathbb{R}^{m \times n}$ *has the following Singular Value Decomposition (SVD)* **Y** = **U**Σ**V**^*T*, *where* $\mathbf{U} \in \mathbb{O}(m),\mathbf{V} \in \mathbb{O}(n)$ *and σ*_*n* *is an m × n diagonal matrix with diagonal entries σ*₁(**Y**), …, *σ*_min{*m*,*n*}(**Y**). *Then we have in* $$\begin{array}{r} {\hat{\mathbf{X}} = \textbf{Prox}_{\lambda} \parallel \cdot \parallel_{p}\left( \mathbf{Y} \right) = \mathbf{U}\textbf{diag}\left( \sigma_{i}\left( \hat{\mathbf{X}} \right) \right)\mathbf{V}^{T},} \\ \end{array}$$ *where* $\sigma_{i}\left( \hat{\mathbf{X}} \right)$ *is defined as the scalar proximal mapping in* : $$\begin{array}{r} {\sigma_{i}\left( \hat{\mathbf{X}} \right) = \textbf{Prox}_{\lambda} \parallel \cdot \parallel_{p}\left( \sigma_{i}\left( \mathbf{Y} \right) \right) = \arg\min\limits_{\sigma_{X} \geq 0}\frac{\left( \sigma_{X} - \sigma_{i}\left( \mathbf{Y} \right) \right)^{2}}{2} + \lambda\sigma_{X}^{p}.} \\ \end{array}$$ In order to be transparent for our proposed approach to solve, we recall two important concepts in convex optimization next. **Definition 4** *(see Chapter 2 p 82 of*) *Let* $\mathbb{R}^{n}$ *be paired by a bilinear functional (inner product)* 〈,〉 *and let* $\left. f:\mathbb{R}^{n}\mapsto\mathbb{R} \right.$ *be any extended real-valued function on* $\mathbb{R}^{n}$. *Then the function f*\* *on* $\mathbb{R}^{n}$ *defined by* $$\begin{array}{r} {f^{*}(y) = \min\limits_{x}f(x) - \left\langle x,y \right\rangle,\mspace{720mu} y \in \mathbb{R}^{n}} \\ \end{array}$$ *is called the Fenchel conjugate of f (with respect to the given pairing). Note that f*\* *is always a closed convex function, regardless of the structure of f*. **Definition 5** *Given the proper convex function* $\left. f:\mathbb{R}^{n}\mapsto( - \infty, + \infty\rbrack \right.$, *the subdifferential of such a function is the (generally multivalued) mapping* $\left. \partial f:\mathbb{R}^{n}\mapsto\mathbb{R}^{n*} \right.$ *defined by* $$\begin{array}{r} {\partial f(x) = \left\{ x^{*} \in \mathbb{R}^{n*} \mid f(z) \geq f(x) + \left\langle x^{*},z - x \right\rangle,\mspace{720mu} z \in \mathbb{R}^{n} \right\}.} \\ \end{array}$$ *The elements x*\* ∈ ∂*f*(*x*) *are called subgradients of f at x. Actually, same definition works for nonconvex f (however, subgradient need not exist)*. A point $x \in \mathbb{R}^{n}$ is a minimizer of a function *f* (not necessarily convex) over $\mathbb{R}^{n}$ if and only if *f* is subdifferentiable at *x* and 0 ∈ ∂*f*(*x*). **Lemma 6** $$\begin{array}{r} {\textbf{Prox}_{\lambda} \parallel \cdot \parallel_{p}(x) = \arg\min\limits_{w}\left\{ (w - x) \right.^{2} + \left. 2\lambda \middle| w \right|^{p}\left.,0 < p \leq 1 \right\}} \\ \end{array}$$ *is closed and convex, but it has no close form solution for general p*. If *p* = 1, it is well-known that the function *ϕ*(*w*) = \|*w*\| is not differentiable but still convex, and can be described by a subgradient (see Section 2.3 of) as ∂*ϕ*(*w*) = *sign*(*w*) and from Lemma 6, we have $$\begin{array}{r} {\textbf{Prox}_{\lambda} \parallel \cdot \parallel_{1}(x) = \left( \mathbf{I} + \lambda\partial\phi \right)^{- 1}(x) = \max\left\{ \middle| x \middle| - \lambda,0 \right\}.} \\ \end{array}$$ Furthermore, we can obtain the following theorem, which improves Theorem 1 in and Theorem 1 in using fixed- point iteration (see Chapter 1 of for details). **Definition 7** *Given a function* $\left. g:\lbrack a,b\rbrack\mapsto\mathbb{R} \right.$, *find ξ* ∈ \[*a*, *b*\] *such that ξ* = *g*(*ξ*). *If such ξ exists, it will be called a fixed point of g and it could be computed by the following algorithm: ξ*^(*n*) = *g*(*ξ*^(*n*−1)), *n* ≥ 1. *And g is said to be a contraction on* \[*a*, *b*\] *if there exists a constant L such that* 0 \< *L* \< 1 *and* \|*g*(*x*) − *g*(*y*)\| \< *L*\|*x* − *y*\| *for any x, y* ∈ \[*a*, *b*\]. **Theorem 8** *Let us denote* $$\begin{array}{r} {\aleph_{\lambda,p} = \max\left\{ \frac{2 - p}{2(1 - p)}\left( 2\lambda(1 - p) \right)^{\frac{1}{2 - p}},\frac{1}{2}\left( \lambda p(1 - p) \right)^{\frac{1}{2 - p}} + \lambda\left( \lambda p(1 - p) \right)^{\frac{p - 1}{2 - p}} \right\},} \\ \end{array}$$ *then for* 0 \< *p* \< 1, *we have* $$\begin{array}{ll} {\mathbf{Pro}\mathbf{x}_{\lambda} \parallel \cdot \parallel_{p}(x)} & {= \text{arg}\mspace{360mu}\underset{w \geq 0}{\text{min}}\left\{ \left( {w - x} \right)^{2} + 2\lambda|w|^{p} \right\}} \\ & {= \begin{cases} 0 & {|x| \leq \aleph_{\lambda,p}} \\ \left. fixed - point\mspace{360mu} of\mspace{360mu} the\mspace{360mu} contraction\mspace{360mu} w\mapsto|x| - \lambda p\frac{w}{|w|^{2 - p}} \right. & {|x| \geq \aleph_{\lambda,p}.} \\ \end{cases}} \\ \end{array}$$ ## 2.2 Spatially Adaptive Fixed Point Iteration (SAFPI) denoising algorithm In, the authors proposed a group sparse representation framework and a Schatten-*p* norm minimization framework for image denoising. In Theorem 1, we have shown these two approaches are equivallent. From combining Theorem 3 and Theorem 8, we obtained a fixed point iteration solution of in Theorem 1, which is more rigorous than. After grouping a set of similar patches $\mathbf{Y} = \left\lbrack \mathbf{y}_{1},\mathbf{y}_{2},\ldots,\mathbf{y}_{n} \right\rbrack \in \mathbb{R}^{m \times n}$, the denoising problem becomes the recovery problem of **x**_*i* from **y**_*i*. And as was shown in Theorem 1, the Schatten-*p* norm minimization problem converts the denoising problem to recover the low-rank matrix **X** from the non-low-rank matrix **Y**, and thus filtering out the noise of the structure set. And the second identity in can be solved using Theorem 3 and Theorem 8. Wavelet-based image denoising assumes that the wavelet coefficients obey the Laplace distribution, and the threshold method is used to filter the noise in the image. The prior distribution of the block matrix singular values can also approximate the Laplace distribution in space. The parameter λ for each group that balances the fidelity term and the regularization term should be adaptively determined for better denoising performance. Using the Spatial Adaptive Laplacian Transcendental as appeared in, the threshold parameter can be set to $\lambda_{i} = \frac{2\sqrt{2}\sigma_{w}^{2}}{\sigma_{i}},$ where *σ*_*i* denotes the locally estimated variance at the position *i*. Now the second identity in becomes $$\begin{array}{r} {\textbf{Prox}_{\lambda_{i}} \parallel \cdot \parallel_{p}\left( \mathbf{Y} \right) = \arg\min\limits_{X}\frac{1}{2}\left| \middle| \mathbf{Y} \right. - \left. \mathbf{X} \middle| \right|_{F}^{2} + \frac{2\sqrt{2}\sigma_{w}^{2}}{\sigma_{i}}\left| \middle| \mathbf{X} \middle| \right|_{S_{p}}^{p}.} \\ \end{array}$$ From, if **Y** has singular value decomposition **Y** = **U**Σ**V**^*T*, we have $\hat{\mathbf{X}} = \mathbf{U}\hat{\Sigma}\mathbf{V}^{T}$, where $\hat{\Sigma} = \textbf{diag}\left\{ {\hat{\varepsilon}}_{1},\ldots,{\hat{\varepsilon}}_{\min\{ m,n\}} \right\}$. And ${\hat{\varepsilon}}_{i}$ can be computed using Theorem 3 and Theorem 8. Recently some developed iterative regularization techniques in offers an alternative approach toward spatial adaptation. The basic idea of iterative regularization is to add filtered noise back to the denoised image i.e., $$\begin{array}{r} {\mathbf{y}^{(k + 1)} = {\hat{\mathbf{x}}}^{(k)} + \delta\left( \mathbf{y} - {\hat{\mathbf{x}}}^{(k)} \right)} \\ \end{array}$$ where *k* denotes the iteration number and *δ* is a relaxation parameter. Besides, we can execute the above denoising procedures for better results after several iterations. In the *k* + 1-th iteration, the iterative regularization strategy in is used to update the estimation of noise variance. Then the standard deviation of noise in *k* + 1-th iteration is adjusted as $$\begin{array}{r} {{\hat{\sigma}}_{\omega}^{(k + 1)} = \gamma\sqrt{\sigma_{\omega}^{2} - \left| \middle| \mathbf{y} \right. - \mathbf{y}^{(k)}{||}_{2}^{2}}} \\ \end{array}$$ where *γ* is a scaling factor controlling the re-estimation of noise variance and the local estimated variance at the *i*-th position is $$\begin{matrix} {{\hat{\sigma}}_{i}^{({k + 1})} = \sqrt{\text{max}\left( {\left( \chi_{i}^{(k)} \right)^{2}/n - \left( {\hat{\sigma}}_{\omega}^{(k)} \right)^{2},0} \right)}.} \\ \end{matrix}$$ where *χ*_*i* is the *i*-th singular value of image **y**. The higher the structural similarity of the blocks in the structure group is, the more correlative the column vectors in the block matrix will be, which means that it has a low rank property corresponding to the noise-free matrix. The information is mainly concentrated in those largest singular values. During the proximal operation, selecting the appropriate threshold parameter for those with larger singular value makes the processed singular value closer to the noise- free singular value, which can well preserve the useful information in the image while filtering out the noises. Therefore, choosing blocks with more similar structure will help to improve the image denoising effect. There are many commonly used similarity measures such as Euclidean distance, cosine angle, and correlation coefficient. The traditional block similarity measure function have some shortcomings in measuring the similarity between blocks. Euclidean distance simply calculate the difference between the pixel gray value of the blocks, and then add up as a standard measure of the degree of similarity. Although this method is simple and easy to implement, it only treats the blocks as isolated pixels and neglects the statistical relevance between local pixels, which leads to the inaccuracy of similarity measure. This is because the blocks are not in an Euclidean space. There is a very strong correlation between the pixels in the block. The local pixel correlation carries important structural information of the blocks. In order to solve this problem, Structural SIMilarity (SSIM) index is often used to evaluate the image quality. SSIM is defined as $\left( 2\mu_{X}\mu_{\hat{X}} + c_{1} \right)\left( \sigma_{X,\hat{X}} + c_{2} \right)/\left\lbrack \left( \mu_{X}^{2}\mu_{\hat{X}}^{2} + c_{1} \right)\left( \sigma_{X}^{2}\sigma_{\hat{X}}^{2} + c_{2} \right) \right\rbrack$, where $\mu_{X},\mu_{\hat{X}}$, $\sigma_{X}^{2},\sigma_{\hat{X}}^{2}$, and $\sigma_{X,\hat{X}}$ denote the average of *X*, the average of $\hat{X}$, the variance of *X* and the variance of $\hat{X}$, respectively. *c*₁ and *c*₂ are two variables to stabilize the division with weak denominator. A detailed step-by-step description of Spatially Adaptive Fixed Point Iteration (SAFPI) denoising algorithm is given by Algorithm 1 **Algorithm 1** Image Denoising via SAFPI Algorithm **Require:** Initialization: $\hat{\mathbf{x}} = \mathbf{y}$;  Iterate on *i* = 1,2, …, *iter* 1. Iterative regularization: $\mathbf{y}^{(k + 1)} = {\hat{\mathbf{x}}}^{(k)} + \delta\left( \mathbf{y} - {\hat{\mathbf{x}}}^{(k)} \right)$ and compute its variance $\sigma_{\omega}^{(k + 1)}$; 2. Divide **y**^(*k*+1) into several blocks, the SSIM is used to classify the blocks with structural similarities into one structural group **Y**_*i*; 3. Noise variance update: re-estimate *σ*_*ω* from **y**^(*k*+1) via ${\hat{\sigma}}_{\omega}^{(k + 1)} = \gamma\sqrt{\sigma_{\omega}^{2} - \left| \middle| \mathbf{y} \right. - \mathbf{y}^{(k + 1)}{||}_{2}^{2}}$; 4. SVD for each noisy data matrix **Y**_*i*: (**U**_*i*, Σ_*i*, **V**_*i*) = *SVD*(**Y**_*i*), where Σ_*i* = **diag**{*ε*₁, …, *ε*_min{*m*,*n*}}; 5. Thresholds update: compute λ_*i* using $\lambda_{i} = \frac{2\sqrt{2}\sigma_{\omega}^{2}}{\sigma_{i}^{(1/p)}}$ and ${\hat{\sigma}}_{i} = \sqrt{\max\left( \Sigma_{i}^{2}/n - \sigma_{\omega}^{2},0 \right)}$; 6. Compute $$\begin{array}{r} {\aleph_{\lambda_{i},p} = \max\left\{ \frac{2 - p}{2(1 - p)}\left( 2\lambda_{i}(1 - p) \right)^{\frac{1}{2 - p}},\frac{1}{2}\left( \lambda_{i}p(1 - p) \right)^{\frac{1}{2 - p}} + \lambda_{i}\left( \lambda_{i}p(1 - p) \right)^{\frac{p - 1}{2 - p}} \right\},} \\ \end{array}$$ 7. Application of proximal operator: Updating the *ε*_*i* value by using the follow formula $$\begin{array}{r} {\varepsilon_{i} = \left\{ \begin{array}{ll} 0 & {|\varepsilon_{i}\left| \leq \right.\aleph_{\lambda_{i},p}} \\ {|\varepsilon_{i}\left| - \right.\lambda_{i}{p{(|}}\varepsilon_{i}\left| - \right.\lambda_{i}\left. p \right|\varepsilon_{i}\left| {}^{p - 1} \right)^{p - 1}} & {|\varepsilon_{i}\left| \geq \right.\aleph_{\lambda_{i},p},} \\ \end{array}\operatorname{} \right.} \\ \end{array}$$ with computed λ_*i* and *ε*_*i* from step 5 and 4; 8. Image update: obtain an improved denoised image ${\hat{\mathbf{x}}}^{(k)}$ by weighted averaging all denoised patches ${\hat{\mathbf{X}}}_{i} = \mathbf{U}_{i}{\hat{\Sigma}}_{i}\mathbf{V}_{i}^{T}$, where ${\hat{\Sigma}}_{i} = \textbf{diag}\left\{ \varepsilon_{1},\ldots,\varepsilon_{\min\{ m,n\}} \right\}$;  Output: ${\hat{\mathbf{x}}}^{(k)}$. # 3 Results and discussion In recent years, many denoising algorithms have been developed and the adaptive image removal algorithms is a hot trend in signal and image denoising. To demonstrate the effectiveness of the proposed denoising algorithm, in this section, we compared the denoising performance with recently proposed state-of- the-art denoising methods, such as BM3D, WNNM, WSNM, Expected Patch Log- likelihood (EPLL), Spatially Adaptive Iterative Singular-value Thresholding (SAIST), Patch-Based Near-Optimal image denoising (PBNO), Global Image Denoising (GID), iterative denoising system based on Wiener filtering (WIENER), and Linear Complex Diffusion Process (LCDP). We have used some well known images that are commonly used in the literature such as. We added noise to them, and test the proposed denoising algorithm with different power *p* under different noise levels. The experimental images are shown in. There are several image quality evaluation indicators measuring success of denoising such as kurtosis, low signal-to-noise-ratio(SNR). Low kurtosis indicate superior performance and it is defined as $k(X) = C_{4}(X)/C_{2}^{2}$, where *C*_*k*(.) is the *k*-th cumulant function. In our work, we evaluated the performance with three criterion: Structure Similarity Index (SSIM), kurtosis and Peak Signal-to-Noise Ratio (PSNR) which defined as $10\log_{10}\frac{M^{2}}{MSE}$, where *M* denotes the maximum intensity of the underlying image and $MSE = \frac{1}{n_{1} \times n_{2}}\sum_{i = 1}^{n_{1}}\sum_{j = 1}^{n_{2}}\left( X_{i,j} - {\hat{X}}_{i,j} \right)^{2}$ is the mean squared error between the denoised image $\hat{X}$ and the noiseless image *X*. All the experiments were carried out on Matlab (R2016a) of a PC with Intel(R) Xeon(R) CPU *E*5 − 1630 *V*4@3.7*GHz* and 32GB RAM. ## 3.1 Analysis of over-shrinkage problem and optimal power *p* Firstly, we noticed that not all values of power *p* applied well to the proposed Spatially Adaptive Fixed Point Iteration (SAFPI) algorithm. It would conduct an approximation deviation with the solved singular values and produce excessive contraction, when the value of *p* is not suitable. As shown in, we tested SAFPI to process low rank approximation on the red patch in with the noise level be 50, which is randomly marked from “Monarch”. In, $\left\{ \sigma_{i}^{(p)} \right\}$ represents the singular values of the denoised similar patches with different power *p*. The ground-truth line (denoted by blue line) is the singular value connection line for the similar blocks of the noiseless red patch in. Now we can see that $\left\{ \sigma_{i}^{(0.8)} \right\}$ (shown on green line) is more close to the ground-truth line. This means that the other $\left\{ \sigma_{i}^{(p)} \right\}$’s (denoted by black, red, blue lines) conducted a serious over-shrinkage problem. In this case, setting *p* = 1 as in WNNM in denoising will lead to bad processing results. So the advantage of SAFPI algorithm is to overcome the over-shrinkage problem, in case we can find the optimal value of power *p*. Secondly, in order to find the optimal values of *p* under different noise levels for SAFPI algorithm, we randomly chose 10 test images in for our experiments and set the values of power *p* to be from 0.05 to 1 with an interval of 0.05. The zero mean additive white Gaussian noise levels were set to be *σ*_*n* = {20, 30, 50, 60, 75, 100}, and the other parameters were the same as WSNM. The results are shown in, the horizontal coordinate denotes the different values of *p* and the vertical coordinate represents the average value of PSNR under given noise level. And the red dots are the optimal points for each given noise level. We can see that the best values of power *p* are 1.0, 0.90, 0.85 and 0.6, when the noise levels are low or medium 20, 30, 50 and 60, respectively. While handling very high noise levels 75, 100, the average PSNR values decrease firstly and then increase, the best values of *p* are 0.95 and 0.9 respectively. To sum up, we find that the optimal value of *p* is inversely proportional to the noise level except for high level of noise, where the best values of *p* are 1 and 0.95. And then we applied the best empirical values for the next experiments. ## 3.2 Performance comparison with different methods We set *p* = {1.0, 0.9, 0.85, 0.6, 0.95, 0.9} for *σ* = {20, 30, 50, 60, 75, 100} in our proposed SAFPI algorithm. And then we compared the performance with seven standard algorithms (BM3D, WNNM, WSNM, EPLL, SAIST, PBNO, GID, WIENER, LCDP) from 13 widely used images from. The results (thanks to the source codes provided by the authors) are in Tables, and. It can be seen from that our algorithm always obtains the best average values of PSNR under different noise levels. The proposed approach achieves 0.3dB to 0.51dB improvement on average over the BM3D, when the noise levels are between 20 and 100. It also achieves 0.02dB, 0.06dB and 0.14dB improvement on average over the WSNM, when the noise levels are 30, 50 and 100, respectively. And our average values of SSIM are the best when the noise levels are 20, 30, 50 and 60. To sum up, for every given low and medium noise level, our algorithm attains the best denoising performance on the values of SSIM and PSNR for all noise levels. This leads to a better image denoising performance and high robustness to noise strength in comparison to several existing denosing algorithms. For visual quality, some comparative images are shown in Figs, and. As shown in, our algorithm resumed the structure of the ear (which is magnified in the highlighted red window) better than other algorithms. When the noise level is very high, as shown in the zoom-in window in, our algorithm could reconstruct clear texture structures, while the competing methods get more blurred textures. Other visual improvements can be seen in Figs and. Sometimes the variation of noise is too big and too small in the same image (in different parts of the image). To demonstrate our method, we randomly selected two small pieces from the given image. Although their local noise level would be different, our algorithm always gets the best visual texture. Now we could conclude that the proposed SAFPI algorithm can display excellent denoising performance, producing good visual effect and rebuilding better textures. If noise is non-Gaussian, one popular method is to transform the non-Gaussian noise into a more tractable Gaussian model such as the generalized Anscombe transformation (GAT). In this paper, we deal with non-Gaussian noise using the proposed algorithms. In the first experiment, we assumed the noise was a mix of Gaussian noise (*σ*_*n* = 20, 50, 100) and speckle noise (the density is *d* = 1\*10⁻³). Then we used the SAFPI algorithm directly to remove the noise. We randomly selecteded six images (Lena, Monarch, Barbara, Cameraman, House, Peppers) on for experimental verification and compared with some excellent denoising algorithms which has been mentioned in the previous experiments. The results are shown on. In the second experiment, we assumed the noise was mixed Poisson-Gaussian noise. Then we transformed the Poisson-Gaussian hybrid noise into an approximate Gaussian noise using the GAT algorithm and obtained the repaired images by using the proposed denoising algorithm and the exact unbiased inverse GAT. We used Lena and C.man as the test images and set eight different peak values to be (1, 2, 5, 10, 20, 30, 60, 120). The Poisson- Gaussian noise were set to be *σ* = peak/10. We compared with BM3D, SAIST, WNNM and recorded the average PSNR and kurtosis parameters of these two images. The results are shown on. All bold numbers represent the best evaluation index values. From, we can see when the standard deviation is not big (*σ*_*n* = 20, 50), our proposed algorithm almost achieved the best values of all three quality evaluation indicators, and obtained the best PSNR values on all of hybrid noises experiments. From, it can be seen that the SAFPI algorithm almost get the highest averaged PSNR value under all peaks experiments. In most cases, our proposed algorithm obtained relatively good kurtosis metrics and optimal PSNR value, which is 0.2 to 0.3dB higher than the BM3D algorithm and about 0.1dB to 1.11dB higher than WNNM. Finally, we bravely attempted to discuss the complexity of SAFPI algorithm. We assume each patch size is *A* \* *A*, where *A* represents the length or width of each block, and *k* is the number of similar patches in each structural group **y**_*i*. Now calculating SVD (step 4 in Algorithm 1) needs $\mathcal{O}\left( \min\left( A^{2}*K^{2},\mspace{720mu} A^{4}*K \right) \right)$ flops in each iteration. And it also costs $\mathcal{O}(K)$ to compute the singular values in step 6. Next since the image **y**^(*k*+1) can be divided into *N* blocks in step 2, then it needs $i*N*\mathcal{O}\left( \min\left( A^{2}*K^{2},\mspace{720mu} A^{4}*K \right) + K \right)$ flops, where *i* is the number of iterations in Algorithm 1. Then we recorded the execution times of several excellent denoising algorithms spent on the above experiments with the standard deviation *σ*_*n* of the white Gaussian noise to be 20: SAFPI 4843.339s, WSNM 5453.311s, WNNM 4410.991s, SAIST 923.9837s, BM3D 17.6242s and EPLL 1550.607s. Our algorithm did not take much longer time while maintaining the best denoising results. # 4 Conclusions In this paper, a fixed-point iteration scheme was developed for sparse optimization in *ℓ*_*p* space with *p* ∈ (0, 1\] by using proximal operator. We showed that group sparse coding was equivalent to Schatten-*p* norm minimization problem, and thus the sparse coefficient of each group were measured by estimating the singular values of each group. When analyzing the optimal value of power *p*, we can find that the optimal value of Schatten *p*-norm is related to the noise level. As the noise level increases, the optimal value of *p* decreases gradually. And if the noise reaches a high level, the optimal value of *p* will be close to 1. The developed SAFPI algorithm can obtain higher PSNR indices and is able to retain promising texture structure information and visual quality. The methods developed in this paper leads to a better image denoising compared to other competing denoising algorithms. There are several future research directions. We are further exploring other non- convex optimization strategies for more effective convergence and further improvement. The convolutional neural networks(CNN) based denoising methods become more and more popular now and we will investigate CNN architectures for the denoising of images in the future. # 5 Appendix **Proof 9 (Proof of Theorem 1)** *Let* **D** = **U** *and* **A** = Σ**V**^*T* *in*, *where* $\Sigma = \textbf{diag}\left\{ \varepsilon_{1},\ldots,\varepsilon_{K} \right\}\left( K = \min\left\{ m,n \right\} \right) \in \mathbb{R}^{K \times K}$ *is a diagonal matrix and each column of* **V** *in* $\mathbb{R}^{m \times K}$ *is decomposed of* **v**_*i* = (*α*^*i*)^*T*/*ε*_*i*. *Then we have* $$\begin{array}{r} {\left( \mathbf{U},\Sigma,\mathbf{V} \right) = \arg\min\limits_{\mathbf{U},\Sigma,\mathbf{V}}\frac{1}{2}\left| \middle| \mathbf{Y} \right. - \mathbf{U}\Sigma\mathbf{V}^{T}{||}_{F}^{2} + \left. \lambda \middle| \middle| \mathbf{A} \middle| \right|_{p,2}^{p}.} \\ \end{array}$$ *Let σ*_*i* *denotes the standard deviation of the sparse coefficients α*^*i* *in the i-th column, then the sum of standard deviations associated with sparse coefficient vector in each column is* $$\begin{array}{r} {\alpha_{i,1}^{2} + \cdots + \alpha_{i,m}^{2} = m\sigma_{i}^{2}.} \\ \end{array}$$ *And then it is not hard to see* $$\begin{array}{r} {\left| \middle| \mathbf{A} \middle| \right|_{p,2} = \sqrt[p]{\sum\limits_{i = 1}^{K}\left( \alpha_{i,1}^{2} + \cdots + \alpha_{i,m}^{2} \right)^{p/2}} = \sqrt{m}\sqrt[p]{\sigma_{1}^{p} + \cdots + \sigma_{K}^{p}}.} \\ \end{array}$$ *Using* *and the unitary property of* **V**, *we have* $$\begin{array}{r} {\sigma_{i}^{2} = \frac{1}{m}{||}\alpha^{i}{||}_{2}^{2} = \frac{1}{m}{||}\varepsilon_{i}\mathbf{v}_{i}^{T}{||}_{2}^{2} = \frac{\varepsilon_{i}^{2}}{m}.} \\ \end{array}$$ *Then it is ready to see* $$\begin{array}{r} {\left| \middle| \mathbf{A} \middle| \right|_{p,2} = \sqrt[p]{\varepsilon_{1}^{p} + \cdots + \varepsilon_{K}^{p}}.} \\ \end{array}$$ *By substituting* *into* *we could obtain* $$\begin{matrix} \left( \mathbf{U},\Sigma,\mathbf{V} \right) & {= \arg\min\limits_{\mathbf{U},\Sigma,\mathbf{V}}\frac{1}{2}\left| \middle| \mathbf{Y} \right. - \mathbf{U}\Sigma\mathbf{V}^{T}{||}_{F}^{2} + \lambda\sum\limits_{i = 1}^{K}\varepsilon_{i}^{p}} \\ & {= \arg\min\limits_{\mathbf{X} = \mathbf{U}\Sigma\mathbf{V}^{T}}\frac{1}{2}\left| \middle| \mathbf{Y} \right. - \left. \mathbf{X} \middle| \right|_{F}^{2} + \left. \lambda \middle| \middle| \mathbf{X} \middle| \right|_{S_{p}}^{p},} \\ \end{matrix}$$ *which appears to be better approximation to the rank function by using the Schatten-p quasi-norm*. **Proof 10 (Proof of Lemma 6)** *Let ϕ* = \|*w*\|^*p*, *by using subdifferential and Corollary 2.59 of*, *one has* $$\begin{array}{r} {\textbf{Prox}_{\lambda} \parallel \cdot \parallel_{p}(x) \subseteq \left( \mathbf{I} + \lambda\partial\phi \right)^{- 1}(x).} \\ \end{array}$$ *If ϕ is convex, this is an equality*. *If we further define* $\Pi(w) = \left. \lambda \middle| w \right|^{p} + \frac{w^{2}}{2}$, *since* $$\begin{array}{r} {\arg\min\limits_{w}\left\{ (w - x) \right.^{2} + \left. 2\lambda \middle| w \right|^{p}{\} = \arg}\min\limits_{w}\left\{ \Pi(w) - wx \right\},} \\ \end{array}$$ *it is ready to see* $$\begin{array}{r} {\textbf{Prox}_{\lambda} \parallel \cdot \parallel_{p}(x) = \Pi^{*}(x).} \\ \end{array}$$ *For* Π\*(*x*) *is the pointwise minimum of the collection of affine functions, we know it is closed and convex. Thus* **Prox**_λ ∥ ⋅∥_*p*(*x*) *is also closed and convex*. *But it is easy to see* Π\*(*x*) *is a discontinues mapping, so generally there is no closed form expression for it*. **Proof 11 (Proof of Theorem 8)** *For* 0 \< *p* \< 1, *let ϕ*(*w*) = \|*w*\|^*p*, *we have* ∂*ϕ*(*w*) = ∅ *when w* = 0. *In order to overcome the singularity of* (\|*w*\|^*p*)′ = *pw*/\|*w*\|^2−*p* *near w* = 0, *following Section 4 of*, *we consider for* 0 \< *ϵ* \<\< 1 *the approximation* $$\begin{array}{r} {\partial\phi(w) \approx \frac{pw}{\max\left( \epsilon^{2 - p},|w|^{2 - p} \right)}.} \\ \end{array}$$ *It is important to observe that* **Prox**_λ ∥⋅∥_*p*(*x*) = 0 *if* $$\begin{array}{r} {|x|^{2} \leq (w - x)^{2} + 2\lambda w^{p},} \\ \end{array}$$ *which is equivalent to* $$\begin{array}{r} {x \leq \min\frac{w + 2\lambda w^{p - 1}}{2} = \frac{2 - p}{2(1 - p)}\left( 2\lambda(1 - p) \right)^{\frac{1}{2 - p}},} \\ \end{array}$$ *equality obtained when* $w = \sqrt[{p - 2}]{2\lambda(1 - p)}$. *Otherwise, the necessary optimality condition is given by* $$\begin{array}{r} {w - x + \frac{\lambda p}{\max\left( \epsilon^{2 - p}, \middle| w \middle| {}_{2 - p} \right)}w = 0.} \\ \end{array}$$ *To solve* *for nonnegative w, let* $$\begin{array}{r} {g(w) = |x| - \frac{\lambda p}{\max\left( \epsilon^{2 - p}, \middle| w \middle| {}_{2 - p} \right)}w,} \\ \end{array}$$ *we consider the iteration* $$\begin{array}{r} {w^{(n)} = g\left( w^{(n - 1)} \right) = |x| - \frac{\lambda p}{\max\left( \epsilon^{2 - p}, \middle| w^{(n - 1)} \middle| {}_{2 - p} \right)}w^{(n - 1)}.} \\ \end{array}$$ *One can easily see that* $$\begin{array}{r} {|g^{\prime}\left. (w) \middle| = \lambda p(1 - p) \right.w^{p - 2} < 1,} \\ \end{array}$$ *if and only if* $|w| \geq \sqrt[{2 - p}]{\lambda p(1 - p)}$. *Notice that the first and second order derivatives of f*(*w*) = (*w* − *x*)² + 2λ\|*w*\|^*p* *are* $$\begin{array}{rc} {f^{\prime}(w)} & {= w - x + \lambda pw^{p - 1}} \\ \end{array}$$ $$\begin{array}{rc} {f^{''}(w)} & {= 1 + \lambda p(p - 1)w^{p - 2},} \\ \end{array}$$ *and one can easily verify that f*(*w*) *is concave in the range of* $\left( 0,\sqrt[{2 - p}]{\lambda p(1 - p)} \right)$, *and is convex in the range of* $\left( \sqrt[{2 - p}]{\lambda p(1 - p)},\infty \right)$. *By using the Contraction Mapping Theorem (Theorem 1.5 on page 11 of*), *we have* $$\begin{array}{r} {w^{(1)} = |x| - \left. \lambda p \middle| x \right|^{p - 1},w^{(2)} = |x| - {\lambda p\left( \middle| x \right|} - \left. \lambda p \middle| x \right|^{p - 1})^{p - 1},\ldots.} \\ \end{array}$$ *The iteration will eventually converge to a fixed-point, which is the root w* = *g*(*w*) in the interval $\left( \sqrt[{2 - p}]{\lambda p(1 - p)},\infty \right)$. *Moreover, by noting that the derivative of x*(*w*) = *w* + λ*pw*^*p*−1 *is* $$\begin{array}{rc} {\overset{˙}{x}(w)} & {= 1 + \lambda p(p - 1)w^{p - 2},} \\ \end{array}$$ *by solving* $\overset{˙}{x}(w) = 0$, *we have* $$\begin{array}{r} {\overline{w} = \sqrt[{2 - p}]{\lambda p(1 - p)},} \\ \end{array}$$ *then* $$\begin{array}{r} {\min x(w) = x_{\overline{w}} = \overline{w} + \lambda p{\overline{w}}^{p - 1} = \left( \lambda p(1 - p) \right)^{\frac{1}{2 - p}} + \lambda p\left( \lambda p(1 - p) \right)^{\frac{p - 1}{2 - p}},} \\ \end{array}$$ *and* $\left. |w| \geq \sqrt[{2 - p}]{\lambda p(1 - p)}\Rightarrow|x| > \left( \lambda p(1 - p) \right)^{\frac{1}{2 - p}} + \lambda p\left( \lambda p(1 - p) \right)^{\frac{p - 1}{2 - p}} \right.$. *Combined with*, *we denote* $$\begin{array}{r} {\aleph_{\lambda,p} = \max\left\{ \frac{2 - p}{2(1 - p)}\left( 2\lambda(1 - p) \right)^{\frac{1}{2 - p}},\frac{1}{2}\left( \lambda p(1 - p) \right)^{\frac{1}{2 - p}} + \lambda\left( \lambda p(1 - p) \right)^{\frac{p - 1}{2 - p}} \right\}.} \\ \end{array}$$ *Thus the proof is completed*. This research was supported in part by the National Natural Science Foundation of China (61201392),in part by the Natural Science Foundation of Guangdong Province, China (No. 2015A030313497), and in part by the NSF140928 of the United States. [^1]: The authors have declared that no competing interests exist.

# Introduction With one of the fastest growing incidence rates in Asia and a total of 690,000 people living with HIV, HIV is a big problem in Indonesia. Despite a relatively low overall prevalence (0.27%), the prevalence among HIV risk-groups is much higher, reaching up to 25.8% among Men who have Sex with Men (MSM), 24.8% among transgender people, and 28.8% among injecting drug users (IDU). Both lack of knowledge on one hand, and stigma towards people living with HIV and their relatives on the other hand are amongst the main causes of the growing epidemic. Several strategies are available to bring a halt to the HIV epidemic in Indonesia, including outreach, harm-reduction community meetings (HRCMs) and Information, Education and Communication (IEC) programs. Outreach focuses on reaching hidden populations of HIV risk groups to engage them in the process of reducing HIV risk behaviors. HRCMs aim at increasing knowledge and awareness about several HIV related topics among IDUs. IEC is a program developed to reduce stigma and improve knowledge in several population groups, like mothers and pregnant women. Although many HIV/AIDS interventions are available, not all of them can be implemented. In fact, when implementing all HIV interventions described in the strategy of Indonesia in 2014, an amount of US\$ 208 million would have been required, whereas only US\$ 97 million was available. This US\$ 111 million funding gap emphasizes the need for prioritizing strategies in this country. Multi-Criteria Decision Analysis (MCDA) is a method to support such priority setting in resource-limited settings with optimal use of available resources. One important criterion considered in an MCDA, is cost-effectiveness. Determining this criterion requires comprehensive and reliable data on the costs of implementing HIV/AIDS interventions. This type of data is however scarce and highly dependent on the specific type and setting of a program. Thus, a reliable cost analysis of all available HIV/AIDS interventions in Indonesia is needed. This study therefore aims to assess the societal costs of outreach programs to MSM and transgenders, harm-reduction community meetings for IDUs, and information, education and communication programs at maternal and child health posts in Bandung, Indonesia in 2016. # Materials and methods ## Study setting and population This study was conducted in Bandung, the capital of West Java province. The HIV/AIDS epidemic that Bandung is facing is a concentrated epidemic comparable to the national estimate. ### Outreach to MSM and TG The outreach to both MSM and TG living in Bandung is conducted by a non- governmental organization (NGO) in Bandung, called Srikandi Pasundan (SP). The outreach program of SP is focussing on informing these key populations about HIV/AIDS and referring them to HIV testing facilities. The program is performed 26 times a month by both paid workers and volunteers, who visit HIV hotspots like malls, parks, and gyms and have a chat, hand out information packages, or refer individuals at risk to HIV testing facilities. ### Harm reduction community meetings for IDUs The HRCMs in Bandung are organized by a NGO called Grapiks for IDUs from around Bandung with the main goal of increasing IDU’s knowledge about the effects of drug use. Several topics are discussed, including HIV/AIDS, tuberculosis, hepatitis, legal issues of drug use, and sterilization of needles. The meetings are provided by staff members of Grapiks and community health centres (the so- called puskesmas) at either community-health centres (11 times a month) or HIV/AIDS clinics (once a month). ### IEC at maternal & child health posts Maternal & child health posts (MCHPs) (the so-called posyandus) are monthly- organized health facilities in Indonesia, where pregnant women or mothers with their children can come for a health check. Besides health measurement, the MHCP also offers IEC about a wide range of health topics, among which HIV/AIDS. The HIV/AIDS IEC program consists of a monthly IEC stand and an IEC presentation, which is organized three to four times a year. The program is coordinated by Warga Peduli AIDS (WPA), a civil society organization concerned with HIV/AIDS, with each district having one or more WPA volunteers who provide the program. ## Data collection and cost estimation Data collection took place in April and May 2017 and focused on the costs made in 2016. Costs were estimated from a societal perspective according the World Health Organization’s (WHO) guidelines for cost analysis in primary health care. We made a distinction between health care costs and non-health care costs. Costs were collected in Indonesian Rupiah (IDR) and converted to US\$ using the official 2016 annual conversion rate. Data were both registered and analysed in Microsoft Office Excel 2007. A detailed description of collection and estimation of costs can be found elsewhere. Data on health care costs were collected by interviewing stakeholders involved in the program, being staff from the coordinating organization (four employees from Grapiks, the program manager and one of the fieldworkers from Srikandi Pasundan, and four employees from the MCHP’s) or staff from the Bandung Aids Commission (KPA Bandung). Written informed-consent was obtained prior to all interviews with these stakeholders and participants received a reimbursement of IDR 100.000 for their participation afterwards. Health-care costs were further divided into capital costs and recurrent costs. Capital costs were defined as costs incurred for resources that last longer than one year. Recurrent costs were defined as costs incurred for resources that are purchased regularly. All inputs related to the programs were identified, classified, and quantified using a micro-costing approach. Capital costs were estimated annually with a discount rate of 3%. Annual costs were based on the working life of the capital resource and the costs of purchasing that resource in 2016. Working life of buildings was assumed to be twenty years, of trainings to be ten years and of furniture and equipment to be five years, based on general agreements. Purchase costs were based on documented data, market prices or expert opinion alternatively. Recurrent costs were calculated by multiplying the costs of a resource unit by the yearly quantity of usage of the resource unit. Data on quantity of usage were based on documented data or expert opinion alternatively. Costs of a resource unit were based on documented data, salary registers, market prices or expert opinion. Household costs were the only included non-health care costs and were further divided into productivity loss costs and travel costs. Productivity loss costs were defined as the income that the visitors miss because of spending their time at the program and were based on the value of their leisure time. Data on these subjects were obtained with a survey conducted among 16 MSM and 13 TG of the outreach program, 23 IDUs visiting the HRCMs and 35 visitors of the MCHPs (see : survey questionnaires). Verbal informed consent was obtained from all participants prior to filling in the questionnaire and participants received a small reimbursement in the form of a pack of biscuits afterwards. The survey contained questions about monthly income, daily working hours, monthly expenditure, travel time, and travel costs. We determined productivity loss costs by measuring the amount of time people spent attending a program and multiplying this by the value of their leisure time. For employed visitors, the leisure time value was calculated based on their self-reported monthly salary, whereas non-employed visitors were assumed to have a leisure time value equal to the minimum salary. Additionally, various attendants of the IEC presentation indicated a desire to increase the frequency of presentations. A scale-up scenario analysis was thus performed determining the annual societal costs if the IEC presentation would be scaled up, i.e. providing it every month instead of three to four times a year. ## Sensitivity analysis As several cost calculations relied on assumptions, a sensitivity analysis was performed to determine the impact of variable uncertainty on the societal costs. The analysis was performed on the biggest assumptions (i.e. underlying the largest cost categories of each intervention) with a plausible 30% uncertainty range being applied. Assumptions included for outreach were duration of meetings, number of referrals and distance travelled, whereas expert opinion and duration of meetings were included for HRCMs. For the IEC program, assumptions on the value of leisure time of WPA volunteers and unemployed visitors were analysed. # Results In 2016, the SP outreach program reached 4500 MSM and 210 TG, resulting in a total number of 4710 people being reached. When looking at HRCMs, a total number of 133 meetings were conducted in 11 community-health centres and one HIV/AIDS clinic. The HIV/AIDS IEC stand was visited by 8 people a month and each presentation was visited by 22 people on average. The average yearly number of visitors per MCHP of both the stand and the presentation was 641. The societal costs paid in 2016 for all programs and the share of total costs for all categories are shown in. Outreach was the most expensive program with a total of US\$ 347,199, followed by HRCM costing US\$ 48,618 and IEC being the least expensive program with US\$ 337 in total societal costs. The costs for reaching out to one individual every week during the entire year of 2016, were US\$ 73.72. HRCMs costed US\$ 365.55 per meeting and the societal costs for providing IEC to one visitor were US\$ 0.51. Non-health care costs accounted for the biggest share in total costs in all three programs, followed by personnel costs for outreach and IEC, and by transport cost for the HRCMs. The additional upscale scenario of the IEC program at MCHPs showed that when intensifying the frequency of the presentation from three to four times a year to 12 times a year, the annual societal costs of the program will become US\$ 788.76, which is US\$ 451.63 more than in 2016. Moreover, the societal costs paid for providing the IEC program to one visitor, will become US\$ 0.47, compared to US\$ 0.51 in 2016. A 15% over- or underestimation in duration of outreach resulted in a deviation of US\$ 5.18 (7.0%) in reaching out to one individual and US\$ 24,396 in societal costs. Both number of referrals and distance travelled seemed to have no significant effects on costs. For HRCMs, a deviation of 15% in values estimated by experts resulted in an in- or decrease of societal costs with US\$ 4,275.00 (8.8%). A 15% and 30% decrease of the duration of HRCMs resulted in a decrease of the societal cost of respectively US\$ 3,589 (7.4%) and US\$ 7,179 (14.8%). In the societal costs of IEC, a 15% over- and underestimation of the value of leisure time of unemployed volunteers and visitors resulted in a US\$ 18.88 and US\$ 0.03 deviation of total costs and social costs per visitor respectively. # Discussion ## Main findings This study examined the societal costs of outreach to MSM and TG, harm reduction community meetings for IDUs and information, education and communication programs at maternal and child health posts in 2016. The societal costs for the interventions in 2016 were US\$ 347,199, US\$ 48,618 and US\$ 337.13 respectively. Besides, this study discovered that the costs for reaching out to one MSM or TG for one year were US\$ 73.72. For organizing one harm reduction community meeting, US\$ 365.55 had to be paid. Besides, the costs for providing IEC to one visitor of the MCHP were US\$ 0.51. For both the outreach- and IEC program, the second largest cost category was personnel. Despite being a big cost item of the SP outreach program, saving on personnel costs should be done with caution as it might reduce the effectiveness of the intervention. Cutting on personnel would likely result in outreach workers being less able to really take time for their clients and by means of that affect the special bond between workers and clients. Saving costs of personnel, or in fact any cost category of IEC, does not seem very preferable as costs are already low and savings could hamper effectiveness. Transport costs could be a more suitable category for cutting costs in both outreach and HRCMs. These costs are mostly determined by transportation fees given to either outreach workers or participants of HRCMs. These fees are often higher than the transport costs that are actually made. Adjusting this system of transportation fees in a way that fees approximate the actual money spent, seems therefore a reasonable way of cutting costs. For the HRCMs however, reducing this transportation fee might make the HRCMs less attractive for the IDUs to visit and might therefore reduce the effectiveness of the program. Savings should thus be done with caution. Productivity loss costs formed a large component of the overall societal costs in all three interventions, eliciting an ongoing economic debate on the value of leisure time. In order to gain a comprehensive societal cost overview, we chose to base visitors’ leisure time values on market wage rates. Although this decision largely determined the productivity loss costs, the financial burden on visitors should not be neglected when adopting a societal perspective. The low annual frequency combined with the expressed desire by IEC visitors to increase the intensity of the IEC program, raised questions about the costs of intensifying the program. Upscaling IEC resulted in an extra amount of US\$ 451.63 to be paid annually. However, when considering the costs per visitor, it was shown that the upscale scenario would be as efficient as current practice, paying US\$ 0.47 per visitor compared to US\$ 0.51 in current practice. Nevertheless, the question remains whether the increase in annual societal costs is affordable. Upscaling of outreach and HRCMs was deemed irrelevant, since these interventions were already implemented on a frequent basis and large scale. ## Limitations Although the methods of our study were largely based on the guidelines as described in the WHO manual for cost analysis in primary health care, this study had some limitations. First, some cost components were solely based on expert opinion as this was often the only way of calculating costs in absence of documented data. Prices that were the result from expert opinion were however checked with local market prices or follow-up interviews if possible and, for HRCMs, expert opinion was included in the sensitivity analyses to show influences on outcomes. Secondly, the cost analysis of outreach and HRCMs were both conducted for one single organisation in Bandung, meaning conclusions on costs can only be drawn for this single organisation. It is however expected that the way outreach and HRCM activities are undertaken in Bandung is comparable to other regions. The results of this study might therefore be considered a rough cost estimate of other comparable programs in Indonesia. This does not apply to the IEC program as this was done for an organization which implements the program throughout the whole country. The small sample size of three MCHPs in this study however means extrapolation should be done with caution as well. Thirdly, the non-health care costs calculations were based on questionnaires filled in by 16 MSM and 13 TG, 23 IDUs, and 35 women visiting MCHPs meaning non- health care costs found in this study might not represent the health care costs of the entire population. Lastly, assumptions needed to be made to complete the cost analysis. This might cause a deviation from the true costs if assumptions were not in line with reality. However, sensitivity analyses showed the influence of over- or underestimations of the biggest assumptions being made is low (\<10%). ## Available literature & future research No other cost analyses of these specific HIV/AIDS interventions in Indonesia have been done so far, hampering any comparison to other results in similar settings. A comparable micro-costing study in the African republic of Benin however reported the costs of outreach through peer educators to be 38,95 US\$ per person reached. These considerably lower costs of outreach in Benin compared to outreach for MSM and transgender in Indonesia are most likely attributable to the service provider’s perspective being chosen excluding non-health care costs, amounting \>56% of total costs for outreach in this study, from the analysis. Additionally, the economic costs of IEC for the general public in Andhra Pradesh state, India, were measured to be 0.16 US\$ per person. As these costs do not include non-health care costs either, these findings are comparable with our findings for the costs of IEC at MCHPs of 0.51 US\$ per person. More importantly however, only three out of the many available HIV/AIDS interventions in Bandung were included in this study. Several other programs in Bandung have already been analysed and significant efforts are made to review all other programs, but a full overview is still lacking. Additional micro- costing studies should thus be conducted until this full overview, ready to be used in an MCDA, is available. MCDA however often considers cost-effectiveness instead of costs, meaning data on effectiveness of all interventions should be available as well. In contrast to cost data, there is hardly any data on effectiveness of interventions meaning studies determining the effectiveness of all interventions should be conducted as well as soon as possible. Additionally, Siregar et al. showed that support of peers/family is important in avoiding loss-to-follow-up of HIV/AIDS patients. Encouraging family members to support their relative might thus help to avoid a possible decrease in effectiveness caused by implementing savings in transportation fees for HRCMs. Lastly, other studies support our finding that scaling up of a HIV/AIDS intervention within Bandung could prove to be a valuable step with low additional costs. # Conclusion The total societal costs of outreach to MSM and Transgender, HRCMs to IDUs, and IEC at MCHPs are US\$ 347,199, US\$ 48,618, and US\$337.13 respectively. It can be concluded that the IEC program is a low-cost program and that costs of the outreach program are relatively high with the HRCMs being not the most expensive nor the cheapest program. ## Recommendations As this study was performed to inform the currently ongoing MCDA process in Bandung, our first recommendation is to combine our cost data with results on the effectiveness of the interventions and use them for this process. Similarly, cost and effectiveness data of other HIV/AIDS interventions in Bandung should be included in the MCDA as well. Considering costs and the contribution of each cost category to the societal costs, a few recommendations can be made as well. First, it is recommended to critically examine and possibly revise the system of transportation fees for both outreach workers and participants of HRCMs. The effectiveness of both programs should however always be in scope to e.g. prevent the scenario in which less clients will come to HRCMs. As the costs of the IEC program are already relatively low, it is not recommended to reduce any cost category. In fact, we recommend to implement the upscale scenario as you reach more visitors and by means of that possibly increase the effectiveness of the program. Before doing so, it should however be investigated if any organization involved in the program is willing and/or able to fund the upscaling of IEC presentation. # Supporting information Our gratitude goes to all staff of Srikandi Pasundan, Grapiks, WPA, and KPA Bandung, who participated in and facilitated most of the interviews. Besides, we would like to thank all our colleagues from the PRISMA team, but in particular Indra Yudha Mambea and Muhammad Putra Hutama, who put a lot of efforts in assisting us. Finally, our special gratitude goes out to both Prof. Dr. R. Baltussen and Dr. A.Y.M. Siregar for their supervision whenever needed during our study. [^1]: The authors have declared that no competing interests exist. [^2]: ‡ These authors also contributed equally to this work

# Introduction Galectin-3 (LGALS3) is a lectin and member of the galectin family of carbohydrate binding proteins that have an affinity for beta-galactosides. Galectin-3 plays a role in fibrosis, inflammation, and proliferation. Galectin-3 is secreted into the systemic circulation by unknown mechanisms and is increasingly recognised as a potential biomarker with clinical value. Increased galectin-3 levels have been associated with various diseases, including cancer, immunological disorders, and cardiovascular traits. Plasma galectin-3 levels are even being considered as a marker of response to cancer treatment. To enhance our knowledge on galectin-3 biology we performed the first genome- wide association study (GWAS) on circulating galectin-3 levels and observed two loci associated with circulating galectin-3 levels. One locus harbours *LGALS3* the gene encoding galectin-3 and the other locus harbours the *ABO* gene which has previously been associated with inflammatory markers, lipids and haematological parameters. # Results We performed a GWAS analysis of 2,269,099 genotyped or imputed autosomal SNPs (HapMap 2 build 36 CEU panel) in 3,776 subjects of the PREVEND cohort (**Table S1**). All included subjects were of European descent. The quantile-quantile plot for association is shown in. There were 2 loci significantly associated with galectin-3 levels (P\<5×10⁻⁸) and 11 SNPs showing suggestive evidence (P\<5×10⁻⁶ and P\>5×10⁻⁸). We performed further testing of the lead-SNP of the loci using an additional subset of 3,516 independent subjects derived from the PREVEND cohort. Using inverse-variance fixed effect meta-analysis we combined the evidence. None of the suggestive, but both 2 P\<5×10⁻⁸ loci of the discovery phase, were confirmed in the independent samples. One locus harbours the *LGALS3* gene and the other locus harbours the *ABO* gene. The *LGALS3* locus accounted for 25.6% of the phenotypic variance. The *ABO* locus explained 3.8% and together *LGALS3* and *ABO* explained 29.2% of the phenotypic variance. Of note, common genetic variation explained twice the amount of the variation of circulating galectin-3 levels compared to age, age squared (age²), gender, and body mass index combined (11.6%). ## Putative causal genetic variants The lead SNP (rs2274273) of the *LGALS3* locus lies in high LD with two non- synonymous variants (rs4644; r² = 1.0 and rs4652; r² = 0.91). As these variants were not present on our platform and not well imputed we wet-lab genotyped these variants and confirmed their association. We next considered potential confounding by the specific galectin-3 assay used and noticed the epitopes of the antibodies used are directed against the region harbouring the non-synonymous variant. Therefore, this variant might affect the affinity of the antibody and not represent a true difference in circulating galectin-3 levels. We did not find variants in high LD (r²\>0.8) associated with the lead SNP (rs644234) of the *ABO* locus. Next, we searched for eQTLs in 1,469 samples from peripheral blood for which gene expression levels were obtained using illumine HT12V3 and illumine H8v2 platforms. rs2274273, rs4644, and rs4652 were all associated with *LGALS3* gene expression levels and rs2274273 and rs4644 were also the strongest SNP associated with that particular *LGALS3* probe. Finally, to gain further insights we queried the catalogue of published genome wide association studies for our loci and observed no previous associations for the *LGALS3* locus but many previous genome wide associations findings have been reported for the *ABO* locus. Previous SNP associations in or near *ABO* are in high linkage disequilibrium with our lead SNP and include associations with inflammatory markers, lipids and haematological parameters as well as diseases such as cancer and coronary heart disease (**Table S2**). ## Relevance of LGALS3 variant for prognostic value of plasma galectin-3 levels To study the relevance of the rs2274273 and rs4644 (r² = 1) variant in the *LGALS3* locus for the prognostic value of the galectin-3 assay on mortality in the general population we repeated our earlier reported analyses with and without rs2274273 as a covariate in the model. Knowledge of the genotype did not appear to change the prognostic value of plasma galectin-3 levels. # Discussion We report the first genetic association study on galectin-3 levels and identified 2 genome-wide significant loci; one including the galectin-3 encoding gene (*LGALS3*) and the other gene being *ABO*. Galectin-3 is a member of the galectin family that comprises of lectins with affinity for beta-galactosidases containing carbohydrates. The galectin gene family is evolutionarily ancient and can be found in vertebrates, invertebrates, and even in protists suggesting an important role in biology. All galectins have a carbohydrate-recognition domain (CRD) consisting of many conserved sequence elements and each galectin has an individual carbohydrate-binding preference. Galectin-3 is an unique galectin as it contains a non-lectin N-terminal region which is connected to the CRD. Galectin-3 is therefore referred to as a chimera- like galectin. Galectin-3 does not contain a signal sequence and is primarily localised within the cytoplasm. It can be externalized by a mechanism independent of the endoplasmic reticulum (ER)-Golgi complex. Galectin-3 has high affinity for lactose and N-acetyllactosamine but can also interact with a wide array of other carbohydrates, membrane and extracellular matrix proteins. Upon ligand binding, galectin-3 (and its ligands) forms cross-links, is involved in strengthening cell-cell interactions, and is associated with stiffening of the extracellular matrix and fibrogenesis. Galectin-3 has been shown to play a role in inflammatory diseases, cancer and heart failure,. Little is known about the regulation of galectin-3. The galectin-3 promoter contains several responsive elements, including Sp-1, AP-1 and cAMP responsive elements. We now report the first 2 genome wide associations with circulating galectin-3 levels. The strongest locus is within the *LGALS3* gene. The lead SNP (rs2274273) is in full LD with two non-synonymous SNPs (rs4644 and rs4652) which were confirmed by follow-up genotyping. Both rs2274273 and rs4644 affected *LGALS3* gene-expression providing a potential explanation for the observed effect. In the current study we also tested whether knowledge of the lead variant in the *LGALS3* gene might obscure the association of plasma galectin-3 levels with outcome but it did not alter our previously published associations further supporting a true effect of these variant on galectin-3. However, some note of caution is warranted. Associations of coding SNPs (e.g. rs4644 and rs4652) that structurally change the properties of its encoded protein can give rise to false positive associations when that protein is also the phenotype under investigation. The non-synonymous SNPs identified in our study also lies within or near the epitopes of the antibodies used for the galectin-3 assay. These antibodies might have different affinities for the amino acid change and therefore this association could also be artifactual. Interference of antibody based assays with epitopes directed against regions harbouring non-synonymous variants are not novel and have previously been reported for the *NPPA-NPPB* locus when ANP levels were measured. Although gene-expression analyses and association with outcome are suggesting a true effect, additional work will be required to define the precise mechanisms of our reported association at the *LGALS3* locus. Our second genome wide locus is the *ABO* locus. The *ABO* locus is becoming an increasingly complex and pleiotropic locus. Variants in *ABO*, and in high LD with our lead SNP (rs644234), have been associated by genome wide association studies with various blood measured traits and diseases. This includes several inflammatory markers, lipids, hematological parameters, cancer, inflammatory diseases, and cardiovascular diseases (**Table S2**). Interestingly, galectin-3 levels also are associated with many of these conditions. Galectin-3 can indeed bind to polysaccharides of the ABO epitopes and even more strongly to the A- or B-histo-blood group epitopes versus the O group. However, this does not explain how the *ABO* gene variant affects circulating galectin-3 levels. In summary, we performed a GWAS on plasma galectin-3 levels and identified two genome wide significant loci, one including the *LGALS3* gene and the other the *ABO* gene. The origins of these associations should be further validated by means of functional experiments. # Materials and Methods ## Study population We studied subjects included in the PREVEND cohort. The PREVEND cohort has been described in detail elsewhere. In brief, 8,592 subjects were enrolled in the PREVEND cohort in 1997–1998. Subjects were asked to refrain from eating and drinking prior to their visit (fasting) in the outpatient clinic (between 8:00 a. um and 1:00 pm) and blood samples were drawn and stored at −80C. The PREVEND study was approved by the local medical Ethical Committee, and is conducted in accordance with the guidelines of the Declaration of Helsinki. All subjects provided written informed consent. ## Galectin-3 Measurements For 7,968 subjects plasma was available to measure plasma galectin-3 levels. The galectin-3 assay is an enzyme-linked immunosorbent assay (BG Medicine, Inc., Waltham, USA). This assay quantitatively measures the concentration of human galectin-3 levels in EDTA plasma. This assay has high sensitivity (lower limit of detection 1.13 ng/mL) and exhibits no cross reactivity with collagens or other members of the galectin family. Commonly used medication like ACE- inhibitors, beta blockers, spironolactone, furosemide, acetylsalicylic acid, warfarin, coumarines, and digoxin have no interference with the assay. All samples were assayed in duplicate. Two standard controls were included in all runs: a lower control (expected value: 13.0–23.1 ng/mL) and a higher control (expected value: 48.9–81.5 ng/mL). The average lower control results were 16.65±1.13 (coefficient of variance: 6.8%), and the average higher control results were 68.17±3.20 (coefficient of variance: 4.7%). ## Genotyping, Quality control & Imputation Genotyping in 4,016 of the total number of participants in PREVEND was carried out using Illumina HumanCytoSNP-12 arrays. SNPs were called using Illumina Genome Studio software. Forty-seven subjects were excluded from analyses because call rates were \<0.95. Another 65 subjects were excluded because they were closely related as judged based on Identity-By-Descent estimation using PLINK v1.07. Population structure was assessed using PCA based on 16,842 independent SNPs. Based on this analysis, an additional 2 samples were excluded that diverged from the mean with at least 3 standard deviations (Z-score \>3) for the first 5 PCAs. Another 35 subjects were excluded based on sex inconsistencies. We excluded samples with a genetic similarity \>0.1. Of 87 subjects no phenotype was available because of missing plasma samples for assessment of Galectin-3. As a consequence 3,776 (1,927 males, 1,849 females) were available for GWAS analysis. SNPs were excluded with a minor allele frequency of \<0.01, call rate \<0.95, or deviation from Hardy Weinberg equilibrium (P\<1×10⁻⁵). Genome wide genotype imputation was performed using Beagle v. 3.3.1, 232,571 genotyped SNPs were imputed up to 2,269,099 autosomal SNPs with NCBI build 36 of Phase II HapMap CEU data (release 22) as reference panel. Replication genotyping was performed by KBiosciences (KBiosciences, Herts, UK) utilizing the SNPline system in an additional 3,516 independent subjects of the PREVEND study. ## Gene-expression analyses We investigated whether each of the associated variants had an effect on gene expression levels by mapping *cis*-expression quantitative trait loci (*cis*-eQTL) in 1,469 samples from peripheral blood, for which gene expression level measurements were obtained using Illumina HT12v3 and Illumina H8v2 platforms. Since the genotypes were imputed using the CEU population of HapMap 2 release 24 as reference, eQTL effects were tested using the imputation dosage values. Effects for SNPs (MAF \>5%, HWE \>0.001) were considered *cis*-eQTLs when the distance between the SNP and the midpoint position of the probe was smaller than 1 MB. As multiple testing correction, we controlled the false discovery rate (FDR) at 0.05, by comparing observed p-values to the null distribution obtained from permuting the expression phenotype labels relative to genotype labels 100 times. We also determined the top eQTL SNP for each given probe and tested whether the GWAS SNP had an independent effect on the associated gene expression probe after removing the effect of the top eQTL SNP. ## Statistical analysis Galectin-3 was non-normally distributed and was log transformed before regression analyses. We calculated residuals of galectin-3 levels after adjustment for age, age², and gender. GWAS analyses were performed on residuals using an additive genetic model in PLINK (v 1.07). The most significant (P\<5×10⁻⁸) SNPs (lead SNP) at each locus was taken forward for further testing. The explained variance of the significant associations was analysed using the directly genotyped variants from the replication stage. Fixed-effect meta-analysis was performed using the variance weighting method of the METAL software package to calculate the overall p-value. The Cox proportional-hazards model was used to calculate the hazard ratio and 95% confidence intervals (CI) of galectin-3. Based on our previous work, sequential models were fitted without and with the SNP of interest. The first model including no covariates (unadjusted) and the second model adjusted for age and gender and the third model adjusted for: age, gender, previous myocardial infarction, previous stroke, hypertension, hypercholesterolemia and diabetes. The assumptions underlying the proportional hazards model were tested and found valid. Analyses were performed using STATA version 11.0 for Windows software (StataCorp LP, College Station, TX, USA). # Supporting Information [^1]: We have read the journal's policy and have the following potential conflicts of interest: BG Medicine, Inc., has certain rights related to galectin-3 measurements. BG Medicine Inc. provided an unrestricted research grant to the Department of Cardiology of the University Medical Centre Groningen. Drs. R.A. de Boer and D.J. van Veldhuisen have received consulting and speaker's fees from BG Medicine, Inc. [^2]: Conceived and designed the experiments: RAdB NV IML PvdH. Performed the experiments: RAdB NV HJW SJLB ACMK LF IML PvdH. Analyzed the data: RAdB NV HJW SJLB ACMK LF IML PvdH. Contributed reagents/materials/analysis tools: RAdB NV DJvV HJW SJLB RTG ACMK WHvG LF IML PvdH. Wrote the paper: RAdB NV IML PvdH. Read and commented on the earlier drafts of this manuscript: RAdB NV DJvV HJW SJLB RTG ACMK WHvG LF IML PvdH.

# Introduction Systemic sclerosis (SSc) is a chronic multisystem connective tissue disorder characterized by fibrotic events, vascular damage and autoantibody production. Two main clinical subtypes have been defined based on the extent of skin involvement, limited cutaneous scleroderma (lcSSc) and diffuse cutaneous scleroderma (dcSSc). Recent candidate gene and genome-wide association studies (GWASs) clearly suggest that an important genetic component underlies this disease. In this regard, an increasing number of *loci* have been reported to be convincingly associated with the susceptibility and clinical manifestations of SSc in the last years. However, the causal functional mutations responsible for these associations have not been unambiguously identified yet in most cases. Outside the HLA region, interferon (IFN) pathway genes, which encode cytokines with critical modulatory effects on innate and adaptive immunity, have been shown to represent a key component of the genetic network leading to autoimmune processes. Interestingly, a misregulated expression of type I IFN genes, also referred to as IFN signature, have been observed in peripheral white blood cells patient subsets of several autoimmune diseases, thus suggesting that the IFN signaling plays a crucial role in autoimmunity. Indeed, multiple single- nucleotide polymorphisms (SNPs) of the IFN regulatory factor 5 gene (*IRF5*), a major regulator of the type I IFN induction, have been associated with different rheumatic disorders such as SSc, systemic lupus erythematosus (SLE), rheumatoid arthritis (RA), and Sjögren’s syndrome (SS). The *IRF5* association with SLE was narrowed down to three different haplotype blocks that seem to have independent functional consequences, including 1) alteration of the protein stability, 2) creation of a donor splice site in intron 1 resulting in transcription of an alternative exon 1B, and 3) modification of the 3′UTR length which affects expression levels. Subsequent studies in SSc patients suggested that genetic variation within *IRF5* correlate with SSc severity and survival. Based on the above, we decided to explore whether the functional haplotype blocks described by Graham *et al*. were also susceptibility signals affecting SSc development and progression. For that purpose, we analysed the allele frequencies of three representative *IRF5* genetic variants that have been previously associated with SSc, in five large Caucasian European cohorts and performed allelic combination and dependency tests. # Patients and Methods ## Study Population We recruited a total of 3,361 SSc patients and 4,012 unaffected controls of Caucasian origin from five different European countries, including an initial cohort from Spain and four replication cohorts from Germany, The Netherlands, Italy and United Kingdom. Case and control sets were matched by geographical origin and ethnicity. Written informed consent from all participants and approval from the local ethical committees of all centres involved in the study were obtained in accordance with the tenets of the Declaration of Helsinki. All SSc patients fulfilled the classification criteria by Leroy *et al*.. The clinical features of the different SSc cohorts are shown in. Case sets were further subdivided based on their skin involvement into limited cutaneous scleroderma (lcSSc) and diffuse cutaneous scleroderma (dcSSc) subgroups, and by autoantibody status according to the presence/absence of anti-centromere antibodies (ACA) or anti-topoisomerase antibodies (ATA), which were detected using standard procedures. Pulmonary fibrosis (PF) was diagnosed by high resolution computed tomography (HRCT). ## SNPs Selection and Genotyping Samples were genotyped for three *IRF5* tag SNPs, rs10488631, rs2004640, and rs4728142, representative of three different haplotype blocks (refers to as Groups 1–3, respectively) which have been reported to have functional roles in SLE patients : Group 1 includes SNPs tagging a 30-bp in-frame INDEL variant of exon 6 that alters protein stability; Group 2 includes an exon 1B splice site variant; and Group 3 corresponds to genetic variants located in a conserved polyadenilation signal sequence that alters the length of the 3′UTR, thus affecting expression levels. Genomic DNA was obtained from peripheral blood cells using standard procedures, and genotyping was performed using TaqMan® 5′ allele discrimination assays (IDs: C\_\_\_2691242_10, C\_\_\_9491614_10, and C\_\_\_2691222_10), in a 7900 HT Fast Real-Time PCR System (Applied Biosystems, Foster City, California, USA). ## Statistical Analysis The statistical power of the study was calculated with Power Calculator for Genetic Studies 2006 software (<http://www.sph.umich.edu/csg/abecasis/CaTS/reference.html>), which implements the methods described in Skol *et al*.. PLINK v1.07 (<http://pngu.mgh.harvard.edu/purcell/plink/>) was used to carried out all statistical analyses of allele frequencies. *P*-values were obtained by performing 2×2 contingency tables and χ² test and/or Fisher’s exact test, when appropriate. Since the association between *IRF5* and SSc has been confirmed in several independent studies , we considered appropriate to set the significance threshold at *P  = *0.05. Odds ratios (OR) and 95% confidence intervals were calculated according to Woolf’s method. Breslow–Day (BD) test method was used to estimate the homogeneity amongst populations. Pooled analyses were performed by Mantel-Haenszel test under fixed effects, or DerSimonian-Laird if the BD test reached statistical significance. Dependency of association between the studied genetic variants was determined by conditional logistic regression analysis as implemented in PLINK, and the allelic combinations were tested using PLINK, StatsDirect (V.2.6.6; StatsDirect, Altrincham, UK), and Haploview (V.4.2). To analyse whether allelic combinations would better explained the possible association than the genetic variants independently, we compared the goodness of fit of both models using PLINK. For that purpose, we calculated the deviance (defined as −2 × the log likelihood), which follows a χ² distribution, to assess the significance of the improvement in fit. If statistically significant differences in the improvement of fit were observed when the haplotype effect was considered, we assumed that this model was more informative explaining the putative association. # Results The overall statistical power of the study, based on previous *IRF5* reports, to detect associations with OR  = 1.2 at 0.05 significance level was 100% for rs2004640 and rs4728142, and 93% for rs10488631 (**Table S1 in**). Additionally, no significant departure from Hardy-Weinberg equilibrium was observed either in cases or controls in each analysed population (*P*  = 0.05). ## Allele Test The results of the global analyses of the discovery cohort and the four independent replication populations separately are shown in **Table S2 in**. Since the Breslow-Day test evidenced no heterogeneity of the ORs amongst the different cohorts (*P*  = 0.05), a combined meta-analysis was performed to test the overall effect of the *IRF5* genetic variants in the whole dataset. The pooled analysis showed that the three SNPs were strongly associated with the global disease (rs4728142: *P*  = 1.34×10⁻⁸, OR  = 1.22, CI 95%  = 1.14–1.30; rs2004640: *P*  = 4.60×10⁻⁷, OR  = 0.84, CI 95%  = 0.78–0.90; rs10488631: *P*  = 7.53×10⁻²⁰, OR  = 1.63, CI 95%  = 1.47–1.81). Highly significant *P*-values were also yielded when the different phenotype subgroups were compared against the control population (**Table S3 in**). However, no statistical significance was observed in the comparisons amongst SSc patients accordingly with the presence/absence of the different clinical characteristics and autoantibody profile, *i.e.* lcSSc vs. dcSSc (rs4728142: *P*  = 0.564; rs2004640: *P*  = 0.971; rs10488631: *P*  = 0.086), SSc ACA+ vs. SSc ACA- (rs4728142: *P*  = 0.359; rs2004640: *P*  = 0.357; rs10488631: *P*  = 0.449), SSc ATA+ vs. SSc ATA- (rs4728142: *P*  = 0.154; rs2004640: *P*  = 0.128; rs10488631: *P*  = 0.259), and SSc PF+ vs. SSc PF- (rs4728142: *P*  = 0.934; rs2004640: *P*  = 0.397; rs10488631: *P*  = 0.945), thus indicating the three *IRF5* polymorphisms are indeed associated with the global disease and there was no phenotype-specific association. ## Conditional Logistic Regression We decided to perform pairwise conditioning analyses to test whether there could be any dependency amongst them. The analysis showed that every SNP maintained its statistical significance after conditioning to the other two except for rs2004640, which was dependent of rs4728142. The moderate linkage disequilibrium between them (r2∼0.68) could explain this fact (**Table S4 in**). ## Haplotype Analysis We also analysed the allelic combinations of the *IRF5* genetic variants included in this study according to the global disease. The most associated haplotype was that containing the risk alleles of the three SNPs (rs4728142\*A-rs2004640\*T-rs10488631\*C: *P*  = 9.04×10⁻²², OR  = 1.75, CI 95%  = 1.56–1.97). The protective haplotype also showed a very significant *P*-value (rs4728142\*G-rs2004640\*G-rs10488631\*T: *P*  = 2.48×10⁻⁷, OR  = 0.84, CI 95%  = 0.78–0.89). When comparing the haplotype model with the independent SNP model, we observed a statistically significant improvement of the goodness of fit compared to rs4728142 (likelihood *P*-value  = 1.23×10⁻¹⁷), rs2004640 (likelihood *P*-value  = 1.94×10⁻¹⁹), or rs10488631 (likelihood *P*-value  = 1.48×10⁻⁴) individually. On the other hand, we also performed a sub-phenotype analysis of allelic combinations to test whether some haplotype could influence a specific clinical condition (**Table S5 in**). This analysis only showed a residual *P*-value for a low frequency haplotype in the PF+/PF- comparison (rs4728142\*G-rs2004640\*T-rs10488631\*T: *P*  = 0.041, OR  = 1.28, CI 95%  = 1.02–1.61). The rest of allelic combinations did not reach statistical significance in any other comparison. # Discussion GWAS data have confirmed *IRF5* as one of the strongest associated signals with SSc. In addition, it has been proposed that particular *IRF5* functional genetic elements contribute to SLE pathophysiology through their relationship with auto- antibodies and IFNα production. These data indicate that this gene may represent a crucial member of the genetic component underlying this type of autoimmune diseases. Previous published data suggested that two different *IRF5* haplotypes influence specific SSc phenotypes. It was hypothesised that these haplotypes may explain a possible *IRF5* association with dcSSc and PF, likely by tagging an intronic 5-bp biallelic insertion-deletion polymorphism (INDEL), which would represent the real causal functional variant. However, our results are not in agreement with this idea, since we did not find evidence for a specific genetic association between *IRF5* and any of the major clinical manifestations, despite the fact that two out of the three genetic variants comprising the previously described risk haplotype were covered in our analysis (rs2004640 and rs10954213 that is highly correlated with rs4728142). A similar discrepancy was also observed by Sarif *et al*., who failed to replicate the *IRF5* haplotype effect on PF described by Dieudé et al.. Our data are, however, consistent with recent GWAS follow-up studies that did not show a phenotype-specific association of *IRF5* with SSc, but a clear association with the overall disease. It should be noted that one of the SNPs included in this study, rs4728142, has been shown to correlate with longer survival and a milder pulmonary involvement in SSc patients. Taking this together with our results, it could be speculated that, although the risk variants of *IRF5* do not predispose to develop PF, they may influence the severity of some clinical features like PF. In any case, it is important to note that whereas antibody profile and disease subtypes are clearly a dichotomous outcome, PF can range from mild-stable to severe-progressive involvement (and the utilised approach for definition of PF does not differentiate the different severity scales of this disease manifestation). As stated before, the tag SNPs analysed here are representative of three haplotype blocks that have been reported to affect the function of the protein in different ways, including production of an alternative spliced isoform, alteration of polyadenylation sites that leads to a shorter messenger RNA, and reduction of protein stability. These three polymorphisms showed strong association signals in our study, supported by a high statistical power. Therefore, the functional alterations caused by their risk alleles may be also of high relevance in the pathogenic mechanisms that lead to SSc. However, the rs2004640 signal was dependant of that from rs4728142 in our study cohort. Hence, rs2004640 might not be an independent SSc susceptibility *locus* although it is functionally relevant, which suggests that not all functional variants in a determined risk gene may play an independent role in the associated disease. In any case, as described in SLE, we observed a significant additive effect amongst the three analysed SNPs because the haplotypes containing both the risk and protective alleles better explained the association between this *locus* and SSc. Hence, although no functional studies have been carried out yet to unmask the possible implication of the *IRF5* risk alleles in the SSc pathophysiology, it is likely that each one of the protein alterations described above influence the development of SSc individually, and that carrying all the three risk alleles results in a critically reduced protein function that highly increases SSc susceptibility. In conclusion, this study clearly shows that a haplotype of three different functional genetic variants within the *IRF5* region confer susceptibility to SSc. The fact that this association is shared with SLE adds another piece of evidence to the common genetic background of both diseases, and provides a new perspective for the study of the type I IFN pathway and its implication in the development of autoimmune conditions. # Supporting Information The authors thank Sofía Vargas, Sonia García, and Gema Robledo (from Instituto de Parasitología y Biomedicina ‘López-Neyra’, CSIC, Spain) for their excellent technical assistance, and all the patients and healthy controls for kindly accepting their essential collaboration. Banco Nacional de ADN (University of Salamanca, Spain) is thanked for supplying the Spanish controls. [^1]: ¶ Membership of the Spanish Scleroderma Group is provided in File S1. [^2]: The authors have declared that no competing interests exist. [^3]: Collection of data: LB CPS PEC JLC MFC LSC EB MTC MVE PA RS CL NH GR TW AK JHWD RM PS JML CF CD AH JW AJS MCV AEV TRDJR. Analysis and interpretation of data: FDC JEM JM LB CPS PEC JLC MFC LSC EB MTC MVE PA RS CL NH GR TW AK JHWD RM PS JML CF CD AH JW AJS MCV AEV TRDJR. Critical revision of the manuscript draft: JEM JM LB CPS PEC JLC MFC LSC EB MTC MVE PA RS CL NH GR TW AK JHWD RM PS JML CF CD AH JW AJS MCV AEV TRDJR. Conceived and designed the experiments: FDC JEM JM. Performed the experiments: FDC JEM. Analyzed the data: FDC JEM. Contributed reagents/materials/analysis tools: LB CPS PEC JLC MFC LSC EB MTC MVE PA RS CL NH GR TW AK JHWD RM PS JML CF CD AH JW AJS MCV AEV TRDJR. Wrote the paper: FDC.

# Introduction Group B streptococci (GBS) are the most frequent cause of early-onset neonatal infection, which is associated with significant morbidity and mortality among infants. The incidence rate of early-onset GBS infection ranges from 0.5 to 3.0 per 1,000 live births, with 4–10% mortality. In their guidelines, Centers for Disease Control and Prevention (CDC) endorse universal culture-based antenatal screening for GBS colonization in all pregnant women between 35 and 37 weeks of gestation to identify women who should receive intrapartum antibiotic prophylaxis. However, a national cohort study found that samples for culture screening between week 35 and 37 of gestation were negative for 81% of the mothers of babies who developed early-onset neonatal group B streptococcal disease. These data suggest that a change in colonization status may have occurred at the time of birth implying that antepartum sampling and culture are not optimal methods to provide a relevant GBS colonization status at delivery, resulting in missed opportunities to avoid GBS transmission from mother to infant during birth. A rapid nucleic acid amplification test performed at the time of delivery may constitute an alternative screening method. Previous studies suggest that the polymerase chain reaction (PCR) test at delivery may accurately reflect intrapartum GBS colonization status. A European consensus conference on intrapartum GBS screening and antibiotic prophylaxis recommends intrapartum antimicrobial prophylaxis based on a universal intrapartum GBS screening strategy using a rapid real-time testing. However, a time-consuming broth enrichment step precludes the use of rapid on-site PCR detection of GBS carriage at the time of delivery. The aim of this study was to compare an intrapartum culture for GBS with the diagnostic performance of BD MAX and GenomEra PCR assays without initial broth enrichment to achieve rapid PCR detection of vaginal carriage of group B streptococci at delivery and identify women who should receive intrapartum antibiotic prophylaxis. # Material and methods ## Ethical approval The clinical study was approved by the Regional Scientific Ethical Committees for Southern Denmark (S-20130089) and the Danish Data Protection Agency (2008-58-0035). All participants provided written informed consent. ## Study design This study was designed to compare new GenomEra GBS PCR results on the same set of samples as previously obtained BD MAX GBS PCR results and culture of vaginal swab samples. Originally, for practical reasons, culture was done on fresh samples and BD MAX GBS PCR on the same samples after one freeze-thaw cycle. Subsequently, GenomEra GBS PCR was performed on the frozen samples with an additional freeze-thaw cycle. Professional, fully qualified laboratory technicians at the Department of Clinical Microbiology performed culture, BD MAX PCR and GenomEra PCR assays. All sample aliquots used throughout the study, per included woman, came from the same ESwab sample tube. The culture results were blinded for the laboratory technicians prior to PCR analysis. In a small number of cases for both assays, the specimens were initially undetermined because of inhibition, reagent failure or system errors, which led to additional testing of the sample and repeating the DNA extraction and PCR assay. The results from both tests were interpreted and produced by the respective PCR equipment software as a qualitative response, either positive or negative for GBS. The results of the GBS culture, BD MAX GBS and GenomEra GBS assays were read and recorded separately by independent laboratory technicians. Discrepant PCR results between BD MAX and GenomEra were examined by a repeat PCR test of samples with both assays. ## Study population and sample collection From a previous prospective observational study including 902 pregnant women giving birth at Lillebaelt Hospital between 2013 and 2014, 91 culture positive and 279 culture negative intrapartum vaginal swab samples kept at -80°C were randomly selected and used to assess the performance of the GenomEra GBS PCR assay. The GenomEra results were compared with results from the original vaginal intrapartum culture and BD MAX GBS. The collection of original specimens was as described by Khalil et al. 2017: Briefly, during labor, the midwife obtained one vaginal ESwab (Copan Diagnostics, Brescia, Italy) sample for both culture (reference standard) and PCR assays for GBS. All samples were cultured immediately as described below at the Department of Clinical Microbiology, Lillebaelt Hospital, Vejle, Denmark, and the specimen tubes containing the vaginal intrapartum ESwab sample medium were, for practical reasons, subsequently frozen at -80°C for later PCR analyses. The PCR analyses were performed as direct tests without initial enrichment of the specimens in a culture broth in order to mimic a rapid on-site intrapartum PCR test. ## Culture of original specimens Direct plating was carried out by streaking the ESwab specimen on a selective Granada agar plate (BioMérieux, Spain). The Granada agar plates were incubated immediately after seeding in CO₂-enriched atmosphere at 35°C. The Granada agar plates were read after one and two days of incubation. All GBS-like colonies (identified by their orange color) were routinely confirmed as *Streptococcus agalactiae* (GBS). The colonies were identified using the Microflex LT MALDI-TOF system (Bruker Daltonik, Germany). One colony from GBS positive cultures was dispersed into 1 mL of broth medium supplemented with 10% glycerol (Statens Serum Institut, Denmark) and stored in a -80-degree freezer. ## BD MAX real-time GBS PCR The BD MAX System (Becton Dickinson, USA) automatically extracts the nucleic acid using a combination of heat, lytic enzymes and magnetic capture beads. The BD MAX GBS PCR assay (Cat. No. 441772) amplifies a section of the *cfb*-gene target sequence of the GBS chromosome. The assay has previously been evaluated against culture for GBS on clinical specimens from routine prenatal screening of women in USA. The BD MAX assay includes an Internal Process Control to monitor for the presence of potential inhibitory substances as well as system or reagent failures that may occur during the process. A sample volume of 300 μL was determined empirically by the laboratory of the Department of Clinical Microbiology as the optimal volume for GBS detection. The assay run takes approximately 120 minutes including reporting of results. Results were interpreted according to the manufacturer’s instructions. ## GenomEra GBS PCR assay The GenomEra CDX system (Abacus Diagnostica, Finland) is a molecular diagnostic analyzer consisting of an integrated thermal cycler and a time-resolved fluorometer. The GenomEra GBS PCR kit targets an internal region of the *cfb*-gene and, based on in-silico analysis of published GBS genomes and experimental data on a selection of GBS strains, is expected to detect all clinical GBS isolates (personal contact with Abacus Diagnostica, 2018). The GenomEra GBS PCR assay is clinically validated and CE-IVD-marked for use only with pre-enrichment broth culture of samples (GenomEra package insert). All the reagents required to perform the amplification and detection steps are readily contained in dry form in the Test Chips, including an Internal Amplification Control (IAC) of a non-naturally occurring DNA sequence to monitor for assay inhibition. The GenomEra GBS assay kit also includes Z-tubes containing zirconium particles for specimen dilution and cell disruption. In this study, we modified the manufacturer’s instructions as we applied direct swab samples instead of a 4-hour pre-enrichment broth culture of samples, and a sample volume of 60 μl of ESwab medium instead of 10 μl of enrichment culture medium. The volume was increased in order to achieve a sample volume more comparable to the one used in the BD MAX assay. Abacus Diagnostica recommended the change to 60 μl to maximize the sensitivity of the GenomEra assay and concomitantly ensure that the Test Chips were not overloaded, as overloading would produce invalid results. The 60 μl of ESwab medium was added to 1,000 μl of the GenomEra buffer supplied for swab samples. Samples were lysed by vortexing for five minutes. From this mixture, 35 μl was transferred to the Test Chip and analyzed on the GenomEra CDX system. The assay takes approximately 60 minutes including the final reporting of results. Results were interpreted according to the manufacturer’s instructions. ## Combined reference standard Discrepant PCR results on samples were examined by repeat testing of samples with both assays so the BD MAX and GenomEra assays were given equal opportunity. A combined reference standard was established by defining true positives as either culture positive samples or culture negative samples where one or more PCR test results (initial or repeat tests) with both the BD MAX GBS and the GenomEra GBS PCR assays were positive. ## Statistics StataCorp. 2015. Stata Statistical Software: Release 14. College Station, TX: StataCorp LP was used for the statistical analysis. # Results ## Performance of the BD MAX GBS and the GenomEra GBS assays with culture as reference Initial (first run of) PCR tests resulted in 102 positive samples with the BD MAX GBS and 84 with the GenomEra GBS assay. Seventy-seven (85%) of the culture- positive specimens were positive with BD MAX and 65 (71%) with GenomEra. Twenty- five of the 279 culture-negative samples were BD MAX GBS-positive, thus a specificity of about 91% was achieved. The corresponding figures for the GenomEra GBS assay were 19 PCR positives of the 279 culture negative samples, and a specificity of about 93%. Differences in sensitivity, specificity and predictive values between BD MAX and GenomEra with vaginal culture as the reference standard were not statistically significant. ## Discrepancy analysis Twenty of the total 370 specimens gave discordant results between BD MAX and GenomEra. Eight of these 20 specimens were culture-negative and 12 culture- positive. All 20 specimens were re-tested (second run) with both BD MAX and GenomEra. One sample was lost for follow-up due to limited sample volume. For several of the specimens, repeat testing yielded a different result than that obtained in the first round. Notably, seven of the eight culture-negative specimens were initially positive with BD MAX, while only one was positive with GenomEra. Four samples continued to be positive in repeat testing with BD MAX, while one specimen changed status from positive to negative with GenomEra. Six initially BD MAX-positive samples were negative in a repeat test with BD MAX, while six samples changed status from negative to positive with GenomEra. GBS isolates frozen from the 12 original culture-positive samples that were GenomEra-negative were re-cultured, re-identified as GBS by MALDI-TOF mass spectrometry, and tested using the GenomEra system to establish whether the original PCR-negative results were due to the lack of the target for the GenomEra PCR assay within the GBS strains cultured. All 12 samples tested positive with the GenomEra PCR assay, indicating that the original negative PCR results on the clinical samples were probably due to poor quality GBS DNA in the frozen clinical samples. ## Performance of the BD MAX GBS and the GenomEra GBS assays with a combined standard as reference Using the combined reference standard, 110 of the 370 samples in the study were classified as true positives. These comprised 91 culture-positive samples, 18 culture-negative samples where both PCR assays were initially positive, and one culture-negative sample, for which both PCR assays had one or more positive test results in initial or repeat assays, respectively. Distributions of results are shown in. shows the performance of cultures of GBS, BD MAX GBS and GenomEra GBS compared to the combined reference standard. Differences in sensitivity, specificity and predictive values between vaginal culture, BD MAX GBS and GenomEra GBS were not statistically significant. # Discussion This study was designed to compare the diagnostic accuracy of the BD MAX GBS assay and a new GenomEra GBS assay for rapid intrapartum PCR detection of vaginal carriage of group B streptococci using direct sample material from vaginal swabs. Based on a direct comparison with culture, BD MAX GBS had slightly better sensitivity, but lower specificity compared to GenomEra GBS, although the differences in performance were not statistically significant. Both PCR assays failed to detect GBS in all culture-positive vaginal samples, although they did detect GBS in several culture-negative specimens. Compared to a defined combined reference standard, there were no statistically significant differences between the respective performance characteristics of culture, BD MAX GBS PCR and GenomEra GBS PCR. The sensitivities of the three assays for detection of GBS in intrapartum vaginal ESwab samples were 83.0%, 87.3%, and 79.1%, respectively. Similar sensitivities were seen in a smaller study examining intrapartum GBS colonization status using GenomEra GBS PCR assay compared to culture. The strength of our study is that the vaginal swab samples used are randomly chosen from a large cohort of prospectively sampled women giving birth at Lillebaelt Hospital. Furthermore, we employed the GBS PCR assays without broth pre-enrichment steps prior to the PCR analyses in order to create a realistic screening scenario for intrapartum testing during labor, since enrichment cultures preclude the practical use of intrapartum PCR assays for detection of GBS in women giving birth. Discrepancies between the results of culture and the two PCR assays on the clinical sample material may indicate true differences in clinical sensitivity and specificity but could also be due to the limitations of the study design and differences in the PCR protocols. By employing a combined reference standard, three repeatedly BD MAX positive samples are regarded as false positives. On the other hand, six of twelve initially BD MAX positive and culture positive samples were not confirmed positive by a repeat PCR test. In contrast, seven of twelve initially GenomEra-negative and culture-positive samples were positive by the repeat test. One weakness of our study is that, for practical reasons, the two PCR assays were compared based on frozen sample material rather than concomitantly with the culture of fresh samples for GBS. This may explain the PCR-negative results for both the BD MAX and GenomEra assay on 14 culture- positive samples, a conjecture underpinned by the fact that the majority of culture-positive samples showed growth of only a few GBS colonies. This may have led to a low number of available targets and less scope for a positive result between the first and second tests of independently prepared samples of a specimen. Notably, the discrepancy analysis seems to indicate that both PCR assays were affected by freeze-thawing because some of the BD MAX samples that were initially positive, changed status to negative despite a high number of GBS-colony-forming units in the samples, based on the semi-quantitative culture results. It is also relevant to note that, in a recent clinical study using fresh vaginal-rectal swab specimens, the BD MAX GBS PCR assay detected 61 specimens with and 62 without enrichment culture, respectively, among 62 culture-positive specimens. Another weakness is that the GenomEra PCR assay was applied after an additional freeze-thaw cycle of the clinical samples. It could be expected that the sensitivity of the GenomEra GBS PCR assay is higher when the test is performed on fresh specimens. It could be argued that the volume of sample material used in the PCR assays is partial to the BD MAX GBS assay, especially in samples with low GBS DNA load or poor DNA integrity. Among positive culture samples with low GBS colony numbers, a positive PCR result may be less likely due to the limited sample volumes applied in PCR assays. BD MAX GBS PCR includes concentrating sample preparation with magnetic capture beads and thus all the original 300 μl of GBS sample material may end up in the amplification reaction. By way of contrast, in the GenomEra assay, the original sample of 60 μl is diluted in 1,000 μl of buffer, and 35 μl of this dilution is then used for PCR. The volume of sample material used in BD MAX may be higher by order of magnitude, even if the sample volume used with the GenomEra assay is increased from 10 to 60 μl. However, before the clinical study, the utilization of frozen specimens and an altered volume was not validated on clinical samples material or reference strains of GBS. The direct and rapid use of BD MAX GBS is clearly a limitation because 3.4% of all specimens were initially undetermined due to either inhibition of the amplification of the Internal Process Control or technical failure of the PCR test. For specimens tested by GenomEra, the corresponding figure was only 2.0%. Omitting LIM broth enrichment of the specimen before a PCR assay for vaginal GBS detection implies a small but statistically significant reduction in sensitivity (92.7% versus 99.1%), compared to the use of the same PCR test with prior LIM broth inoculation. However, an LIM broth enrichment step 18 hours prior to the PCR assay precludes the use of a rapid on-site GBS PCR test during birth. Screening women for GBS during pregnancy at week 35–37 is known to fail to identify a large number of women with intrapartum carriage of GBS. Thus, direct, rapid intrapartum GBS PCR screening seems preferable to antepartum GBS screening by culture or PCR with prior LIM broth enrichment of the specimens. Several studies have confirmed that intrapartum PCR tests detect GBS more accurately than conventional GBS culture methods. Screening of both vaginal and rectal specimens for GBS increases the yield of GBS-positive antepartum culture samples and the predictive values for intrapartum vaginal colonization status. Simultaneous intrapartum sampling of vagina and rectum for a rapid PCR assay could increase the yield of detection at birth, as the gastrointestinal tract is the natural reservoir for GBS. This may also increase the number of women given intrapartum prophylactic antibiotic treatment. However, the added clinical benefit of intrapartum and rectovaginal detection of GBS, compared to vaginal detection of GBS, to prevent EOGBS, is not yet clarified. Based on the combined reference standard, the 14 PCR-negative but culture- positive samples appear counterbalanced by a higher number of PCR-positive samples from both PCR assays for the 19 culture-negative samples. This indicates that, for the detection of GBS in vaginal samples, there is no significant difference in sensitivity and specificity between PCR tests and culture. However, none of the three assays employed detected all GBS colonized women in this study. Run time is critical because a rapid test result is needed in order to make a fast decision whether or not to administer prophylactic antibiotics before delivery in a childbirth unit. The difference in run time between the BD MAX and GenomEra PCR assays is only one hour (120 minutes vs. 60 minutes, respectively). Notably, the BD GeneOhm StrepB assay is a rapid and sensitive PCR assay (\~ 1 hour). However, for optimal performance, the BD system still requires specially trained laboratory staff. With thorough training and maintenance of midwives’ qualifications, the GenomEra system is less complicated and more readily applicable as a point-of-care test in a clinical department than the BD MAX system. In few cases the intrapartum time can be less than 60 minutes and therefore administering prophylactic antibiotics to prevent infection has to rely on other information, e.g. risk factors. A future prospective study evaluating the capabilities of GenomEra GBS PCR on intrapartum fresh specimens performed on-site in a maternity ward is desirable in comparison with culture and/or another rapid and easy-to-use PCR assay. # Conclusion Both PCR assays performed comparably to culture of the intrapartum vaginal samples. In particular, the GenomEra GBS PCR assay is potentially as an easy and rapid on-site test for intrapartum detection of vaginal carriage of GBS at a maternity ward to identify women in labor who should receive intrapartum antibiotic prophylaxis. # Supporting information Our thanks to the staff at the Departments of Gynecology and Obstetrics, and Clinical Microbiology at Lillebaelt Hospital for collecting and testing specimens. [^1]: The authors have declared that no competing interests exist.

# 1 Introduction The caesarean section (CS) rate has been increasing during the last decades and the rate of CS varies among hospitals. In Switzerland CS rates reach 33% and more. As many different classifications exist, the heterogeneity prevents valid comparisons between countries and hospitals. A lack of clarity regarding indication and relevant obstetric history can be observed. A commonly accepted classification of CS and its indications would allow an evaluation and comparison of the contributors to the CS rate and would make comparison between hospitals, regions, and countries possible. The Robson classification of CS shows the CS rates in specific groups, see and, to help identifying possible reasons for this variation. The Robson classification is recommended by the World Health Organization (WHO). It is based on pre labour, intrapartum and postpartum data. Currently it is not in use in Switzerland although it has been highly recommended. As multiple births as well as elective CS numbers are rising in Switzerland and worldwide, the necessity of a valid benchmark of comparable groups, meaningful data and outcome measures becomes evident. Meta analyses support the interest in the Robson classification. The possibility to derive information from data is developing very fast, as routinely collected administrative and health data have accumulated during the last years and recent technology grants a higher degree of accessibility. The Insel Gruppe Berne with approximately 62000 inpatient stays and more than 2200 annual births at the department of obstetrics and gynecology serves as a tertiary care center being obliged to treat high-risk patients (multiple pregnancies, preterm deliveries and repeat CS). To overcome different national specifications in statistics and lacking classifications for valid benchmark we conducted this study by evaluating different methods of classification and grouping with regard to costs and outcome parameters. The technical approach offered the opportunity to develop a proof-of-concept for a Structured Query Language (SQL) query based medical classification process using routinely collected health data. The novel approaches of this study are: i by using the technical capabilities and the amount of data from the clinical data warehouse extracting data with a minimum of requests (this study provides one of the first complex queries to the data), ii evaluating a medical classification by routinely collected health data and iii searching missing information by a novel text mining tool, which was tested (searching the electronic health record) and iv mapping the mother’s Robson class to the corresponding newborn. The study was conducted to elaborate a proof of concept of a complex medical classification which is based on routinely collected health data and which can be applied automatically. Moreover, it should be demonstrated that outcome related classes can be used for a standardized benchmarking. We hypothesize i that routinely collected health data can be used to apply complex medical classifications and ii that the Robson classification is highly capable of classifying mothers and their corresponding newborn into meaningful groups with regard to outcome. # 2 Methods The study was conducted at the coding department and the department of obstetrics and gynecology Inselspital, University Hospital of Berne, Switzerland. The clinical data warehouse at the Inselspital contains administrative and medical data of all patients from the department of obstetrics and gynecology and the neonatology division. The data include the diagnoses codes (International Statistical Classification of Diseases and Related Health Problems 10th version, ICD) and procedure codes (Swiss classification of procedures, “CHOP”) of inpatient cases and clinical data as the APGAR values (outcome related score to assess the neonate’s status) or laboratory results. Inclusion criteria: Inpatient cases at the Inselspital Berne, discharges from 2014–2017 (224’331); all cases with a procedure code for caesarean section as procedure encoded (2’700), see. Filtering of the datasets was performed to make sure, that only classifiable and classified individuals remained in the data. The presence of a ‘null’–class of a relevant value, applied to individuals without the required information for classification, led to a removal of 3 entries of mother cases. After extracting the inpatient cases with CS, the corresponding newborns were mapped by a linkage code using the case identity number. Thus, it was possible to assign the mother’s Robson class to the corresponding neonate(s). Stillborn cases were excluded, as their diagnoses are not coded due to Swiss coding regulations. After data extraction and before analysis, the cases were anonymized (2’697 mother cases, 3’086 newborn cases), (total of cases see). Data were extracted from the clinical data warehouse and mapped. Outcome variables (referring to the clinical situation) were defined referring to literature (distinct ICD codes, intensive care treatment, ventilation hours, transfusion, 5-minute-APGAR score, Base Excess and pH value). An algorithm (SQL query) to apply the Robson classes using claims data (e.g. procedure codes, ICD codes, Diagnosis related groups (DRGs), costs) and otherwise routinely collected health data (laboratory, text, APGAR scores) was set up. The case related cost data were obtained from the REKOLE^®- based cost-unit accounts, according to national standard. SwissDRG DRG type (related to length of stay, current Swiss inpatient reimbursement system of diagnosis related groups) and case related costs were used as surrogates of economic outcome variables in addition to the clinical outcome variables. The method of outlier calculation was executed according to the SwissDRG’s annually revised standard (batch grouper). The Robson classes were applied to both mother and child. Descriptive statistics for log10 case related costs and several clinical indicators (defined by e.g. ICD code, intensive care, ventilation, interventions/procedures) was conducted. The algorithm to propose Robson classes to the mothers’ cases was programmed in Transact-SQL, querying the necessary patient and case information and creating entries which represent the Robson class. The Ethics Committee of the Canton Bern approved the study (KEK-Nr. Req-2017-00927) for quality assurance purpose. No informed consent was necessary. According to the regulations of the Bernese Ethics Committee no data combining a set of diagnoses and laboratory values on patient level (potentially identifying or sensitive patient information) can be labeled as fully anonymized. Supporting data, which can be given unrestricted access to, can be found in the supporting information file (contact information: <https://www.gef.be.ch/gef/de/index/direktion/organisation/kek/kontakt.html>). Data sharing is partly restricted as the original dataset contains de- identifying sets of coded diagnoses on patient level. Further data requests can be send to Dominique Furrer (<dominique.furrer@insel.ch>), local data protection manager of the institutional data access of the Insel Data Science Center, University Hospital of Bern, Berne, CH. # 3 Results The Robson classification demonstrated to be a highly usable method to aggregate relevant obstetric information corresponding to clinical indicators for both mother and child. The manual revision of 100 cases showed a high validity of the method, see. The method itself successfully produced an application of the Robson criteria to the data of cases with CS. Furthermore, benchmarking cases of mother and child with cephalic on term pregnancies for outcome and complications was made possible by associating outcome information. The distribution of clinical indicators, ICD codes, intensive care unit treatment, APGAR score and DRG type SwissDRG (see Figs –) could be successfully mapped to Robson classes of cases with CS and the corresponding newborn. Case related costs could be mapped to the distinct classes and a comparison of the distribution of costs and relevant differences between DRGs and Robson classes could be conducted (see, and Tables, and Figs). The results showed a relevant contribution to cost based grouping only in a few Robson groups. Analyzing the DRG type (length of stay) of mother cases per Robson class it could be demonstrated that the distribution of inlier and high outliers differs in the Robson classes: in class 10, 5, 3 and 1 with inliers distributed to Robson class 10 in 18.97% of cases, class 5 in 18.07%, class 3 in 7.98% and class 1 in 18.72% of the cases respectively, see. High outliers were distributed as follows: 46.21% to Robson class 10, 3.03% to class 5, 0.76% to class 3, and 5.03% to class 1 (see Figs). # 4 Discussion and conclusion With rising numbers in CS it is essential to improve the data base for benchmarking by implementing valid classifications to work with, especially for elective CS and births at term with cephalic lies. We successfully demonstrated, that a query on routinely collected health data could serve for a complex medical classification. Even with variables from several sources including text mining for missing information it was possible to achieve a complete database for classification (missing values in only 3 cases), a method for benchmarking and monitoring of quality and outcome. With a few Robson groups still showing unexpected results concerning the distribution of cases compared to other studies, a thorough validation of the programmed algorithm referring to international standards and the coding rules will be planned for a future study. This must include an analysis of cases with spontaneous delivery. Having conducted these further studies, it will be possible to compare case related costs of the Robson classes to SwissDRGs in order to compare the consistency of the groups. However, as the Robson classification is meant to benchmark quality and outcome it might not be capable of contributing to recovery of costs in DRG systems. Risk-stratification might be conducted using this data set and should be the next step in order to evaluate the Robson criteria and outcome indicators. The analysis of outcome data might influence quality monitoring, benchmarking. Applying the method to the group of spontaneous deliveries could be helpful, in order to benchmark caesarean rates. The method could be easily adopted to the data of spontaneous births’ cases by (change of selection criteria). ## 4.1 Strengths The routinely collected dataset showed a very good quality when referring to completeness and consistency. The connection of laboratory, medical, financial and administrative data on the individual patient’s level with regard to mother and the corresponding child demonstrates a novel approach. The method itself can be universally adapted because an international catalogue of diagnosis codes (ICD) and standardized variables as gestational age are used for the query. ## 4.2 Weaknesses As national coding rules differ and coding rules in Switzerland underwent changes during the last years, the numbers per Robson class show a limited comparability to those of recent studies extracting the relevant information for classification manually from the patients’ health record. This has to be taken into account when implementing the method using ICD coding information. ## 4.3 Limitations The verification of the method is based on data and patient groups from one single hospital in the context of Swiss national coding regulations. Moreover, as the study served as a technical proof of concept, only the patient group of CS was analyzed. Therefore, the study is limited to conclusions concerning the distribution of Robson classes within the group of CS cases. No conclusions can be drawn concerning the distribution of CS regarding all births’ modes. Risk- stratification and differences to coding approaches elsewhere have to be elaborated before using this method for international benchmarking. ## 4.4 Conclusion It is possible to set up an automated method to categorize patient groups according to complex medical classifications based on routinely collected health data. This study might enhance the discussion to adopt the automated classification on routinely collected health data in Switzerland and elsewhere for benchmarking, as outcome variables could be successfully associated to the specific Robson classes and the classification of cases can be conducted efficiently. The method proofed to be capable of applying outcome variables to the complex classification. # Supporting information Reto Baumgartner, SAP Team, Insel Gruppe, Bern, for helping to extract the data and technical support; Roger Mathys and Martin Meister, Insel Data Science Center IDSC, Inselspital, University Hospital of Bern, for programming the code. [^1]: The authors have declared that no competing interests exist.

# Introduction Covalent modification of proteins by SUMO is an essential process in mammals, as evidenced by the embryonic lethality of mouse mutants lacking the SUMO conjugating enzyme, Ubc9. SUMO modification regulates numerous nuclear processes, including transcription, DNA replication and repair, and chromosome segregation; and also plays a role in the transport of proteins through the nuclear pore. SUMO is a member of the larger family of ubiquitin like proteins, which shares about 18% sequence identity to ubiquitin, and is structurally quite similar,. Like ubiquitin, SUMO is covalently attached to lysine residues within the target protein, although in the majority of cases SUMO is attached to a lysine within the ψKxE consensus site (where ψ is hydrophobic and x is any residue). In mammals there are four SUMO isoforms, including SUMO1, which most closely resembles the single yeast Smt3. SUMO2 and -3, which are very similar to each other, contain an internal sumoylation consensus site and more readily form poly-SUMO chains. SUMO4, variants of which have been linked to diabetes, is more similar to SUMO2 and 3 than SUMO1. Attachment of SUMO to lysine residues within a target protein is mediated by a conserved enzymatic pathway. SUMO is first cleaved at a di-glycine motif close to the carboxyl-terminus, to generate the mature form of the protein. SUMO processing is carried out by members of a family of SUMO proteases, termed SENPs (for SUMO/sentrin specific peptidase), that can also catalyze the removal of SUMO from lysines within target proteins. The processed SUMO is transferred from the heterodimeric E1 enzyme to Ubc9, which is the sole SUMO E2 conjugating enzyme. The loading of Ubc9 with SUMO is the only energy requiring step in the modification pathway, and results in the attachment of SUMO to the catalytic cysteine of Ubc9 via a thioester linkage. SUMO-loaded Ubc9 can modify substrate proteins directly, resulting in the covalent attachment of SUMO to the acceptor lysine. Although there is no absolute requirement for an E3, a number of SUMO E3s have been identified, including members of the PIAS (protein inhibitor of activated STAT) family, the polycomb protein, Pc2/Cbx4, and RanBP2/Nup358. SUMO E3s interact with both E2 and substrate and promote the transfer of SUMO from Ubc9 to the target lysine in the substrate. In addition to the covalent attachment of SUMO to lysine residues in target proteins, recent work has identified specific motifs that mediate non-covalent interactions with SUMO. In general these motifs consist of a hydrophobic core, which is often flanked by acidic residues. The best characterized of the SUMO interaction motifs (SIMs) have the consensus sequence, V/I-x-V/I-V/I or V/I-V/I-x-V/I/L, where position two or three can be any amino acid. Structural studies have demonstrated that these hydrophobic SIMs bind to the second β-strand and the first α helix of SUMO. SIMs can bind to SUMO in two opposite orientations, and the presence of acidic residues either upstream or downstream of the hydrophobic core has been suggested to contribute to the orientation of binding by forming interactions with basic residues in SUMO. With increased attention focused on the non-covalent interaction of SUMO with SIM containing proteins, it has become clear that there is also some flexibility in the precise sequence of the hydrophobic core of the SIM. For example, the corepressor CoREST1 binds to SUMO2, but not SUMO1, via a SIM, with a five amino acid core in which positions 1, 3 and 5 are hydrophobic residues. Additionally, it has been shown that CK2-mediated phosphorylation of serine residues surrounding the hydrophobic core of the PIAS1 SIM can increase SUMO binding. Clearly, the somewhat variable nature of the consensus SIM, combined with the fact that such a short amino acid sequence will be found by chance quite frequently, emphasizes the importance of testing the functionality of potential SIMs. Interestingly, consensus SIMs are found in several components of the sumoylation machinery, including a subunit of the E1 heterodimer, members of the PIAS E3 family, RanBP2/Nup358, and Pc2. Additionally, SIMs are found in some SUMO substrates. This clearly raises the possibility that components of the modification pathway interact non-covalently with SUMO to facilitate its transfer to substrates. In support of this, the SIM in RanBP2/Nup358 is directly adjacent to the minimal IR1-IR2 domain that has E3 activity. However, although this SIM has been shown to bind to SUMO, it appears not to be essential for E3 activity, at least *in vitro*. Interestingly, analysis of the SIM in PIAS1 demonstrated that it can indeed bind to both SUMO1 and SUMO2, and that the SIM is required for the transcriptional regulatory activity of PIAS1. However, the PIAS1 SIM has not been shown to play any role in the ability of PIAS1 to act as a SUMO E3. For some substrates, including BLM, USP25 and Daxx, it appears that the presence of a SIM within the substrate itself can increase sumoylation by facilitating the recruitment of SUMO-loaded Ubc9, via non-covalent interaction of the Ubc9-bound SUMO with the SIM. This promotes binding of Ubc9 to the target modification site in the substrate, allowing for transfer of the SUMO to the substrate. In addition to playing possible roles in SUMO modification, SIMs may also act as protein interaction modules which facilitate the formation of multi- protein complexes. The importance of sumoylation for the maintenance of PML bodies has been known for some years. However, it has now been shown that the presence of a SIM in the PML protein is also required for the recruitment of sumoylated proteins to PML bodies, and for PML body formation. Other proteins, such as the Ring finger ubiquitin E3, RNF4, have multiple SIMs which appear to facilitate the binding of poly-sumoylated proteins. The binding of RNF4 to sumoylated proteins, including PML, allows for the RNF4 mediated ubiquitylation of the sumoylated substrate protein. The polycomb protein, Pc2/Cbx4, is a component of the PRC1 (polycomb repressive complex) complex, and was first identified based on similarity with the *Drosophila* Pc protein. In flies there is a single Pc, whereas mammals have five Cbx paralogs, in addition to the HP1 proteins. Outside the conserved amino- terminal chromodomains there is relatively little sequence similarity between *Drosophila* Pc and mammalian Cbx proteins, except for a short hydrophobic region at the carboxyl-termini of all but the HP1 proteins. We have shown that Pc2 is a SUMO E3 for the transcriptional corepressors CtBP1 and CtBP2, and several other proteins have been identified as substrates for Pc2 E3 activity, including the kinase, HIPK2, the transcriptional regulator, SIP1, a DNA methyltransferase and the CTCF insulator protein. Structure function analyses have revealed at least two domains that contribute to Pc2 E3 activity, and it is intriguing to note that each of these two sub-domains contains a single consensus SIM. However, the functional relevance of the SIMs in Pc2 has not been tested directly. *In vitro* Pc2 appears to have quite weak E3 activity, whereas its E3 activity in transfected cells, at least for CtBP, is quite robust. We, therefore, decided to analyze the *in vivo* contributions of the Pc2 SIMs to its E3 activity. Here we show that both SIMs in Pc2 contribute to non-covalent SUMO binding and are required for full E3 activity, as well as for covalent modification of Pc2 itself by SUMO. Analysis of mutant forms of both Pc2 and SUMO1 or SUMO2 demonstrates that recruitment of SUMO to Pc2 containing sub-nuclear foci requires non-covalent SUMO-SIM interactions. This work provides clear evidence that SUMO-SIM interactions are required for the E3 activity of Pc2. # Results ## Pc2 Contains Two SIMs Inspection of the amino acid sequence of Pc2 reveals the presence of two conserved SIMs. One of these (SIM2) is in the carboxyl-terminal region of Pc2, very close to the CtBP interaction motif and the primary sumoylation site on Pc2. Comparison of this region of Pc2 with homologs from a number of vertebrate species reveals that it is conserved in mammals, birds and frogs, but is not present in Pc2 homologs from two fishes. In contrast, the adjacent CtBP binding motif (PxDL\[R/S/T\]) is present in all of the species shown in. The more amino terminal SIM (SIM1) is again in a region of relatively high sequence conservation ( and), and is present in all vertebrate species examined. These two SIMs consist of a hydrophobic core, with the more carboxyl terminal one in the reverse orientation, such that position 3 is a non-hydrophobic residue. Examination of the sequences surrounding the two Pc2 SIMs reveals that SIM2 has several acidic residues in close proximity to it, as well as a conserved serine that is within a consensus CK2 phosphorylation site. SIMs in several other proteins share this property of being surrounded by acidic and potentially phosphorylatable residues. Interestingly, the more amino terminal SIM1 is in a very highly conserved block of sequence, but this almost completely lacks acidic residues. To confirm that the consensus SIMs in Pc2 could indeed mediate SUMO binding, we incubated bacterially expressed fusion proteins encoding amino acids 250–560 of Pc2 with GST fusions to SUMO1 or SUMO2, which had been purified form bacteria and immobilized on glutathione agarose. As shown in, the wild type Pc2 fusion bound to both SUMO1 and SUMO2, whereas deletion of either SIM1 or SIM2 alone (or both together) clearly decreased the interaction with both SUMO1 and SUMO2. Thus, it appears that Pc2 is capable of interacting directly with SUMO1 and SUMO2, and that the consensus SIMs in Pc2 are required for this interaction. Based on the apparent difference from other SIMs, and the fact that SIM1 is in a region of Pc2 which we have previously shown to have E3 activity, we decided to analyze this SIM functionally. As a first test of whether SIM1 is important for Pc2 E3 function, we created two mutant forms of Pc2 in which two lysines or three alanines had been introduced into the IVIV of SIM1. COS1 cells were cotransfected with T7-tagged CtBP and six histidine-tagged SUMO1 (H6-SUMO1), together with wild type or mutant Pc2, or a control vector. Cells were then lysed and subjected to direct western blotting for T7-CtBP and the sumoylated form of the protein. As shown in, coexpression of wild type Pc2 increased the amount of sumoylated CtBP compared to cells transfected with CtBP and SUMO1 alone, whereas, this increase was not seen with either of the mutant Pc2 constructs. To verify that these two mutants were otherwise functional, we coexpressed them together with T7-CtBP and precipitated proteins on anti-Flag agarose. T7-CtBP coprecipitated to a similar degree with Flag-tagged wild type Pc2 and each of the mutants, whereas no CtBP was visible in a control precipitate. These data suggest that the Pc2 SIM1 is important for Pc2 E3 activity *in vivo*. ## Identification of a Minimal Functional Domain Surrounding SIM1 To simplify analysis of SIM1, we created a fusion protein in which the amino- terminal 290 amino acids of Pc2 were fused to a carboxyl-terminal consensus sumoylation motif (VKPE), followed by a stop codon, and an amino-terminal Flag epitope tag. To verify that this construct \[(2-290)-VKPE\] could indeed be sumoylated in cells, we coexpressed it in COS1 cells with H6-SUMO1. Cells were then lysed in 6M guanidine HCl, and histidine tagged proteins isolated on cobalt agarose. As shown in, when the cobalt bound fraction was western blotted with a Flag antibody sumoylated forms of the VKPE fusion were clearly visible, but were not seen with a similar construct lacking the VKPE peptide. It has previously been shown that addition of a sumoylation consensus and a nuclear localization sequence (NLS) is enough to drive some sumoylation of a reporter protein. We therefore compared the level of sumoylation of such constructs to Pc2(2-290)-VKPE, to begin to test whether its sumoylation was due to more than simply driving nuclear localization of a sumoylation consensus site. As shown in, sumoylation of Myc-epitope tagged pyruvate kinase fused to two peptides from IκBα was readily detected by cobalt precipitation and Myc western blot. However, the sumoylated form of this reporter protein represented only a very small fraction of the total protein, when analyzed by direct western blotting of cell lysates. In contrast sumoylated Pc2(2-290)-VKPE constituted a much larger proportion of the total (compare the lower panels). Thus the Pc2(2-290)-VKPE fusion is becoming sumoylated to a greater degree than would be expected for a protein that simply contains an NLS and a sumoylation consensus site, suggesting the presence of an additional activity within this fusion protein. We next tested a series of deletion constructs within the context of the VKPE fusion. A carboxyl-terminal truncation to amino acid 240, which removes SIM1, greatly reduced modification of the VKPE fusion. In contrast, amino-terminal deletions either to amino acid 68 or 170 actually increased the proportion of sumoylated protein, and modification by endogenous SUMO was readily apparent. To test the importance of SIM1 in this context, we analyzed two versions of the (2-290)-VKPE fusion, in which the SIM1 had either been mutated to KVIK or deleted. As shown in, both of these mutations essentially abolished modification of this construct. To verify that the sumoylation we were seeing was occurring within the VKPE motif, we also tested a wild type fusion to a VRPE peptide. No sumoylation of this construct \[(2-290)-VRPE\] was observed. To test whether the region of sequence conservation surrounding SIM1 was sufficient to promote sumoylation we tested a construct which contained only 55 amino acids from Pc2 \[(236-290)-VKPE\]. As shown in, this construct was robustly sumoylated, whereas a version in which the SIM had been deleted, or a fusion to a VRPE peptide were unmodified. Together, these data suggest that the conserved region surrounding SIM1 is sufficient to promote modification of a linked consensus sumoylation site, and that the SIM is essential for this activity. ## Mutational Analysis Around SIM1 The core of the 55 amino acids surrounding SIM1 are highly conserved across diverse vertebrate species. To identify other amino acids which might be important for SUMO binding, we performed alanine scanning mutagenesis across this region, within the context of the Flag-Pc2(2-290)-VKPE fusion protein. Eight double alanine mutants were tested, in which we primarily focused on charged and hydrophobic residues. Seven of the eight were expressed at similar levels to the wild type protein, and are shown in. We next tested each of these seven mutants for modification by SUMO1. As shown in, alteration of lysines 278 and 280 to alanine effectively abolished the activity of this construct, whereas mutation of lysines amino terminal to the SIM had little or no effect (compare lane 17 with lanes 2, 4 and 8). Mutation of two amino-terminal hydrophobic residues (IV 252/253) also completely abolished activity, whereas the VI 277/279 mutation had a more minimal effect (lanes 6 and 15). Additionally, mutation of two asparagines (255/257) as well as the EN 271/272 mutation somewhat reduced activity (lanes 11 and 13). A similar set of analyses was also performed using SUMO2 in place of SUMO1, with similar results. This analysis demonstrates that several conserved amino acids surrounding the SIM1 are essential for promoting sumoylation in the context of these fusion proteins, and suggests that they may play a role in SUMO binding. To begin to compare the importance of the two SIMs in Pc2 for E3 activity, we created a series of expression constructs in which the IVIV of SIM1 and the VILL of SIM2 had been deleted. Single and double SIM deletions were generated in the context of full length Pc2, or a truncated form (encoding amino-acids 2-531) that lacks the carboxyl-terminal 29 amino acids (the C-box), and is delocalized from polycomb bodies. Additionally, we transferred the two most severe alanine mutations (IV 252/253 and KK 278/280) to both full length Pc2, and the Pc2 deletion lacking the C-box. We first verified that SIM deletions did not affect interaction with CtBP by coimmunoprecipitation from transfected COS1 cells. As shown in, SIM deletions either alone or in combination had no affect on CtBP interaction, whereas deletion of the PIDLR CtBP interaction motif significantly weakened CtBP binding. We next tested the effect of deleting the SIMs on interaction with a non-conjugatable version of SUMO3 (GFP-nc-SUMO3) in transfected COS1 cells. Flag-tagged Pc2 proteins were collected on anti-Flag agarose, and probed for the presence of co-precipitating GFP-nc-SUMO3. As shown in, deletion of either SIM1 or SIM2 reduced the interaction of Pc2 with nc- SUMO3. In contrast, alteration of the primary sumoylated lysine in Pc2 to arginine (K494R) had little effect. Similar experiments using the IV 252/253 and KK 278/280 mutant forms of full length Pc2 also revealed a decrease in the interaction of these Pc2 mutants with nc-SUMO3. To further compare SUMO binding by SIM mutant forms of Pc2 we tested interaction of GST-SUMO1 or GST-SUMO2, purified from bacteria, with Flag-tagged Pc2 expressed in COS1 cells. As shown in, Pc2 and the SIM1 deletion mutant bound to GST-SUMO1, whereas binding of the SIM2 or double SIM mutant was clearly reduced. As with the results from co- transfection of Pc2 with nc-SUMO3, both single SIM mutants reduced binding to GST-SUMO2. Together, this data suggests that both SIMs in Pc2 can contribute to non-covalent SUMO binding, but that SIM1 plays less of a role with SUMO1. ## An *In Vivo* Role for SIM1 and SIM2 in Pc2 E3 Activity We next analyzed the effects of SIM mutations on SUMO modification of Pc2 and CtBP. Each of the Pc2 SIM deletion constructs was coexpressed in COS1 cells with SUMO1, and analyzed by direct western blotting for the Flag tag on Pc2. Mutation of SIM1 in the context of full length Pc2 had little effect on Pc2 sumoylation, but completely abolished it in the context of the delocalized Pc2(2-531) (compare lanes 2, 4 in each panel). In contrast, mutation of SIM2 alone or in combination with SIM1 completely abolished Pc2 sumoylation in either context. To test modification of a recruited substrate protein, we performed a similar set of analyses in the presence of coexpressed CtBP. As shown in, mutation of SIM2 alone reduced CtBP sumoylation, whereas deletion of SIM1 had little effect. However, deletion of both SIMs in combination, greatly reduced E3 activity towards CtBP. In the context of the delocalized Pc2(2-531) truncation mutant, deletion of either SIM alone abolished E3 activity towards CtBP. Thus it appears that both SIMs contribute to full E3 activity towards CtBP, whereas there is a lesser requirement for SIM1 for auto-sumoylation of Pc2. Analysis of the two alanine mutants (IV 252/253 and KK 278/280) in the context of full length Pc2 revealed no discernible effect on sumoylation of either Pc2 or CtBP (data not shown). However, in the context of the de-localized Pc2(2-531) construct, modification by SUMO1 of both Pc2 and CtBP was reduced by these point mutations. Since the alanine mutations surrounding SIM1 appeared to affect modification by both SUMO1 and SUMO2, in the context of the amino-terminal Pc2 fusions, we next compared modification of CtBP and full length Pc2 by SUMO1 and SUMO2 in parallel. COS1 cells were transfected with CtBP and Pc2 or one of the SIM deletion mutants, together with either SUMO1 or SUMO2. As shown in, deletion of SIM1 did not reduce sumoylation of Pc2 and in this assay only slightly reduced CtBP modification by SUMO1, whereas deletion of SIM2 alone or in combination with SIM1 dramatically reduced SUMO1 modification of both proteins. Interestingly, with SUMO2 deletion of SIM1 alone clearly reduced modification of both Pc2 and CtBP. Together, these data suggest that SIM2 is required for Pc2 E3 activity with either SUMO1 or SUMO2, whereas, in the context of full length Pc2, SIM1 appears to contribute primarily to E3 activity with SUMO2. ## SIM Deletions Delocalize SUMO from Pc2 Foci Pc2 can recruit CtBP and SUMO1 or SUMO2 to sub-nuclear foci, termed polycomb bodies. This provides a useful assay to examine the requirements for assembly of the Pc2-containing sumoylation complex, and subsequent substrate sumoylation, in living cells. To test the effects of SIM mutations on the recruitment of SUMO1 and SUMO2, we created eYFP-tagged versions of the single and double SIM deletion mutants of Pc2. COS1 cells were transfected with individual eYFP-Pc2 constructs, imaged at 22 hours after transfection, and the distribution of Pc2 fluorescence scored as soon as a significant number of transfected cells were readily visible. By analyzing the cells only 22 hours post-transfection we hoped to avoid problems due to high level over-expression of the transfected proteins. The wild-type and each of the three mutant forms of Pc2 all formed sub-nuclear foci, in the majority of cells, although deletion of SIM1 resulted in an increase in the proportion of cells with foci that were larger than those seen with wild type Pc2 (and data not shown). We did not observe an increase for any of the mutants in the proportion of cells with a diffuse nuclear pattern of Pc2 localization, as seen with the C-box deletion. To test the ability of these mutants to recruit other proteins, we coexpressed the eYFP-tagged Pc2 constructs together with eCFP-tagged CtBP. Cells with Pc2 foci were then selected and the presence of eCFP foci scored, to determine what proportion of cells with typical Pc2 localization had co-localization of the interacting partner. Deletion of either SIM alone, or both together had minimal effect on CtBP recruitment to Pc2 foci, whether they were the large or small foci. Thus it appears that all three SIM mutant forms of Pc2 localize normally and can recruit a partner protein. We next created fusions of SUMO1 and SUMO2 to the monomeric Cherry fluorescent protein (referred to here as mC-fusions). Additionally we created non-conjugatable mutants (ΔGG mutants) of SUMO1 and -2, in which the di-glycine motif that is required for removal of the carboxyl- terminal tail had been altered to two alanines. As shown in, the ΔGG mutants of both SUMO1 and SUMO2 were unable to be attached to either CtBP or Pc2 when expressed in COS1 cells. To test whether deletion of the SIMs in Pc2 affected localization of SUMO1 or SUMO2 to polycomb foci, we coexpressed wild type or SIM mutant eYFP-tagged Pc2 together with mC-SUMO fusions in COS1 cells. Cells were examined as for CtBP co-localization and scored for colocalization of the SUMO fusion protein with Pc2. mC-SUMO1 co-localized with Pc2 in almost 80% of cells with Pc2 foci, whereas for SUMO2, the degree of colocalization was less than 60%. Examples of the different kinds of Pc2 and SUMO localization observed are shown in. For both SUMO1 and SUMO2, deletion of SIM1 had little effect on the proportion of cells with clear colocalization, although with SUMO2 it did drop to below 40%. Deletion of either SIM2 alone, or in combination with SIM1 dramatically reduced colocalization of either SUMO1 or SUMO2, consistent with the more important role for SIM2 in E3 activity. We next tested whether the ΔGG SUMO mutants were able to colocalize with Pc2. For SUMO1, the proportion of cells in which we observed colocalization decreased to less than 30% with the ΔGG mutant, and for SUMO2 colocalization was effectively abolished. Together, these data suggest that the observed localization of SUMO to Pc2 foci is SIM dependent, and requires the SUMO to be competent for processing and substrate modification. ## SIM Interaction Mutant SUMOs Are Not Competent for Pc2 E3 Activity The residues within SUMO, which contribute to non-covalent interaction with SIMs, have been identified. Recent work demonstrated that for certain substrates, the presence of a SIM was required for efficient modification, and that mutation of amino acids 30, 31 and 33 of SUMO2 to alanines abolished SIM- dependent modification. To test the requirement of SUMO-SIM interaction for Pc2 E3 activity, we generated a similar triple alanine mutant form of SUMO2 (QFI mutant, see), as well as a double point mutant form of SUMO1, in which phenyl alanine 36 and valine 38 of SUMO1 were converted to alanines (SUMO1 FV mutant, see). Each of these mutants was created in the context of a processed SUMO, which ended with the di-glycine motif at its carboxyl-terminus, and either a six-histidine tag or monomeric Cherry fusion at the amino-terminus. Colocalization of the mC fusions of these SUMO mutants with eYFP-Pc2 was analyzed as before. Coexpression of mC-SUMO2(QFI) with eYFP-Pc2 resulted in essentially no colocalization of the mutant SUMO2 with Pc2. For the SUMO1(FV) mutant, colocalization was observed in less than 10% of cells, below the level seen with the double SIM mutant Pc2 and wild type SUMO1. We next analyzed the FV and QFI mutants for their ability to be covalently attached to both Pc2 and CtBP in transfected COS1 cells. As shown in, both the SUMO1(FV) and SUMO2(QFI) mutants were severely impaired in their ability to be attached to Pc2. However, when we also over-expressed Ubc9 in these cells this was able to drive the modification of Pc2 by both of the SUMO mutants, to a level similar to that seen with the wild type proteins, suggesting that these SUMO mutants are functional for substrate modification. We also compared modification of CtBP by wild type and SIM-interaction mutant SUMOs in the presence of co-expressed Ubc9 and Pc2. When SUMO1 or the FV mutant form were co-expressed with Ubc9, similar levels of CtBP modification were observed. In the presence of co-expressed Pc2, SUMO1 modification of CtBP was clearly visible, whereas with the FV mutant no modified CtBP was observed. Similar results were obtained with SUMO2 and the QFI mutant, suggesting that both SUMO mutants can be conjugated to CtBP, but are not functional for Pc2 E3 activity. Together, these data suggest that the SIM interacting residues in both SUMO1 and SUMO2 are required for Pc2 E3 activity *in vivo*, and support the idea that SIMs in Pc2 contribute to E3 activity. # Discussion Here we show that Pc2 contains two non-covalent SUMO-interaction motifs that are required for *in vivo* SUMO E3 activity. Consistent with this we show the SIM- interacting domains of both SUMO1 and SUMO2 are required for Pc2-dependent sumoylation of CtBP, but are dispensable when sumoylation is driven by high levels of Ubc9. This provides the first *in vivo* evidence for a role for non- covalent SIMs in SUMO E3 activity. SUMO interaction motifs generally consist of a short hydrophobic core sequence, which often has acidic residues surrounding it. In addition, recent evidence has suggested that phosphorylation of CK2 sites adjacent to a SIM can increase the efficiency of SUMO binding. The best defined SIMs consist of a hydrophobic block of four residues, where position 2 or 3 can be variable. Both SIMs identified in Pc2 conform to this consensus, having a stretch of four hydrophobics with no interruption. One of these, SIM2, has several acidic residues in the surrounding sequence, as well as a consensus CK2 phosphorylation site, although the importance of this remains to be tested. In contrast to SIM2, of the 20 amino acids surrounding SIM1 only one is acidic. Our mutational analyses have identified two pairs of residues surrounding SIM1 that contribute to SUMO interaction and Pc2 E3 activity, although it should be noted that mutating either pair alone has a more subtle effect on E3 activity than mutating or deleting the hydrophobic core. In the context of the isolated amino-terminal half of Pc2, mutation of either of these pairs of amino acids, one of which consists of two hydrophobic residues, the other two basic residues, had more effect on E3 activity than mutating the single acidic residue within the proximity of SIM1. No other convincing candidate SIMs are present in Pc2, although SIM consensus sequences can be somewhat variable. In Pc2 other small patches of hydrophobic residues are present, including three isoleucines close to the extreme carboxyl-terminus of the protein, which have been suggested to be important for E3 activity (A.D.S. and S.H.Y., unpublished). Additionally, it is possible that SUMO might form non-covalent interactions via other binding surfaces. Much of our analysis of the effect of Pc2 SIM mutations has relied on CtBP sumoylation as a readout for Pc2 E3 activity. Analysis of the sequence of CtBP reveals several groups of hydrophobic amino acids that resemble the SIM core consensus. However, examination of the structure of CtBP suggests that none of these regions are surface exposed, such that they are unlikely to be able to make contact with SUMO. However, for some SUMO substrates, the presence of a hydrophobic SIM can drive sumoylation, and it has been suggested that the SIM acts to recruit SUMO-loaded Ubc9 to the substrate. In our assays SIM-interaction mutants of both SUMO1 and SUMO2 are competent for modification of CtBP when driven by over-expression of Ubc9, suggesting that in the absence of Pc2 SIM- SUMO interactions do not contribute to CtBP modification. Among the first proteins to be shown to have hydrophobic SIMs were members of the PIAS family and RanBP2/Nup358. This clearly raised the possibility that SIMs in components of the sumoylation machinery may contribute to sumoylation, and in the case of PIAS and RanBP2/Nup358 might be required for full E3 activity. However, the minimal domain of RanBP2/Nup358 required for *in vitro* E3 activity does not include the SIM, although it is directly adjacent to it, and it has been shown to bind SUMO *in vitro*. For the PIAS family the case for a role of SIMs in E3 activity is even less clear. Conserved SIMs are found in all members of the mammalian PIAS family, but we (J.C.M., M.H.K. and D.W., unpublished) and others have shown that the SIM in PIAS1 is not required for E3 activity. Intriguingly the SIM in PIAS1 has been shown to contribute to its activity as a transcriptional coregulator. Thus the PIAS1 SIM may be functioning more like the SIM in CoREST1, to recruit transcriptional regulators that are themselves targets for covalent SUMO modification. We have not analyzed the effect of Pc2 SIM mutations on transcriptional regulation by Pc2 since they had clear effects on E3 activity. However, if E3 activity is required for transcriptional regulation by Pc2, such mutations would be expected to affect Pc2 mediated repression. Alternatively, recruitment of sumoylated proteins to Pc2 containing PRC complexes might be in part dependent on SUMO interactions with the Pc2 SIMs. This work clearly demonstrates that non-covalent SUMO binding by Pc2 is required for E3 activity *in vivo*, raising the possibility that SIMs in other components of the sumoylation pathway may play a role. The fact that the SIM-interaction mutations of both SUMO1 and SUMO2 result in similar defects in Pc2-dependent CtBP modification, but do not completely inactivate SUMO provides further support for the idea that the E3 activity of Pc2 requires non-covalent interactions with SUMO. One possibility, is that one or both of the Pc2 SIMs play a role in recruiting SUMO-loaded Ubc9. Thus active Ubc9 with SUMO attached to the catalytic cysteine would bind preferentially to Pc2 and once the SUMO had been transferred to substrate, this interaction would be weakened, allowing for exchange of the unloaded for loaded Ubc9. So far we have not been able to show any effect of the SIM mutations in Pc2 on Ubc9 recruitment, but this remains an attractive model. Recent evidence has suggested that Ubc9 itself can be covalently modified with SUMO on lysine 14, and that this allows for preferential recruitment of sumoylated Ubc9 to some substrates. This would also provide a potential mechanism for the role of SIMs in Pc2 activity – sumoylated (on K14) Ubc9 might be preferentially recruited in part via SUMO-SIM interactions, thereby promoting the transfer of the loaded SUMO from the catalytic cysteine to substrates such as CtBP. However, we have been unable to show any effect of mutating lysine 14 in Ubc9 on Pc2 modification, E3 activity or on Ubc9 recruitment (data not shown). An alternative possible explanation for the importance of SIM-SUMO interactions in Pc2 E3 activity is that these interactions do not promote recruitment of Ubc9, but that the interaction of the SIM with SUMO promotes the transfer of SUMO from Ubc9 to substrate by positioning the SUMO appropriately. Our data suggest that SIM2 is the more important of the two SIMs in Pc2, although we have not been able to clearly show that one SIM performs a function that the other does not. However, it is possible that one SIM might contribute preferentially or even exclusively to modification of some Pc2 E3 substrates. In this context, SIM2 does appear to play a larger role in sumoylation of both CtBP and Pc2 itself than does SIM1, although there is clearly a role for SIM1. An additional possibility is that at physiological levels of expression SIM1 might promote only the transfer of SUMO2 to substrates, since it appears less able to work with SUMO1, even when over- expressed. This is consistent with data comparing binding of multiple different SIMs to both SUMO1 and SUMO2, which suggests that negatively charged amino acids surrounding the hydrophobic core contribute to SUMO1 binding, but much less so to binding of SUMO2. In summary, this work provides the first clear demonstration of a requirement for non covalent SUMO-interaction motifs for SUMO E3 activity. Additionally, it appears that Pc2 may function differently from members of the PIAS family, which do not appear to require SIMs for E3 activity. # Materials and Methods ## Cell Culture COS1 cells were cultured in DMEM with 10% bovine growth serum. For live cell imaging and protein expression, COS1 cells were transfected with Fugene and with LipofectAMINE respectively, according to the manufacturers' instructions. ## Plasmids Pc2, and CtBP constructs were expressed from modified pCMV5 plasmids with amino- terminal T7, Flag or His6 tags. Fluorescent protein fusions were expressed from modified pCS2 with amino-terminal fusions to eCFP or eYFP (Clontech), or to monomeric Cherry. Pc2 amino-terminal fusions to a sumoylation consensus were created in pCS2, with an amino-terminal Flag tag and a carboxyl-terminal tag encoding VKPE, or VRPE. Pyruvate kinase sumoylation reporters were a kind gift from Ron Hay. Point mutations and internal deletion mutants were generated by standard PCR techniques and verified by sequencing. SUMO expression constructs were created in modified pCMV5 and pCS2 vectors as above. FV and QFI mutations were introduced by PCR, in the context of SUMOs lacking the carboxyl-terminal tail after the di-glycine motif. The ΔGG mutants were generated by PCR, resulting in alteration of the two terminal glycines to alanines, with the carboxyl-terminal tail present. ## Cobalt Affinity Purification Cells were lysed in 6M guanidine-HCl, 50 mM NaH₂PO₄ (pH 8.0), 10 mM Tris-HCl pH 8.0, 100 mM NaCl. His6 tagged proteins were bound to Talon Resin (Clontech) at room temperature. Resin was washed 3 times for 30 min with 8M Urea, 50 mM NaH₂PO₄ pH 7.0, 100 mM NaCl. ## Coimmunoprecipitation and Western Blotting COS1 cells were lysed in PBS with 1% NP-40 (PBSN) and protease inhibitors (Roche). After centrifugation, lysates were immunoprecipitated with Flag-agarose (Sigma) for 2 h at 4°C. Beads were washed 3 times with PBS-N and proteins were separated by SDS-PAGE and transferred to Immobilon-P (Millipore). Proteins were visualized using ECL (Amersham). ## GST Binding Assays GST-SUMO1 or GST-SUMO2 was purified from BL21 cells on glutathione agarose, as previously described. Following elution with glutathione and dialysis, proteins (2 µg GST-SUMO per binding reaction) were rebound to glutathione agarose, blocked overnight with BSA, and incubated with lysates form transfected COS1 cells (in PBS with 0.1% Triton X-100 and protease inhibitors), or with His6-YFP- Pc2 fusions purified from bacteria. Following incubation with rocking at room temperature for 2 hours, the glutathione agarose was allowed to settle for 10 minutes and the supernatant removed. Beads were then washed 3 times with 1 ml of lysis buffer, by gentle mixing followed by settling at room temperature for 10 minutes. Bound Flag-tagged or His6-YFP-tagged Pc2 was detected by western blot, and GST-SUMO was visualized by Coomassie blue staining or western blot. ## Live Cell Imaging For quantification, COS1 cells were plated into 4-well \#1 cover-glass chamber slides (Nunc LabTek) and transfected using Fugene 6 (Roche). After 21-23 hr, cells were imaged with a Nikon Eclipse TE200 inverted fluorescence microscope with GFP, CFP, Rhodamine and DAPI filter sets, at 60x magnification using a water immersion lens (Nikon PlanApo 60x 1.20 WI). To quantify co-localization with Pc2, cells expressing Pc2 in sub-nuclear foci were selected, and scored for colocalization of the other fluorescent protein. In each experiment, at least 50 cells were scored for each combination, and the results of representative assays are shown. Black and white images were acquired using Openlab with a Hamamatsu digital camera (C4742-95) and colorized in Photoshop CS2 (9.0.2). The eYFP signal was false colored to green, and the eCFP or mCherry kept as white on black, as these colors retained the best contrast. We thank R.T. Hay, Y. Chen, M. Dasso, R.Y. Tsien and B.M. Paschal for providing reagents, and B.M. Paschal for advice. [^1]: Conceived and designed the experiments: JM DW. Performed the experiments: JM TM MK SHY. Analyzed the data: JM TM MK DW. Contributed reagents/materials/analysis tools: AS. Wrote the paper: JM DW. [^2]: Current address: Whitehead Institute for Biomedical Research, Cambridge, Massachusetts, United States of America [^3]: The authors have declared that no competing interests exist.

# Introduction The tumor suppressor gene, *Trp53*, which encodes the p53 protein, plays an important role in maintaining genomic integrity in response to a wide range of cellular stresses including DNA damage, hypoxia, ribonucleotide depletion, and oncogene activation. These stress signals stimulate the activation of p53 protein, resulting in effects on multiple cellular processes including apoptosis, cell cycle arrest, and senescence, mediated largely through the activity of p53 in transcriptional regulation of its target genes. Dysfunction of p53 can predispose to the development or progression of cancers. In fact, the p53 gene is the most frequently mutated gene identified in a variety of human cancers with more than 50% of human tumors characterized by p53 mutations. Further evidence for a critical tumor suppressor role of p53 is provided by analysis of p53 knockout mice (p53^−/−), which develop tumors, predominantly thymic T cell lymphomas, at an early age in essentially 100% of homozygous p53-deficient mice. p53^−/− mice, deficient for expression of all p53 isoforms, are developmentally normal but are susceptible to a variety of spontaneous tumors by about 6 months of age. About 75% of the tumors that develop in p53^−/− mice on either C57BL/6 or 129/Sv genetic backgrounds are T-cell lymphomas. Sarcomas also develop in p53 knockout mice, and other malignancies, such as B-cell lymphomas, occur less frequently. To assess the role of p53 in the initiation and progression of other malignancies, we attempted to generate a reversible p53 knockout mouse (p53^rev/rev) by inserting a neomycin resistance gene flanked by loxP sites into the first exon of the p53 gene. Based on current understandings of p53 transcription, it was anticipated that, in the absence of Cre-mediated recombination, this would introduce a transcriptional stop signal, so that p53^rev/rev mice would not express p53 protein and would develop tumors precisely as observed in conventional p53^−/− mice. Surprisingly, however, expression of p53 protein was detected at varying levels in multiple tissues of p53^rev/rev mice, with expression level in thymus similar to that of wild-type (wt) mice, but with reduced levels in spleen, uterus, kidney, liver and heart, and with no detectable p53 in peripheral B lymphocytes. Analysis by 5′RACE revealed that thymi of p53^rev/rev mice expresses a new species of p53 mRNA transcribed from a novel start site utilized in thymocytes but not in p53^rev/rev B cells. p53^rev/rev mice developed splenic B cell lymphomas at high frequency, but did not develop the thymic lymphomas that are characteristic of p53^−/− mice. # Materials and Methods ## Mice p53^+/− mice and CD19-Cre⁺ transgenic mice were purchased from the Jackson Laboratory (Bar Harbor, ME). p53^+/− mice were intercrossed to produce p53^−/− mice. p53^rev/rev mice, generated as described below, were backcrossed with C57BL/6 (B6) mice for at least six generations. Protocols for animal care and use were conducted consistent with the Guide for the Animal Care and Use of Laboratory Animals of National Institutes of Health. The protocol was approved by the Committee of the Animal Care and Use of Laboratory Animals of National Institutes of Health (IACUC protocol number: EIB-029). No surgery was performed, and humane endpoints were followed per IACUC guidelines. All animals were housed at Bioqual (Rockville, MD) and experiments were performed in NCI. ## Generation of Trp53 gene targeted mice A *Trp53* gene targeting vector was constructed from a 5 kb DNA segment including exon 1 of the *Trp53 gene*, and was isolated from 129 mouse genomic DNA by digestion with Kpn I. A neomycin resistance cassette flanked by loxP sites was inserted into the first exon of the gene. A thymidine kinase cassette (TK) was placed further upstream of exon 1 and used as a negative selection marker. Electroporation and selection were performed with the 129 CJ7 embryonic stem (ES) cell line as described by Southon and Tessarollo. Two independent successfully targeted ES cell clones were injected into C57BL/6CR blastocysts, generating chimeras that transmitted the targeted allele to progeny. ## Genotyping for p53^rev/rev ES cells and mice with Southern blot DNA for Southern blot analysis was isolated from ES cells. DNA isolation and Southern blot analysis procedures were as described elsewhere. DNA was digested with Kpn I, electrophoresed on a 0.7% agarose gel, transferred to a nylon membrane, and hybridized with probe A (.) ## Reverse transcription PCR (RT-PCR) A NucleoSpin kit (Clontech, Mountain View, CA) was used to isolate total RNA from various tissues or cells for RT-PCR analysis. Primer pair E1 (5′-TTA GGG GGC ACC TAG CAT TC-3′) and E4 (5′-AGT TGC CCT GGT AAG TTT TTT G-3′) was used to detect conventional p53 mRNA. Primer pair N2F (GACGTA AAC TCC TCT TCA G) and E4 was used to detect the expression of new species of p53 mRNA. The primer pair N1F (GGG CGC CCG GTT CTT TTT GTC A) and N1R (TTG GTG GTC GAA TGG GCA GGT AGC) amplifies transcripts from the neomycin resistance gene. Actin primer pair ACTIN-F (5′-ATG CCA ACA CAG TGC TGT CTG GTG G-3′) and ACTIN-R (5′-CTG ATC CAC ATC TGC TGG AAG GTG-3′) was used to detect the expression of β-actin. RT-PCR reactions were carried out in 50 µl of PCR reaction mix containing 25 µl of PCR buffer (One-step RT-PCR kit, Life Technologies), 1 µg of total RNA and 0.2 µM primers. RT-PCR was carried out at 42°C for 30 minutes, 35 cycles of 95°C for 30 sec, 58°C for 30 sec and 72°C for 1 min. RT-PCR products were visualized by agarose gel electrophoresis with EtBr staining. ## 5′Rapid amplification of mRNA end by PCR (5′-RACE) 5′-RACE was performed according to instructions for the 5′-RACE kit purchased from Invitrogen (San Diego, CA). Primer p53-1165R (ACC ATC ATC ACA CTG GAA GAC) was used to make cDNA. The primer pair adaptor primer and p53-411R (AGT TGC CCT GGT AAG TTT TTT G) was used to amplify the 5′ end of the single stranded cDNA ligated with adaptor, and PCR products were sequenced. ## Western blot analysis for p53 protein Mice or cells were irradiated at 10 Gy. Three hours later, mouse tissues or cells were lysed in buffer containing 50 µm Tris (pH 7.4), 150 µm NaCl, 1 mM Na₂VO₄, 1%NP-40, and protease inhibitor cocktail. Twenty- five micrograms of protein extract from each sample was fractionated by sodium dodecyl sulfate-polyacrylamide gel electrophoresis and transferred to a nitrocellulose membrane. Immunoblotting analysis was carried out with monoclonal antibodies specific for p53 (1C12) (Cell Signaling, Beverly, MA) and actin (AC-15) (Sigma, St. Louis, MO). ## Apoptosis analysis for thymocytes and splenocytes in response to ionizing radiation Thymocytes and splenocytes were treated with ionizing radiation at 10 Gy, cultured at 37°C for 24 hours and stained with PI and annexin V-FITC for flow cytometric (FACS) analysis. ## Histological analysis Tumors were fixed in 4% paraformaldehyde at 4°C overnight, dehydrated through a graded alcohol series, and then embedded in paraffin. Sections of 6–8 µm were prepared and stained with H&E. Tumors were classified according to established criteria. ## Analysis of chromosomal aberrations by Spectral Karyotyping (SKY) Metaphase spreads were prepared and Spectral Karyotyping (SKY) was performed for the identification of chromosomal abnormalities according to standard protocols available at <http://www.riedlab.nci.nih.gov/protocols.asp>. A minimum of 10 metaphases were acquired and analyzed for each cell line-derived tumor. The karyotypic findings are described in accordance with the ISCN nomenclature rules (ISCN, 2005). # Results ## Generation of mice bearing a p53 reversible knockout gene p53 reversible knockout (*p53^rev/rev*) mice were generated with the intent of allowing cell type-specific expression of p53. The first exon of p53 was disrupted by inserting a neomycin resistance cassette (neo) flanked with loxP sites. It was anticipated that insertion of the neo cassette would result in termination of p53 transcription with the polyA signal sequence in the neo cassette before the transcript elongates into intron 1. The gene-targeting vector was transfected into ES cells, and Southern blot analysis was used to identify gene targeted ES cell clones. Wild-type ES cell DNA generated a 5 kb band with probe A after Kpn I digestion, while heterozygous (p53^+/rev) ES cell DNA generated one additional 7 kb band. Four of 85 ES cell clones tested were identified as containing the p53^rev allele. Two of the positive clones were used to make chimeric mice that transmitted the p53^rev allele to offspring. These mice were backcrossed to B6 background for at least 6 generations. p53^+/rev mice were intercrossed to produce p53^rev/rev mice as genotyped by PCR. Development and early survival of p53^rev/rev mice were normal, with 21 p53^+/+, 40 p53^+/rev, and 19 p53^rev/rev mice observed at weaning among 80 offspring from p53^+/rev intercrosses, consistent with Mendelian segregation and indicating that no early lethality was associated with the p53^rev/rev genotype. ## Differential expression of p53 protein in tissues and cell lineages of p53^rev/rev mice: detection in thymocytes but not peripheral B lymphocytes It was anticipated that no p53 protein would be produced in p53^rev/rev mice in the absence of Cre-mediated recombination. Normally, p53 protein is expressed but not easily detected due to its rapid proteasomal degradation under steady-state conditions. To assess p53 protein expression in p53^rev/rev and wt controls, mice were irradiated at 10 Gy to increase p53 protein levels, largely through post-translational stabilization. Tissues taken from these mice 3 hr later were used to make protein lysates that were analyzed by western blotting with p53-specific antibodies. Surprisingly, p53 protein was detected in multiple tissues in both wt and p53^rev/rev mice with the notable exception that its level was markedly reduced in spleen of p53^rev/rev mouse. We then proceeded to study isolated populations of thymocytes and spleen cells for the effects of radiation on p53 expression in vitro. As shown in, comparable levels of p53 protein were induced in response to irradiation of wt and p53^rev/rev thymocytes. In contrast, p53 protein was at or below the limits of detection in p53^rev/rev spleen cells but was readily detected in lysates from spleen cells of wt mice. This prompted us to test purified splenic T cells and B cells for their responses to irradiation. Studies of cells from wt mice revealed striking elevations of p53 levels in irradiated B cells and lesser but still substantial elevations of expression in treated T cells. By comparison, there was no detectable p53 response by B cells from p53^rev/rev mice and the response of treated T cells, while detectable, was considerably reduced from that of irradiated wt T cells. ## Characterization of p53 mRNA: Identification of a novel transcript in thymocytes from p53^rev/rev mice To determine the mechanisms responsible for the unexpected expression of p53 protein in cells of p53^rev/rev mice, we first examined the nature of p53 transcripts by RT-PCR using primers from exon 1 and exon 4. p53 transcripts were readily detected in both thymocytes and spleen cells of wt mice, but were very low in thymocytes and not detectable in spleen cells of p53^rev/rev mice. We then used 5′-RACE to determine if thymocytes from p53^rev/rev mice might express alternative species of p53 transcripts that could be translated into p53 and identified a novel species of p53 mRNA, designated as p53REVmRNA. The 5′-RACE PCR product was sequenced in its entirety and found to include a 176 bp sequence at the 5′ end identical to the 3′-sequence of the neomycin resistance cassette that was used for gene targeting and that included the loxP site. This was followed by a 25 bp sequence identical to the 3′end of exon 1, and then exons 2 and 3 and part of exon 4. The sequences downstream from exon 4 were identical to the sequences of wt p53 cDNA. The p53 protein predicted to be translated from p53REV mRNA would thus be identical to wt p53. illustrates the new transcriptional start site of p53REV RNA and the approach taken for RT-PCR analysis of this transcript. RT-PCR analyses using primers E4 and N2F corresponding to a unique sequence in p53REV, revealed that p53REV mRNA is expressed in thymocytes and splenic T cells but not in splenic B cells of p53^rev/rev mice, paralleling the patterns of p53 protein expression in thymocytes and spleen cells of these mice. In contrast, expression of the gene encoding the neomycin resistance protein, detected using primers N1F and N1R, was comparable in both thymocytes and splenic B cells of p53^rev/rev mice, indicating that the expression of p53REV mRNA was not regulated by the promoter of the neomycin gene. The interpretation of these findings for regulation of p53 expression in p53^rev/rev mice is discussed below. ## Irradiation-induced p53-dependent apoptosis is intact in thymocytes but not spleen cells of p53^rev/rev mice A major function of p53 expressed in response to stress is the induction of apoptosis. To compare the responses of cells from p53^rev/rev and wt mice to stress, thymocytes and splenocytes were irradiated and examined by FACS for apoptosis using analyses of Annexin V and PI staining. After irradiation and overnight culture, only 4% of thymocytes from p53^−/− mice had undergone apoptotic death in comparison to 91% of thymocytes from wt mice, demonstrating the p53 dependence of radiation-induced apoptosis. Under the same conditions, similar proportions of thymocytes from wt and p53^rev/rev mice had undergone apoptosis (92% and 86%, respectively). These results indicated that thymocytes from p53^rev/rev mice that express p53 protein encoded by p53REV mRNA were as susceptible to radiation-induced apoptosis as thymocytes from mice expressing p53 from the wt transcript. Similar analyses of spleen cells showed that 29% of cells from p53^−/− mice underwent apoptosis post irradiation compared to 76% of cells from wt littermates whereas the proportions of apoptotic cells from p53^rev/rev and wt mice were 43% and 82%, respectively. Thus, spleen cells from mice totally deficient in p53 and from p53^rev/rev mice were similarly resistant to radiation-induced apoptosis. ## p53^rev/rev mice spontaneously develop B cell lymphomas p53^−/− mice appear to develop normally, but spontaneously develop tumors by 6 months of age, approximately 75% of which are thymic T cell lymphomas. To determine the effects of p53^rev expression on tumor development, we studied the survival of wt, p53^+/rev and p53^rev/rev mice for one year. The survival of both wt and p53^+/rev mice was 85% at one year with none of the mice dying of cancer. In marked contrast, all p53^rev/rev mice were dead by 48 weeks of age. Twenty-three of 29 p53^rev/rev mice had massive splenomegaly at the time of death. Histopathologic studies of 11 of these mice revealed changes characteristic of splenic marginal zone B cell lymphomas (SMZL) including marked expansion of the marginal zone with invasion of the red pulp, often associated with compression of the white pulp. The lymphomas were limited to spleen except for one case with a subcutaneous metastasis. Of the remaining p53^rev/rev mice, two died with skin lesions, one with hepatomegaly, two with subcutaneous sarcomas, and two from unknown causes. By flow cytometric analyses, the lymphoma cells were large and CD19⁺B 220^loIgM⁺IgD⁻CD5⁺CD21⁻CD2 3⁻. In tumor bearing mice, large B220^loCD5⁺ cells phenotypically similar to those of the splenic lymphoma could also be found in the peripheral blood but not in the blood of young p53^rev/rev or wt mice. B cell lymphomas from p53^rev/rev mice did not express either p53REV mRNA or p53 protein, but were positive for expression of the neomycin resistance gene. Six B cell lymphomas from p53^rev/rev mice were subjected to karyotypic analysis. Only one tumor was karyotypically normal (40, XY). The others exhibited a range of chromosome counts ranging from 2n to 4n and carried a variety of chromosomal anomalies including translocations, insertions and deletions. Four of the tumors were characterized by a single dominant clonal population, while two of the tumors were comprised of two dominant populations. Losses of chromosomes 12 and 19 and gains of chromosome 10 and 15 were often present. However, there was no chromosomal anomaly common to the different tumors that would suggest a unifying molecular mechanism. Of note, none of the B cell lymphomas analyzed had translocations involving Ig loci. ## No B cell lymphomas developed in p53^rev/rev mice expressing B cell-specific CD19-Cre To determine whether p53 protein would be expressed in p53^rev/rev B cells in the presence of Cre expression, CD19-Cre transgenic mice were bred with p53^rev/rev mice to generate p53^rev/revCD19-Cre⁺mice. Splenic B cells from p53^rev/revCD19-Cre⁺ and p53^rev/revCD19-Cre⁻ mice were isolated, irradiated and cultured for three hours at 37°C. The lysates of B cells were immunoblotted with p53 antibodies. shows that B cells from p53^rev/revCD19-cre⁺ mice were induced to express p53 protein while no detectable p53 protein was found in the B cells of p53^rev/revCD19-Cre⁻ mice. No B cell lymphomas were observed in ten p53^rev/revCD19-Cre⁺ mice followed for 9 months, while tumors developed in seven of nine p53^rev/revCD19-Cre⁻ mice. # Discussion The tumor suppressor p53 plays a critical role in maintaining genomic integrity by regulating the expression of genes responsible for mediating apoptosis, cell cycle arrest, DNA repair, and senescence in response to DNA damage or oncogenic stress. More than 50% of human tumors harbor mutations in the p53 gene, and essentially all p53^−/− mice develop tumors, predominantly thymic lymphomas, by one year of age. This study was designed to overcome the limitations on studying the contributions of p53 deficiency to transformation of cells other than T cells imposed by the early development of thymic lymphomas in conventional knockout mice. The approach was to generate mice, termed p53^rev/rev mice, in which expression of p53 was prevented by inserting a floxed neomycin resistance gene into the first exon of p53 gene, and in which cell lineage-specific p53 expression could be induced in the presence of an appropriate lineage-specific Cre. In contrast to these expectations, p53 protein was expressed in multiple tissues of p53^rev/rev mice, including thymus and spleen, in the absence of Cre. Studies of lymphocyte subsets from spleen showed that p53 was expressed by T cells but not B cells and that B cells remained negative for p53 expression, even following induction of DNA damage by irradiation. Strikingly, these mice developed splenic B cell lymphomas but no T cell lymphomas. Precedent for a role of p53 deficiency in B cell transformation exists in earlier studies of conventional p53^−/− mice that were shown to develop MZL at low frequency in the same time frame as thymic T cell lymphomas. Interestingly, studies of human SMZL identified deletions and/or mutations in TRP53 in ∼20% of cases. Although these changes had no prognostic significance, they may indicate shared cross-species mechanisms in the pathogenesis of these poorly understood neoplasms. Interestingly, we had also been unsuccessful in efforts to make another reversible knockout mouse by using a neomycin resistance gene with loxP sites flanking SV40 polyA signal sequence as a STOP cassette. As shown in, p53 expression was reduced by 90% but was not completely abolished when this STOP cassette was inserted into intron 6 of p53 gene. Moreover, this level of expression was sufficient to prevent appearance of tumors in p53^rk1/rk1 mice. The SMZL of p53^rev/rev mice were CD5⁺ and thus resembled a subset of human SMZL comprising about 25% of cases. It is currently not clear if the CD5⁺ cases of human SMZL represent a phenotypic variant of SMZL with aberrant expression of CD5 or if they represent a separate clinico- pathologic entity. In mice, there are several models with lymphomas diagnosed histologically as SMZL that are CD5⁺ and may derive from peritoneal B1a cells rather than splenic MZ B cells. Distinguishing between MZ and B1a cells as possible origins for these lymphomas will require the identification of reliable specific markers for each lineage that can be used for studies of normal and tumor cells. The B cell lymphomas of p53^rev/rev mice were clearly different from the pro-B cell lymphomas observed in non-homologous end joining (NHEJ)-deficient and H2AX-deficient mice crossed onto a p53^−/− background, which are characterized by the presence of T(12;15) translocations. These translocations result in the juxtaposition of the *Myc* oncogene on chromosome 15 under the transcriptional regulation of the IgH promoter on chromosome 12. Interestingly, half of the karyotyped p53^rev/rev tumors had translocations involving chromosome 15 and 2/3 had an extra copy of chromosome 15, similar to what is observed in mouse thymic lymphomas. By achieving altered cell lineage specificity of p53 expression, p53^rev/rev mice have created a novel and instructive model of B cell neoplasia. However, the regulatory mechanisms underlying this lineage-specific change in expression of p53 remain less than fully understood. We identified a transcriptional start site for p53REV mRNA located near the 3′ end of the neomycin resistance cassette that was utilized in thymocytes but not in B cells or B cell lymphomas of p53^rev/rev mice even though the neo gene was expressed at equal levels in these populations. This indicated that expression of p53REV mRNA was not determined simply by a foreign neo promoter. It thus seems likely that insertion of the neomycin gene in exon1 may disrupt the tissue specificity of an alternative p53 promoter, silencing the expression in B cells of p53^rev/rev mice. In initial experiments designed to further probe regulation of p53, we deleted the immediate promoter and partial first exon of the p53 gene in BAC DNA which was introduced as a transgene into p53^−/− mice. Surprisingly again, p53 protein was expressed in both thymocytes and splenocytes. Analysis of cDNA by 5′-RACE demonstrated a transcriptional start site within exon 1 of the p53 gene that is not the classic (common) site but corresponds to a cDNA sequence previously entered in GENEBANK (access number: CJ049635). Our data suggest that the p53 gene might have an unknown promoter that can act at long range to regulate p53 expression, as has now been described for a number of genes. It is worth noting that while p53 protein is absent from the entire B cell population in p53^rev/rev mice, the lymphomas that develop in these mice bear the unique histopathologic features of SMZL and are thus quite distinct from the B lymphomas recently reported to occur in B cell-specific p53 knockout mice ; in that strain, p53-deficient B lineage cells were generated by the activity of mb1-Cre on a floxed p53 allele. The tumors that developed in those mice all expressed CD43, a B lineage marker that is extinguished when normal B cells rearrange the kappa locus during maturation in the bone marrow, suggesting that they all derived from immature B cells. Consistent with an origin in immature or pro-B cells, those tumors expressed translocations involving Ig loci, suggesting aberrant V(D)J rearrangement or class switch recombination. In contrast, the B cell lymphomas derived in our studies from p53^rev/rev mice expressed surface IgM and did not contain translocations involving Ig loci, suggesting that these lymphomas arose after normal and successful V(D)J recombination. In this regard it is noteworthy that SMZL also develop in other models in which p53 function is compromised but at low frequencies. The basis for this differential susceptibility of marginal zone B cells to transformation in these different experimental settings remains to be determined. The preferential development of SMZL in p53^rev/rev mice might reflect the stage of B cell development at which p53 protein expression is terminated in cells of this lineage, rendering this subset exceptionally susceptible to transformation. Analyses of developing B lineage cells in the bone marrow and spleen will be required to approach this question. Previous studies reported that p53 mutations are associated with multiple subtypes of human B lymphomas, including mucosa-associated lymphoid tissue (MALT) lymphomas of marginal zone B cell origin, mantle cell lymphomas, centrocyte-like lymphomas, follicular lymphomas and SMZL. The p53^rev/rev mouse therefore provides a novel animal model for understanding the mechanisms involved in the pathogenesis of the subsets of human B cell non-Hodgkin lymphomas that originate in marginal zone B cells. Importantly, p53 can be re-expressed at wt levels in p53^rev/rev B cell lymphomas when Cre is activated, thus providing a unique system to understand the influence of p53 expression at different stages of B cell lymphoma development and progression. The B1a B cell-like surface markers of p53^rev/rev B lymphomas suggested B1a B cells as possible cells of origin for these lymphomas. Therefore, the p53^rev/rev mouse may be a useful model for studies of MZ and B1 B cell lymphomas. Although the (p53^rk1/rk1) mouse of Supplementary 3 was an unsuccessful reversible p53 knockout mouse, expression of p53 protein in this line is less than 10% of that of wild-type mice in the absence of cre; and it may therefore be a valuable animal model for understanding the effect of suppressed p53 expression level in tumor development. # Supporting Information We are grateful to Drs. Dinah Singer, Izumi Horikawa and Curtis Harris for their critical reading of this manuscript. We thank Eileen Southon and Susan Reid for their help in generating mutant mice. We thank Genevieve Sanchez-Howard and the staff at Bioqual for expert animal care and breeding. We are grateful to Thomas Ried for his support of, and Callie Seaman for her assistance with, the SKY analysis. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YJC RJH. Performed the experiments: YJC MJD. Analyzed the data: YJC MJD HCM RJH. Contributed reagents/materials/analysis tools: YJC MJD LT. Wrote the paper: YJC MJD LT HCM RJH. [^3]: Current address: Division of Cancer Treatment and Diagnosis, National Cancer Institute, National Institutes of Health, Bethesda, Maryland, United States of America

# Introduction At the first station of central odor processing, the main olfactory bulb (MOB), signal processing is regulated by synaptic interactions between glutamatergic and GABAergic inputs of the mitral cells and tufted cells (M/T cells), which are the major projection neurons. The M/T cells receive both excitatory glutamatergic inputs from the olfactory sensory neurons (OSNs) and inhibitory GABAergic inputs from the interneurons, which predominantly consist of periglomerular cells (PGCs) and GCs. The GABAergic inhibitory inputs from the interneurons deliver lateral or recurrent inhibition to the M/T cells to modulate incoming sensory inputs to an optimal level. Because the GCs outnumber the PGCs in the bulb and because only approximately 20% of the PGCs make connections with both presynaptic OSNs and postsynaptic M/T cells, most of the lateral inhibition is mediated by the GCs. The lateral inhibition from the GCs is exerted via reciprocal dendrodendritic synaptic connections with the M/T cells and functions as a contrast enhancer to facilitate the discrimination of correlated ON inputs. Long-term potentiation (LTP) and long-term depression (LTD) are two forms of synaptic plasticity that are considered as the cellular substrates for learning and memory. LTP or LTD in the OSN-MC synapses in the main olfactory bulb can be produced by a brief tetanic or low frequency ON stimulation. Recently, LTP of the field excitatory postsynaptic potentials (fEPSPs) in the glomerulus was induced by theta burst stimulation (TBS) of the ON. Moreover, Hebbian spike timing-dependent plasticity (STDP), a form of plasticity that strictly requires a temporal correlation between the pre- and postsynaptic responses, was also induced in vivo in β–lobe neurons of the locust mushroom body through the electrical stimulation of Kenyon cells and in the excitatory inputs to the GCs in the main rat olfactory bulb. Despite these elegant investigations in different levels and species, there have been a lack of studies that directly examine STDP in the MCs. In the olfactory bulb, the lateral inhibition driven by the GCs tends to affect the dendritic depolarization induced by the excitatory inputs and either prevents or delays firing in the MCs or decreases the firing rate during stimulus presentation. This change in the firing rate begins at the end of the spiking period induced by odor stimulation in vivo, indicating that the recent activity history could affect lateral inhibition between the MCs. However, little is known about whether and how prior persistent modification of the inhibition modifies the predisposition for subsequent induction of long- lasting changes of the excitatory OSN-MC synapses. This form of synaptic plasticity regulation has not been fully investigated in the olfactory system. Because STDP is a form of synaptic plasticity whose induction is heavily dependent on the timing of the incoming spikes and the frequency of postsynaptic spikes when the spike bursts are induced by the pairing protocol, the presence of lateral inhibition onto the postsynaptic MCs may profoundly influence the induction of STDP and the total olfactory bulb output. Moreover, the timing of synaptic inhibition itself may also be regulated. Therefore, the lateral inhibition exerted by the GCs plays critical roles in odor information processing and in olfactory learning and memory. The mechanism by which the frequency of the intra-burst spikes is controlled by lateral inhibition is of great importance in understanding how sensory information is encoded and processed. GCs receive two types of excitatory inputs on their proximal and distal dendrites. The reciprocal dendrodendritic synapses with the MCs are the primary source of distal excitatory inputs, which mediate local dendrodendritic inhibition (DDI). Cortical feedback input is one of the major sources of proximal input and mediates the global top-down modulation of DDI in the olfactory bulb. These two anatomically distinct excitatory inputs display persistent modification with opposing directions when subjected to the same stimulating protocol, which suggests that they may differentially exert their influence on the subsequent induction of synaptic plasticity in MC excitatory synapses and thus have different implications in regulating the total olfactory bulb output. Because the extent of granule cell (GC) function in the local versus global output modes can have an important impact on the computational role GC performs in the olfactory bulb circuit, the modification of the excitatory ON input by prior plasticity of the distinct excitatory inputs to GCs will profoundly influence the encoding and processing of odor information in the bulb. In this study, we found that TBS could elicit differential plasticity of two different excitatory inputs to the GCs. Interestingly, the same protocol also induced a persistent modification in the lateral inhibition to the MCs with opposing directions. This bidirectional modification of the inhibitory inputs differentially regulated the predisposition of the subsequent induction of STDP of excitatory ON inputs to the MCs. Further evidence demonstrated that this regulation was achieved by the regulation of the spike frequency within the bursts employed by the pairing protocol for STDP induction. Thus, our results revealed one of the mechanisms by which the frequency of the intra-burst spikes is controlled in the sensory system. Because fine temporal burst structure is proposed to convey stimulus-related information to postsynaptic cells, this novel form of inhibition-dependent regulation of plasticity may contribute to the encoding or processing of olfactory information in the olfactory bulb. # Materials and Methods ## Ethics Statement The protocols for the animal care and use were approved by the Experimental Animal Ethics Committee at the Nanjing Medical University (permit number 20100582). ## Olfactory bulb brain slice preparation Acute olfactory bulb slices were prepared from P14–21 Sprague Dawley rats. The rats were deeply anesthetized with ketamine (140 mg/kg, ip) and decapitated, and the brain was quickly placed into ice-cold artificial cerebrospinal fluid (ACSF) containing (in mM) 124 NaCl, 3 KCl, 1.23 NaH₂PO₄, 1.2 MgSO₄, 26 NaHCO₃, 10 dextrose and 2.5 CaCl₂ bubbled continuously with 95% O₂/5% CO₂. Horizontal olfactory bulb slices (300 µm thick) were prepared with a vibrating blade microtome (WPI Inc., USA). Fresh slices were incubated in the chamber with carbogenated ACSF and recovered at 30°C for 30 min and then maintained at room temperature. ## Electrophysiological studies Conventional whole-cell recordings in the current-clamp mode were made with patch pipettes containing (in mM) 140 K-methylsulfate, 4 NaCl, 10 HEPES, 0.2 EGTA, 4 MgATP, 0.3 Na₃GTP, and 10 phosphocreatine. The pH was adjusted to 7.4 with KOH. The micropipettes were made from borosilicate glass capillaries (Sutter Instrument Co.) and had resistances in the range of 5–8 MΩ. The cells were viewed under an upright microscopy (Eclipse E600-FN, Nomarsky, Nikon Corp., Tokyo, Japan) and recorded with an Axopatch-200B amplifier (Molecular Devices, Palo Alto, CA). Moreover, the MCs were identified by their morphology, size and location. In the GCs recordings, the cells were selected from the granule cell layer (GCL) based on their small cell-body diameters (\<10 µm). The olfactory bulb slices were perfused with 32°C ACSF that was bubbled continuously with carbogen (95% O₂/5% CO₂). The EPSPs or inhibitory postsynaptic potentials (IPSPs) were recorded in the control ACSF in the absence of any receptor blockers to ensure that both the glutamatergic and GABAergic neurotransmission were intact. The membrane potentials (mV) of the MCs in the current-clamp mode were from −52 to −61 mV. A bipolar stimulating electrode (inside diameter, 25 µm, FHC Co., USA) was placed into the external plexiform layer (EPL) or granule cell layer (GCL), to evoke synaptic responses at the distal or proximal inputs, respectively. The current intensity of the test stimuli (0.01–0.30 mA) was set to produce half-maximal EPSPs (one-peak monosynaptic responses, with amplitudes between 1 and 5 mV). The basal evoked synaptic responses were produced by 100 µs electrical stimulation at 0.05 Hz except during the induction of STDP. The TBS consists of five bursts of five stimulations (intra bursts: 100 Hz; inter-burst 5 Hz) repeated 5 times at 0.1 Hz. The data were low-pass filtered at 2 kHz and acquired at 5–10 kHz. The series resistance was always monitored during the recording for fear that re- sealing of the ruptured membrane would cause changes in both the kinetics and amplitude of the EPSPs. The cells in which the resistance or capacitance deviated by \>20% from the initial values were excluded from the analysis. The data were collected with the pClamp9.2 software and analyzed using Clampfit9.2 (Molecular Devices, Palo Alto, CA). The STDP in the MCs was induced by pairing the EPSPs with postsynaptic spike bursts. The pairings were repeated 60 times at a 0.1 Hz stimulation and the EPSPs were evoked by stimulating the glomeruli that correlated with the recorded MCs. These postsynaptic bursts comprised of three spikes and were elicited by a current injection (intensity: 50–300 pA; duration: 50 ms). The inter-spike interval (ISI) in control was set as 19–21 ms (20.1±0.2 ms, n = 12; frequency approximately 50 Hz) by adjusting the intensity of the injected current. The time interval Δ*t* was defined as the time between the onset of the compound EPSP and the onset of the action potential (AP) burst (first AP in the burst). The positive time window was defined as the EPSPs that occurred before the postsynaptic bursts, whereas the negative time window referred to the postsynaptic bursts that occurred before the EPSPs. The persistent potentiation or depression was defined as the percentage changes in EPSP or IPSP amplitude during the last 10 min of the recording after the repetitive stimulation. To examine whether direct activation of M/T cells also contribute to the changes in STDP, we usually used one or two single stimuli to EPL or GCL before TBS was delivered and examine whether EPSP can be elicited. In most of our recordings, TBS in EPL or GCL with moderate intensity usually could not directly activate M/T cells. Occasionally we observe EPSPs were induced in MCs. We discarded these cells and did not continue to perform further recording on these cells. The data are presented as the mean ± SEM. Paired Student's *t* tests were applied as statistical tests if not indicated otherwise, and the statistical significance was asserted for *p*\<0.05.Within-group comparisons were performed using a two-tailed *t* test, and the difference between the groups was compared using ANOVA post hoc comparisons. An ANOVA *post hoc LSD* test was used when equal variances were assumed. The differences were considered significant when *p*\<0.05 and significance for homogeneity of the variance test was set at 0.1. # Results ## Distinct long-term plasticity of excitatory inputs to the GCs In order to investigate whether persistent modification of the inhibitory inputs regulates the predisposition of the subsequent STDP induction of excitatory ON inputs to the MCs, we need to induce both long-term plasticity of inhibitory inputs and STDP of excitatory ON inputs to the MCs. Since GCs are the main type of interneuron in the olfactory bulb that laterally inhibits the M/T cells, we first tried to induce long-term plasticity of excitatory inputs to the GCs and then determine whether this plasticity could in turn elicit long-term plasticity of inhibitory inputs to MCs. We performed whole-cell patch clamp recordings of evoked EPSPs in the GCs of MOB slices to examine whether long-term plasticity of excitatory inputs onto GCs (excitatory inputs→GCs synapses) could be elicited using a specific induction protocol. A TBS protocol (five 100-Hz bursts of 5 shocks, repeated at 5 Hz) was focally delivered to the distal or proximal excitatory inputs to the GCs to induce long-term plasticity of these inputs. The EPSPs were recorded in the current clamp mode without disturbing GABA_A receptor-mediated inhibitory synaptic transmission. The bipolar stimulating electrode was placed in either the EPL or GCL within a 200–300 µm distance from the cell body of the MCs to stimulate distal or proximal excitatory inputs, respectively. Strowbridge and colleagues had employed a two- photon guided minimal stimulation to selectively activate the two inputs. In order to ensure that we also activated a relatively homogenous population of presynaptic processes when stimulating either of the two sites (i.e., the specificity of the stimulations), we deliberately controlled the intensity of the local stimulation to ensure the independence of the stimulation on distinct sites. The synaptic responses we obtained from these two inputs displayed distinct properties in their current kinetics (rise time, GCL: 19.9±3.7 ms, EPL: 44.5±5.2 ms, *p*\<0.001), similar to the observations obtained from the Strowbridge lab. At the distal excitatory inputs in EPL that mediated local DDI, TBS elicited LTD of these inputs in 6 out of the 7 cells (78.0±2.3%, n = 6, *p*\<0.001, paired-samples t test, *p*-values are between baseline and the last 10 minutes after induction;). In contrast, the same protocol, when delivered to the GCL, produced a LTP of the proximal excitatory inputs in 7 out of the 8 cells (153.2±4.1%, n = 7, *p*\<0.001, paired-samples t test;). These two forms of synaptic plasticity were not due to a rundown of the EPSPs, because no persistent changes in the EPSP amplitude were observed when the TBS was absent (104.7±6.3%, n = 6, *p*\>0.05). In addition, the synaptic plasticity was not associated with any obvious changes in the input resistance or membrane potential. The resting potentials of the GCs detected in this study (approximately −60 mV) were in the normal range of the resting potential (−76 to −54 mV) reported by a previous study, indicating that the recorded GCs were healthy. Thus, distinct long-term plasticity of excitatory inputs to the GCs could be induced with identical TBS delivered to EPL or GCL. ## Modulation of inhibition onto the MCs by plasticity of the GCs Lateral inhibition in the olfactory bulb is largely mediated by reciprocal dendrodendritic synaptic connections between the MC lateral dendrites and the dendrites of inhibitory GCs. It is possible that plasticity at the excitatory inputs onto the GCs may alter their intrinsic property and/or driving force onto the MCs and, consequently, modulate the lateral inhibition onto the MCs. This potential modulation in lateral inhibition might contribute to the refinement of encoding or processing of olfactory information in the MCs. Thus, we further investigated whether the differential plasticity of GC excitatory inputs cause differential long-term plasticity of lateral inhibition in the MCs. The evoked IPSPs were recorded in the MCs by extracellularly stimulating the EPL or GCL (GCs→MCs synapses) before and after TBS (. 2D). These responses appeared to be mediated by GABA_A receptors, as they exhibited a reversed polarity near −70 mV and were blocked by gabazine (10 µM;). We only analyzed the responses from the MCs that showed clear inhibitory responses to single stimulations (see sample traces). The TBS depressed the IPSPs evoked by the EPL stimulation for at least 30 min (65.7±2.8%, n = 6; *p*\<0.001;). Moreover, using the same protocol, the TBS delivered on to the GCL potentiated the IPSPs (124.5±3.3%, n = 7; *p*\<0.001;). These two forms of synaptic plasticity were not associated with obvious changes in the input resistance or membrane potential. As a control, no persistent changes in the IPSP amplitude were observed when the TBS was absent (95.9±8.7%, n = 4; *p*\>0.05;). The resting potentials of the MCs detected in this study (approximately −55 mV) were in the normal range of resting potential (−65 to −47 mV) reported by previous studies , indicating that the recorded MCs were healthy. These results revealed that using an identical protocol, which induced plasticity with opposing directions in GCs when delivered to two excitatory GC inputs, could also elicit distinct long-term plasticity of inhibitory inputs onto the MCs. ## STDP of ON inputs In order to determine whether and how the plasticity of inhibitory inputs mediating lateral inhibition affects STDP of the ON inputs from sensory neurons, we need to further induce STDP in M/T cells. As a Hebbian synaptic learning rule, STDP has been previously demonstrated in various neural circuits, including the olfactory system. Induction protocols for STDP commonly consist of the repetitive pairing of single pre- and postsynaptic spikes at regular intervals. However, neuronal activity in vivo is far more complex, with a spectrum of activity level from almost no activity to short bouts of high- frequency spike bursts. Therefore, the pairing of the presynaptic spikes with the postsynaptic bursts sometimes represents a more natural situation in vivo. The lateral inhibition from the GCs may prevent or delay firing or may decrease the firing rate during stimulus presentation in MCs. Because STDP is a form of synaptic plasticity whose induction is heavily dependent on the timing of the incoming spikes and the frequency of the postsynaptic spikes, the presence of lateral inhibition in the postsynaptic MCs may profoundly influence the STDP induction and the total olfactory bulb output. We initially found that the repetitive pairing of the EPSPs with single spikes at 0.1 Hz failed to induce any persistent changes in the EPSPs of the ON inputs in MCs (100.4±7.0%, n = 7; *p*\>0.05, ; 108.5±6.8%, n = 7; *p*\>0.05). In contrast, the repetitive pairing of the EPSPs with postsynaptic bursts consisting of three postsynaptic APs within a critical time window induced persistent modifications in the EPSPs in MCs (at ON inputs→MCs synapses;). The EPSPs were evoked by stimulating the glomeruli associated with the recorded MCs, and could be blocked by co- application of NMDA- and AMPA-type glutamate receptors antagonists such as AP5 (50 µM) and NBQX (20 µM), indicating that they were mediated by glutamatergic neurotransmission. The postsynaptic APs were elicited by a current injection. The basal level of ISI of the postsynaptic bursts was set to 19–21 ms (20.0±0.3 ms when Δ*t* = +30 ms and 20.1±0.3 ms when Δ*t* = −50 ms; see) by adjusting the intensity of the injected current. The Δ*t* was defined as the time between the onset of the EPSP and the onset of the first AP in the burst. The pairings were repeated 60 times at 0.1 Hz. We found that the EPSPs potentiated in 6 out of the 8 cells when the Δ*t* was set to +30 ms (143.7±2.5%, n = 6; *p*\<0.001;), whereas it was depressed when the Δ*t* was set at −50 ms (74.3±1.3%, n = 6; *p*\<0.001;). These two forms of synaptic plasticity were not associated with obvious changes in the input resistance or membrane potential. The entire time window for the induction of potentiation and depression of excitatory ON inputs was then determined. These results indicate that STDP of excitatory OSN inputs to MCs could be induced by pairing EPSPs with postsynaptic spike bursts within a critical time window. It has been reported that stimulation of ON induced long-lasting depolarizations (LLDs) that lasted as long as several seconds. The LLDs may only occur synchronously in cells whose apical dendrites ramify within the same glomerulus, suggesting that the LLDs involved intraglomerular interactions among the M/T cells. Therefore, glomeruli stimulation in this study might also induce LLDs in the correlated MCs. Indeed, occasionally we observed LLD-like responses in some of our results. These responses were initiated by a fast and graded monosynaptic input from the OSNs followed by a slower component. To ensure that the EPSPs we detected were mediated by monosynaptic neurotransmission, we discarded those EPSPs with multiple peaks and only used those that displayed single-peak responses. In addition, because ON stimulation with high intensities more frequently produces LLDs, we deliberately controlled the stimulation intensity within moderate level in most of our recordings. Thus, the responses we report here represent monosynaptic inputs to the MCs. ## Regulation of plasticity in the ON inputs After confirming the feasibility of the STDP induction, we further investigated whether and how the prior plasticity of lateral inhibition affects the subsequent STDP of ON inputs from the OSNs. We began to collect whole-cell baseline EPSP data in MCs approximately 10 min after the prior stimulation was finished. The baseline EPSP recording in MCs lasted for 10 min and was followed by a STDP induction. Thus, the interval between the end of the prior TBS stimulation and subsequent STDP induction was 20 min. The prior TBS protocol, which was used to induce LTP or LTD of lateral inhibition when delivered to the proximal or distal inputs to GCs, respectively, did not affect the baseline EPSPs of ON inputs in MCs (OSN→MCs synapses; 106.8±3.2%, n = 6; *p*\>0.05, ; 96.7±1.9%, n = 6, *p*\>0.05;). However, it modified the predisposition for the subsequent spike timing-dependent LTP or LTD induction in MCs. The protocol used in EPL to induce the LTD of lateral inhibition facilitated the subsequent LTP of EPSPs on ON inputs by elevating the magnitude of potentiation (186.0±2.4%, n = 6; compared with control, *p*\<0.001; Independent-samples t test;), whereas it suppressed LTD by reducing the magnitude of depression (99.8±1.2%, n = 6; compared with control, *p*\<0.001; Independent-samples t test;). In contrast, when the same stimulating protocol was delivered to the proximal inputs onto the GCs to induce LTP of lateral inhibition, it suppressed the subsequent LTP of EPSPs on ON inputs by reducing the magnitude of potentiation in MCs (109.1±2. 2%, n = 6; compared with control *p*\<0.001; Independent-samples t test;), whereas it facilitated LTD by elevating the magnitude of depression (50.1±1.5%, n = 6; compared with control, *p*\<0.001; Independent-samples t test;). These results reveal the bidirectional regulation of STDP by prior TBS of dictinct inputs onto GCs and suggest the changes in STDP may be caused by plasticity originated in GCs. If the changes in STDP are truly caused by plasticity originated in GCs, then blockade of this plasticity should also suppress the changes in STDP. To test this possibility, we treated the slices with antagonists of AMPA- and NMDA-type glutamate receptor only during TBS to block the glutmatergic neurotransmission. This treatment totally abolished the induction of plasticity in GCs. (EPL 94.0±2.1%, n = 5; compared with baseline *p*\>0.05; ; GCL 91.6±3.5%, n = 4; compared with baseline *p*\>0.05;). As a result, no subsequent changes in STDP were observed (EPL 132.9±8.0%, n = 6; compared with control *p*\>0.05; ; GCL 140.2±5.3%, n = 7; compared with control *p*\>0.05; ANOVA *LSD* test;). Taken together, these data strongly suggest that the prior plasticity history of lateral inhibition driven by the two distinct excitatory inputs to GCs exerted differential regulation on the subsequent plasticity of excitatory inputs from the OSNs. ## Changes in the ISI is the causative factor for regulation of plasticity What is the mechanism that underlies the regulation of plasticity in ON inputs? It has been reported that recent activity history could affect the lateral inhibition between MCs and the lateral inhibition may alter MC spike-timing. In addition, the frequency of postsynaptic spikes within the pairing protocol can affect the induction and magnitude of STDP, most likely due to a change in the back propagation of spike trains. This raises the possibility that the regulation of STDP exerted by prior plasticity in lateral inhibition was achieved by regulating the ISI or frequency of APs in the burst. Therefore, we further examined in MCs whether and how the prior stimulating protocol affected the subsequent spike patterns within the induction protocol for STDP. Here, the TBS on the EPL or GCL was defined as the priming stimulation and the recordings were performed on the MCs. Interestingly, we found that prior EPL stimulation of distal inputs onto the GCs significantly shortened the ISI of the three-spike- burst at the time-point when the subsequent pairing protocol was delivered (ISI 18.0±0.3 ms, n = 8; compared with baseline 20.5±0.2 ms, *p*\<0.001; paired- samples t test;). In contrast, an identical prior stimulating protocol delivered to GCL prolonged the ISI (22.6±0.5 ms, n = 7, compared with baseline, *p*\<0.001;). As a control, the ISI of the spike burst was kept unchanged when no prior stimulation was delivered (20.9±0.5 ms n = 6, compared with baseline 20.8±0.5 ms, *p*\>0.05;). These results suggested that the regulation of the subsequent STDP may be correlated with coincident changes in the ISI within the STDP induction protocol. Although these results suggest that the changes in the ISI and the subsequent regulation of LTP/LTD may be correlated, we still do not know whether the changes in the ISI are the causative factor for the regulation of STDP. It has been reported that partially up- or down-regulating GABA_A receptor function with its agonist or antagonist can bidirectionally regulate the ISI within a spike burst,. Therefore, to examine this possibility, we used a selective GABA_A receptor agonist or antagonist to determine how it affected the pattern of spikes within the induction protocol. Gabazine (GBZ) and isoguvacine (ISO) are highly selective GABA_A receptor antagonists and agonists, respectively. We first investigated the dose-response relationship of these two agents and determined the concentration that matched the ISI regulation exerted by the previous stimulating protocol. We found that a partial blockade of GABA_A receptor function with GBZ at 1.5 µM displayed similar effects on the ISI exerted by a prior stimulation at the EPL, whereas a submaximal activation of GABA_A receptor function with ISO at 3.0 µM produced comparable effects on the ISI exerted by a prior stimulation at the GCL. This partial manipulation of GABA_A receptor function could mimic, to a certain extent, the prior potentiation or depression of lateral inhibition the MCs received from the TBS. It was reported that the GABA_A receptor blockade increased the excitatory MC responses to odors, suppressed the adaptation of MC firing rate and disrupted the animal's ability to distinguish between similar odors. However, the concentration of drugs used in this study were largely decreased compared with previous studies , and we determined that GBZ at this level failed to clearly or persistently affect the cell excitability, manifested by the absence of changes in the membrane potential or synaptic responses (n = 4, ANOVA *LSD* test, *p*\>0.05). Thus, we largely avoided these unwanted side effects. Then we examined how these treatments could affect the induction of the STDP. GBZ (1.5 µM) facilitated spike timing-dependent LTP (tLTP) by increasing the magnitude of potentiation at OSN→MCs synapses (172.3±4.7%, n = 6, *p*\<0.001, Independent-samples t test;), but suppressed spike timing-dependent LTD (tLTD) by decreasing the magnitude of depression (93.6±1.5%, n = 6, *p*\<0.001;). In contrast, the ISO at 3.0 µM suppressed tLTP induction by decreasing the magnitude of potentiation (92.9±1.5%, n = 6, *p*\<0.001;), but facilitated tLTD by increasing the magnitude of depression (58.8±1.2%, n = 6, *p*\<0.001;). These results demonstrate that mimicking ISI modification with a GABA_A receptor agonist or antagonist could cause a similar modification of the predisposition for subsequent STDP induction. To further ensure that the change in the ISI was the causative factor for plasticity modification, we injected three shorter current steps to the recorded MCs so that each current step was sufficiently strong to trigger a single action potential. When we set the ISI to 17.4 ms and 22.9 ms to mimic the bidirectional changes in the ISI following the prior stimulations, we were able to observe similar changes in the STDP (192.4±3.2%, n = 6, compared with control: 165.5±15.4% *P*\<0.001, ; 97.7±2.0%, n = 6, compared with control: 78.8±7.4% *P*\<0.001, ; 95.8±2.4%, n = 6, compared with control: 165.5±15.4% *P*\<0.001, ; 61.3±1.0%, n = 6, compared with control: 78.8±7.4% *P*\<0.001). These results suggest that the changes in ISI may contribute to the regulation of STDP. The regulation of the ISI that was present continually after the TBS did not require acute GC stimulation, suggesting that TBS causes a tonic change in inhibition. To further examine whether the regulation of the ISI depend on a tonic change in inhibition that was continuously present after TBS, we also monitored miniature IPSPs (mIPSPs) before and after TBS and found that the TBS at EPL suppressed the mIPSPs amplitude in MCs (control, 1.68±0.15 mV, after EPL TBS, 1.38±0.09 mV; *p*\<0.01;). These results provided further evidence that the TBS induces tonic and continuous regulation of the ISI and suggest that the changes in the ISI by prior plasticity of lateral inhibition may be the causative factor for the regulation of STDP. Moreover, it also suggests that modification in the GC-to-MC synapse may take a role in changes of inhibition. If the changes in ISI really contribute to the regulation of STDP, then reversing the changes in ISI by resetting ISI back to control level should also abolish changes in STDP. To this purpose, we either employed pharmacological intervention of ISI with GABA_A agonist/antagonist or directly reset the ISI with the three individual APs to get the ISI back to control levels (around 20 ms) after TBS. Following the EPL priming stimulation, which induced LTD of lateral inhibition, we applied 3.0 µM of ISO immediately before and during the pairing stimulation protocol for STDP induction. This treatment elevated the ISI value in MCs, which was already decreased by the prior EPL stimulation, to no-priming control levels (19.6±0.2 ms, n = 6, compared with control, 20.0±0.3 ms, *p*\>0.05;). As a result, the induction with a pairing protocol failed to elicit any persistent changes in the magnitude of spike timing-dependent LTP or LTD at OSN→MCs synapses, similar to controls without prior stimulation (LTP 152.9±1.9%, n = 6, *p*\>0.05, ANOVA *LSD* test; LTD 77.9±1.4%, n = 6, *p*\>0.05;). Similarly, treating with 1.5 µM GBZ following GCL priming stimulation decreased the ISI value, which was already increased by prior GCL stimulation, to no-priming control levels (20.8±0.4 ms, n = 6, compared with control 20.4±0.5 ms, *p*\>0.05;). As a result, the induction with the pairing protocol failed to elicit any persistent changes in the magnitude of LTP or LTD (LTP 141.7±2.9%, n = 6, *p*\>0.05, ANOVA *LSD* test; LTD 73.2±1.4%, n = 6, *p*\>0.05;). Moreover, adjusting the current injection level after prior TBS to convert the ISI back to control levels also reversed the STDP to control levels (147.5±1.9%, compared with control, *p*\>0.05, ; 145.2±1.1%, compared with control, *p*\>0.05). Taken together, these results further confirm that the change in the ISI is the causative factor for regulation of plasticity. # Discussion In the olfactory system, few studies have investigated the regulation of plasticity in olfactory sensory input by prior activity from GCs. This study demonstrates that using an identical TBS stimulation delivered to the distal or proximal inputs could induce persistent modification in the inhibition to MCs with opposing directions. This bidirectional modification of the inhibitory inputs differentially regulated the subsequent synaptic plasticity of excitatory ON inputs to the MCs. Because the modulation of the inhibition onto the MCs could alter the firing rate of spikes, the observed changes in the ISI caused by the plasticity in the excitatory inputs of GCs could be ascribed to several scenarios. First, the differential modification of drives in GCs to MCs is the simplest explanation for the TBS-mediated persistent changes of mitral cell inhibition. Second, the plasticity in the GC-to-MC synapse may also be involved. Third, the regulation of GC spiking by plasticity of excitatory inputs may relieve the Mg²⁺ blockade of NMDA receptors at the dendrodendritic synapses and, thus, dynamically regulate the inhibition of the MCs. The neurons that fire high-frequency bursts of spikes were found in various sensory systems including the olfactory system. The firing of the bursts in response to sensory input depends on the intrinsic cellular mechanisms that function with feedback from higher centers to control the discharge properties of these neurons. A growing number of studies indicate the possibility that the bursts possess a distinct function in the transmission of sensory information. It has been observed that the intrinsic membrane properties favored MCs firing at 40 Hz and the MC discharge was also stabilized at a preferred frequency of 40 Hz. In addition, additional bursts of action potentials may be triggered by the synaptic inputs locked to the air inhalations. Moreover, it has been proposed that neuronal oscillations enhanced stimulus discrimination by ensuring AP precision, and the maintenance of AP precision could be compromised at oscillation frequencies higher than 50 Hz. These findings imply that the natural spike frequency in MCs in vivo that permits high levels of optimal stimulus discrimination is approximately 50 Hz. In this study, the STDP in MCs was induced by the pairing of presynaptic EPSPs with realistic 50 Hz postsynaptic bursts within a critical window of tens of milliseconds. The changes in synaptic strength depend on the inter-burst interval rather than the precise timing of the individual spikes. These burst timing-dependent plasticity rules may be specifically beneficial for the circuits in which the information relevant for synaptic refinement is contained in the timing of the bursts rather than that of the individual spikes. Given the very precise dependence of the magnitude of STDP on the ISI, a slight change in the ISI in MCs may produce significant implications in the output of the olfactory bulb and, thus, may have a particular physiological relevance in the olfactory system. In the developing olfactory bulb, the inhibitory synapses distributed along the secondary dendrites of the MCs can dynamically regulate the extent of spike propagation, with a smaller activation of the inhibitory synapses facilitating the spike propagation. The lateral and recurrent inhibitions in the olfactory bulb play distinct roles in shaping the MC spiking pattern, which is critical to odor information processing. Both the two forms of inhibition are thought to be important in odor discrimination and in the generation and synchronization of odor-evoked rhythmic MC activity. Although the inhibition we observed in this study is largely driven by the excitatory inputs on to the GCs and thus may largely represent the lateral inhibition, we could not exclude the potential contribution of recurrent inhibition. The two types of anatomically distinct excitatory inputs on to the proximal and distal dendrites of GCs may exert different functions through distinct mechanisms. The excitatory input onto the distal dendrites in the EPL mediates local dendrodendritic inhibition, whereas the excitatory inputs onto the proximal dendrites at least partially, if not completely, are cortical feedback inputs that mediate global top-down modification of the DDI. Using a two-photon guided minimal stimulation in acute rat brain slices, Strowbridge and colleagues positioned an extracellular stimulating electrode near a specific dendritic segment and thus were able to activate relatively homogenous populations of presynaptic processes, based on the kinetic properties of the resulting postsynaptic currents. Although we did not use a similar technique, we were able to deliberately control the intensity of the local stimulation to ensure the independence of the stimulation on distinct sites in this study. Moreover, the synaptic responses we obtained from these two inputs also displayed distinct properties in the current kinetics, similar to the findings obtained from Strowbridge's group. Therefore, we could still activate a relatively homogenous population of presynaptic processes when stimulating either of the two sites. Because the extent of GC function in the local versus global output modes could exert a critical effect on the computational role it performs in the olfactory bulb circuit, our present findings of the modification of excitatory ON inputs by prior plasticity of the distinct excitatory inputs to GCs reveal an efficient route to regulate the encoding and processing of odor information in the bulb. # Supporting Information [^1]: Conceived and designed the experiments: WL. Performed the experiments: TM XZ LC NZ SR FJ. Analyzed the data: TM XZ. Contributed reagents/materials/analysis tools: TT. Wrote the paper: WL. [^2]: The authors have declared that no competing interests exist.

# Introduction Hepatic sinusoids are capillary bed in the liver consisted with sinusoidal endothelial cells, stellate cells, and Kupffer cells. Nutrient-rich venous blood and oxygen-rich arterial blood combines and flows through the sinusoids. Sinusoids are separated from adjacent hepatocytes by the space of Disse. The sinusoids run straight between the liver cell cords and communicate with each other though interconnecting sinusoids. The diameter of hepatic sinusoids is 7-15 µm and the diameter of pericentral sinusoid is larger than that of periportal sinusoid. The hepatic sinusoids are special in that they possess the sieve plate-like pores. Unlike non-hepatic capillaries that exchange molecules ranging between 0.5 and 12 nm by endocytosis and transcytosis, hepatic sinusoidal endothelium could exchange molecules ranging up to 150-175 nm through the sieve plate-like pores. Under normal physiological conditions, the ingested toxins are transported into hepatocytes and detoxified by penetrating the sinusoids of the hepatic microcirculatory system. However, under pathological conditions, morphological changes in the sinusoids impair hepatic microcirculation and induce functional damage. Recently, structural alterations such as the diameter, surface area, and fenestrae of the sinusoids have been reported in acute and chronic liver injury and liver cancer. However, there still lack quantitative nondestructive imaging studies on normal and diseased liver models to reveal the structural difference. With increased interest in the significance of the hepatic sinusoids, significant efforts have been made to investigate sinusoidal structure, but light and confocal microscopic studies on the sinusoids are incomplete compared with nondestructive three-dimensional (3D) imaging. To date, nondestructive structural studies of hepatic microvasculature have been performed by micro- computed tomography (micro-CT) using a conventional X-ray source. Conventional micro-CT nondestructively acquires 3D images from thick tissue; however, the image resolution is insufficient for observing the sinusoids. These studies, therefore, are restricted to the hepatic biliary, arterial, and venous systems. Synchrotron radiation micro-CT provides spatial resolution of 1 µm and is a feasible imaging system for studying sinusoids and their alterations in 3D space. Recently, 3D images of the cortex microvasculature in the brain at 1 µm resolution have been visualized, and the differences between wild-type and APP23 knockout mice were successfully observed. Considering the diameter of the capillaries in the mouse brain, the spatial resolution of synchrotron radiation micro-CT is sufficient to detect the sinusoids with diameters of 7-15 µm in the mouse liver. The present study applied synchrotron radiation micro-CT to investigate the 3D structure of hepatic sinusoids and their alterations associated with acute liver injury. # Materials and Methods ## Materials and animals Carbon tetrachloride (CCl₄), corn oil, propylene oxide, and phosphate buffered saline (PBS) were purchased from Sigma (St. Louis, MO). Osmium tetroxide (OsO₄), glutaraldehyde, and epoxy resin were obtained from Electron Microscopy Sciences (Fort Washington, PA). Male, 6-week-old C57BL/6 mice were maintained under specific pathogen-free conditions. All animals received humane care, and the experiments were approved by the Institutional Animal Care and Use Committee at Pohang University of Science and Technology, Pohang, Republic of Korea (approval number: 2011-01-0015). ## CCl₄-induced liver injury and specimen preparation The CCl₄ was prepared as a 50% (vol/vol) solution in corn oil, and a single dose of 2 ml/kg of body weight was administered by intraperitoneal injection (n = 5) to induce hepatic injury. Control animals were injected with corn oil alone (n = 5). Groups of age- and sex-matched mice were sacrificed on day 1 after acute CCl₄ treatment. Liver tissues were obtained from five mice in each group at the indicated times and then fixed with 2.5% glutaraldehyde in PBS. The right hepatic lobe from each mouse was processed for histological analysis and synchrotron radiation micro-CT. Liver sections (5 µm) were stained with hematoxylin and eosin for general histology and examined under a light microscope. For synchrotron radiation micro-CT, liver sections (2-3 mm) were stained in 1% OsO₄ for 4 h, dehydrated in an ethanol series, and embedded in an epoxy resin in 1.5 ml Eppendorf tubes. ## Instrumentation All experiments were performed with polychromatic X-ray beams (4-15 keV energy range) at the 7B2 X-ray microscopy beamline of the Pohang Light Source (2.5 GeV), Republic of Korea. A silicon shutter was used as an attenuator to control total X-ray flux. The detector system consisted of a thin (100-250 µm) and cleaved CdWO₄ single crystal scintillator and a charge-coupled device (CCD)-based video camera. The image displayed on the scintillator was converted to a visible image and then magnified by an optical lens before it was captured by the CCD. Specimens embedded in resin were mounted on a sample holder and typically placed 100-200 mm apart from the detector to achieve optimum contrast. Scanning was performed by rotating the specimen in increments of 0.18^o in an X-ray beam and acquiring an X-ray transmission image at each angle of view. Three-dimensional image reconstruction from 1000 projections was carried out using a standard filtered back projection algorithm (Image Pro Plus; Media Cybernetics, Silver Spring, MD). Reconstructed tomographic slice images consist of 1600 × 1200 pixels in the horizontal and vertical directions and the horizontal field of view is 580 µm. Using synchrotron radiation micro- CT, images with voxel size of approximately 1 µm are achievable. ## Analysis of the reconstructed image data As the X-ray images in this study contained both hepatic microvasculature and surrounding parenchymal cells, the desired area to be displayed and analyzed was segmented. Using the segmentation application of Amira software (Mercury Computer Systems, Chelmsford, MA), each pixel of the image was assigned a label describing the region to which it belonged. Pixels in tomographic slices were denoted as parenchymal area, sinusoidal area, central venous area, or portal venous area. Each labeled region was surface generated for 3D visualization using the Amira software based on human-computer interaction. Briefly, four regions were manually segmented by different gray scale values of the microstructure and its morphological features. In OsO₄-stained liver, the vessels had a lower gray scale value than that of parenchyma, indicating that veins and sinusoids can be distinguished from parenchyma. Moreover, central vein and portal vein were distinguished by the presence of other branches of the portal tracts such as hepatic artery, bile duct, and lymphatic vessels. In 3D volume-rendered images of liver tissues in normal and CCl₄-treated mice, diameters, surface areas, and volumes of pericentral and periportal sinusoids were measured. ## Statistical analyses All values are expressed as means ± S.D. *P* values were calculated from Student’s t tests, based on comparisons with the appropriate control samples tested at the same time. # Results ## Identification of the sinusoid in normal liver We performed synchrotron radiation micro-CT on normal liver tissue stained with or without an X-ray contrast agent, OsO₄ (. In the phase contrast images without OsO₄ staining, the overall structure of the hepatic microvasculature was ambiguous, and we could not detect the hepatic sinusoids (. In absorption contrast images, we observed two different shapes of vessel-like structures on the OsO₄-stained liver tissue (. One was circular with a diameter larger than 10 µm, and the other was rod-shaped with a diameter less than 10 µm. By comparing the representative tomographic slice image and the histological image stained with hematoxylin and eosin, we identified the vessel- like structure larger than 10 µm as a vein and one less than 10 µm as a sinusoid (. These data indicate that the absorption mode of synchrotron radiation micro- CT enables detection of the sinusoids. Additionally, based on the similar diameter and distribution of the sinusoids between the histological and X-ray images, the structures of the hepatic sinusoids were visualized by absorption contrast X-ray imaging. ## Identification of the sinusoids and veins in normal and CCl₄-treated liver We observed necrotic regions and disrupted sinusoidal structures in the histological images of normal and CCl₄-treated liver (. We performed synchrotron radiation micro-CT on OsO₄-stained liver tissue obtained from normal and CCl₄-treated mice, and acquired the absorption contrast images (. Necrotic regions (hepatic sinusoids) could be distinguished from non-necrotic regions on the absorption contrast images from acute CCl₄-injured mice. These findings suggest that structural alterations of the sinusoids in pathologically injured liver are detectable using the absorption mode of synchrotron radiation micro-CT. In particular, disease- specific pathological features such as necrosis and disrupted hepatic sinusoidal structures in necrotic areas were detected on absorption contrast images. ## Identification of pericentral sinusoid disruption in CCl₄-treated liver We next investigated the structural alterations of the hepatic sinusoids in the acute CCl₄ model using the histological and absorption contrast X-ray images from the pericentral and periportal areas ( and. Disrupted sinusoids in the necrotic region were detected in histological images of the pericentral area of CCl₄-injured liver compared with normal liver (. When we segmented the pericentral sinusoids ( and the central vein ( in the absorption contrast images, we found that the sinusoidal structure was disrupted, but that the vein was protected from CCl₄ damage in the CCl₄-injured liver. No significant difference in the veins or sinusoids of the periportal area was observed between normal and CCl₄-treated livers (. Because we used a severe CCl₄ injury model, the midzonal and some other sinusoids were disrupted, as shown in the absorption contrast X-ray images (. These data demonstrated that the particular sinusoidal alterations can be identified using the absorption mode of synchrotron radiation micro-CT. ## 3D structure of the sinusoids and veins in normal and CCl₄-treated liver We next investigated the 3D structure of the hepatic sinusoids and veins in normal and CCl₄-treated mice (. No significant difference in the sinusoidal structure in the periportal area was observed between normal and acutely injured mice (. In contrast, the 3D image of the pericentral sinusoids in CCl₄-treated mice showed sparser and more disconnected vascular network compared with that in the normal mice (. Thus, disrupted structures of pericentral sinusoids can be distinguished from undisrupted periportal sinusoids in 3D images of the hepatic sinusoidal network of acute CCl₄-injured liver, as observed by synchrotron radiation micro-CT. ## Quantitative analysis of hepatic sinusoids in normal and CCl₄-treated liver In 3D volume-rendered images (diameters of pericentral and periportal sinusoids were measured in normal and CCl₄-treated liver (. In normal liver, sinusoidal diameters of pericentral and periportal area were 13.7 ± 1.4 µm and 8.8 ± 2.4 µm, respectively. The diameter of pericentral sinusoid was larger than that of periportal sinusoids. After acute CCl₄ treatment, there was no significant difference in the diameter of periportal and pericentral sinusoids compared with control (. We also performed a quantitative analysis on volume-rendered 3D image of normal and CCl₄-treated liver using morphological parameter including surface area and volume. The results of measured surface area and volume of the sinusoid and parenchyma in normal and CCl₄-treated liver are given in. The total volume of the parenchyma was almost the same in normal and CCl₄-treated liver (both about 20.2 × 10⁶ µm³); whereas the volume of the sinusoid differed; 2.64 ± 0.54 × 10⁶ µm³ (pericentral area) and 2.60 ± 0.27 × 10⁶ µm³ (periportal area) in normal liver and 0.28 ± 0.05 × 10⁶ µm³ (pericentral area) and 0.96 ± 0.25 × 10⁶ µm³ (periportal area) in CCl₄-treated liver. The sinusoidal volume/parenchymal volume ratio in periportal and pericentral area was significantly lower in CCl₄-treated mice compared with control, indicating the reduction in the sinusoidal volume fraction by acute CCl₄ treatment (. However, no significant difference was observed in the sinusoidal volume/sinusoidal surface area ratio between normal and CCl₄-treated mice (. The low value of sinusoidal volume/sinusoidal surface area indicates a decrease in the number of sinusoids as the diameter of hepatic sinusoids appeared to be similar. # Discussion We demonstrated that the absorption mode of synchrotron radiation micro-CT can produce 3D images of hepatic sinusoids and their alterations in normal and injured mouse livers. Furthermore, pathological features of acute liver injury, such as the disruption of pericentral sinusoids in necrotic regions, were detectable in the 3D images from synchrotron radiation micro-CT. Synchrotron radiation X-rays generally use absorption or phase contrast effects when visualizing the angioarchitecture of tissue. Synchrotron radiation X-rays with high coherence can induce a phase contrast effect through edge enhancement. In absorption mode, the angioarchitecture of the tissue can be visualized by microinjecting an X-ray contrast agent into the vessels or staining the surrounding tissues. As conventional X-ray contrast agents, including radiopaque polymers, fill only the lumen side of the blood vessel, the vascular architecture and the interrelationships among vessels are difficult to visualize. Recently, Ritman et al. suggested the use of OsO₄ as an X-ray contrast agent for micro-CT because it enhanced visualization of the coronary artery wall structure and nephron architecture, and their interrelationships. We visualized the 3D hepatic sinusoidal structures using the absorption mode of synchrotron radiation micro-CT with OsO₄ as an X-ray contrast agent. The presence of OsO₄ resulted in lower absorption by vessels than by parenchyma in the absorption contrast images of mouse livers. The liver, which consists of many structural and functional units called lobules, is the main metabolic organ in the body. Detoxification is conducted in each lobule of the liver, and thus understanding microvascular alterations of the lobule in three-dimensions during liver injury and regeneration is significant. Several imaging techniques have been applied to visualize the angioarchitecture of the lobule at 3D space in order to elucidate the microvasculature of the liver lobule under physiological conditions. Although light or confocal microscopy has sufficient spatial resolution to visualize the sinusoids, the penetration into thick specimens is limited. Thus, light and confocal microscopic studies on the liver require thin tissue sections, so the depth of the observable tissue is restricted. The 3D angioarchitecture of the lobule has been investigated at the level of the sinusoids using confocal laser scanning microscopy, but the depth of the specimen is limited to about 150 µm thick liver tissue slices. In contrast, we observed the 3D angioarchitecture of the liver lobule at the level of the sinusoids with liver tissue sections of 2-3 mm thickness using synchrotron radiation micro-CT. Moreover, synchrotron radiation X-rays with a spatial resolution of 1 µm is sufficient to detect the sinusoids with diameters of 7-15 µm in the mouse liver. In our study, we observed that hepatic sinusoids in the normal liver were directly branched off from the portal veins. Moreover, hepatic sinusoids were running between the liver cells and interconnecting sinusoids were also present between the sinusoids. In addition, consistent with previous reports, sinusoidal diameters of pericentral and periportal area were 13.7 ± 1.4 µm and 8.8 ± 2.4 µm, respectively, and pericentral sinusoids were larger than periportal sinusoids. In both the two-dimensional and 3D images of synchrotron radiation micro-CT, we distinguished disrupted pericentral sinusoids from undisrupted periportal sinusoids in the acute CCl₄-injured liver. Injury to the hepatic microvasculature induced by CCl₄ occurs via the cytochrome P450s. These enzymes reside in the parenchyma and generate a concentration gradient from the central to the portal vein. Thus, CCl₄ treatment causes necrosis of pericentral sinusoids. OsO₄ strongly binds to the lipids, and CCl₄ induces lipid alterations in necrotic regions. Therefore, when OsO₄ is used as an X-ray contrast agent, the absorption contrast of necrotic regions (pericentral sinusoids) should be higher than that of non- necrotic regions (periportal sinusoids) in acute CCl₄-injured liver. Recent studies on structural alteration of hepatic sinusoid in acute CCl₄ model have been conducted by light microscopy, confocal laser microscopy, and intravital fluorescence microscopy. After acute treatment with CCl₄, sinusoidal diameter was significantly smaller than that in controls and sinusoidal area was significantly reduced in CCl₄-treated mice. Moreover, pericentral sinusoids are scattered in the necrotic region. Our results of 3D volume rendering image were identical to the previous studies in that the volume and the vessel connectivity of the pericentral sinusoid were significantly reduced and disconnected, respectively. Because our model was acute CCl₄ injury model, we were not able to observe the fibrotic and cirrhotic region of the liver, observed in chronic CCl₄ injury model. These fibrotic and cirrhotic regions can be observed using our system of synchrotron radiation micro-CT combined with OsO₄ staining. Therefore, in the future, we are planning to visualize three-dimensionally, the extensive fibrotic and cirrhotic regions of the chronic liver models in both mouse and human liver. For decades, structural alterations of hepatic sinusoidal endothelial cells having sieve plate-like pores have been studied by scanning and transmission electron microscopy. The limitation of our study in terms of anatomical imaging was that the spatial resolution of synchrotron radiation micro-CT is not sufficient to investigate the unique morphology of the hepatic sinusoidal endothelial cells. Synchrotron radiation micro-CT, on the other hand, is a feasible imaging system to investigate the diameter and distribution of hepatic sinusoids rather than the structural alterations of hepatic sinusoidal endothelial cells. Further improvement of this imaging system with a subnano spatial resolution will make possible the 3D visualization of characteristic morphology of sinusoidal endothelial cells in normal and pathological conditions. In our study, to visualize the hepatic sinusoid, liver tissues had to be postfixed and stained in OsO₄ to give an absorption contrast and embedded in resin to protect the tissue from high flux X-rays. This fixation and staining made simultaneous functional imaging of the liver difficult in our system. In addition, in the processes of fixation, postfixation with OsO₄, and dehydration, the shrinkage and the collapse of morphological structure in the specimen can be induced. However, we could not detect any artifacts in the overall structure of liver tissue when our results were compared with histological images. The combination of histological and confocal imaging of the liver tissues with synchrotron radiation X-ray microscopy can be applied to further examine the possible artifacts. In conclusion, we nondestructively identified the hepatic sinusoids and their structural alterations in acutely injured mouse liver using the absorption mode of synchrotron radiation micro-CT for the first time. The 3D structure and their alterations of the sinusoids could be identified in diverse hepatic diseases including fibrosis, cirrhosis and liver cancer in human. We are grateful to Mr. Gwang Pil Park for editing this manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: YJY YKK JHB YSG. Performed the experiments: YJY SC. Analyzed the data: YJY SC BKK JP YKK JHB YSG. Contributed reagents/materials/analysis tools: JHL JYH YSG. Wrote the manuscript: YJY OYK BKK JHB YSG.

# Introduction Black Americans have faced disproportionately high infection rates, job losses, and loss of life during the COVID-19 pandemic. This population may be at greater risk of severe illness from COVID-19 due to higher prevalence of certain comorbidities compared to non-Black Americans, including high blood pressure and diabetes. Indeed, 33% of non-Hispanic (NH) Black Americans, compared to 27% of NH White Americans, demonstrated risk factors for severe COVID-19 illness. This has likely led to disproportionate rates of hospitalizations and deaths of NH Black Americans throughout the pandemic. Economic inequalities along racial and ethnic lines, worsened by the pandemic, compound these health risks. The Bureau of Labor Statistics reported that higher unemployment rates among non-Hispanic Black Americans persisted through October of 2020. Black Americans also more often hold jobs demanding on-site work: only 19.7% of non-Hispanic Black Americans reported they had the ability to work from home in 2017–2018, compared to 16.2% of Hispanic Americans and 29.9% of non- Hispanic White Americans and 37.0% of non-Hispanic Asian Americans. Intuitively, mental health impacts could follow the unequal disruptions of Black Americans’ physical and economic security. Two separate surveys utilizing non-probability panels with data weighted to U.S. demographic totals found higher rates of suicide consideration, substance use to cope with the pandemic, and suicide behavior risks among Black than White Americans early in the pandemic. Another non-probability study conducted after the stock market drop in March 2020 reported higher levels of anxiety related to economic hardship were among Black Americans than their counterparts. Despite this, no increase in symptoms of depression were observed in nationally representative probability surveys among Black Americans as lockdowns commenced in March and April 2020. Before the pandemic, Black Americans appeared to generally experience lower rates of anxiety and depression compared to other racial and ethnic groups, yet suffered higher prevalence of serious mental health issues. While they had lower lifetime rates of mood and anxiety disorders than White Americans and lower 12-month prevalence of generalized anxiety disorder, panic disorder, and social anxiety, Black Americans were in contrast more likely to meet criteria for PTSD and to have serious psychological distress than White Americans. Even among poor Black Americans, life satisfaction is higher and stress incidence is lower compared to White Americans. It is possible that this difference would continue after the onset of the pandemic, but on the other hand, mental health trends may not maintain the same pattern because of the aforementioned inequalities between racial and ethnic groups. Mental health outcomes during the COVID-19 pandemic should inform public health policy. To our knowledge, the trajectory of both depression and anxiety for racial and ethnic groups during the pandemic in the U.S. have not yet been examined using longitudinal data. The ability to recover from anxiety and depression over time may differ among racial and ethnic groups, considering the differences in economic resources accessible to them during the pandemic. Thus, our primary aims are to (1) analyze differences in odds of having depression and anxiety among Black and non-Black U.S. adults when state lockdowns were first implemented in March; and (2) explore changes in odds of developing depression and anxiety among Black and non-Black U.S. adults over time between March and November 2020. # Methods ## Procedures and methods ### Data Data for this study was from the Understanding America Study (UAS). Established at the University of Southern California (USC) in 2014, the UAS is a nationally representative, probability-based internet panel of U.S. households and consists of approximately 9,000 respondents aged 18 and above. Prior access to the internet is not a prerequisite to be in the panel; respondents without prior internet access are provided with a computer tablet and broadband internet subscription. Respondents complete surveys once or twice a month covering various substantive topics. Partly as a result of this, the UAS comprises a vast amount of background information on UAS respondents, including extensive measures of physical and mental health, income, labor force participation, cognitive functioning, and demographics. ### COVID-19 tracking survey Since March 10, 2020, all active members of the UAS panel have been invited to participate in an ongoing coronavirus tracking survey. Each UAS panel member participating in the tracking survey receives a survey every two weeks. Every day, about 500 respondents are invited to take the tracking survey for a total of about 7,000 respondents over a two-week period. The micro data, questionnaires, codebooks and reports are publicly available to any registered data users at the UAS data website. As of November 30, 2020, a total of 17 survey waves have been completed. Mean response rate across all survey waves is 83.63% (SD = 6%; Min = 76.35%; Max = 96.66%). ## Measures ### Mental health outcome measures The mental health outcome was measured by the well-validated Patient Health Questionnaire-4 (PHQ-4), a brief 4-question survey developed by Kroenke et al. and validated by Löwe et al. (Cronbach’s *α* = 0.82) to screen patients for depression (frequency of “feeling, down, depressed or hopeless,” and “little interest or pleasure in doing things”) and anxiety (frequency of “feeling nervous, anxious, or on edge” and “not being able to stop or control worrying”) over the past fourteen days. Response options for each item included “not at all,” “several days,” “more than half the days,” and “nearly every day.” Scores across the four items can be summed up to analyze the overall mental health outcome. Alternatively, each condition (depression, anxiety) can be scored and analyzed separately. We considered three binary outcomes in this study: showing symptoms for both depression and anxiety disorder if scores on the overall scale are ≥ 6 (probable depression and anxiety); showing symptoms for depression disorder if the scores on depression subscale are ≥ 3 (probable depression); and showing symptoms for anxiety disorder if the scores on the anxiety subscale are ≥ 3 (probable anxiety; 22). These cutoff values are indicative of major depressive disorder and generalized anxiety disorder with high sensitivity and specificity. We include the combined measure of symptoms for anxiety and depression as an indicator of overall mental distress as prior research has done. ### Race variable Respondents reported their race according to one of the following response options: White only, Black only, American Indian or Alaska Native only, Asian only, Hawaiian/Pacific Islander only, or Mixed. Respondents also reported whether they are Spanish, Hispanic or Latino. For comparing the mental health outcome across racial groups, we used these two pieces of information to create five racial groups: NH White (White); NH Black (Black); NH Asian (Asian); NH Other (Other); and Hispanic. We combined Whites, Asians, Others, and Hispanics into one group for comparing Black Americans vs. non-Black Americans. ### Baseline depression symptoms In earlier waves of UAS surveys, respondents also completed the 8-item Center for Epidemiologic Studies–Depression Scale (CES-D), which entered our regressions as control for baseline depression symptoms. Previous research has shown that CES-D scores are reasonably well-correlated with the PHQ-9, from which the PHQ-4 is derived. Scores range from 0–8. We used the most recent CES-D data before the start of Wave 1 of UAS COVID-19 Tracking Survey in March 10, 2020, as some respondents completed the 8 CES-D items more than once as part of the ongoing Health and Retirement Study (HRS) that is fielded in the UAS every two years. The time gap between the latest pre-pandemic CES-D assessment and the start of Wave 1 of UAS COVID-19 Tracking Survey on March 10, 2020 is about 12.75 months (S.D = 8.07 months; min = 0.033 month; max = 58.53 months). This indicates that on average the CESD assessment was taken one year prior to the onset of COVID-19 in March, 2020. We also consider a mid-pandemic CES-D assessment taken in late December 2020. ### Demographic variables We also controlled for a set of covariates, which included: age (18–34, 35–54, 55–64, or 65+); highest educational attainment (high school diploma or below, some college, or college and above), annual household income (\$0–29,999, \$30–59,999, \$60–99,999, or \$100,000 or more), gender (female or male), whether a household has children under 18 or not, employment status (currently employed; currently unemployed; or others that included retired, individuals on sick leave, disabled, others). UAS respondents’ key demographic information are updated every 3 months. More details about how the UAS respondents’ demographic information are updated are available at the following website: (<https://uasdat a.usc.edu/index.php?r=eNpLtDK0qi62MrFSKkhMT1WyLrYytFwwskuTcjKT9XIrM_JLi1Mz8nNSlK xrASlLDmM>). We use the most updated demographic information in the analysis. Information about employment status was collected in every wave of the UAS COVID-19 Tracking Survey since employment status changed more frequently for some subpopulation groups during the pandemic. *Risk perceptions*. To check the robustness of our main results, we also considered whether adjusting for risk perceptions in the models would change the outcome. Particularly, we analyzed responses to the following three risk perception questions asked in the UAS COVID-19 Tracking survey: “What is the chance that you will get the coronavirus in the next three months?”; “If you do get infected with the coronavirus, what is the chance you will die from it?”; and “What is the percent chance that you will run out of money because of the coronavirus in the next three months?” Responses to the three questions were provided on a validated visual linear scale ranging from 0% to 100%. *Data analyses*. All analyses applied survey weights, which adjusted for probabilities of sample selection and survey nonresponse; and aligned the UAS sample with distributions of key demographic variables in the U.S. civilian population in terms of age, gender, race/ethnicity, education, and location (see <https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3502405> for detailed information about how the UAS survey weights were constructed). All regression analyses control for the aforementioned demographic variables and baseline CES-D mental health score. We reported odds ratios, 95% CIs, and level of statistical significance for each logistic regression. Robust standard errors were used in cross-sectional logistic regressions, whereas cluster robust standard errors were used in longitudinal logistic regressions to allow for possible correlation in responses across survey waves within respondents. All analyses were conducted in 2020 using Stata version 15.0. # Results presents weighted distributions of demographic characteristics of the UAS COVID sample (counts within subgroups are in parentheses). In Wave 1, 8,815 UAS respondents were invited. Of those, 7,145 completed the survey, 54 started but did not complete the survey, and 1,616 did not start the survey, yielding a response rate of 81.06%. We conducted non-response analysis and found a slight imbalance in some demographic characteristics between responders and non- responders; this difference was not big enough to alter the conclusion of our study findings. Wave 1 respondent and non-respondent data can be assessed at the following UAS data webpage: <https://uasdata.usc.edu/index.php>. For comparison, we also provide weighted distributions of the same characteristics from the 2019 American Community Survey (ACS) for U.S. adults aged 18 and above. ACS data are 1-year estimates ([https://data.census.gov/mdat/ #/search?ds=ACSPUMS1Y2019](https://data.census.gov/mdat/%2523/search?ds=ACSPUMS1 Y2019)). The ACS is conducted by the U.S. Census Bureau every month, every year, and samples about 3.5 million U.S. households a year to create accurate population estimates. Overall, the ages of sample respondents ranged from 18 to 101 years old with a mean age of 48 (in terms of age groups, 24% of sample respondents were 18–34, 37% were 35–54, 18% were 55–64, and 20% were 65+). About 38% had a high school diploma or GED or below, 28% went to college but did not earn a degree, and 34% earned at least a college degree. As for racial and ethnic composition, 63% identified themselves as White, 12% as Black, 5% as Asian, 3% as Other, and 17% as Hispanic. About 27% came from households with annual household income less than \$30K, 27% between \$30K-\$60K, 23% between \$60K-\$100K, and 23% above \$100K. 52% of sample respondents are female, with 65% coming from households with no children under 18. The mean baseline CES-D score was 2.24 (SD = 2.01; Min = 0; Max = 8) and did not differ significantly between Black Americans (2.27) and non-Black Americans (2.24) (p-value = 0.757). In contrast, the mean mid-pandemic CES-D score was 3.05 (SD = 1.66; Min = 0; Max = 8) and was significantly lower for Black Americans (2.91) than non-Black Americans (3.06; p-value = 0.04), showing that both Black and non-Black Americans began in a similar position in terms of depression, but then diverged after the onset of the pandemic. presents distributions of the same demographic characteristic, split by non-Black vs. Black Americans. ## Cross-sectional analysis In, we presented the prevalence of probable anxiety and depression, anxiety, and depression for each of the subgroups. In, we present our estimates of adjusted odds ratios (AOR) for having probable anxiety and depression, anxiety, and depression. Data are from Wave 1 of the UAS COVID Tracking Survey, fielded between March 10–30, 2020. All logistic regression models adjusted for age (18–34, 35–54, 55–64, 65+), highest educational attainment (high school diploma or GED and below, some college, college and above), household income (\$0–29,99, \$30,000–59,999, \$60000–99,999, \$100,000 or more), gender (male, female), whether a household has children or not, employment status (employed, unemployed, others), and baseline pre-pandemic CES-D score. As previously mentioned, probable anxiety and depression was defined as scores on the overall scale ≥ 6, and probable depression (anxiety) was defined as scores on depression (anxiety) subscale ≥ 3. Demographic data were missing for some respondents in Wave 1 data, but we found no difference in PHQ-4 score between respondents with complete demographic data (n = 6,456) and respondents with incomplete demographic data (n = 476). As such, we treated the missing values as missing at random. Additionally, risk perception items were excluded from the final model as they did not affect the final results. Estimates from logistic regressions show that during the pandemic, Black Americans were significantly less likely than non-Black Americans to have probable anxiety and depression (AOR = 0.487; 95% CI = \[0.308, 0.771\]; P \< 0.01); probable anxiety (AOR = 0.474; 95% CI = \[0.319, 0.704\]; P \< 0.001); or probable depression (AOR = 0.522; 95% CI = \[0.330, 0.824\]; P \< 0.01). More directly, the odds of Black Americans reporting symptoms of either anxiety or depression are about half that of non- Black Americans, and the differences are statistically significant. We also compared odds of having probable anxiety and/or depression among non-Black racial and ethnic groups during the pandemic (i.e., Whites, Hispanics, Asians, and non-Black others) but found no significant differences among them adjusting for the same covariates presented in (results not shown). Odds ratios for covariates have expected signs: for instance, the odds of having probable anxiety and depression during the pandemic decline with age, household income, and increase with baseline CES-D scores and being unemployed or not working. With respect to the baseline CES-D score, our results indicate that for every one unit increase in the baseline CES-D score, the odds of having probable anxiety and depression, anxiety, or depression in a given month during the pandemic increases by about 35%, 31%, and 39%, respectively, consistent with findings from prior studies. ## Longitudinal analysis Since the UAS COVID Survey collected data on the same individuals over time, we were able to assess whether differences in mental health outcome between Black Americans and non-Black Americans we observed in March persisted over time during the pandemic. In, we present our estimates of AORs from logistic regressions for having probable anxiety and depression in the first column; probable anxiety in the second column; and probable depression in the third column, for each month from March to November. Data are from all 17 waves of the UAS COVID Tracking Survey, fielded between Mar 10, 2020 and November 30, 2020. Clustered robust standard errors were used in logistic regression. Results are weighted. All logistic regression models adjusted for month of the year, age (18–34, 35–54, 55–64, 65+), highest educational attainment (high school diploma or GED and below, some college, college and above), household income (\$0–29,99, \$30,000–59,999, \$60000–99,999, \$100,000 or more), gender (male, female), whether a household has children or not, employment status (employed, unemployed, others), and baseline pre-pandemic CES-D score. Each of the demographic variables were interacted with the time variable to model potential change over time in the estimated coefficients. The coefficients of the demographic variables interacted with the time variable were omitted in. For comparison with logistic regressions, we also estimated AORs from a multilevel fixed-effects model using an unstructured covariance option, which allows for all variances and covariances to be distinctly estimated, but results are not shown; estimates from fixed-effects multilevel model estimation are quantitatively as well as qualitatively similar to those from logistic regression. The mean item non-response rate for the PHQ-4 across all waves (wave 1 through wave 17) is about 1% (S.D = 0.43%; Min = 0.44%; Max = 1.85%). We did not find any systematic relationship between the baseline CES-D score and item non-response for the PHQ-4. We conducted attrition analysis in the UAS panel and found that the average ann\`ual attrition rate is about 9.4% and that that attrition was not selective by education, age, gender, income, race and ethnicity. The UAS panel is refreshed continuously throughout the year. Our results in indicate that, as in March, Black Americans were significantly less likely than other racial groups to have probable anxiety and/or depression in the following months after March, even after controlling for age, education, household income, whether a household has children or not, gender, employment status, and (most importantly) baseline CES-D score. Again, the odds of Black Americans reporting symptoms in a given month are about half that of non-Black Americans, and the differences are statistically significant in all months. Our results also indicate that overall probable anxiety and depression among the U.S. adult population was high in March and April, and then gradually began to improve starting in May. When broken down by type of disorder, the results also reveal that anxiety showed a higher degree of fluctuation from March to November in the U.S. adult population compared to depression. This may in part reflect differences in specificity for the measurements of depression (specificity of 0.92) and anxiety (specificity of 0.83; 19, 20). # Discussion Previous work has documented mental health disparities across racial and ethnic lines in times of disaster. Understanding how these differences manifest during the COVID-19 pandemic is critical to provide appropriate mental health interventions and implement effective policies. Thus, our goals in this study were to understand racial disparities in odds of having anxiety and depression among Black and non-Black Americans aged 18 and above when state lockdowns began in March of 2020, and to explore changes in this disparity over time between March and November. We found the odds of Black Americans having depression and anxiety was lower on average throughout the pandemic than that of non-Black Americans. In contrast, we found no significant differences in mental health outcomes among non-Black racial and ethnic groups (e.g., Whites, Asians, Hispanics, and Others). We also found that the monthly gap in depression between Black and non-Black Americans began to widen in September of 2020. This is further supported by our finding that the baseline CES-D scores did not differ significantly between Black and non-Black Americans before the pandemic, but did differ significantly mid- pandemic by December of 2020. All analyses adjusted for sociodemographic variables and baseline pre-pandemic CES-D score. These findings are paradoxical given that Black Americans disproportionately face some types of mental health stressors compared to other racial and ethnic groups during the pandemic as previously discussed and that the physical health of Black Americans is significantly worse than non-Black Americans. To explore this result, we comment on our findings of mental health disparities during the COVID-19 pandemic in light of the mental health of Black Americans in general as well as mental health disparities during other times of disaster, drawing on findings from prior studies. A few studies have considered racial and ethnic differences in mental health specifically during the pandemic, reporting heterogeneous findings. One study using a representative probability panel examined levels of psychological distress using the Kessler 6 scale before (in February 2019) and during the pandemic (in May 2020) among the same respondents. The authors found an increase in psychological distress among 12.8% of respondents, which was only more pronounced among Hispanic Americans compared to other racial and ethnic groups. However, the research utilized a far smaller sample size than we analyzed (N = 1,870), so real differences between groups may not have been detectable. A second probability-based study measuring mental distress in the same way found no differences in prevalence by race/ethnicity between April and July of 2020, but again was limited by a small sample size (N = 1,337; 32). On the other hand, a longitudinal study in the UK found that racial and ethnic minorities experienced a greater increase in mental distress after the onset of the pandemic as measured by the General Health Questionnaire-12, possibly due in part to lockdowns and physical distancing. A separate study in the Netherlands found no increase in anxiety or depression at all in the general population after the pandemic began and noted a stable employment situation relative to the U.S. which may play a large role; the Netherlands’ universal healthcare system and better social safety net (relative to the US) may also explain this difference. With further investigation, these differences may shed some light on the factors contributing to the racial and ethnic disparities in mental health in the US. The similarity between mental health outcomes of Black Americans during COVID-19 and in other times of disaster varies. Black Americans remain more vulnerable to the physical effects of both the COVID-19 pandemic (as previously established) as well as disasters in general, but limited research on psychological impacts precludes any conclusive interpretations. For example, hurricanes Ike and Katrina exacted a greater toll on Black persons, who disproportionately populated evacuation shelters as evacuees after Katrina and reported greater personal impact and loss of services after Ike compared to those who were White. As expected, Black Americans in hard-hit Mississippi counties reported higher levels of depression and psychological distress than White Americans in the aftermath of Hurricane Katrina, and Black Americans living in Texas after Hurricane Ike were more likely to meet criteria for PTSD compared to White and Hispanic Americans. In these situations, racial disparities in psychological impact reflected those in physical impact, unlike what we observed during the COVID-19 pandemic. In contrast, a year after 9/11, Black persons living in New York City during the disaster were less likely than those who were White to meet criteria for major depression or have poor mental health; this is expected following our previous logic as the representation of Black lives lost during this disaster was less than that of the New York City population. As discussed in the introduction, other research has shown that Black Americans experienced lower rates of anxiety and depression outside of the context of disasters before the pandemic, but had higher rates of serious conditions like PTSD or serious psychological distress. However, as previously discussed, we found no differences in depression (CES-D score) before the pandemic (although we had no ability to make comparisons with anxiety), and our main results showed lower rates for both anxiety and depression among Black Americans during the pandemic. Further research is required to determine whether racial and ethnic mental health disparities in terms of depression and anxiety in general differ from disparities in serious psychological distress during crises like COVID-19. Marginal evidence indicates that Black Americans may systematically underestimate their risks compared to their counterparts. Previous research has found that Black Americans had the same odds as White Americans to believe that those of lower socioeconomic status and those who were Black were more likely to die of COVID-19 complications; however, Black Americans had lower odds than White Americans to hold this belief about those who were older and those who had chronic health conditions. But this potential explanation was not supported by our data. We analyzed responses to risk perception questions in the UAS COVID-19 Tracking survey about the chance to get the coronavirus in the next three months, the chance of death if infected with the coronavirus, and the chance of running out of money because of the coronavirus in the next three months. The data revealed that, instead of underestimating the risks, Black respondents perceived the same risk as non-Black respondents of getting infected with COVID-19 in the next 3 months and actually perceived a higher risk than non- Black respondents of death and running out of money due to COVID-19 in the next three months. This finding held even after controlling for demographics and baseline depression symptoms. Most importantly, our main findings were insensitive to whether we controlled for responses to the three risk perceptions, in addition to demographics and baseline depression symptoms (results not shown). Alternatively, some might argue that the mental health advantage of Black Americans over non-Black Americans is due to higher religious involvement among Black Americans and higher importance Black Americans place on religion, compared to non-Black Americans. However, the empirical evidence supporting the religious explanation has been limited. Racial differences in reporting styles might also explain our results. The outcome measures used in this study are inherently subjective and thus are vulnerable to response scale biases. Different racial groups could use different response scales or attach different meanings to a verbal label when self- reporting their mental health. For instance, Black Americans might have a higher standard over what it means to be mentally stressed, especially during crises like COVID-19 pandemic. If this is the case, then the differences in self-report of mental health reflects differences in response scale used by different racial groups rather than differences in underlying mental health conditions. Future research should investigate this possibility. # Conclusion Unlike any public health crisis in living memory, the COVID-19 pandemic has thrown the U.S. into upheaval. Black Americans have faced disproportionately high rates of illness, job losses, and death in its midst. The mental health of Black Americans thus remains a major concern as the pandemic and its effects unfold. Hence, in this study, we assessed the mental health of the U.S. adult population through probable anxiety and depression by race over time during the pandemic. We found no mental health differences among non-Black racial and ethnic groups, but discovered significantly better mental health among Black Americans in comparison. This result raises several questions given numerous prior reports suggesting that Black Americans are disproportionately affected by the pandemic in terms of mortality, hospitalization, and job loss. Understanding the significance of this result requires further research into how Black Americans think about their mental health and perceive health risks during public health crises. The project described in this paper relies on data from survey(s) administered by the Understanding America Study, which is maintained by the Center for Economic and Social Research (CESR) at the University of Southern California. The content of this paper is solely the responsibility of the authors and does not necessarily represent the official views of USC or UAS. The collection of the UAS COVID-19 survey data is supported in part by the Bill & Melinda Gates Foundation. The authors thank Yajuan Si for her encouragement and advice in developing this paper. [^1]: The authors have declared that no competing interests exist.

# Introduction Enzymatic hydrolysis has been the most common mode of bacterial resistance to β-lactam variants, as β-lactamases are apt to expand their substrate spectrum with simple mutations. Since the identification of the first extended-spectrum β-lactamases (ESBLs) in the 1980s, members of the ESBLs have been steadily increasing following the release of new man-made drugs. Analyses of amino acid substitutions in these enzymes have enhanced our current understanding of substrate spectrum expansion in class A β-lactamases. However, much of the molecular basis that underlies the adaptability of these enzymes to new antibiotics remains to be understood. Even in the most extensively studied TEM- type ESBLs, only a small number of amino acid substitutions have been confirmed to have a direct role in the enzyme’s expanded substrate spectrum: ten amino acid substitutions against β-lactam antibiotics, six against β-lactamase inhibitors, and two against both. To profile the adaptability of class A β-lactamases to a new antibiotic, we conducted a large-scale mutant selection with the chromosomally encoded PenA from *B. thailandensis* (BTH_II1450 from *B. thailandensis* strain E264) to investigate its adaptation to an oxy-imino-cephalosporin, ceftazidime. *B. thailandensis* is resistant to penicillins, and therefore is a good system for a substrate expansion study of this kind. Furthermore, because *B. thailandensis* is a non-pathogenic soil saprophyte, it is unlikely that this bacterium has been heavily exposed to ceftazidime. We expected that with *penA* we would be able to observe fresh mutational responses to the challenges posed by exposure to the new ceftazidime drug. Furthermore, *penA* is a closely related ortholog to genes in the pathogenic species *B. pseudomallei, B. mallei,* and the *Burkholderia cepacia* complex (Bcc), , and the antibiotic regimen used to treat infections by these bacteria generally includes ceftazidime. *B. pseudomallei* is the etiological agent of septicemic melioidosis, which is endemic in Southeast Asia and Northeastern Australia. *B. mallei*, the cause of glanders, is a species derived from a clone of *B. pseudomallei*. The Bcc, which is a complex composed of more than 10 *Burkholderia* species, including *B. cepacia*, *B. cenocepacia*, and *B. multivorans*, is a group of nosocomial pathogens of increasing concern as these bacteria cause respiratory and systemic infections in patients with cystic fibrosis (CF) or chronic granulomatous disease, and in other immuno-compromised patients. To date, only a few cases of ceftazidime resistance in *B. pseudomallei* and Bcc have been reported and resistance in most of these cases pointed to a single gene. Two single amino acid substitutions, Pro167Ser and Cys69Tyr (amino acid residue numbering following Ambler *et al.*), have been described in PenA of ceftazidime resistant *B. pseudomallei* isolates. Similarly, two orthologs of PenA, PenB2 and PenB3, with 7 and 2 amino acid alterations, respectively, were shown to be associated with ceftazidime resistance in clinical isolates of *B. cenocepacia*. # Results and Discussion ## All 516 Ceftazidime-resistant Mutants had a Mutation in the *penA* Gene We measured the intrinsic level of the susceptibility of the wild-type *B. thailandensis* to ceftazidime to determine the appropriate antibiotic pressure for the experiment. Then, a concentration (5 µg/ml) slightly lower than three times the wild-type MIC value (1.75 µg/ml based on the E-TEST) was used in the mutant selection process. From a total of 1×10⁸ cells on a selection plate, 1 to 10 cells survived to form visible colonies after 16–20 h incubation, resulting in an estimated mutation frequency of 10⁻⁸ to 10⁻⁷. To investigate the mutation patterns of *penA*, we sequenced the gene in resistant *B. thailandensis* isolates. Intriguingly, all 516 isolates contained a single point mutation in the coding region of *penA*. To confirm that *penA* genes with a mutation (hereafter referred to as *penA\**s) were responsible for the ceftazidime-resistance that developed in the mutants, we inactivated the *penA\** alleles of four randomly selected mutants containing Cys69Phe, Arg164His, Glu166Lys, or Ala172Thr substitutions by replacing a region spanning 196 bp in the middle of the coding region with a *tet*^r cassette. All four isolates with inactivated *penA\** (that are, with Δ*penA\**s) lost resistance to ceftazidime (MIC ≤1.75 µg/ml), having an MIC comparable to the level of the wild-type strain with Δ*penA* (i.e. E264ΔP). When a mutant strain with Δ*penA*\* was provided with an intact copy of the *penA\* in trans*, ceftazidime resistance was restored, indicating that *penA\** was the factor responsible for ceftazidime resistance. This *penA\** selection method conducted in its native (wild-type) host simply under a high level of antibiotic pressure without an added mutagen (that is, without an unnaturally elevated mutation rate, which may result in simultaneous multiple mutations), enabled us to investigate the functional adaptability of PenA at the level of single mutations with clear phenotypes. Furthermore, a short incubation time of 16–20 h under selection, that was the same as the normal growth time required for the wild-type strain, ensured that the mutant enzyme sustained a normal growth rate under high antibiotic pressure. In fact, the same growth rate as that of the wild-type was observed with all of the mutants. Therefore, the set of acquired PenA\* enzymes provided information about the extent to which changes in amino acid residues are able to confer high hydrolytic activity against ceftazidime, while maintaining the necessary integrity of the enzyme. In contrast, approaches employing artificial mutagenesis often generate multiple mutations, and in these cases interpreting the contribution from each mutation is not straightforward. Furthermore, many such mutations confer only weak hydrolytic activities and/or cause low enzyme stability, thus making the mutations unsuitable for selection in natural and/or clinical settings. Although clinically occurring mutations that evolve in a patient treated with antibiotics may not have issue of enzyme stability, many of them do not have clear phenotypes associated with substrate spectrum expansion and/or only confer weak β-lactam hydrolyzing activity. ## Twelve Amino Acid Positions in PenA were Substituted, each with Distinct Patterns A total of twelve amino acid positions were repeatedly found substituted in 516 mutated PenAs. Ten of the substituted amino acids were located in the omega loop, a structural domain constituting part of the active-site pocket. One of the four conserved domains in class A β-lactamases, ¹⁶⁶EXXLN¹⁷⁰ (numbering following Ambler *et al.*), is located in the omega-loop. Notably, all three conserved residues in this domain were substituted in the set of PenA mutants. The other two substitution positions were located close to, but not in, two other conserved domains, ⁷⁰SXXK⁷³ and ¹³⁰SDN¹³². However, no substitution was found near the fourth domain, ²³⁴KTG²³⁶. Each of the twelve positions exhibited one to four different substitutions out of five to seven substitution choices considering all possible single nucleotide changes in the codons, totaling twenty nine substitutions. Most (nine of the twelve) positions had more than two types of substitutions, reflecting that the major impact of the substitutions is more caused by which residues are replaced rather than which replaced the original residues. However, the fact that not all possible residues were found in the substitutions at each position suggests that there are specific qualifications required by the substitute residues in each substitution. In this regard, it is notable that positions Asn170 and Ala172 exhibited a preference for a molecular weight (MW) range such that the substitutes were the largest three or four among the candidates, respectively, regardless of the different R-group properties. Assuming that the MW roughly correlates with occupying volume by a residue, these substitution patterns suggest that steric factor may play a role in these substitutions. Furthermore, the two substitutes for position Cys69 share an intriguing commonality; both are aromatic residues. In contrast, three positions Leu162, Thr171, and Asp179 had only one-type of substitutions. At least at positions Leu162 and Asp179 that had a large number of associated isolates (26 and 54, respectively), the selected substitute residues may be the only qualified ones. Notably, these residues Leu162 and Asp179 are present at the either side of the outer boundary of the omega loop, suggesting that these specific locations may limit diverse substitutions at these positions compared to those inside the omega loop. Although structural data will be needed to understand the specific changes made by each substitution, our data formulated by a large number of mutants clearly unveiled the possible evolutionary paths of the class A β-lactamase during substrate spectrum expansion. ## Each Substitution Exhibited a Distinct MIC Level and Frequency of Occurrence The Glu166Lys and Leu169Gln substitutions, which occurred in the ¹⁶⁶EXXLN¹⁷⁰ domain located in the omega loop, resulted in the highest ceftazidime-hydrolytic activity. Interestingly, the Cys69Phe substitution, which occurred outside the omega loop directly upstream of the essential catalytic Ser70, also resulted in one of the highest MICs for ceftazidime. That the Glu166Lys substitution in PenA exhibited high ceftazidime- hydrolyzing activity (Fig. 1ab) is intriguing, because no substitution at the position Glu166 has been found in clinical isolates of class A β-lactamases. Glu166 has been reported to play a key role in β-lactam hydrolysis catalyzing the deacylation step after the acylation step catalyzed by Ser70 leading to the formation of an acyl-enzyme intermediate with the β-lactam acylated to the side chain of Ser70. This deacylation-deficient PenA\* with Glu166Lys may produce resistance to ceftazidime by a covalent-trapping mechanism as suggested by a recent study with a deacylation-deficient mutant TEM with a Glu166Arg substitution. Among more than two substitutions at some positions, there were those that better served in terms of the ceftazidime-hydrolytic activity. Most notably, the three positions in the conserved domain ¹⁶⁶EXXLN¹⁷⁰, Glu166, Leu169, and Asn170, showed distinct MIC levels depending on substitutions, and those with dissimilar residues appeared to result in high impacts. Most notably, the substitution with highly dissimilar Lys (basic) at position Glu166 (acidic) showed a much higher level of MIC compared to the substitutions with nonpolar Gly or Asp, which is also acidic (Fig. 1ab). At position Leu169 (nonpolar), the substitution with nonpolar Pro showed the lowest MIC compared to the substitutions with Gln (polar) or Arg (basic). In addition, two of the three substitutions of Asn170 were with aromatic residues, resulting in higher MICs than the third substitution with Lys (Fig. 1ab). In class A β-lactamases, the terminal amide group of Asn170 has been proposed to play a critical role in the catalysis of β-lactams by positioning a water molecule in the vicinity of the catalytic Ser70 through a hydrogen-bond network. As the Asn170 mutants may not have a normal hydrogen bond network, it remains to be studied how these enzymes maintain their catalytic activity, and further, how the aromatic substitutes result in high hydrolytic activity against ceftazidime. Our data demonstrate that the MIC levels and the relative frequency in the occurrence of substitutions did not correlate well with each other. This low correlation may be because the difference in ceftazidime-hydrolyzing activity levels may not be a critical factor once all of the mutants surpass a threshold- level of hydrolytic activity required for survival. In fact, our growth experiment showed the same growth rate for all mutants, which is comparable to that of the wild-type strain. It has been known that the intrinsic nucleotide- level mutational biases present in bacteria; G→A and C→T transitions have been found to be the most common non-synonymous mutations. Additionally, a higher prevalence of the transversion A↔T than of A→C or T→G has previously been observed in the *B. mallei* genome. In this regard, it is notable that three of the most frequent nucleotide substitutions at the twelve positions were indeed the transitions G→A (Glu166Lys and Asp179Asn) and C→T (Ala172Val). In addition, at positions Cys69, Arg164, Ala172, and Asp176, where multiple substitutions occurred, the transitions G→A and C→T were highly preferred. However, the preference in the transversion A↔T was not observed in the PenA substitutions. Nine of the twenty nine substitutions were found in clinical isolates of class A β-lactamases, and five of these substitutions (Cys69Tyr, Arg164His, Arg164Ser, Asp179Asn, and Leu169Arg) have been verified to cause substrate spectrum expansion in *B. pseudomallei* PenA or TEM- or SHV-type enzymes. Among the twelve substitution positions, only one (Asp179) had additional substitutions that were found in TEM- or SHV-type ESBLs but not obtained in our PenA mutation set. As Asp179 could be further substituted, besides by Asn, by Gln in TEM- and by Gly and Ala in SHV-type ESBLs, it appears that substitution at position 179 is affected by specific local environment within each β-lactamase. On the other hand, Met is present instead of Cys at position 69 of TEM-1 and SHV-1. Therefore, derivatives of these enzymes displayed substitution patterns different from those in PenA (TEM and SHV Tables summarizing variants in each group can be found at <http://www.lahey.org/Studies/>). Position 171 of TEM-1 and SHV-1 also has a different residue (Glu instead of Thr) compared to PenA, however no derivatives of the enzymes have been identified at this position. Except for positions 69 and 171, all other ten substituted positions in PenA have the same well-conserved residues as in the other enzymes, including PenA homologs in *B. pseudomallei* and *B. mallei* and the representative enzymes, TEM-1, SHV-1, and CTX-M-9, that do not have activity against ceftazidime. The sequence conservation at these positions suggests a pivotal role played by these residues, and this in turn suggests that the expansion of the substrate spectrum in class A β-lactamases may follow a common evolutionary trajectory. On the other hand, Pro is present at position 167 in *B. pseudomallei* instead of Thr as in *B. thailandensis* PenA, and therefore the Pro167Ser mutation found in clinical isolates of *B. pseudomallei* was not obtained in our set of mutants. Although the effect of this mutation has been confirmed in *B. pseudomallei*, whether the same affect applies to *B. thailandensis* PenA remains to be verified, as the substitute Ser167 is similar to the original residue Thr167 in *B. thailandensis* PenA. ## The Twelve Amino Acid Positions are Co-localized in the Enzyme and Appear to cause a Perturbation in the Omega Loop To investigate the structural changes in PenA that correlate with the substrate spectrum expansion to ceftazidime, we conducted modeling analyses using CTX-M-9 covalently linked to cefoxitin (PDBID: 1YXM, 58% of amino acid identity to PenA) as a template. We chose CTX-M-9 for the simulation because of its high homology to PenA and it does not hydrolyze ceftazidime similar to wild-type PenA. Simulated 3D models of PenA showed that the two substituted amino acid positions 69 and 136 co-localized at the active-site pocket area of the enzyme along with the ten positions in the omega loop. In our simulations, the size and shape of the binding pocket did not show significant correlation with each amino acid substitution. Instead, we observed that the most substitutions of Arg164 or Asp179 (except Arg164Cys), disrupting the stabilizing salt bridge between the two residues that clamp the omega loop structure at both ends, resulted in increased distances between positions 164 and 179. This is consistent with previous studies that demonstrated that disruptions of the ionic bond between residues 164 and 179 destabilize the omega loop. Notably, the Asp179Asn mutant, which had a higher MIC than the Arg164 mutants, also had a longer 164179 distance in our analysis. In addition, PenA\*s with Asn136Asp, Glu166Lys, or Asn170His, also showed significantly increased distances between positions 164 and 179, suggesting a possibility that the altered local environment created by some other substitutions may also affect the interaction between Arg164 and Asp179. Although each of the amino acid substitutions may have caused diverse local disturbances, all twenty nine mutations had the same congruent effect. Specifically, we noted that the space in the omega loop (calculated as the sum of the distances between amino acid residues 163 and 174 and between 164 and 173) increased in all PenA\*s, including those with substitutions outside the omega loop, compared to the wild-type PenA. We postulate that increased space in the omega loop may accompany increased flexibility of the loop in the mutant enzymes. Then, this flexibility would in turn relieve steric hindrance between the omega loop and the bulky 7β-side chain of ceftazidime, thereby increasing accessibility of ceftazidime to the binding pocket. ## PenA\*s have Altered Activities for Hydrolyzing Various β-lactam Antibiotics Our data comparing the MICs of the wild-type strain E264 and the various mutant strains showed that PenA is responsible for the intrinsic moderate insensitivity of the wild-type strain E264 to various classes of β-lactam antibiotics, including amoxicillin, ceftriaxone and cefotaxime (a 3^rd-generation cephalosporin), and cefepime (a 4^th-generation cephalosporin). To investigate the extent to which the structural changes in the PenA\*s that were adjusted to ceftazidime affect the enzyme activity towards other β-lactam antibiotics, we measured the MICs for selected β-lactam antibiotics. These antibiotics included the four that the wild-type enzyme was able to hydrolyze and a carbapenem antibiotic (meropenem) and a β-lactamase inhibitor (clavulanic acid with amoxicillin). All PenA\*s generally exhibited decreased levels of resistance to the original substrates as observed in many mutants of the class A β-lactamase that acquired activity to 3^rd generation cephalosporins. The hydrolytic activities of all PenA\*s against amoxicillin (MIC range: 3 ∼ 24 µg/ml) were effectively inhibited by clavulanic acid. In addition, resistance to meropenem was not observed in the wild-type or the mutants. ## A Second-phase Selection did Not Result in a Mutation in the Coding Region but Rather in the Promoter of *penA* To investigate whether PenA\*s accumulate more mutations under an elevated level of ceftazidime, we conducted a second-level selection with three randomly chosen PenA mutants, containing Cys69Phe, Glu166Lys, or Asp179Asn. The selection process did not result in any additional changes to the coding region, but the same single point mutation, G→A in the region upstream of *penA*, was observed in seven of the eleven second-phase mutants derived from all three mutants. Intriguingly, this mutation also has been identified in a clinical isolate of *B. pseudomallei*, and was noted conferring CAZ resistance. The G→A point mutation lies in the putative -10 sequence 5′-TACGCT-3′, changing it to 5′-TACACT-3′, which is closer to the consensus sequence for the -10 sequence in *E. coli*, 5′-TATAAT-3′. One of these mutants, E166K1-F1, derived from the mutant E166K1 with Gln166Lys, was further investigated. This mutant showed a significant increase in hydrolytic activity against ceftazidime compared to the parental strain E166K1. The double mutant *penA\** fragment from E166K1-F1 carried by pRK415K conferred an approximately three-fold increase in MIC value compared to a similar clone with a fragment from E166K1 in the E264(Δ*penA*) background. To determine whether the point mutation resulted in increased *penA\** expression, we compared the transcriptome profiles of E166K1-F1 and E166K1 obtained before and 1 h after ceftazidime challenge. Intriguingly, the basal level expression of *penA\** was approximately four-fold higher in E166K1-F1 than in E166K1, but there was no further induction in response to the addition of ceftazidime. In contrast, no other genes were differentially expressed as highly as *penA*\* in E166K1-F1 compared with E166K1 (data not shown). In addition to the promoter mutations, there were a smaller number of other isolates (four of the eleven) for which we could not find a mutation in the promoter or in the coding region of *penA\**. This finding suggests that there may be an additional determinant that contributes to resistance to ceftazidime in *Burkholderia thailandensis*. ## Perspectives In our mutant selection, a large proportion of the mutations in the bacterium were expected to occur in *penA*; however, the finding that all 516 ceftazidime- resistant isolates that we collected had a mutation in this gene was intriguing. This result strongly suggests that the class A β-lactamase PenA is a predominant determinant for the development of ceftazidime resistance in *B. thailandensis* and perhaps more widely in *Burkholderia* spp. This notion is supported by the fact that most of the validated mutations associated with ceftazidime resistance in *B. pseudomallei* and *B. cepacia* have to date been located in *penA* homologs. In the second-phase mutant selection with an increased antibiotic challenge, we obtained mutations that were not associated with *penA* but were present elsewhere in the genome, demonstrating the presence of other genetic determinants. In clinical *Pseudomonas aeruginosa* isolates, which often lack class A β-lactamase genes, the development of ceftazidime-resistance could be a result of various mechanisms with multiple determinants. There certainly is a possibility that homologs of these *P. aeruginosa* genes, including those coding for class B and D β-lactamases, efflux pumps, and porins, may have mutations associated with ceftazidime resistance. The levels of ceftazidime resistance in clinical *B. pseudomallei* isolates, and in Bcc, have been determined to be low. However, the continued use of ceftazidime in clinical settings suggests the potential for the increased emergence of ceftazidime resistance in these bacterial groups. Furthermore, as SHV and CTX-M enzymes are thought to have originated from chromosomal β-lactamases of *Klebsiella pneumoniae* and *Kluyvera* species, respectively, *Burkholderia penA* also has the potential to cause a global impact if it becomes mobile. Substitutions in the omega loop, especially at positions 164, 166, and 179, have been reported to seriously affect the stability of TEM β-lactamases. However, the Met182Thr substitution was shown to restore the stability of these enzymes. As PenA\*s with a perturbation in the omega loop produced high ceftazidime- hydrolyzing activity without obvious growth retardation of the host bacteria, it is intriguing to note that PenA contains a Thr residue at position 182. However, it remains to be investigated if the Thr182 in PenA contributes as a global stabilizer for the vast array of substitutions we obtained in the active site area. Mapping twelve positions and their twenty nine single amino acid substitutions in a class A β-lactamase capable of altering the substrate spectrum of the enzyme is unprecedented. Notably, only seven of these twelve positions and nine of the twenty nine mutations have been reported in clinical isolates of TEM- or SHV-type β-lactamases, or in the PenA homologs of other *Burkholderia* spp., and not all of these have been reported with confirmation in their effects on the substrate spectrum expansion. These data have substantial value, representing a comprehensive repertoire of enzyme adaptability that can occur in clinical settings. These mapped amino acid positions and substitutions with clear ceftazidime resistance phenotypes provide an invaluable resource for the development of preventive strategies and treatments against not only ceftazidime resistance in pathogenic *Burkholderia* species but also in ESBLs in general that share functional mechanisms in their evolution against new drugs using common features of conserved residues. # Materials and Methods ## Bacterial Cultures All *Escherichia coli* strains were grown in Luria Bertani (LB) media, and all *B. thailandensis* strains were grown in LB or AB minimal media containing 0.25% glucose (ABG) at 37°C. The concentrations of antibiotics used for *E. coli* were as follows: tetracycline, 10 µg/ml; kanamycin, 50 µg/ml; and ampicillin, 100 µg/ml. For *B. thailandensis*, the concentrations of tetracycline and kanamycin used were 50 µg/ml and 250 µg/ml, respectively. ## Determination of the MIC (Minimal Inhibitory Concentration) Values The MIC values were measured by the E-test, following the manufacturer’s instructions (AB Biodisk, Solna, Sweden). Briefly, each strain was grown on Müller-Hinton agar plates at 37°C for 2 days. For the strains harboring pRK415K-derived plasmids, Müller-Hinton agar supplemented with kanamycin (250 g/ml) was used. Single colonies from the plates were suspended in 1 ml PBS until the turbidity reached 0.5 MacFarland standard. Using sterile cotton swabs, cell suspensions were spread on the Müller-Hinton agar plates, the E-test strips were placed, and the plates were incubated at 37°C for 16∼18 h. The lowest concentrations at the E-test strips where no visible growths were observed were recorded as the MICs. Average values were calculated from triplicate experiments. The agar dilution method was conducted as described by Wiegand *et al.* , except that LB agar was used instead of Mueller Hinton agar. Briefly, a single colony of each strain grown on ABG minimal medium agar was used to inoculate 2 ml of LB broth, and the culture was incubated overnight with shaking (250 rpm) at 37°C. The overnight cultures were serially diluted with fresh LB and dispensed into the wells of a 96-well microtiter plate. Using a multi-channel micropipette, 1 µl of the diluted bacterial suspensions was placed on each LB agar plate containing 0.5 to 8 µg/ml of ceftazidime and incubated at 37°C for 16 to 20 h. The lowest concentration of antibiotic such that there was no visible bacterial growth was observed in the spot containing approximately 10⁴ CFU (colony forming unit), which was recorded as the MIC. The number of CFUs in serially diluted bacterial suspensions was determined by spreading 100 µl of the appropriately diluted bacterial suspensions on LB agar plates, followed by incubating the plates for 24 h at 37°C, and counting the viable cells. The average values were calculated from triplicate experiments. ## Isolation of Mutants Resistant to Ceftazidime A single colony of the wild-type *B. thailandensis* strain E264 was grown overnight in 2 ml of LB broth at 37°C with shaking (250 rpm). The overnight culture was pelleted by centrifugation (2,600 X g) for 5 min at 4°C, and the cells were resuspended in fresh LB broth to approximately 10⁹ CFU/ml. Cell suspensions (100 µl) were spread on LB agar plates containing 5 µg/ml (slightly lower than three times the wild-type MIC) ceftazidime, which were subsequently incubated for 48 h at 37°C until ceftazidime resistant mutants formed visible colonies. We picked two to three colonies from each selection plate to collect up to 516 isolates for this study. A second-round selection was conducted with an elevated concentration of ceftazidime (96 µg/ml) with selected ceftazidime-resistant mutants. ## Localization of the Mutations in *penA* To map the mutations in *penA* (BTH_II1450 in the strain E264 genome), we first set out to sequence the gene. Genomic DNA of each ceftazidime-resistant mutant was purified using a Wizard Genomic DNA Purification Kit (Promega, Madison, USA) and was used as the template for PCR amplification of *penA* and short flanking regions for a 1,386 bp amplicon (271 bp upstream from the start codon of the gene to 230 bp downstream of the stop codon). PCR reactions were conducted in a 50 µl reaction mixture containing 2.5 U of HotStar HiFidelity polymerase (Qiagen, Hilden, Germany), 50 pmol of the primers penA-F (5′-CGTCAATCCGATGCAGTACC-3′) and penA-R (5′-GCCGTTATCGCACCTTTATC-3′), 100 ng template DNA, 10 µl of 5X Q solution, and 10 µl of 5 X HotStar HiFidelity buffer. The reaction consisted of the following three steps: initial enzyme activating step (95°C for 5 min), amplification step (35 cycles of 94°C for 15 sec, 61°C for 1 min, and 72°C for 1.4 min), and final extension step (72°C for 10 min). Gel-purified 1.4 kb PCR products were sequenced using a 3730XL DNA analyzer (Applied Biosystems, Foster City, CA, USA) in both directions using primers penA-F and penA-R. ## Construction of *penA-*null Mutants The wild-type and mutated *penA* genes were disrupted as follows. First, the wild-type *penA* was PCR-amplified, using the primer pairs penA-KF (5′-ATATAT<u>GGTACC</u>CGTCAATCCGATGCAGTACC-3′) and penA-KR (5′-ATATAT<u>GGTACC</u>GCCGTTATCGCACCTTTATC-3′), containing a *Kpn*I recognition site (underlined) at the end. Then the 1.4 kb PCR product was digested with *Kpn*I and ligated into pUC19, which was digested with the same enzyme. The resulting plasmid was double-digested with *Xho*I and *Pfl*FI to remove an internal region (position 124 bp to 319 bp) of *penA,* which is 885 bp long, blunt ended with T4 DNA polymerase (NEB, Ipswich, MA, USA), and were inserted with a tetracycline-resistance (*tet*^r) cassette by ligation. The *tet*^r cassette was previously amplified from pRK415K, using HotStar HiFidelity polymerase (Qiagen, Hilden, Germany) and primers tetR-F (5′-ATATAT<u>CTCGAG</u>GTGAGGCTTGGACGCTAGG-3′) and tetR-R (5′-ATATTT<u>CTCGAG</u>CTTGGATCAGACGCTGAGTG-3′), containing a *Xho*I recognition site (underlined), *Xho*I-digested, and blunt-ended. The construct was transferred into *B. thailandensis* strains employing a modified method of natural transformation. Briefly, 3 ml of a defined medium (DM) prepared as described by Thongdee *et al*. was inoculated with a single colony freshly grown on LB agar and was incubated overnight with shaking at 250 rpm at 37°C. The overnight culture of 200 µl was diluted with 10 ml of fresh preheated DM and the culture was grown with shaking (250 rpm) at 37°C to an OD₆₀₀ of approximately 0.5. The culture was then concentrated 20-fold in 500 µl fresh DM and 50 µl aliquots of the concentrated cells were mixed with 0.5 µg of plasmid DNA. The mixture was incubated for 30 min on ice and 2 ml DM preheated to 37°C was added and the mixture was incubated overnight at 250 rpm at 37°C. After washing the culture with 1 ml of fresh DM and resuspending the pellet in 250 µl of fresh DM, 100 µl of the cell suspension was plated on ABG medium containing 50 µg/ml tetracycline and incubated at 37°C for 48 hrs to select for *tet*^r cassette-containing constructs. An obtained *penA* null mutant was verified by PCR using a primer pair penA_LF (5′-AACAGATCGCCGAGATGG-3′) and penA_LR (5′-GCGAACGTTGCCCGATAC-3′) that hybridize to the genomic regions outside the DNA sequence used for mutant construction. ## Complementation of the *penA* Mutations *Kpn*I-treated PCR products of *penA\** (*penA* with a single nucleotide mutation), which was amplified using the primers penA-KF and penA-KR as described above, was ligated with *Kpn*I-treated pRK415K. This plasmid was transferred into the *E. coli* strain S17-1 by using a conventional transformation method. The transformed S17-1 strain was conjugated with a *B. thailandensis penA-*null mutant on ABG agar plates containing 250 µg/ml kanamycin, and the plates were incubated at 37°C for 2 days to select for transconjugants. A successful conjugation was confirmed by purifying the plasmid from transconjugants and examining the characteristic restriction patterns of the plasmid. ## Modeling Analysis Because the structure of PenA is not known, the structures (for both mutants and wild-type) were predicted by homology modeling using CTX-M-9 (PDBID: 1YXM, 58% of identity) as a template with SYBYL-X (Tripos Inc., St. Louis, MO, USA). With the predicted structures, molecular dynamic simulations were conducted using OpenMM Zephyr 2.0.3 for 500 ps (0.002 ps/step) at 303.15 K with the Amber03 force field to release any structural constraints originating from the template. Then, energy minimization was performed using SYBYL -X (Tripos Inc., St. Louis, MO, USA) with a Tripos force field until the energy gradient reached 0.001 kcal/(mol•A). ## Gene Expression Study Gene expression was assayed using microarray technology. The bacterial culture was incubated in LB media until the OD₆₀₀ reached 0.5, then 2 ml of the culture was taken and mixed with 4 ml of RNAprotect bacteria reagent (Qiagen, Hilden, Germany) to prevent alterations in the transcriptome. For cultures that were induced by ceftazidime, a sub-MIC level of ceftazidime (24 µg/ml) was added, the samples were withdrawn after 1 hr, and were immediately mixed with two volumes of RNAprotect bacteria reagent. Total RNA isolation and labeling with Cy3 and Cy5 fluorescence dyes were conducted following PFGRC SOP (<http://pfgrc.jcvi.org/index.php/microarray/protocols.html>). Labeled samples were hybridized to the *B. thailandensis* whole genome microarray (MYcroarray, Ann arbor, MN, USA) in a hybridization chamber in a 42°C water bath for 16 to 20 hrs. Microarray slides were washed twice with 1 x SSPE for 3 min and then with 0.25 x SSPE for 30 sec, and were dried immediately by spinning in a micro-slide centrifuge for 1 min. Hybridized slides were scanned using a GenePix 4000B microarray scanner (Molecular Devices, LLC, Sunnyvale, CA, USA). The scanned images were analyzed using TIGR SPOTFINDER (<http://www.tm4.org>) to obtain relative transcript levels. LOWESS (locally weighted scatterplot smoothing) normalization of data was executed using the software tool MIDAS (<http://www.tm4.org>). The resulting data were visualized and further explored by using TMEV (<http://www.jcvi.org/cms/research/software>). # Supporting Information [^1]: Conceived and designed the experiments: HSK HY. Performed the experiments: HY KHC YSC KK. Analyzed the data: HSK HY KHC WCN. Contributed reagents/materials/analysis tools: HSK. Wrote the paper: HSK HY. [^2]: The authors have declared that no competing interests exist.

# Introduction Insects are important model organisms for studying the evolution and mechanisms of immunity and host-pathogen interactions. For example, experimental approaches have been established for oral inoculation of natural bacterial pathogens for the main insect model, the fruit fly *Drosophila melanogaster*, thereby adding a vital tool to the methodological repertoire of insect immunology. This has enabled the successful in-depth study of the pathology of bacterial infections. The red flour beetle *Tribolium castaneum* (Herbst 1797) has developed into a fully-fledged insect model organism. The value of *T. castaneum* as an alternative insect model lies in the fact that, as a coleopteran, it shows a number of distinct differences to the fly and since it is evolutionarily more basal, it can be regarded as being more representative of other insects. The availability of an expanding genetic and genomic toolbox that includes well- functioning systemic RNAi , has made *T. castaneum* an upcoming model for a number of research fields, including immunity and host-parasite interactions. Furthermore, *T. castaneum* is a serious pest species in many areas of the world, leading to substantial losses in the nutritional value of stored agricultural products. Therefore, there is a strong interest in research on pest management for this species. *Bacillus thuringiensis* Berliner 1915 (*Bt*) is a Gram-positive bacterium that forms highly resistant endospores when nutrients in the environment become limiting. One of its main characteristics is that it produces plasmid-encoded crystalline inclusions (Cry proteins) during the sporulation phase, which are toxic to specific insect orders upon ingestion. The nomenclature of Cry toxins is based on amino acid identity. Cry3 toxins are active towards some coleopterans and cross-order activity has been reported for some of the lepidopteran-specific Cry toxins (reviewed). The vast majority of studies have focussed on the toxicity of Cry toxins, and several mechanisms for its mode of action have been proposed (reviewed). However, many insects, including *T. castaneum* have been shown to be refractory to purified toxins, and mortality is observed only when bacterial spores are added to the diet. The ingestion of spores and the following infection process that takes place in the gut and subsequently the haemolymph is considered a natural infection route for *Bt*. Investigations on how the bacteria behave inside the host after infection and processes that act in addition to the toxins are therefore highly interesting from the viewpoint of host-parasite coevolution. We exposed *T. castaneum* to *Btt* bacteria via oral route, and moreover made use of a genetically well characterised *Bt* strain. Since both the host and the pathogen are accessible to genetic manipulation, the system will enable detailed genetic analyses of the infection process and host-pathogen interactions. Importantly, *Bt* itself is an organism of utmost importance for basic and applied sciences. Currently studied natural insect hosts of *B. thuringiensis* are mostly lepidopterans, such as the diamondback moth (*Plutella xylostella*), the tobacco hornworm (*Manduca sexta*), and the cotton bollworm (*Helicoverpa armigera*) for which the full repertoire of genetic and genomic tools is not yet available. Likewise, even though *D. melanogaster* has been shown to die from exposure to *Bacillus* species, including *B. thuringiensis*, to our knowledge it has not been established as an experimental host for *Bt*. Transgenic *D. melanogaster* carrying lepidopteran (*M. sexta*) Cry receptor have been shown to become susceptible to *Bt*, suggesting a role for this specific receptor. However, such a system would not allow addressing the natural infection process or the genetic variation in the full range of factors that are relevant for susceptibility to *B. thuringiensis*. Our first objective was to verify the most suitable bacterial strain for the investigation of this host-pathogen interaction. We identified *Bt morrisoni* bv. *tenebrionis* (*Btt*) as infective to *T. castaneum*, and further characterised the susceptibility of geographically diverse populations of *T. castaneum* to this strain. We then investigated the behaviour of the bacteria in the host and the time course of the infection. We also demonstrate the transfer of plasmids from *Btt* to a non-pathogenic but genetically characterised *Bt* strain, which thereby became able to successfully infect *T. castaneum*. The availability of such a genetically accessible strain will be most useful for a more in-depth analysis of this interaction in the future. The *T. castaneum – Bt* system proposed here shows the potential for in-depth experimental analyses of a coleopteran insect model host's interaction with this important pathogen. # Results ## Insecticidal Activity of Different *Bt* Strains to *T. castaneum* Larvae We analysed the infectivity of four different *Bt* strains towards three different *T. castaneum* populations, the laboratory populations San Bernardino (SB) and Georgia 2 (GA-2) and the recently wild-collected Croatia 1 (Cro1) population. When comparing the survival of the naïve group to the other treatments, only the *Btt* strain was able to induce significant mortality of *T. castaneum* larvae from all beetle populations. All other bacterial strains induced no significant mortality above the background level of the control insects. Larvae were kept constantly on the spore-containing diet (flour discs with spores in a 96 well plate), but the majority died within the first 24 hours after the exposure had started, with low mortality during the following days. Mortality was dependent on the spore concentration used to prepare the diet (1×10⁹ mL⁻¹: z = 4.463, p = ***\<***0.0001, 1×10¹⁰ mL⁻¹: z = 6.870, p = \<0.0001), and SB and GA-2 population differed in their responses to the dietary spores (z = 2.484, p = 0.013). Note that the total spore number that each larva was confronted with was approximately 4×10⁷ for the 1×10⁹ mL⁻¹ and 4×10⁸ for the 1×10¹⁰ mL⁻¹ concentration of the original suspension used to prepare the diet (see for details). ## Dose Response Curves for *Btt* Infection The infection system allows for exposure to precise doses of dietary bacterial spores by adding different concentrations of spores per mL to the flour the experimental animals are kept on. This enabled us to study in more detail how the infection success of *Bt* depends on the spore exposure dose. For this, we used the *Btt* strain since it was the only strain causing significant mortality of *T. castaneum* larvae, and we used SB, GA-2 and Cro1 insect populations to test whether dose-response curves are population specific. For spore concentrations above a threshold concentration of 10⁸ spores per mL, all three populations showed a clear dose-dependent mortality, but the populations differed in the dietary concentration of spores required to kill a certain proportion of larvae. Over a broad range of spore concentrations, the wild population Cro1 was found to be around 30–40% less susceptible than the two laboratory populations. The lowest of the tested spore concentrations that resulted in reduced survival of larvae in all three populations was 5×10⁸ mL⁻¹ (z = 3.643, p = 0.0003). When fed on the highest spore concentration tested (5×10¹⁰ mL⁻¹), some larvae of the laboratory populations SB and GA-2 were still alive at day seven, but all had died by day 13 (data not shown). ## Differences in Susceptibility to *Btt* among ten Beetle Populations Data obtained from the previous two experiments indicated that beetle populations may differ in their susceptibility to *Btt*. We therefore further compared the susceptibility of ten beetle populations to test this finding in more depth. Our ten populations showed substantial differences in their susceptibility to *Btt*, varying from 40%–85% survival after seven days of constant exposure to spores. When compared to the standard laboratory population (SB), populations Cro1, Cro2, 50, 57, and 61 (Cro1: z = −2.527, p = 0.011, Cro2: z = −5.696, p = \<0.0001, 50: z = −1.948, p = \<0.0001, 57: z = −5.005, p = \<0.003, 61: z = −3.004, p = 0.004) had higher survival rate when fed on *Btt* spore-containing diet (5×10⁹ mL⁻¹). The majority of larvae died during the first day of exposure; mortality was strongly reduced on the second day, and on the third day only a small percentage of the larvae died. In most of the populations, no mortality was recorded thereafter. ## Adult Susceptibility to *Btt* and *Btk* Strains Despite our observation that the adults (SB, GA-2 and Cro1 population) fed on the *Btt* spore-containing diet (5×10⁹ ml⁻¹), no mortality was recorded during seven days of exposure. This experiment was repeated twice with the same results. It has previously been shown that adult *T. castaneum* are susceptible to purified toxin formulations of *Bt kurstaki* (*Btk*). We therefore tested the susceptibility of adults of the SB beetle population to *Btk* spores (5×10⁹ mL⁻¹), however, similarly to *Btt*, no mortality was observed. ## Plasmid Exchange between *Btt* and the Non-pathogenic *Bt* 407*gfpcry* ⁻ We were able to transfer pathogenicity factors from *Btt* (naturally neomycin resistant) to the non-pathogenic, green fluorescent protein (GFP)-expressing *Bt* 407*gfpcry* ⁻ that is erythromycin resistant. After conjugation and selection on neomycin and erythromycin, we identified a number of double resistant clones. The selected clones were all of the 407 genetic background, which was confirmed by Rep-PCR and which would imply that the *gfp* carrying plasmid was not transmittable from *Bt* 407*gfpcry* ⁻ to *Btt*. We tested for the presence of the *cry3A* gene with a *cry3A*-specific PCR. *Btt* carries two plasmids, a smaller one with unknown virulence factors and a large *cry*-carrying plasmid. A large proportion (around 90%) of the tested clones was *cry* negative. Since the negative clones were able to grow on neomycin, this indicated that the resistance for this antibiotic may be present on the smaller plasmid. These clones were denoted as *Bt* 407*gfp*-*neocry* ⁻. We were not able to test the toxicity of these clones, since the bacteria did not sporulate in the presence of both antibiotics, even after two weeks of growth in spore-culturing conditions. Of the double resistant clones, five tested positive for *cry3A* and were denoted as *Bt* 407*gfp-neocry* ⁺. One clone was chosen for further analyses. This conjugated strain *Bt* 407*gfp- neocry*⁺that had the large *cry*-carrying plasmid was able to induce considerable mortality in SB and Cro1 beetles. Mortality was lower compared to the original *Btt* strain). The mortality pattern during the seven days of spore exposure was similar to the *Btt* strain, with the majority of larvae dying on the first day. We noticed that the large *cry*-carrying plasmid was rather stably retained in the conjugated *Bt* 407*gfp-neocry*⁺strain. In the majority of cases where the conjugated strain was raised in the absence of antibiotics, the plasmid remained present. However, upon repeated freezing and thawing of glycerol stocks, the plasmid was lost at a higher rate. We observed plasmid loss also when the strain was raised with erythromycin alone, or with both antibiotics (erythromycin and neomycin) together. However, when raised with neomycin alone, the *cry* gene was retained (as detected by PCR), but the GFP signal was lost, suggesting that harbouring all three plasmids comes with a cost for the cells. ## Limited Exposure Time to *Btt* Spore-containing Diet In the previous experiments, larvae were continuously kept on spore-containing flour. However, since most larvae died on the first day of exposure, continuous exposure may not be necessary to achieve mortality. Therefore, to analyse in more detail the behaviour of the ingested pathogen and the infection dynamics in the host, we limited the exposure time to the spore-containing diet. We therefore tested the exposure time necessary to induce mortality, and kept *T. castaneum* larvae (SB population) on spore-containing flour (*Btt*, 5x10⁹ spores ml⁻¹) for between 30 and 180 minutes before transferring them to spore-free diet and followed their survival. Larval mortality 24 hours post initial exposure (PIE, here defined as the start of the 180 min. exposure period) occurred with only 60 minutes of exposure, although it was significantly different from the control treatment after 120 minutes (z = 2.311, p = 0.021). After 180 minutes of exposure, mortality reached values equivalent to continuous exposure and no mortality was recorded 48 hours PIE. This suggests that the number of spores required to induce mortality is possibly rather low and that the first physiological changes in both the host and the parasite that contribute to mortality probably occur quite early in the process of infection. ## Larval Mortality Rate We used the information from this experiment to obtain a more complete picture of the course of larval mortality following the infection. We exposed larvae of the SB population for 180 minutes to 5x10⁹ spores ml ⁻¹ (*Btt*, *Bt* 407*gfp-neocry*⁺and the control strains) and then transferred them to fresh flour without spores. We subsequently screened survival every hour until twelve hours PIE, and then again at 24h and 48 hours PIE. A small number of larvae (4%) had already died during the 180 min. of exposure. The survival curves for both pathogenic strains (*Btt* and *Bt* 407*gfp-neocry* ⁺) followed the same mortality trend. Larvae started dying seven to eight hours PIE and mortality was more strongly induced at 10 and 12 hours PIE. However, most larvae died between 12 and 24 hours PIE. Although the survival curves of the pathogenic strains showed similar mortality rates, *Bt* 407*gfp-neocry*⁺again induced lower mortality in comparison to *Btt* (z = 5.164, p = 0.007). ## Infection Dynamics of *Btt* Infection To describe the infection process in more detail, we monitored bacterial growth in the host (SB beetle population) hourly until 13 hours PIE, since *Btt* induced fast mortality in previous experiments. Larvae were exposed for a maximum of three hours to *Btt* spores (5x10⁹ spores ml⁻¹), and observations started at two hours PIE. At two hours PIE, only germinating spores, which appeared dark under phase contrast, were observed in the midgut. After about two to three hours PIE, the spores started to elongate into growing cells. Already at four to six hours PIE, 23% of infected larval midguts had a high load of vegetative cells, rising to 51% of the larvae seven to nine hours PIE. The midgut was entirely filled with bacteria, which seemed to be retained inside the midgut since they were not entering the surrounding buffer through the midgut wall after dissection. We were not able to observe bacteria in the haemolymph at any time point, although the haemolymph appeared darker in colour in some individuals suggesting the activation of an immune response (phenoloxidase reaction). After the originally ingested spores had germinated, only vegetative cells were observed during the following time- points. The formation of new spores was only observed in larvae that had been dead for one to two days, and after seven days the vast majority of bacteria inside the larvae had sporulated. Overall, in about 30% of the cases, neither germinated spores nor vegetative cells could be observed in the midgut. ## Spore Load of Cadavers after Infection with *Btt* We measured the spore load of larval cadavers that had died from *Btt* infection. The mean total spore number per larvae was 1.83×10⁷, but it varied considerably among cadavers. We compared the spore load of larvae that had died on different days after being constantly exposed to the dietary spores. The spore load was significantly higher in larvae that had died on the first day as compared to those that had died on the third day after spore exposure (Wilcoxon test, χ² = 9.06, df = 2, p = 0.011, error bars 1+/− SE). The spore load recovered from cadavers was much higher than the amount of spores larvae had originally ingested, which is indicative of successful replication of *Btt* inside *T. castaneum*. # Discussion The red flour beetle *T. castaneum* and the bacterium *B. thuringiensis (Bt)* provide a useful oral infection model system for experimental studies of host- pathogen interactions. The system enables the simultaneous study of bacterial infection strategies and responses of the host in a well-studied insect model organism. The four *Bt* strains that we tested carry different Cry toxins and have previously been shown to be able to induce mortality in coleopterans. We found that only the Cry3Aa producing strain, *Btt*, resulted in significant mortality of *T. castaneum* larvae when exposed to spore - toxin mixtures. The main factor causing pathogenesis of the *Btt* strain is not fully clear. *Btt* produces the Cry3Aa toxin, which when applied in its purified form (without the addition of spores), caused mortality to the yellow mealworm beetle *Tenebrio molitor* but not to *T. castaneum*. A recent study reported a very low mortality in *T. castaneum* when exposed to spore-crystal mixtures of Cry3Aa producing strain. The different mortality observed in this study might come from a different bacterial chromosomal background or from differences in spore concentrations that were provided in the diet. Heimpel and Angus (1960) categorised *Bt* susceptible lepidopteran insects into three types. The first and second types are susceptible when they are subjected to the toxin preparations alone, with differences in the speed with which mortality is induced. The third type of insects are not susceptible to the toxin alone, but a spore-toxin formulation is required for pathogenesis. Since the purified toxins from the *Btt* isolate are not sufficient to induce mortality in *T. castaneum* larvae, infection success here as well might rely on spore - crystal synergism, as suggested previously by Li et al. Further research is necessary to determine the exact role of the Cry protein and the possible relevance of spore - crystal interactions in the infection process in this system. *Btt* was originally isolated from a larva of *T. molitor*, a species which is closely related to *T. castaneum*. The nucleotide sequence of *Btt*’s *cry* gene, is identical or very similar to the toxin genes of other isolates that have been found to induce mortality of coleopterans –. *T. castaneum* larvae were shown to be susceptible to *Bt* isolates from Egypt and to strains isolated in Pakistan, but no further information on these isolates is available. To the best of our knowledge, the present study is the first showing *T. castaneum* susceptibility to the *Btt* strain. In contrast to oral infections, studies where *Bt* has been introduced into the haemocoel via septic wounding, non- coleopteran *Bt* strains were able to induce significant mortality. Such septic infection, which may also occur in nature, circumvents the infection processes in the gut, where specificity is mediated through Cry proteins, which bind specifically to host receptors in the gut epithelium. We were able to transfer pathogenicity from *Btt* to a non-coleopteran strain of *Bt* (*Bt* 407*gfpcry* ⁻) through the transfer of plasmids. Mortality induced by the conjugated strain was somewhat lower than with *Btt*, suggesting that additional virulence factors might reside on the chromosome of the original pathogen. Alternatively, if the Cry protein plays a crucial role in acting synergistically with the spores to cause mortality, the lower virulence might be caused by a lower copy number of the transferred plasmid or reduced expression of the *cry* toxin gene in the conjugated strain. The exchange of plasmids from *Btt* to a non-pathogenic strain may also be interesting to assess the role of the exchange of genetic material for the evolution of pathogens and as a factor contributing to the maintenance of genetic diversity and virulence in natural *Bacillus* populations. Plasmids can easily be exchanged in some *Bt* strains, which may broaden the host range of these bacteria. Most parts of the gut of *T. castaneum* are far less basic than typical lepidopteran guts, such that spores probably start to germinate immediately after ingestion, as they do in vitro. For this reason, infection can potentially depend on the early toxin-induced damage and reduced gut peristalsis, which enables the bacteria population to grow and remain in the midgut, but this needs to be experimentally verified. Note that food passage from the mouth to the ileum is only 60 minutes in *T. castaneum*. We observed rapid bacteria proliferation in the gut only a few hours after the feeding had started, which was not observed within the spore and flour/yeast mixture, so this cannot be attributed to the bacteria feeding on the beetle diet. It would be interesting to investigate in more detail the reasons for differential susceptibility to *Btt* of larvae from different populations. Such differences may be due to a number of reasons, including differences in the immune responses of the different populations against the bacteria. Resistance may be related to genetic diversity of the host, since both recently captured populations and the outcrossed line (OC) showed rather high resistance, compared to most of the laboratory lines. Alternatively, populations may differ in their associated microbiota, which might play a role, even though the midgut microbiota is not the sole reason for the infection success of *Bt*. Adults seemed to be resistant when subjected to the same dose of spore-crystal preparations as larvae (5×10⁹ spores ml⁻¹). By examining the spore discs, we observed that the adults did not avoid the infectious diet. The potential reasons for adult resistance may include superior processing of the toxin in the midgut, but also immune responses that are more efficient against *Bt*. Moreover, there are morphological differences between the midgut of larvae and adults, with adults having numerous regenerative crypts along the surface of the midgut. This could potentially confer resistance through faster regeneration of epithelial cells as suggested by Ferre and van Rie, but this hypothesis needs further investigation. Interestingly, previous studies showed that commercial formulations of purified toxins from a lepidopteran specific strain *Bt kurstaki*, induced mortality in *T. castaneum* adults. However in our study, spore - toxin preparations of *Btk* induced mortality neither in adults nor in larvae. Most of the *Bt* infection scenarios have been described for lepidopteran insects – whose midgut physiology markedly differs from coleopteran insects. *Bt* has been reported to cause death through general septicaemia, by invading host tissues from the midgut. This may involve repeated resporulation of the vegetative cells in the midgut, which facilitates the production of highly concentrated Cry crystals. The course of *Bt* infection in *T. castaneum* seems to differ from the infection process described for lepidopteran hosts, and does not follow the expectations for a typical intoxication process as observed in other insects. We could not detect bacteria in the hemolymph of live larvae, nor did we observe the formation of new spores as long as the insect was alive, although it is possible that numbers of bacteria in the haemolymph or spores in the gut could have been below our detection limit. A possible reason as to why resporulation may not be required is that vegetative cells also express the *cry* gene in the *Btt* strain. The infection process in the *T. castaneum* larval gut was fast and mortality was rapidly induced. Fast killing may be advantageous for *Btt* since as long as the host is alive it has to overcome its immune system. By killing quickly, *Btt* can exploit the hosts' nutrients and sporulate, which enables further infections and pathogen propagation. This strategy may explain our somewhat puzzling observation that the pathogen achieved lower spore load in larvae that died later than those that died on the first day after exposure, which differs from the observations made in lepidopteran insects where insects that died on the third day had more spores than those that died earlier or later after infection. Different host species were shown to offer a more favourable environment for *Bt* replication than the others, a phenomenon which is present within one species and is time-of death dependant; nevertheless, it is different for different *Bt* species that might have evolved specialisations for different insect orders. The ecology of *Bt* is not completely understood since most of the *Bt* spores are abundantly found where the target hosts are not always present. Although transmission of *Bt* in nature is not well characterised, a higher prevalence of entomopathogenic (toxin-carrying) *Bacillus* isolates in soil was correlated with the presence of insect cadavers in a field trial, and specialisation of a certain isolate for lepidopteran insects has been suggested. Of particular relevance to the ecology of *T. castaneum* is the observation that *Bt* has been isolated from animal food mills and that *Bt* strains isolated from granaries have been shown to be able to induce mortality in *T. castaneum*. Moreover, a *Bt* isolate that has a highly similar *cry* gene sequence to the sequence of the *cry3Aa* gene from the *tenebrionis* strain was isolated from dead *Tribolium* sp.. The cannibalistic nature of *T. castaneum* and other *Tribolium* species provides the opportunity for them to come into contact with a high dose of spores if they cannibalise larval cadavers. In support of this hypothesis, we observed that some larvae that were allowed to feed on three week old infected larval cadavers for two days, subsequently died and their bodies were loaded with *Btt* (data not shown). Although more detailed experiments are needed to verify this observation, it tentatively suggests that the reproductive cycle of *Bt* can be completed in *T. castaneum* in nature. # Materials and Methods ## Insects, Bacteria and Infection Protocol ### Insects Our study is based on eight laboratory and two wild populations of *T. castaneum*. Genetic differentiation between the populations and some degree of inbreeding might be expected in the laboratory populations due to potential genetic bottlenecks at the time of collection and the time for which they have been kept in the laboratory. The San Bernardino population (SB) originates from Alexander Sokoloff, California. The outcrossed population OC Münster was produced in our laboratory by crossing 10 different laboratory populations: 43, 50, 51, 52, 53, 55, 57, 58, 59, 61, which had been provided by Michael Wade (Indiana University, Bloomington, USA), together with the GA-2 population. The Croatia 1 (Cro1) and Croatia 2 (Cro2) populations are presumably the most genetically diverse since they were collected recently (Croatia, May 2010: Cro1∶45° 48′ 55.98′′, 16° 17′ 12.7968′′, Cro2∶46° 0′ 11.9628′′, 15° 50′ 39.195′′), and were established from multiple individuals from random mating pairs (165 pairs for Cro1 and 27 pairs for Cro2). The offspring of the pairs were used to establish the stock populations. Both of the wild populations are kept as large stock populations (ca. 10,000 individuals each) and were allowed to adapt to laboratory conditions for about 14 generations (1 year and 6 months) before the experiments started. All beetles were kept on heat-sterilised (75°C) organic white flour (type 550) with 5% brewer’s yeast at 30°C, 70% humidity and a 12h/12h light-dark cycle. ### Bacterial strains In this study, the susceptibility of *T. castaneum* to *Bt* bacterial strains was investigated in order to find the most suitable strain for investigation of host-pathogen interactions. Strains for the infections were chosen according to their Cry toxins. *Bt tolworthi* and *Bt kumamotoensis* both carry toxins that are toxic against coleopteran insects. *Bt morrisoni* bv. *tenebrionis* is toxic to coleopterans,. *Bt kurstaki* is a lepidopteran-specific strain although purified toxins were found to be active against *T. castaneum* adults. All *B. thuringiensis* strains were provided by the *Bacillus* Genetic Stock Center (BGSC, Ohio State University, USA) except for the strains *Bt* 407*cry* ⁻ and *Bt* 407*gfpcry* ⁻, the latter of which carries a green fluorescent protein (GFP) marker. These strains were kindly provided by Dr. Christina Nielsen-Leroux, Institut National de Recherche Agronomique, La Minière, 78285 Guyancourt Cedex, France. ### Production of spore-crystal preparations Spores were freshly produced before each infection using a modified version of a previously described protocol. Vegetative cells and spores were cultured at 30°C. Bacteria from a glycerol stock (stored at -80°C) were plated on LB agar and grown overnight. This was done freshly before each infection to prevent loss of pathogenicity by long-term storage of bacteria on LB agar plates. The following day, 5 ml of BT medium (w/V–0.75% bacto peptone (Sigma), 0.1% glucose, 0.34% KH₂PO₄, 0.435% K₂HPO₄) was inoculated with one bacterial colony with the addition of 25 µL of salt solution (w/V–2.46% MgSO₄, 0.04% MnSO₄, 0.28% ZnSO₄, and 0.40% FeSO₄) and 6.25 µL of 1M CaCl₂×2H₂O and allowed to grow overnight on a bacterial shaker at 200 rpm. The following day, the resulting bacteria suspension, 5 mL of salt solution and 250 µL of 1M CaCl₂×2H₂O were added to 1 L of BT medium, and it was further incubated for a total of seven days in darkness. On day four, another 5 mL of salt solution and 250 µL of 1M CaCl₂×2H₂O were added. After seven days the suspension was centrifuged at 4000 rpm for 15 minutes, washed once in phosphate buffered saline (PBS) and then resuspended in PBS. The spores were counted with a Thoma counting chamber. Such spore preparations together with their crystals (spore-crystal preparations) were stored for a maximum of three days at room temperature and protected from light until they were used in experiments. ### General infection protocol For the infection of *T. castaneum* larvae, a modified protocol from Oppert (2010) was used. The desired spore concentrations were adjusted by adding PBS, and 0.15 g of heat-sterilized flour with yeast was added per ml of spore suspension. Forty microliters of the resulting liquid diet was pipetted into each well of a 96-well plate (Sarstedt, Germany) under sterile conditions. The diet for the control insects was made in the same way but without the addition of spores. The open 96-well plates were then placed in plastic boxes (Tupperware), three in one box. Six holes were punctured in the lids of boxes (3 cm diameter) and plugged with foam stoppers (K-TK e.K., Germany) (4.2 cm diameter) to allow the air to circulate. The boxes were placed in a 50°C oven overnight to allow the spore-crystal discs to dry. Once the spore-crystal suspension had been mixed with flour it was only used on that same day in order to prevent spore germination and bacteria growth in the medium. The drying process did not allow for any spore germination and bacterial proliferation in the disc, which was confirmed by examining the disc under the microscope (400×magnification) before the infection. After the drying process, one larvae was added per well and the 96-well plates were sealed with transparent self- adhesive tape and holes were punctured to allow air circulation in each well. The 96-well plates were placed back into the plastic boxes and were kept as described before at 30°C and 70% humidity for infection. This protocol minimises the risk of contamination with spores and is suitable for rapid infection of a large number of individuals. The larvae remain constantly visible, which enables easy screening of survival (up to 3000 individuals per person, per hour). For laboratory surface sterilisation, 4% Incidin Active (Ecolab) was used. For each infection, 13–14 day old larvae (approximately 4 mm long) descending from approximately 200–300 one month old parents were allowed to feed for varying amounts of time, depending on the experimental setup. Since the *Btt* spores were homogenously mixed into the flour, the larvae are unlikely to selectively avoid taking up the spores from their food. However, an avoidance strategy could be to stop feeding when food is recognised as infectious. *T. castaneum* larvae can tolerate starvation for a maximum of 2 weeks, such that it would be possible that the exposed larvae that did not die early on during the exposure stopped feeding and died from starvation later on. To exclude this possibility, we verified that larvae had fed during the days of exposure by examining the flour feeding discs. The majority of larvae had eaten; however the feeding rate seemed reduced in comparison to control animals. Adult beetles (approximately two weeks post eclosion) were infected in the same way. Dead larvae were recognisable by the black body colour, or their immobility when touched with the tip of an injection needle and the relaxation of their legs. The concentration of spores used in the experiments is expressed as concentration of spores per mL of the original suspension that was used to prepare the diet. Since the liquid evaporates during the overnight drying process of the flour discs, spores per mL can be expressed as spores per 150 µg of flour with yeast, which was the amount of flour that was added per mL of suspension. Each larva was confronted with 40 µL of the spore-containing liquid diet, therefore the total spore number per disc is approximately the spore concentration per mL divided by 25. Furthermore, larvae eat a small portion of this diet, which would indicate that the number of spores necessary to cause mortality is potentially low. ## Experimental Design ### Insecticidal activity of different *Bt* strains to *T. castaneum* larvae We analysed the susceptibility of three beetle populations (SB, GA-2 and Cro1) to four different *Bt* strains whose toxins or spore-toxin preparations have previously shown toxicity towards coleopteran insects: *Btt*, *Btk*, *Btkm*, *Bttw*. Spore concentrations of 1×10⁹ and 1×10¹⁰ ml⁻¹ were tested for each bacterial strain. Larvae were kept constantly on spore-containing diet and the survival was assessed daily for seven days. Forty eight larvae were used for each of the treatment and the control groups. ### Dose response curves for *Btt* infection To test the insecticidal activity of different spore concentrations of *Btt*, a dose response curve was performed using the following concentrations of spores per ml⁻¹∶1×10⁶, 1×10⁷, 1×10⁸, 5×10⁸, 1×10⁹, 3×10⁹, 5×10⁹, 7×10⁹, 1×10¹⁰ and 5×10¹⁰. Larvae from the SB, GA-2 and Cro1 populations were kept constantly on spore-containing diet and survival was assessed daily for seven days and then on the 13^th day. Forty eight larvae were used for each of the treatment and the control groups. ### Differences in susceptibility to *Btt* among ten beetle populations To test the susceptibility of ten beetle populations that were collected from different regions of the world, the *Btt* spore concentration was adjusted to 5×10⁹ ml⁻¹. Larvae were kept constantly on spore- containing diet and the survival was measured daily for seven days. Ninety six larvae were used for each of the treatment and the control groups. ### Adult susceptibility to the *Btt* and *Btk* strains The susceptibility of adults (SB, GA-2 and Cro1) was tested with a *Btt* spore concentration of 5×10⁹ ml⁻¹. In a previous study it was shown that *T. castaneum* adults are susceptible to purified toxins of the *Btk* strain, therefore in a separate experiment, we tested the susceptibility of beetles from the SB population to *Btk* spores (5×10⁹ ml⁻¹). The beetles were kept constantly on spore-containing diet and survival was assesses daily for seven days. Forty eight adults were used for each of the treatments and for the control group. ### Plasmid exchange between *Btt* and the non-pathogenic *Bt* 407*gfpcry ^–* The *Bt* 407*gfpcry* ⁻ strain is cured of a large Cry-carrying plasmid and carries a GFP marker linked to erythromycin resistance (pHT315-*paphA3':gfp*). The strain is well genetically characterised and can be easily genetically manipulated. Since the strain does not induce mortality in *T. castaneum,* we transferred plasmids via conjugation from the *Btt* strain in order to test whether we could also make it pathogenic. *Btt* carries two plasmids, a smaller one and a large plasmid that carries the *cry* gene together with other potential pathogenicity factors. *Btt* is naturally neomycin resistant. Bacterial conjugation was performed as described previously. The donor and recipient strains were grown separately at 30°C, 200 rpm, in Luria Broth (LB) medium with appropriate antibiotics overnight and were subsequently diluted 1∶100 into 7 ml of pre-warmed LB medium. Cultures were grown to an optical density (OD 600) of 0.5, and 250 µl of each strain were mixed together and incubated at 30°C and 180 rpm for 3 hours. To select for transconjugants, the suspension was plated on LB agar plates with neomycin (15 µg/mL) and erythromycin (10 µg/mL) and grown overnight. Individual colonies were screened by colony PCR (1. 2′–94°C, 2. 20′′–94°C, 3. 20′′–57°C, 4. 40′′–72°C (2.–4.×35), 5. 3–72°C), using the primers Col1A and Col1B. Before each experiment with the conjugated strain, the Cry3A gene was confirmed as present by heating 5 µl of spore suspension for 20 minutes at 90°C and the same PCR protocol as above was used with 2 µL of spore suspension. The genomic background of bacterial strains obtained after the conjugation was confirmed by repetitive extragenic palindromic sequence-based PCR analysis (Rep-PCR) as previously described, this is a DNA fingerprinting technique based on the generation of distinctive electrophoretic patterns via primers designed for Rep sequences. ### Bioassay with the conjugated strain The toxicity of the conjugated strain *Bt* 407*gfp-neocry*⁺was tested on larvae from SB and Cro1 beetle populations using the general infection protocol as mentioned previously. Strains that were used in this bioassay are summarised in. Besides *Btt* and the newly created *Bt* 407*gfp-neocry* ⁺, *Bt* 407*cry* ⁻ and *Bt* 407*gfpcry* ⁻ were used to control for the presence of different plasmids. Larvae were kept constantly on the spore-containing diet for seven days and survival was assessed daily. A sample size of ninety six larvae was used for each of the treatments. ### Limited exposure time to *Btt* spore-containing diet As observed in vitro in LB medium, spores germinate and elongate into vegetative cells in about 2.5 hours. We therefore expected the earliest formation of vegetative cells in the midgut to start at about 2.5 hours after the start of the exposure time. To analyse the infection dynamics after exposure to the spore-containing diet (5×10⁹ spores ml ⁻¹) we limited the exposure time, i.e. larvae of the SB beetle population were allowed to feed for 30, 60, 90, 120 and 180 minutes, after which time they were transferred to spore-free flour. Their survival was assessed daily for three days. Forty eight larvae were used for each of the treatment and the control group. ### Larval mortality rate Larval death rate was measured hourly to obtain a more detailed picture of the mortality dynamics. To test whether the *Btt* and the *Bt* 407*gfp- neocry*⁺differ, both strains were used in this experiment together with the control *Bt* strains. Larvae of the SB beetle population were exposed to spore-containing diet (5×10⁹ spores ml ⁻¹) for three hours after which they were transferred to spore-free flour. Survival was assessed hourly until the twelfth hour post initial exposure (PIE). A sample size of 48 larvae was used for each of the treatment and the control groups. ### Infection dynamics of *Btt* infection To analyse infection dynamics, larvae of the SB beetle population were exposed to *Btt* spores (5×10⁹ spores ml ⁻¹) for three hours after which they were transferred to spore-free flour. Larvae were collected hourly until the thirteenth hour PIE. To observe whether *Btt* bacteria are able to invade the haemolymph from the midgut, each larva was first punctured dorsally with the tip of an injection needle (0.3 mm diameter) between the first and the second segment and the haemolymph was collected with a 1 µL capillary (Hirschmann GmbH). The amount of haemolymph that could be collected was on average about 0.1 µL. In several cases the haemolymph extraction was unsuccessful because the infection had already progressed so far that the body had become soft. The haemolymph was added to a droplet of PBS buffer on a microscope slide and was observed under the microscope using phase contrast (400×magnification). The same larva was then ice anesthetized, placed on a Petri dish and the first and the last segment were removed with a razor blade. A drop of PBS was added and the gut was carefully pulled out with a pair of forceps. Bacteria were observed after the midgut was homogenised with a pair of needles (200×and 400×magnifications). In total, 120 larvae were used for the analysis. Because the alimentary canal isolation and the analysis itself were time consuming, the same time points were done across different days. The observed characteristics of infection dynamics were similar for the same time point when analysed on different days. On average twenty four larvae were analysed per time point. ### Spore load of cadavers after infection with *Btt* Larvae of the Cro1 population were kept constantly on spore-containing diet (*Btt*, 5×10⁹ spores ml ⁻¹) and to quantify the spore load, larvae that died on the first, second and third day (n = 6 for each day) were separated daily. To ensure complete sporulation, cadavers were used that were ten days old. Cadavers were individually homogenised with a pestle in 200 µL of PBS. The suspension was subsequently pushed through a cell strainer with a 40 µm nylon mesh (BD Biosciences) by using a pipette. The spores were counted with a flow cytometer (BDFacsCanto II) using 4.5 µm green fluorescent beads (Polysciences) as a reference, and analysed using BD FACSDiva Software. To estimate the mean total spore number in cadavers, 56 larval cadavers were randomly picked after seven days of constant exposure of larvae to the spores (*Btt*, 5×10⁹ spores ml ⁻¹). ## Statistical analyses Survival experiments were analysed using the R statistical package (R Development Core Team, 2011) version 2.11.1. Within R we used the Cox proportional hazard model (‘survival’ library: Therneau \[and original Splus-\> R port by Lumley\] 2011) to test the effect of the treatment on survival. In some cases the control or treatment groups had 100% survival. Because no event occurred, there was no contribution to the likelihood, and a cox model could not be fitted. Therefore we denoted one individual in the group as dead at the first timepoint, allowing us to fit the model. In the experiments testing different bacterial strains and concentrations, and for the dose response curve, we started with the full model (e.g., *fullmodel \<-coxph (Surv (timeofdeath, censor) ∼ Beetle population \* Bacteria concentration)* and then performed model simplification by backwards elimination of non-significant terms. The data for the spore load comparison and the regression analysis were analysed with JMP version 9 for Mac. The data for the comparison of cadaver spore load were not normally distributed (Shapiro-Wilk test) and did not have equal variances (Levene test). We therefore performed a nonparametric Kruskal-Wallis test to analyse the effect of treatment (day of death). Pairwise comparisons were then done for each of the three days in turn (Wilcoxon test) and to reduce the probability of type 1 errors we performed a Bonferroni correction (α = 0.0169). # Supporting Information We would like to thank Gunther Jansen (Department of Evolutionary Genetics, CAU Kiel, Germany) and Christina Nielsen-Leroux (Institut National de Recherche Agronomique, La Minière, 78285 Guyancourt Cedex, France) for providing the *Bt* 407*cry* ⁻ and *Bt* 407*gfpcry* ^– strains. We would further like to thank the members of *Bacillus* cluster of the DFG-SPP 1399 Host-Parasite Coevolution priority programme for their helpful suggestions and input during the development of this infection system. Special thanks to Hinrich Schulenburg (Department of Evolutionary Genetics, CAU Kiel, Germany) for his comments on the experimental data of the manuscript. We would also like to thank the group of Paul Schmid-Hempel (ETH Zürich, Switzerland) for providing *T. castaneum* laboratory populations and to Kristina Duvnjak, Igor Franić, Marijan Iveković, Djurdjica Bohucki and Sanja Jukić for their help in collecting the wild *T. castaneum* populations in Croatia. We are grateful for the comments by two anonymous reviewers and the editor. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BM RP SAOA JK. Performed the experiments: BM CS RP. Analyzed the data: BM CS SAOA JK. Wrote the paper: BM SAOA JK.

# Introduction Brain derived neurotrophic factor (BDNF) plays a crucial role in activity- dependent brain plasticity, especially in relation to learning and memory, such that long term potentiation and memory consolidation purportedly declines as we age. One variant, rs6265, also known as Val66Met, is the frequently studied functional single nucleotide polymorphism (SNP) in the BDNF gene at 11p13, resulting in a valine/methionine substitution. The Val allele of this SNP has been linked to exercise-related memory enhancement in humans. In a similar vein, exercise-induced memory enhancement in rats is linked to BDNF signalling in the hippocampus, and in the striatum. Adults carrying the Met allele have associated smaller brain activation in mesolimbic regions, and hippocampal function (e.g. long term potential and memory consolidation) is strongly associated with cellular mechanisms that are mediated by BDNF. Additionally, increased Body Mass Index (BMI) in children and adults is associated with BDNF, and excessive food intake may significantly reduce the production of BDNF, leading to detrimental effects on cognition. In addition, healthy young Chinese adults carrying the homozygous Met allele have smaller brain volume in frontal, temporal, cingulate and insula cortices and Japanese Met carriers, particularly females, have age- related atrophy in the bilateral dorsolateral prefrontal cortex. Moreover, in two separate experiments in young and old cohorts, it has been shown that elderly Met/Met carriers, in comparison to a younger cohort, recall fewer items on an associative memory test than Val carriers. This suggests that the genetic effects on cognition are more pronounced as cognitive resources diminish during the ageing process, particularly for those carrying the Met allele. However, elderly people have also been shown to have better Stroop performance and enhanced brain activation when carrying the Met allele and recent evidence links the homozygous Val allele of the Val66Met BDNF gene to deficits in working memory processes in healthy elderly. Thus, the data linking BDNF on brain structure and function in elderly is inconsistent. In this study, we focus on the commonly studied functional rs6265 in relation to brain volume and cognition. Three additional highly linked BDNF SNPs were genotyped (7103411, 7124442 and 2049045). Variants in rs7103411 and rs6265 have been linked to poorer cognitive performance on a battery of neuropsychological tests, particularly in memory for rs7103411 and for the met allele on rs6265 in elderly females, although perceptual speed for rs6265 met was shown to be significantly better than for Val carriers. Furthermore, rs7124442 is associated with the development of anorexia and bulimia nervosa, both weight-related psychiatric diseases also characterized by cognitive dysfunction. The fourth SNP rs2049045 has not been extensively studied. However, we did not include these in our analyses with brain volume and cognitive function. We decided only to examine the rs6265 SNP because it is the only functional polymorphism of the 4 genotyped, it has been extensively studied in other populations, and was highly linked to the other three SNPs in our cohortthe three non-functional SNPs were highly linked in our cohort to the functional polymorphism rs6265.. Against this background, here we aim to examine, for the first time, the functional BDNF SNP in our cohort, and its relation to memory-related cognition and brain structure in an elderly population of Swedish men and women from the Prospective Investigation of Vasculature in Uppsala Seniors (PIVUS) study. We hypothesise that variations on the functional BDNF allele in an elderly population will be associated with poorer cognitive function and structural brain differences in global brain volumes, and regions associated with memory (e.g. hippocampus, prefrontal cortex). # Methods ## Subjects A total of 367 elderly Swedish men (n = 181) and women (n = 186) between 70 and 75 years of age, who had had a brain scan, from the Prospective Investigation of the Vasculature in Uppsala seniors (PIVUS) were included in this study, which was conducted in Uppsala, Sweden. The study was approved by the Ethics Committee of the University of Uppsala and all participants gave written informed consent. The participants included in this study were a subset chosen from the total PIVUS cohort of 1016 people based on several inclusion/exclusion criteria described further below. From the original 1016, 409 participants had participated in a brain scan. Of these 409, 27 had either suffered a stroke between the ages 70 and 75, or there were indications of cortical infarcts, and were excluded. 2 subjects who had a Mini Mental State Score (MMSE) of 24 or below at the time of the MRI scan were excluded. Of the remaining 380 cognitively healthy subjects, 13 were excluded due to missing genotype data, leaving a total of 367 subjects available for further analyses. The maximum time the subjects could score on the TMT-B was 600 seconds. Subjects who reached this score were not included in analysis because a ceiling effect would weaken the analysis of variance. Finally, 12 subjects lacked TMT-B data, thus 345 subjects was used. See for demographics and genetic data of included subjects, and for PRISMA flowchart of exclusion/inclusions. ## Cognitive measures ### Trail Making Test The TMT is a test of various aspects of executive function (e.g. mental speed, divided attention and working memory) with varying difficulty \[i.e. TMT version A (TMT-A) and TMT version B (TMT-B)\], which is associated with frontal lobe and hippocampal function. During the TMT-A, participants are required to ‘join the dots’ representing 25 targets of ascending numbers (1-2-3-4-5…) on a sheet of paper. For the TMT-B, participants respond in the same manner, but instead alternate between a set of numbers (1-13) and a set of letters (A-L), again in ascending order (1-A-2-B-3-C-4-D-5…). In essence the TMT-B task incorporates working memory and set-shifting capabilities, as participants are required to follow mentally the progression of two consecutive yet differing lists. The participants must start their trial on the circle marked ‘Begin’, and continue until they reach the endpoint, a circle marked ‘End’. The goal is to complete the task as quickly as possible, and the outcome measurement is time in seconds. ### Verbal fluency test Performance on the verbal fluency test required participants to recall verbally and spontaneously as many different animal names as possible within 60 seconds. The time it took to give responses was recorded, with repeated answers discounted from the total score. ## Covariates In all assessments we used Body Mass Index (BMI), total matter (gray+white) volume (TMV), gender and education attainment as covariates, due their potential confounding effects on brain structure and function. We chose to include both gray and white matter volume because we are examining both volume and function (e.g. in terms of Trail Making performance), and gray matter is inextricably linked to white matter. For example, previous research links weight to brain structure, e.g., and to worse working memory performance and smaller brain volumes in the elderly. Furthermore, global and regional brain volume reduction is commonly observed in the elderly. Females have also been shown to have lower brain volumes than males. However, we chose to exclude cerebrospinal fluid (CSF), as there is evidence that in some cases, the measurement of CSF is most susceptible to error in SPM. Finally, education level was used as a covariate, given that it can impact on brain volume, particularly the hippocampus, and cognition in the elderly. The educational level (basic level = 1, college = 2, university = 3) for each subject was assessed by means of a standardized questionnaire. ## Genotyping and linkage disequilibrium analysis Four BDNF SNPs were genotyped: the functional rs6265 (position: chr11:27, 636, 492), rs7103411 (position: chr11: 27,656,701), rs7124442 (position: chr11: 27,633,617) and rs2049045 (position: chr11:27, 694,241). However, linkage disequilibrium analysis revealed high similarity between the four genotypes (D′\>0.9) and thus only the functional snp was subjected to further analysis in our cohort. We focused on the functional 6265 polymorphism, given that it has been extensively studied in other cohorts previously, and also because it is the functional variant of this gene, and thus perhaps more relevant to brain function. For rs6265, the A allele encodes for methionine (Met) and the G allele encodes for valine (Val). The A allele is less frequent in rs6265, making methionine the uncommon amino acid at position 66 of the BDNF protein. Genetic information is found in. The SNPs genotyped in our cohort have previously been associated with Body Mass Index (BMI) and cognition in elderly populations. BMI was considered a confound, given previous research on its link to brain structure, e.g., and used as a regressor in our analyses. Furthermore, we have previously shown in the same population, that being overweight is linked to worse working memory performance and smaller brain volumes. Linkage information was calculated using R statistical computing in with the “genetics” package installed and is found in. Genotyping of the four SNPs was carried out at the SNP technology platform at Uppsala University ([www.genotyping.se/](http://www.genotyping.se/)) using an Illumina Golden Gate Assay (Illumina Inc., San Diego, CA, USA). Genotyping was conducted by investigators who were blind to the purpose of the study, and deviance from the Hardy-Weinberg equilibrium was assessed using a chi-squared test, and rs6265 can be considered to be in equilibrium (p\<0.05).. ## MRI acquisition Global and regional measures of brain volume were acquired with magnetic resonance imaging (MRI) and processed using voxel based morphometry (VBM) in SPM8 (<http://www.fil.ion.ucl.ac.uk/spm/software/spm8/>), a technique that uses an algorithm to determine local concentrations of gray matter (GM), white matter (WM) and cerebral spinal fluid (CSF) densities on a voxel by voxel basis. A high resolution three-dimensional T1-weighted volumetric turbo field echo scan was acquired using a Philips 1.5 Tesla scanner (Gyroscan NT, Philips Medical Systems, Best, The Netherlands). The three dimensional gradient echo sequence was used with scan parameters: repetition time (TR) 8.6 ms, echo time (TE) 4.0 ms, and flip angle 8 degrees. Sagittal slices with a field of view of 240 mm, a slice thickness of 1.2 mm, inter-slice interval of 0.9 mm and an in-slice resolution of 0.94×0.94 were reconstructed. ## MRI processing Morphological changes in GM were calculated using SPM8, by segmenting GM from white matter and cerebrospinal fluid using the unified segmentation approach. Following the segmentation procedure, probability maps of GM were ‘modulated’ in order to control for the effect of spatial normalization by multiplying the probability value of each voxel by its relative volume in native space before and after warping. We then normalized GM images, based on probability maps, into Montreal Neurological Institute (MNI) standard space for further contrast analyses in an additive model for each allele. Modulated images were smoothed with an 8-mm full width half maximum (FWHM) Gaussian kernel, in line with recent VBM studies by our group and others. This smoothing kernel was applied prior to statistical analyses to increase the signal to noise and to correct for image variability. ## Statistical analyses Statistical analysis was conducted on the functional BDNF SNP (rs6265) and its relationship to the dependent variables (TMT-B, verbal fluency) was done using R. We controlled for BMI as weight may be a significant confound on cognition, particularly hippocampal-related memory performance. Furthermore, sex may significantly alter the size and structure of the brain particularly in the hippocampus and other subcortical brain regions, and particularly in the elderly. Additionally, level of education attainment during adolescence may impact on cerebral volume. Thus it was important to identify weight, sex, education and total matter volume (TMV) as confounding variables. ## VBM analyses All VBM analyses were carried out using the SPM8 VBM GUI function (VBM 8.1 version 1.20; [www.fil.ion.ucl.ac.uk/spm](http://www.fil.ion.ucl.ac.uk/spm)). Our VBM analyses consisted first of contrasts between any A allele (Met) versus G (Val), and second, of linear regressions examining the association between genotype and brain volumes: all analyses corrected for the confounding variables. We chose to conduct both a contrast analysis and regression analyses, so that we could test both a dominant model and an additive model. The contrast analyses allow for the testing of a dominant model, that is, comparing any A allele (the presence of any Met allele) with the G allele (the presence of a Val only allele). However, with the dominant model approach we may fail to observe a more discrete linear relationship between gene and brain volume (with covariates) that is better examined with an additive model (accounting for 3 types of genotype: AA, AG, GG). We also conducted a post-hoc region of interest regression analysis in the bilateral hippocampus, due to its association with BDNF and memory e.g.. We selected the hippocampus mask using the PickAtlas anatomical labeling (aal) toolbox function in SPM8 (<http://fmri.wfubmc.edu/software/PickAtlas>), in order to avoid the problem of ‘double-dipping’. We ran a contrast analysis between any A allele (Met) versus G allele (Val) using the hippocampus mask, correcting for BMI, sex and total matter volume (TMV) by adding these variables as regressors. Subsequently we conducted a regression analysis with the hippocampus mask, using TMT-B score as the dependent variable, again correcting for BMI, sex and total matter volume (TMV). Unless otherwise stated, the activations we report are at this exploratory stage are uncorrected. # Results ## Participant demographics The A (Met) allele in rs6265 was slightly more common in females (67%). However, we demonstrate no significant gender differences (independent of SNP variation) between BMI, trail making B score, education and total matter volume.. We correct all subsequent analyses for gender (as well as BMI, total matter volume and education attainment). See. ## Association between SNPs and cognitive scores (Verbal fluency and TMT-B score) We examined associations separately between rs6265and scores on the Trail Making Task B (TMT-B) and Verbal Fluency task. Covariates to correct for potential confounds were total matter volume (TMV), BMI, sex and education level. Linear models investigating association with verbal fluency did not yield significant results. The BDNF 6265 SNP showed an uncorrected significant association with the TMT-B scores as shown in. depicts the association between TMT-B and the most commonly studied and functional rs6265 SNP. ## VBM results ### Whole Brain Contrast analysis in rs6265 See. By contrasting those with any A, representative of the met allele (e.g. AA, AG) versus no A, for rs6265, and correcting for matter volume, sex, education and BMI, we found that those with any A (Met) allele had increased volume in the right medial prefrontal cortex (MNI coordinates x = 20, y = 74, z = 10,t = 3.49, p\<0.001) and right caudate body (MNI coordinates x = 6, y = 16, z = 4,t = 2.74, p = 0.001), and decreased volume in the right occipito-temporal gyrus (MNI coordinates x = 52, y = −38, z = −16, t = 3.22, p = 0.001) and right orbitofrontal cortex (MNI coordinates x = 18, y = 20, z = −18, t = 2.31, p = 0.01). ### Whole Brain Linear Regression analyses in rs6265 See. In our first linear regression model, with total matter volume, sex BMI and education as covariates we examined how allele frequency for rs6265 was associated with brain volumes. We found greater bilateral cerebellar volume that most significantly regressed with the A (met) allele of rs6265 (MNI coordinates x = 58/−45, y = −64/−50, z = −38/−32, t = 3.74/3.12, p = 0.01/p = 0.01). Additionally, we found a FDR corrected peak level smaller brain volume in the right thalamus/brainstem (MNI coordinates x = 12, y = −8, z = −4, t = 3.14, p = 0.001) associated most strongly with the A (met) allele. See. In a second linear regression model independent of allele status, but this time to examine the association between TMT-B (working memory) scores and brain volume, we again used matter volume, BMI, education and sex as covariates of no interest. We found negative associations (the lower, and thus better the TMT-B scores, the larger the brain volume in the left cerebellum (MNI coordinates x = −38, y = −36, z = −34,t = 4.16, p\<0.001), right precuneus (MNI coordinates x = 26, y = −52, z = 72, t = 3.32, p = 0.03), and left superior frontal gyrus (MNI coordinates x = −16, y = 30, z = 56, t = 3.84, p = 0.008). We found positive associations (the higher, and thus worse the TMT-B scores, the larger the brain volume) in the left brainstem/pons (MNI coordinates x = −12, y = −30, z = −40, t = 3.95, p = 0.05) and bilateral posterior cingulate cortex (MNI coordinates x = −22, y = −44, z = 46,t = 3.07, p = 0.001/x = 34, y = −60, z = 22, t = 3.03, p = 0.001). As an additional step we re-ran the analyses to include allele code for rs6265 but this did not alter the results observed. ### Region of interest (ROI) Linear Regression analyses in rs6265 Finally, given the known relationship between hippocampus, BDNF expression and memory function e.g., we conducted post-hoc regression analyses, correcting for total matter volume, BMI, sex and education, with a region of interest (ROI) mask for the bilateral hippocampus to examine specifically the association with rs6265. As a first step with the ROI mask, we found a positive association in line with the A (met) allele and greater brain volume in the left hippocampus (MNI coordinates x = −24, y = −14, z = −16, t = 1.68, p = 0.02). In other words, those with the any A (met) allele had larger left hippocampal volume. As a second step, to examine the association between TMT-B performance and hippocampal volume, we ran a second regression using the same covariates and found a negative correlation (the lower, thus better the TMT-B score, the larger the brain volume) in bilateral hippocampus (MNI coordinates x = −16, y = −4, z = −22, t = 3.52, p\<0.001; x = 32, y = −12, z = −24, t = 3.13, p = 0.001). See. As a final step, to link allele variation status, TMT-B scores and brain volume in the hippocampi, we re-ran the above masked regression analysis but this time included a regressor to denote allele code (e.g. AA = 1, AG = 2, GG = 3). There was no significant correlation found for rs6265. # Discussion We show that the risk allele in the functional rs6265 (Val66Met) BDNF polymorphism is associated with deficits in memory-related cognition (working memory), and differences in regional brain volumes in an elderly population of Swedish men and women. We found that the any A allele (Met) on rs6265 is linked to faster working memory performance in this elderly population. Additionally, in an additive model regression analysis we showed that those elderly people possessing the seemingly beneficial any A (Met) allele on rs6265 had larger brain volumes in the bilateral hippocampus andbilateral cerebellum, and reduced volume in the right thalamus. In a dominant model contrast analysis, we found significantly larger volumes in the right caudate body, right medial prefrontal cortex (regions linked to memory, motor coordination and decision making), and smaller volumes in the right occipito-temporal lobe and right thalamus (regions linked to somatosensory responses and arousal). Furthermore, greater bilateral hippocampal, cerebellum and precuneus, with smaller brainstem and bilateral posterior cingulate were associated with better working memory performance independent of allele variation, in an additive model. No association was found between the 6265 BDNF functional SNP and verbal fluency performance or with Body Mass Index (BMI) in our elderly cohort, and no differences were found in global brain volumes. Thus, our data suggests that variation on the BDNF gene, specifically the rs6265 SNP contribute to cognitive and anatomical variation, and that the Met allele may be more beneficial to cognitive function in the elderly. Collectively, our data suggest that larger brain volumes in memory, decision-making and motor regions, combined with smaller volumes in somatosensory/arousal regions may indicate mediation effects by BDNF leading to greater cognitive control and less arousal interference (e.g. via peripheral autonomic responses) in elderly people. We examined the relationship between rs6265 and brain volumes using both additive (regression) and dominant (contrast) analyses, so that we could explore linear relationships, as well as extreme contrasts in brain volume between the allele variations on this gene. Consequently, we find different, though not contradictory results following these analyses, demonstrating that brain volume changes in some regions are more susceptible to the additive model, which are not observed in the more extreme dominant model. The dominant contrast model shows that prefrontal, occipito-temporal and caudate regions are altered by the presence of any Met allele, whereas the bilateral cerebellum and thalamus show differences when measuring the presence of none, 1 or 2 copies of the Met allele, in a more discrete additive regression model. Given the link between higher order learning, memory and the 6265 BDNF SNP (e.g. Mattson et al., 2004), our dominant contrast analyses were only able to show differences in brain regions associated with higher order learning and memory (e.g. PFC, occipito- temporal gyrus, caudate body), which may be the strongest link to the BDNF gene in terms of brain function. Whereas the additive regression models in our analyses showed links between the 6265 SNP and regions associated with motor coordination, arousal and emotion recognition systems (bilateral cerebellum, brainstem, precuneus and bilateral posterior cingulate), arguably regions that are more prone to individual variation, particularly in an ageing population with vast life experience differences. We found both larger and smaller gray matter volume in the rs6265 BDNF polymorphism, which may have functional significance. Specifically, greater brain volumes were found in relation to those with the Met allele, in the right medial prefrontal cortex, right caudate and bilateral cerebellum. Additionally, better working memory task performance in those with the Met allele was associated with larger left cerebellum, right precuneus and bilateral hippocampal volumes. These brain regions are perhaps collectively involved in better memory function, particularly in those elderly people who have the Met allele (especially the homozygous variant). Previous research has linked cognitive function in older adults to declining volume in the cortico-cerebellar pathways, and so having at least one copy of the Met BDNF allele may help to preserve these brain regions from age-related atrophy and cognitive decline. Although it must be noted that the Met allele of the 6265 BDNF SNP is associated with worse cognitive function. However, there is also some evidence that the 6265 BDNF Met allele influences hippocampal glutamatergic activity, which may positively influence memory in older adults. Thus, the effects on brain structure and function of having at least one, if not two copies of the Met allele in older age, is still not yet clear. Smaller brain volumes in those with the Met allele were also observed in our older cohort, in the right occipito- temporal gyrus, right OFC, and right thalamus, which may be associated with reduced function in attention and arousal brain networks, especially in relation to emotional processing. Perhaps those older adults carrying the Met allele are less aroused by emotional stimuli, which ordinarily interferes with memory processing. Consistent with this notion, better working memory performance was associated with reduced volumes in the brainstem and bilateral posterior cingulate, regions also associated with emotional arousal. BDNF regulates neuronal growth and brain plasticity in memory-related brain structures, such as the hippocampus, but the beneficial effects of BDNF on learning and memory seem to diminish with age. Our data suggest that poorer performance on a working memory task and related brain differences are specifically associated with the Val allele rs6265 BDNF allele in an elderly Swedish cohort, and that carrying the Met allele may have neuro-protective effects. This supports recent findings linking the homozygous Val allele of the Val66Met BDNF gene to deficits in working memory processes in healthy elderly. Furthermore, elderly Met carriers have been shown to be better at the Stroop task and show better brain function during tasks as measured by event related potentials. Our study expands on these findings by measuring brain volume in the elderly. However, ours and the data described above contradict previous reports that it is Met, not Val allele carriers who have poorer performance on a cognitive task and alterations in brain volume. Thus, genetic variation effects on brain structure and function (particularly memory) are obviously complex, and likely also involve the effects of many other factors, such as age, life stressors, diet, exercise etc. For example, in rs6265, young adults carrying the Met allele have deficits in brain function, young Met carriers have less exercise-related memory enhancement and elderly Met carriers have been shown to recall fewer items during a memory test than Val carriers, a difference that was not observed in a younger cohort. Furthermore, healthy young Chinese adults carrying the Met allele have smaller brain volumes in frontal, temporal, cingulate and insula cortices. Thus, it could be that the beneficial effects of carrying the Met rs6265 allele are more apparent in the elderly than in younger persons, or that other environmental factors play a large part. We found that those elderly people in our Swedish cohort carrying the Val allele, who were slower during the working memory task, had smaller right and left cerebellum volumes, caudate body and right medial prefrontal cortex, as well as larger thalamus, occipito-temporal and orbitofrontal volumes. Smaller cerebellar volume likely entails less efficient motor responses during this paper-based working memory task requiring motor dexterity, and indeed the cerebellum is involved in motor coordination, specifically sensorimotor control, with greater volumes linked to better performance on a working memory task. In support of this interpretation, Laing and colleagues reported that elderly carriers of the Val rs6265 allele had slower perceptual speed during paper-based memory tasks but they did not also measure brain volume in the elderly as we did. The caudate body is also involved in motor coordination, and is one of the main regions of atrophy underlying degenerative motor impairment. In a similar vein, smaller caudate volume in our Val elderly carriers predicted slower motor responses on the working memory task, thus BDNF-related deficits may specifically target dopaminergic neurons in the caudate in the elderly, leading to motor impairment. Furthermore, age-related alterations in connectivity between medial prefrontal and caudate regions are linked to memory impairments, suggesting that reduced medial prefrontal cortex observed in our Val carriers also contributes to working memory deficits. Larger thalamus volume may indicate plasticity due to greater sensory neural traffic, which in turn might interfere with cerebellar motor coordination, given the connectivity between these brain regions. Thus, smaller thalamus and greater cerebellar, caudate and medial prefrontal volumes in elderly persons carrying the Met allele may underlie better sensorimotor control and enhanced working memory performance, as seen in our cohort. We also observed that those elderly people carrying the rs6265 Val allele, who had worse performance on the working memory task, had larger occipito-temporal and orbitofrontal volumes. The occipito-temporal cortex is part of the ventral visual processing stream, serving recognition, visual thought, planning and memory (Milner, 2012), whereas the orbitofrontal cortex is involved in decision making. Greater volume in these regions may be indicative of greater effort to sustain effective cognitive performance in the elderly, especially if increased plasticity in regions underpinning peripheral autonomic arousal (e.g. heart rate, perspiration, muscle contraction) leads to interference effects on focused attention, such that heightened arousal may complicate decision making processes. In a similar vein, larger bilateral posterior cingulate cortex volume was associated with worse working memory performance in our cohort, a brain region involved in the default mode network, e.g. a network in which function is increased during idle periods, but reduced during focused attention. Functional decreases in the posterior cingulate cortex are associated with better cognitive performance following cognitive training. Thus, it might be, if we consider our other volumetric findings, reduced arousal responses in the brains of elderly subjects (e.g. heart rate, perspiration, muscle contraction) are linked to better cognitive and sensorimotor control. Reduced plasticity in midbrain and default mode network (e.g., thalamus, posterior cingulate cortex), but greater plasticity in cerebellum, medial prefrontal cortex and caudate, which are supported by BDNF expression, may reduce arousal interference and strengthen cognitive control. Elderly carriers of the Met allele may have hypoactivation in mesolimbic regions, as a consequence of reduced plasticity, which might be detrimental to motivated cognitive performance in the young. From this perspective, it might be beneficial for elderly, less agile subjects, to rely on cognitive, rather than arousal brain systems, which might explain the switch in the beneficial effects of the Met SNP between young and old. Faster working memory performance correlated with greater brain volume in the bilateral hippocampus in our elderly cohort but perhaps most significantly on the left side. This compliments previous extensive research that BDNF regulates brain plasticity and memory function, particularly in relation to the hippocampus e.g.. But to our knowledge, we are the first to demonstrate a link between hippocampus size and working memory performance in an elderly cohort, whose enhanced working memory performance was also associated with the Met allele of the rs6265 BDNF SNP. ## Strengths and limitations The main strength of this study is that the sample studied here was on a homogenous sample of elderly Swedish men and women from one town in Sweden. Thus, our cohort would have had similar educational and life experiences, which might otherwise lead to major confounds in a less homogenous population. We also took measures to ensure that our cohort were otherwise healthy in terms of cognitive ability and neuroanatomical infarcts. Furthermore, we corrected for sex, Body Mass Index, education and total matter volume which are all factors that may influence brain function and structure. Furthermore, unlike many other studies into BDNF in humans, we examined four highly linked BDNF SNPs and their influence on two cognitive measures and brain volume. However, we were only able to report uncorrected data in all but one cluster, which for an exploratory study such as ours is acceptable, but does highlight that greater statistical power is needed to detect more robust findings. Future VBM studies that can report FWE- or FDR-corrected clusters, to correct for multiple comparisons and to rule out spurious findings, will progress this work further. Additionally, due to a lack of longitudinal data, we cannot speculate on whether a specific BDNF SNP is more or less detrimental to cognitive function and brain structure in the elderly compared to younger populations. Finally, we included only measures of verbal fluency, mental speed, divided attention and working memory (using verbal fluency and Trail Making tasks), thus preventing an exploration into how variation on the BDNF allele might influence other cognitive functions, which could be explored further, by adding additional neuropsychological tasks in future studies. In summary, our data suggest an association between BDNF rs6265 (val66met) and working memory ability in an elderly population of Swedish men and women. Furthermore, greater brain volume was associated with the Met allele of rs6265 in the right medial prefrontal cortex, cerebellum, caudate body and hippocampi (regions linked to memory, motor coordination and decision making), as well as smaller volume in the right occipito-temporal gyrus and thalamus (regions linked to somatosensory responses and arousal). Additionally, larger bilateral hippocampal, cerebellum and precuneus, with smaller brainstem and bilateral posterior cingulate cortex were associated with better working memory performance independent of allele variation. Thus, the Met allele of rs6265 appears to be more beneficial to the elderly than the Val allele, in terms of brain health and memory function. Given their role, these brain regions suggest that the Met allele of BDNF promotes plasticity in functions associated with an ability to exert control over one's sensorimotor processes. The link to brain volume and working memory performance was independent of BDNF polymorphisms, and therefore it seems likely that other environmental factors are at play. Nevertheless, susceptibility to develop memory deficits in later life may be exacerbated by one's genetic predisposition. Given that exercise is linked to greater levels of BDNF in the brain, adopting healthier lifestyle choices, e.g. increasing one's daily exercise regime, may help to counter some of these detrimental cognitive effects, particularly in elderly persons. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SJB EN LL. Performed the experiments: SJB EN LL. Analyzed the data: SJB EN JAJ LL. Contributed reagents/materials/analysis tools: SJB LL EN. Wrote the paper: SJB EN DJS RF LL HBS.

# Introduction Myosin VI (encoded by the *Myo6* gene), a member of the actin filament-based molecular motor proteins, is the only myosin known to move towards the minus end of the actin filament thus far and appears to be involved in a wide range of cellular functions such as clathrin-mediated endocytosis, cell migration, vesicular membrane traffic, cell migration and mitosis. *Myo6* is expressed in the actin-rich cuticular plate of inner and outer hair cells of the ear and is fundamental for the development and maintenance of stereocilia. At least three mutations in *Myo6* have been associated with non- syndromic deafness in humans probably because of disruptions to the structure and function of stereocilia. Bats use echolocation, usually involving ultrasonic frequencies for orientation and often for foraging. Echolocating bats perhaps have the most sensitive high-frequency hearing among mammals, and echolocation calls emitted by most echolocating bats range in dominant frequency from 11 kHz to over 200 kHz. Such excellent auditory performance making echolocating bats fascinating mammals for studying genes associated with hearing. Recently, many studies have revealed that some genes associated with hearing have undergone positive selection in echolocating bats and cetaceans. Considering the important role of *Myo6* in hearing, it is reasonable to hypothesize that the *Myo6* may also be a target gene for positive selection in bats that use laryngeal echolocation compared with species that do not use laryngeal echolocation (the Old World fruit bats in the family Pteropodidae). *Myo6* is also expressed abundantly in the photoreceptor cells and retinal pigment epithelial (RPE) cells in the retina. Moreover, evidence from myosin VI functional null *Snell’s waltzer* (*sv*/*sv*) mice showed that myosin VI contributes to the normal functioning of retinal electrophysiology. Thus myosin VI also plays an unknown but important role in vision. It was suggested that laryngeal echolocating bats use echolocation rather than vision as major means of perceiving their environment. However, in addition to using olfaction, pteropodids without the ability of laryngeal echolocation presumably rely more on vision for orientation and finding food than other laryngeal echolocating bats. Thus it is also reasonable to hypothesize that positive selection may act on *Myo6* in pteropodids, as species in this lineage use vision primarily for orientation. Moreover, myosin VI is highly expressed in polarized epithelial cells such as kidney proximal tubule cells and intestinal enterocytes, where it is associated with clathrin-mediated endocytosis. In the kidney, large amounts of glomerular- filtered serum proteins are reabsorbed by proximal tubule cells relying on a process called receptor (megalin/cubulin)-mediated endocytosis. A recent study revealed that myosin VI plays an important role in this process via vesicle formation and the transportation of vesicles towards early endosomes. Renal proximal tubule reabsorption is very important for health, being responsible for the clearance of the vast majority of proteins filtered by the glomerulus. The impairment of this process will cause proteinuria, an excess of serum proteins in the urine. More importantly, as many serum proteins are carrier proteins binding many essential components including vitamins and trace elements, receptor-mediated endocytosis also accounts for the preservation of many essential serum components such as vitamin D, vitamin B₁₂ and iron. Among the bats (Chiroptera), Old World fruit bats and New World fruit bats (Phyllostomidae) have independently evolved a high-carbohydrate but low-nitrogen diet comprising mainly fruit and/or nectar. Although some New World fruit bats are known to supplement their diet with insects, and Old World fruit bats also accidentally or even deliberately consume insects, recent studies with stable- isotope analyses showed that Old World fruit bats and New World fruit bats in the subfamily Stenodermatinae are predominantly frugivorous. Most fruits that frugivorous bats mainly eat contain only 0.2–1.4% dry weight protein. Nectar also contains only traces of amino acids. Besides, fruits are also lacking many essential vitamins, especially fat-soluble vitamins such as vitamin D and vitamin B₁₂. Captive pteropodids fed on fruit-only diets show nutrient deficiencies and even neurological impairment caused by vitamin B₁₂ deficiency. Hence, how Old World fruit bats subsist on diets with low contents of protein and essential nutrients has long been of interest, and earlier studies focusing on behavior and physiology revealed some clues. Because *Myo6* is involved in the critical importance of receptor-mediated endocytosis for the preservation of protein and essential nutrients such as vitamins, we hypothesized that the *Myo6* gene may have undergone positive selection in the Old World fruit bats and/or the New World fruit bats. In this study, we sequenced the coding region of the *Myo6* gene from 15 bat species, and studied the molecular evolution of this gene in bats and other mammals. We also used real-time PCR assay to determine patterns of gene expression in a range of organs, including the eye, cochlea and kidneys, to understand where in the body proteins with potential functional significance are expressed. With the combination of molecular evolutionary analyses and real-time PCR assay, we intend to test our hypotheses that *Myo6* has undergone positive selection in echolocating bats for their high-frequency hearing, in Old World fruit bats for their effective vision and/or in Old World fruit bats and New World fruit bats for the preservation of protein and essential nutrients. # Materials and Methods ## Ethics Statement Our procedures involving animals were in accordance with the guidelines of Regulations for the Administration of Laboratory Animals (Decree No. 2 of the State Science and Technology Commission of the People’s Republic of China on November 14, 1988). Bats were sacrificed by cervical dislocation and their tissues were sampled immediately. Protocols were approved by the Animal Ethics Committee of East China Normal University (ID No: 20101002). The neotropical bat species were sampled in Mexico during April, 2010 for our previous study with assistance from Professor Rodrigo A. Medellín and Dr. Rafael Avila-Flores of the Instituto de Ecología, Universidad Nacional Autónoma de México, under the scientific collecting license number FAUT-0250, issued in the name of their collaborator, Dr. Gerardo Suzán. ## Taxonomic Coverage We sequenced the coding region of the *Myo6* gene from 15 bat species covering eight of the 17 extant chiropteran families. From the suborder Yinpterochiroptera, we included three Old World fruit bats from the family Pteropodidae that do not use laryngeal echolocation (*Cynopterus sphinx*, *Rousettus leschenaultii* and *Eonycteris spelaea*). Also from the Yinpterochiroptera we sampled five insectivorous echolocating bats from sister families to the Old World fruit bats: *Megaderma lyra* (Megadermatidae), *Rhinolophus ferrumequinum* and *R. pusillus* (Rhinolophidae) and *Hipposideros pratti* and *H. armiger* (Hipposideridae). For echolocating bats from the suborder Yangochiroptera, we studied two New World fruit bats *Artibeus lituratus* and *Leptonycteris yerbabuenae* (Phyllostomidae) and five insectivorous bats representing three families: *Tadarida plicata* (Molossidae), *Myotis ricketti* and *Pipistrellus abramus* (Vespertilionidae) and *Mormoops megalophylla* and *Pteronotus parnellii* (Mormoopidae). The family Mormoopidae is sister to the New World fruit bats. All new *Myo6* sequences were deposited to GenBank and accession numbers are JX023444-JX023458. We also obtained available published *Myo6* sequences of nine other mammal species from GenBank to provide a greater phylogenetic coverage for molecular evolutionary analyses: *Homo sapiens* (NM_004999), *Pan troglodytes* (XM_001144940), *Mus musculus* (NM_001039546), *Rattus norvegicus* (XM_001061392), *Bos taurus* (NM_001206072), *Canis familiaris* (XM_862495), *Ailuropoda melanoleuca* (XM_002923922), *Equus caballus* (XM_001503608) and *Sus scrofa* (NM_214021). Details on echolocation and food habits for all bat species and their corresponding *Myo6* accession numbers are listed in. ## Isolation, Amplification and Sequencing For these 15 bat species, total RNA was isolated from brain tissue (stored at −80°C) using Trizol reagent (Invitrogen) following the standard protocol, and 5 ug total RNA was reverse-transcribed into cDNA by SuperScript™ III Reverse Transcriptase kit (Invitrogen). The coding sequence of the *Myo6* gene was divided into four overlapping fragments (∼1000 bp for each), and four pairs of primers were designed to amplify these four fragments. All PCR products were isolated using a 1% agarose gel and purified with a Gel Extraction Kit (Qiagen), then ligated into pGEM-T easy vector (Promega), cloned and sequenced using the Terminator kits (Applied Biosystems) on an ABI 3730 DNA sequencer. Then the sequences of four fragments were assembled together to obtain the full length of *Myo6* coding sequences. Considering the specific expression of *Myo6* isoforms in brain and polarized epithelial cells such as kidney proximal tubule cells reported in a previous study, we cloned the fourth fragment of *Myo6* coding sequences from kidney cDNA for six bat species: *C*. *sphinx* and *R*. *leschenaultii* for the Old World fruit bats, *R*. *ferrumequinum* and *H*. *armiger* as representatives of yinpterochiropteran laryngeal echolocating bats and *M*. *ricketti* and *Scotophilus kuhlii* for yangochiropteran laryngeal echolocating bats. *S. kuhlii* (a close relative of *P. abramus* also from the family Vespertilionidae) was used because we did not have kidney tissue from *P. abramus*. Comparisons of isoforms from brain and kidney tissues showed that the only differences between kidney and brain isoforms was that isoform expressed in the brain contained a 9aa small insertion and those from kidney contained a 32aa large insertion, which was congruent with the results from a previous study. ## Phylogenetic Reconstruction The nucleotide sequences of 24 species were aligned using ClustalX and coding sequences were translated to amino acids using MEGA4. Then the Bayesian phylogenetic tree was reconstructed based on the aligned nucleotide sequences using MrBayes 3.1.2 with the TIM3+I+Γ nucleotide substitution model selected by jModelTest0.1. For the Bayesian analysis, we performed 10,000,000 generations of MCMC and sampled every 100 generations, with the first 2,000,000 generations discarded as burn-in, since the standard deviations of split frequencies were stable below 0.01 after 2,000,000 generations of MCMC performances. All other options and priors were the default settings of MrBayes 3.1.2 software. ## Molecular Evolution Analyses Since the potential occurrence of recombination could adversely affect the power and accuracy of phylogenetic reconstruction and the detection of positive selection, we first detected whether there was evidence of recombination in our dataset before molecular evolutionary analyses. We conducted GARD in the HyPhy package to detect the evidence of statistically supported recombination breakpoints (thus recombination). A species topology with 24 mammals was constructed based on the accepted species relationships. To test for positive selection in *Myo6*, we estimated the rate of nonsynonymous substitutions (d_N) and synonymous substitutions (d_S) using PAML CODEML. We first performed two-ratio models, in which the d_N/d_S ratios (termed as omega or ω) was allowed to differ between the foreground and the background. Firstly, to test selection pressure of *Myo6* gene in echolocating bats, separate two-ratio models were conducted with the foreground branch set as the ancestral branch leading to Chiroptera (all bats), Yinpterochiroptera, Yinpterochiroptera echolocating bats and Yangochiroptera, respectively. We also conducted two-ratio models on the ancestral branch leading to Old World constant frequency (CF) rhinolophids (species from the families of Hipposideridae and Rhinolophidae), because species of this lineage could emit CF echolocation calls at high-duty-cycles and strong evidence of positive selection has recently been found acting on a key hearing gene (*Prestin*) in this lineage. Finally, we also conducted two-ratio models on the ancestral branches leading to the Old World fruit bats and the New World fruit bats, respectively. In all cases, the one-ratio model, which assumes the equal *d*_N/*d*_S ratio among all branches, was performed as a null model. We also applied the test 2 of branch-site model A to detect positively selected sites along particular branches. In this model, the phylogeny is separated into two portions: the fixed branches are set as foreground and the remaining branches as background. Four site classes of codons are assumed (class 0, class 1, class 2a and class 2b). Class 0 and class 1 codons are assumed to evolve under purifying selection (0\<ω₀\<1) and neutral selection (ω₁ = 1) respectively throughout the phylogeny. Class 2a and class 2b evolve under, respectively, purifying and neutral selection on the background, but are grouped together as class 2 and allowed to evolve under positive selection (ω₂\>1) on the foreground. The null model was the modified branch-site model A with the ω₂ fixed as 1. We applied the test 2 of branch-site model A to the above seven ancestral branches which were tested by two-ratio models, i.e., ancestral branches leading to Chiroptera, Yinpterochiroptera, Yinpterochiroptera echolocating bats, Yangochiroptera, Old World CF rhinolophids, Old World fruit bats and New World fruit bats. All the results of alternative and null hypotheses were compared using likelihood ratio tests (LRTs). In addition to the methods of PAML, two alternative approaches were used to detect the evidence of positive selection in our dataset. First, the random effects branch-site model (Branch-site REL) was applied in the HyPhy package to examine site-wise variation across branches. Since no uniform section pressure was enforced across all background branches assumed to be not under positive selection, this method might be more robust to errors. Second, using the Datamonkey web server (<http://www.datamonkey.org/>), we conducted the GA-Branch method, in which a range of ω classes were assigned to branches without *a priori* designation of focal lineages of interest. ## Real-time PCR Real-time PCR assays were performed to determine the expression of the *Myo6* gene in major bat tissues. Three representative bat species were used: *C*. *sphinx* (a frugivorous bat from the Old World fruit bats), *R*. *ferrumequinum* (an insectivorous yinpterochiropteran echolocating bat) and *M*. *ricketti* (an insectivorous yangochiropteran echolocating bat). Adult individuals of these three bat species were sampled from the wild in China. Individuals were sacrificed humanely after capture and brain, eye, cochlea, heart, lung, liver, stomach, intestine, kidney and pectoral muscle tissues were stored in liquid nitrogen immediately for RNA preservation. For each species, three adult individuals were used for replication. Total RNA was isolated using Trizol reagent (Invitrogen) following the protocol, and cDNA was synthesized from 5 ug total RNA using SuperScript™ III Reverse Transcriptase kit (Invitrogen). Gene expression was analyzed using SYBR®Premix Ex Taq™ (TaKaRa) in the ABI PRISM 7300 real-time PCR system (Applied Biosystems) following the protocol (See for primer information). The amount of cDNA template for each tissue was fixed to 100 ng. The PCR products were sequenced for confirmation. The amount of *Myo6* were normalized to the *Gapdh* gene, and arbitrarily calibrated to pectoral muscle for each species using the 2^−ΔΔCt method. The Kruskal-wallis test was used to test for significant differences of the *Myo6* mRNA expression levels within ten tissues for each species. And pairwise comparison was done by the two-tailed nonparametric Mann-Whitney U test, when the Kruskal-wallis test yielded a statistically significant value (*P*\<0.05). *P*-value \<0.05 was considered as significant. # Results Our final *Myo6* gene sequence dataset comprised 24 taxa, including three Old World fruit bats (Pteropodidae), two frugivorous New World fruit bats (Phyllostomidae) and other ten insectivorous laryngeal echolocating bats. The alignment of *Myo6* coding sequences comprised 3861 nucleotides, equating to 1287 amino acids, of which 169 amino acids (∼13%) were variable in eutherian mammals. Our Bayesian phylogenetic reconstruction based on the coding sequences of *Myo6* revealed a tree in which the major groupings agreed with the accepted mammal species tree. Hence the species of Pteropodidae (*C. sphinx*, *R. leschenaultii* and *E. spelaea*) grouped with species in the family Rhinolophidae (*R. ferrumequinum* and *R*. *pusillus*), Hipposideridae (*H. pratti* and *H*. *armiger*) and Megadermatidae (*M. lyra*) to comprise the clade Yinpterochiroptera \[Bayesian posterior probability (BPP) = 1.00\]. Other bats (*T. plicata*, *M. megalophylla*, *P. parnellii*, *M. ricketti*, *P. abramus*, *A. lituratus* and *L. yerbabuenae*) grouped together and comprised the clade Yangochiroptera (BPP = 1.00). We found no evidence of recombination breakpoints (i.e. recombination) in our dataset and hence excluded any potentially adverse influences of recombination in our phylogenetic reconstructions and subsequent molecular evolutionary analyses. In order to detect selection pressures acting on the *Myo6* gene in echolocating bats, Old World fruit bats and New World fruit bats, we conducted a number of two-ratio model tests to seven branches (branch A∼G). Results of all two-ratio model tests are shown in. Our results of two-ratio models which set the foreground as the ancestral branch leading to Chiroptera, Yinpterochiroptera, Yangochiroptera, Yinpterochiroptera echolocating bats and Old World constant frequency (CF) rhinolophids (branches marked with A, B, C, D and E), respectively, showed no elevated ω values on these ancestral branches compared with the ω values of the corresponding background branches. These results indicated that no selective pressure change of *Myo6* was occurred in those major focal branches leading to echolocating bats. The two-ratio model test for New World fruit bats (the branch marked with G in) exhibited similar results, with the two-ratio model which set the ancestral branch leading to New World fruit bats as the foreground showing no significantly better fit than the null (one-ratio) model \[likelihood ratio test (LRT) statistic (2Δ*ℓ*) = 0.012, df = 1, *P*\>0.05\]. However, the two-ratio model which designed the ancestral branch of Old World fruit bats (the branch marked with F) as foreground was a significantly better fit to the dataset than the one-ratio model (2Δ*ℓ* = 19.611, df = 1, *P*\<0.001). The estimated ω value on the ancestral branch of Old World fruit bats was an order of magnitude greater than that of background (0.135 versus 0.037, respectively), indicating a selection pressure change acting on *Myo6* in the Old World fruit bats. Then we performed the test 2 of branch-site model A to detect the positively selected sites on the above seven ancestral branches leading to different lineages of bats. Results of all tests 2 of the branch-site model A are shown in. No evidence of positive selection was detected on five focal branches leading to echolocating bats (branch A, B, C, D and E) with the exception of the ancestral branch leading to yinpterochiropteran echolocating bats. Statistically supported evidence of positive selection (2Δ*ℓ* = 8.21, df = 1, *P = *0.004) was detected on the ancestral branch leading to Yinpterochiroptera echolocating bats, however, only one positively selected site was found \[260Q, Bayes Empirical Bayes (BEB) values = 0.998\]. However, a strong signature of positive selection was detected by branch-site model A test on the ancestral branch of Old World fruit bats (2Δ*ℓ* = 4.39, df = 1, *P = *0.036). Twelve positively selected sites were detected on the ancestral branch of Old World fruit bats, of which two had BEB values \>0.95 (677V and 1165I). As a comparison, we also performed the branch-site model A test on the ancestral branch of New World fruit bats, considering this taxon had evolved feeding habits similar to the Old World fruit bats. However, no evidence of positive selection was detected on that branch. Alternative methods based on branch-site (REL) and branch (GA- branch) comparisons failed to detect positive selection in the Old World fruit bats and other branches tested above (data not shown). We also calculated the posterior probabilities of positively selected sites for each amino acid of *Myo6* on the above seven branches which tested both by two- ratio model tests and branch-site model A tests. In accordance with the results of branch-site model tests, significant evidence of positive selection was found on ancestral branch leading to Old World fruit bats. In order to examine the distribution pattern and potential influences of the positively selected sites detected in the ancestral branches leading to yinpterochiropteran echolocating bats (one amino acid site) and Old World fruit bats (12 amino acid sites), respectively, we mapped these amino acid sites onto the myosin VI protein secondary structure constructed in previous studies. Of the 13 positively selected sites, nine were distributed on the motor domain (147V, 260Q, 300H, 521S, 534S, 560V, 677V, 678G and 791Y), two on the coiled- coil region (903V and 913G) and two on the globular domain after large and small insertions (1165I and 1169K). We also performed Real-time PCR assays to determine the expression of *Myo6* mRNA in ten tissues of three representative bat species: *C*. *sphinx*, *R*. *ferrumequinum* and *M*. *ricketti*. *Myo6* was ubiquitously expressed in all these ten tissues of three bat species. For *C*. *sphinx*, expression levels of the *Myo6* gene among ten tissues were significantly different (*P*\<0.001, df = 9, Kruskal-wallis test), and the highest expression of the *Myo6* gene occurred in the kidney. For *R*. *ferrumequinum* and *M*. *ricketti*, the *Myo6* expression levels within ten tissues were also significantly different (*P*\<0.001 and *P*\<0.01, respectively, df = 9, Kruskal-wallis test), and the highest expression of the *Myo6* gene were also found in the kidney. However, for *R*. *ferrumequinum*, the expression of the *Myo6* in the kidney was not significantly higher than that of the stomach (*P* = 0.275, Mann-Whitney U test). And for *M*. *ricketti*, the expression of the gene in the kidney was not significantly higher than that of the lung (*P* = 0.127, Mann-Whitney U test). These results indicating that the *Myo6* gene might play an important role in kidney function, especially in the frugivorous Old World fruit bat, *C*. *sphinx*. # Discussion Results from two-ratio model tests showed that no selection pressure change was found in major focal branches leading to echolocating bats and also the New World fruit bats. However, the results of two-ratio model tests revealed a significantly greater value of ω (d_N/d_S) on the ancestral branch of Old World fruit bats compared with the rest of the tree, indicating a selection pressure change acting on *Myo6* in this lineage. The subsequent branch-site model A test revealed that *Myo6* has undergone adaptive evolution in the ancestral branch leading to Old World fruit bats. Our results of real-time PCR showed the highest expression of *Myo6* gene in the kidney in *C*. *sphinx*, indicating that proteins produced by this gene may be involved in kidney function in frugivorous Old World fruit bats. However, we could not show the expression levels of *Myo6* isoforms separately in each tissue, since the primers we used were designed based on the sequences of the conserved tail domain after both large and small insertions. However the primer design is unlikely to influence *Myo6* expression patterns shown with our real- time PCR results, considering the *Myo6* isoforms with 32aa large insertion and 9aa small insertion were specifically expressed in polarized and unpolarized cells, respectively. Our results show discrepancies with a previous study, in which the highest levels of *Myo6* expression was found in the brain, pancreas, prostate, testis and small intestine but not in the kidney of human tissues studied by northern blot methods. The discrepancy might be caused by methodological differences, but is more likely to reflect differences in expression patterns in bats compared with humans. That the *Myo6* gene functions in hearing is supported by overwhelming evidence, as mutations cause non-syndromic deafness in humans. The evolution of echolocation and its associated high-frequency hearing makes laryngeal echolocating bats a fascinating mammal group for molecular evolutionary studies of hearing genes. Many genes associated with hearing have recently been proved to have undergone positive selection in echolocating bats and also in echolocating cetaceans,. However, our results from the branch-site model A tests showed no significant evidence of positive selection in the major focal branches leading to echolocating bats, although one statistically supported positively selected site (260Q) was found in the ancestral branch of Yinpterochiroptera echolocating bats. These results indicate that although the *Myo6* gene is fundamental for the development and maintenance of stereocilia, this gene was not a potential target of positive selection and thus may not contribute to the evolution of high-frequency hearing in echolocating bats. On the contrary, our results of branch-site model A tests revealed strong evidence for positive selection acting on the ancestral branch of Old World fruit bats, a lineage that does not possess laryngeal echolocation. These results suggest that the *Myo6* gene may be involved in other important physiological functions rather than hearing in Old World fruit bats. Some studies have reported that the *Myo6* gene is also abundantly expressed in the retina and is required for the normal functioning of photoreceptor cells. In accordance with these observations, our real-time PCR assays also detected the expression of the *Myo6* gene in the eyes of *C*. *sphinx*, *R*. *ferrumequinum* and *M*. *ricketti*, although the expression levels were all relatively low. Since they lack the ability of laryngeal echolocation, Old World fruit bats are more dependent on vision for perceiving their environment than are their echolocating relatives. Thus we could not role out the possibility that the positive selection of the *Myo6* gene may have been driven by the requirement for more effective vision in Old World fruit bats. In photoreceptors, the *Myo6* gene is highly expressed in the margins of the inner segments and the outer surface of the ellipsoid mitochondrial mass. *Myo6* gene might be involved in the localization of mitochondria which may be important for dark current generation by the sufficient supply of ATP to Na⁺, K⁺-ATPases on the inner segment plasma membrane. In retinal pigment epithelial (RPE) cells, *Myo6* might play an important role in transportation of lysosomes which are responsible for the normal daily digestion of photoreceptor disc membrane. Considering those potential important roles of the *Myo6* gene in retinal, we could not role out the possibility that the positive selection of this gene in Old World fruit bats might relate to the evolution of their effective visual system to enhance their visual sensitivity at dim-light environment. Because the evidence for the exact role that *Myo6* plays in vision is still unclear, more studies on its function are needed to determine the possible role of this gene in vision in Old World fruit bats. A large number of studies have recently identified an important role for myosin VI in clathrin-mediated endocytosis in polarized epithelial cells such as kidney proximal tubule cells and intestinal enterocytes (see Introduction). Thus another possible explanation for positive selection acting on *Myo6* in Old World fruit bats is the adaptation of receptor-mediated endocytosis in kidney proximal tubule cells. *Myo6* is highly expressed in the kidney where it located in the brush border of renal proximal tubule cells, which is also supported by our real-time PCR results. Moreover, the study of myosin VI functional null *Snell’s waltzer* (*sv*/*sv*) mice revealed that myosin VI plays an important role in kidney proximal tubule protein reabsorption, as *sv*/*sv* mice showed an albuminuria phenotype. The significantly high expression of the *Myo6* gene in the kidney of *C*. *sphinx* strongly indicated the important role of this gene in kidney function in Old World fruit bats. Although some studies observed that Old World fruit bats accidentally or even deliberately consume insects, a predominantly frugivorous food habit has recently been confirmed by studies with stable-isotope analyses (see Introduction). Thus, the low protein content in the special diets of Old World fruit bats, highlights the significance of protein preservation by this mechanism in Old World fruit bats compared with insectivorous, omnivorous and carnivorous bats. Although renal proximal tubule reabsorption is highly efficient, small quantities of protein would still naturally be lost in urine in healthy humans and other animals. Thus, it is plausible that Old World fruit bats have evolved a more efficient receptor- mediated endocytosis for protein preservation to adapt to their low protein diets. Alternatively, the positive selection of *Myo6* in Old World fruit bats might relate to the preservation of essential nutrients such as vitamin D \[25-(OH) vitamin D₃\] and vitamin B₁₂ rather than proteins. 25-(OH) vitamin D₃ is filtered by glomeruli as a complex with vitamin D-binding protein (DBP) and then reabsorbed into proximal tubule cells by receptor-mediated endocytosis. Then 25-(OH) vitamin D₃ is transported to mitochondria and hydroxylated into 1,25-(OH)₂ vitamin D₃, which is the active form of vitamin D for the regulation of metabolism. Thus, receptor-mediated endocytosis is of great importance for the preservation and activation of vitamin D, since defects in this process would cause vitamin D deficiency and disease. Similarly, vitamin B₁₂ is filtered and then reabsorbed as a complex with transcobalamin (TC). The reabsorption and accumulation of vitamin B₁₂ by receptor-mediated endocytosis is of great importance for vitamin B₁₂ homeostasis. Since fruit is known to be devoid of both these fat-soluble vitamin components, Old World fruit bats (especially frugivorous bats) are thought to be naturally in a state of vitamin D and vitamin B₁₂ deficiency. Thus, it is reasonable to assume that the adaptive change of *Myo6* in Old World fruit bats might relate to the evolution of their efficient receptor-mediated endocytosis for the preservation and homeostasis of essential fat-soluble vitamins or other similar components which are deficient in plant food resources. The 12 positively selected sites detected by branch-site model A tests in the Old World fruit bats are mainly distributed in the motor domain region (eight sites), the coiled-coil domain (two sites) and the globular tail domain (two sites). The motor domain is known to be important for the minus-end directed movement, and mutations in this domain could cause motor activity impairment and subsequent transportation blocking. Notably, for these eight sites distributed in the motor domain region, we found that the residue 147V was located close to the nucleotide-binding site and 677V and 678G were within the predicted actin- binding site. Residue 300H was distributed in the putative motor domain unique insert 1 region, which is thought to be involved in the modulation of nucleotide binding and release. Residue 791Y was distributed in the reverse gear region which is involved in the lever-arm redirection and thus responsible for the reverse movement of myosin VI. The coiled-coli region is thought to mediate the dimerization of myosin VI, and the globular tail domain is necessary for cargo- targeting. Thus, the positively selected sites might influence the activities of these regions and therefore the efficiency of the myosin VI protein. Among these 12 positively selected sites, nine amino acid changes at positions 147, 534, 560, 677, 678, 791, 903, 913 and 1169 were found to be specific to the Old World fruit bats compared with other bats and mammal groups. The amino acid change from hydrophilic Threonine (T) to hydrophobic Valine (V) at position 677 might enhance the capability of binding/recognition of hydrophobic ligands. The amino acid change from Cysteine (C) to Tyrosine (Y) at position 791 involved the introduction of an extra aromatic side chain, and might enhance the ability of stacking interactions with other aromatic side chain. The amino acid change from Glutamine (Q) to Lysine (K) at position 1169 introduced a positively-charged amino group, and might be involved in the binding of negatively-charged ligands. In the absence of direct biochemical evidence, we are unable to assign more significance to these amino acid changes in the Old World fruit bats. More studies are needed to determine what functional changes in the myosin VI protein may result from mutations at these sites. In addition, the New World fruit bats which belong to the family Phyllostomidae, especially the members of the subfamily Stenodermatinae, have also evolved a predominantly frugivorous feeding habit similar to that of Old World fruit bats. However, no positively selected site was detected on the ancestral branch of this lineage, and also in the species of *A. lituratus* which belongs to the subfamily Stenodermatinae (data not shown). These results might reflect the fact that the Old World fruit bats evolved such special diets slightly earlier than did the New World fruit bats (nearly 28mya versus almost 20mya). Moreover, another plausible explanation for these results might be that fruit available to the Old World fruit bats and the New World fruit bats was different during the time of their radiations. Neotropical regions possess a greater diversity of food plants and more stable plant food resources than do Palaeotropical regions. Thus it is possible for species of New World fruit bats to selectively ingest mixed fruit species to obtain sufficient proteins without the evolution of an adaptive mechanism for the preservation of protein and essential nutrients as found in Old World fruit bats. It is interesting to note that, other molecular evolutionary studies focusing on genes in bats (and other mammals) in relation to their diets also show a similar discrepancy between the Old World and the New World fruit bats. For instance, the glucose transporter 4 (GLUT4, encoded by *Slc2a4*) which plays a crucial role in glucose homeostasis has undergone adaptive changes only in the Old World fruit bats in relation to their high sugar diet but not in the New World fruit bats. In conclusion, our results show that the *Myo6* gene, which was widely considered as a hearing gene, has undergone adaptive evolution in the Old World fruit bats without laryngeal echolocation and associated high-frequency hearing. Positive selection on *Myo6* in Old World fruit bats may be related to their specialized diets. In combination with the high levels of expression of *Myo6* in kidney tissue, our results provide evidence that *Myo6* has undergone adaptive evolution in Old World fruit bats in relation to receptor-mediated endocytosis for the preservation of protein and essential nutrients. As receptor mediated-endocytosis is a multistep process and involves many molecules other than myosin VI, more studies are needed to delineate what genes in addition to *Myo6* may also contribute to the specialized dietary adaptations of Old World fruit bats. # Supporting Information We thank Yang Liu for helpful comments on molecular evolutionary analyses, Yao Chen for help in the lab and Guimei He for help in real-time PCR analyses. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BS XH SZ. Performed the experiments: BS XH. Analyzed the data: BS XH. Contributed reagents/materials/analysis tools: SZ. Wrote the paper: BS GJ SJR SZ.

# Introduction Upon encounter with cognate antigen presented by antigen presenting cells, naïve CD4 T cells proliferate and differentiate into effector and memory subsets. These activated (effector or memory) CD4 T cells then recirculate through various tissues to aid infection clearance and protect against pathogen re- exposure. Emerging literature suggests that T cell differentiation occurs in several steps. Initially, naïve and central memory CD4 T cells are primed in secondary lymphoid organs. These primed but not fully differentiated cells are released from the secondary lymphoid organs and migrate to the sites of inflammation. Full CD4 T cell differentiation then occurs at sites of infection and inflammation after they have migrated from the secondary lymphoid organs. Furthermore, it is becoming increasingly clear that the activated CD4 T cells are not all terminally differentiated and the majority can actually remain flexible and be re-polarized in different inflammatory environments. Therefore, it is clear that the myriad of inflammatory signals that effector and memory CD4 T cells receive from the environment play key roles in influencing their subsequent responses. One of the signals controlling CD4 T cell function is the inflammatory cytokine IL-12. This cytokine is produced in response to pathogens, but is also present at inflammatory sites in human disorders. The current paradigm is that IL-12 primarily promotes the differentiation of naïve CD4 T cells into Th1 cells. Also, the presence of IL-12 during priming of naïve CD8 T cells has been shown to promote strong effector functions and memory development. Interestingly, the majority of these studies have focused on the effects of IL-12 if it is present during priming of naïve T cells (IL-12 present before or following TCR stimulation). However, after the initial priming, activated CD4 T cells that are not fully differentiated will also be exposed to IL-12 as they migrate through the circulatory or lymphatic system before entering the sites of infection, where they will be further activated via TCR induction by antigen presenting cells. The effects of IL-12 signals in altering the response of activated, non- fully differentiated T cells to subsequent TCR stimulation remains poorly understood. Recent murine studies, suggest that IL-12 alters the responses of activated T cells to subsequent TCR stimulation, suggesting that the modulation of CD4 T cell responses by IL-12 is more complex than simply inducing the differentiation of Th1 cells. In this regard, Richer et al. and Kim et al. demonstrated that murine effector/memory CD8 or secondary effector CD4 T cells exposed to *in vivo* inflammatory signals, driven primarily by IL-12 and/or type I interferons, have an altered response to *ex vivo* re-challenge with antigen. In these studies, exposure to IL-12 decreased the dose of antigen required to stimulate the maximal T cell response (also known as functional avidity). Likewise, murine memory CD8 T cells conditioned with IL-12 and IL-18 *in vitro* have enhanced cytokine production and cytotoxic activity upon TCR re- stimulation. In addition, previous studies from us and others has demonstrated that prior exposure to IL-7, IL-15 or a TLR5 ligand increases the responsiveness of human T cells to TCR stimulation. Collectively these studies suggest that prior exposure to different cytokines or inflammatory signals alters how T cells respond to TCR stimulation. Although these studies provide insight into murine T cell biology, whether IL-12 similarly regulates the function of human T cells and the precise molecular mechanism by which IL-12 alters subsequent TCR-mediated responses has not been fully elucidated. To explore these questions we used a system consisting of human peripheral blood CD4 T cells that have been activated under non polarizing conditions, which models primed, but not fully differentiated, human CD4 T cells that are released from the secondary lymphoid organs into circulation. We found that prior exposure to IL-12 elevated the response of human activated CD4 T cells to stimulation via the TCR. The IL-12 mediated increases in responses to TCR stimulation seemed to be mediated by two distinct mechanisms: increased activation of select TCR signaling molecules and increased metabolic respiration. This data suggest that the regulation of CD4 T cell function by IL-12 is more complex than simply driving Th1 differentiation. Instead it seems that IL-12 is continually shaping human CD4 T cell responses in a context- specific manner. Based on our results we propose a model in which IL-12 present in blood, infection sites, and/or at inflammatory sites primes human effector or memory CD4 T cells that are not terminally differentiated, allowing them to respond faster when they encounter their cognate antigen at sites of infection and be more easily polarized depending upon the cytokine milieu they encounter. # Materials and Methods ## Human samples Peripheral blood mononuclear cells (PBMCs) were obtained from anonymous donors as previously described. Blood donors at the DeGowin Blood Center at the University of Iowa Hospitals and Clinics between 18 and 55 years provided written informed consent for cells not used for transfusion to be used for research. The consent process was approved by the University of Iowa’s Institutional Review Board. The signed written consent forms are maintained by the DeGowin Blood Center. The completely deidentified samples were then provided to investigators at the University of Iowa. Because all cells were obtained from discarded products, the donors approved for the research use of their cells, and the donors were de-identified, we did not required further Institutional Review Board approval to use these blood samples. All human subject studies were in compliance with the Declaration of Helsinki. ## Isolation and cytokine pretreatment of human activated CD4 T cells CD4 T cells were negatively selected from PBMCs using the human CD4 T cell enrichment kit (Stem cell Technologies) to provide \>98% CD4 T cells (data not shown). The cells were activated for 5 days with bead-bound anti-TCR/CD28 antibodies in the presence of IL-2. This method results in cells that are 100% positive for CD4 and 90–96% positive for CD45RO. The cells were rested without stimulation for 24 h in complete RPMI (RPMI 1640 with 10% FBS, penicillin/streptomycin, and l-glutamine). The activated CD4 T cells were then treated with or without recombinant cytokines (R&D Systems) for different times and doses in complete RPMI. After pretreatment, cells were washed with RPMI 1640 three times to remove the cytokines. ## Cytokine production measured by ELISA After cytokine pretreatments, cells were resuspended in complete RPMI and stimulated with plate-bound anti-TCR (anti-CD3 OKT3 clone)for 24 h. For the rest experiments, pretreated cells were rested for various times before anti-TCR stimulation (anti-CD3 OKT3 clone). For the oligomycin inhibitor experiment, pretreated cells were TCR-stimulated (anti-CD3 OKT3 clone) in the presence of oligomycin (2.5 μM). For the inhibition of glycolysis experiment, pretreated cells were TCR-stimulated (anti-CD3 OKT3 clone) in the presence of 2-DG (5 mM). Cytokine release was measured in triplicate using standard TMB based ELISA. To account for the variations between human donors and the conditions of independent experiments, data were normalized for each donor as: Fold-increase over no cytokine = (concentration of treated sample ÷ maximum concentration of no cytokine sample). For functional avidity, data were normalized as: Percent maximum response = (concentration of sample ÷ maximum concentration of respective treatment) x 100%. Normalized data was fitted to a sigmoidal curve to calculate EC₅₀. However, absolute values obtained from the ELISAs are shown in, and Figs as a reference. ## Intracellular cytokine staining and flow cytometry For intracellular staining, pretreated cells were resuspended in complete RPMI and stimulated with 6 μg/mL of plate-bound anti-TCR (anti-CD3 OKT3 clone) for 8, 18, and or 24 h, in the presence or absence of brefeldin A or Monensin (BioLegend) added for the last 6 h. Cells were washed in FACS buffer (PBS, 10% FBS, and 0.05% sodium azide), and then stained for cytokines per manufacturer’s suggestion (Biolegend). All antibodies were purchased from Biolegend. For surface molecule staining, pretreated cells were washed in FACS buffer and stained on ice with the primary and/or secondary antibodies, followed by FACS analysis. Live lymphocytes were gated based on forward and side scatter. Quadrants were set so the baseline cytokine production of non-TCR stimulated cells was less than 1%. For survival assays, pretreated cells were washed in Annexin V binding buffer and stained with annexin V and propidium iodine (PI) per manufacturer’s suggestion (BD Biosciences). Annexin V and PI double negative cells were considered as viable cells, annexin V positive but PI negative cells as apoptosis undergoing cells, and annexin V and PI double positive cells as dead cells. ## Quantitative Real-time PCR After cytokine pretreatments, cells were stimulated with 6 μg/mL of plate-bound anti-TCR (anti-CD3 OKT3 clone) for 8 and 18 h. cDNA was synthesized from total RNA and real-time RT-PCR was performed on an Applied Biosystems Model 7000 using SYBR Green. The expression of mRNA was normalized to that of mRNA encoding β-actin and quantification of fold induction of treated *vs* untreated was analyzed by the 2^{-Δ ΔCT} method. ## Immunoblotting Activated CD4 T cells were treated with IL-12 (50 ng/mL) for different times and immunoblotting was performed as described previously. Alternatively IL-12 pretreated cells were stimulated with 2 μg/mL of crosslinked anti-TCR (anti-CD3 clone OKT3) and anti-CD4 (clone RPA-T4) and immunoblotting was performed. We use anti-TCR in combination with anti-CD4 because this provides the best enhancement of proximal and distal TCR signaling. The intensity of the immunoblotting bands was determined using the Licor Odyssey v3.0 software. Normalization of the phospho-protein intensity to the actin intensity was conducted as described previously. ## Antibodies The following antibodies were used for immunoblotting, cell-surface, and intracellular stains: The anti-LAT Y191 and anti-LAT from Millipore. The anti- LCK pY505, and anti-SLP-76 pY128, anti-IL-12Rβ1, and annexin V from BD Biosciences. The anti-PLC-γ1 pY783, anti-p38 pT180/Y182, anti-AKT pT308, anti- ZAP-70 pY319, anti-SRC pY416, anti-PLC-γ1, anti-LCK, anti-SLP-76, anti-AKT, anti-STAT4, anti-STAT4 pY693, anti-CD25, and anti-ERK 1/2 antibodies were purchased from Cell Signaling Technologies. The anti ERK1/2 pTpY185/187 and anti-AKT pS473 were from Invitrogen. The anti-GAPDH was from Meridian Life Science. The DyLight 800 and DyLight 680 labeled secondary antibodies were obtained from Thermo Scientific. The FITC anti-IFN-γ (4S.B3), Alexa Fluor 647 anti-IL-10 (JES3-9D7), Alexa Fluor 647 anti-IL-4 (8D4-8), Alexa Fluor 647 anti- TNF-α (Mab11), and APC anti-IL-13 (JES10-5A2) all from BioLegend. The anti-CD3 (OKT3), anti-CD4 (RPA-T4), FITC anti-CD28 (CD28.2), anti-CD-2 (RPA-2.10), anti- CD49d (9F10), anti-CD11a (HI111), FITC anti-CD278, anti-CD150, and DyLight 488 IgG (Ply4053) antibodies and propidium iodine were obtained from Biolegend. Anti-IL-12 Rβ2 was purchased from R&D Systems. ## Seahorse and ECAR/OCR measurements After cytokine pretreatments, cells were resuspended in XF media (DMEM with glucose, pyruvate and glutamine) and were plated onto poly-L-lysine-coated XF-96 plates. Cells were then incubated for 30 min in a non-CO₂ incubator and their metabolic profiles examined using an XF-96 Extracellular Flux Analyzer (Seahorse Bioscience). The optimal concentrations of mitochondrial inhibitors were Oligomycin 2.5 μM, FCCP 1.5 μM, Rotenone 5 μM and Antimycin A 5 μM (data not shown). OCR and ECAR values were normalized to cell numbers as previously described. Metabolic parameters were then assessed as described before using: (1) Basal respiration = (Basal OCR before adding Oligomycin)–(OCR following Rotenone/Antimycin A); (2) Maximal respiratory capacity = (OCR peak rate following FCCP)–(OCR following Rotenone/Antimycin A); (3) Basal ECAR = Basal ECAR rate before adding Oligomycin; and (4) Maximal ECAR = ECAR rate following addition of Rotenone/Antimycin A. ## Glucose uptake assay After cytokine pretreatment, cells were washed in glucose free-medium and then cells were resuspended in glucose free medium containing 2-NBDG for 40 minutes at 37°C. Cells were then washed and analyzed by flow cytometry. ## Statistical analysis Statistical analysis between the groups was assessed using GraphPad Prism. Specific tests for statistical significance are indicated in the figure legends. Differences were considered significant when p values were below 0.05. # Results ## Prior exposure to IL-12 alters human activated CD4 T cell responses to subsequent TCR stimulation To address if prior exposure to inflammatory cytokines altered how human activated CD4 T cells respond to TCR stimulation, primary human CD4 T cells were isolated from PBMCs and then activated for 5 days under non-polarizing conditions with anti-TCR/CD28 antibodies and recombinant IL-2. Following stimulation, cells were rested without stimulatory signals for 24 h before further analysis of downstream functions. Our system is intended to mimic the *in vivo* situation of a primed, but not fully differentiated, human CD4 T cell entering a site of infection, where it will encounter inflammatory signals and antigen. Activated CD4 T cells were then incubated with recombinant cytokines, extensively washed to remove the cytokines and stimulated with plate-bound anti- TCR antibodies (anti-CD3 clone OKT3). No additional costimulatory signals were provided to focus exclusively on the effects of the inflammatory cytokine on TCR-induced T cell activation. Pretreatment with IFN-γ, IFN-β, TNF-α, or IL-4 had little to no effect on TCR- induced IFN-γ production in comparison to cells treated in media alone. In contrast, pretreatment with IL-12 significantly enhanced the TCR-induced production of IFN-γ compared to untreated cells, but did not cause measurable IFN-γ production in the absence of TCR stimulation. No synergistic or antagonist effects on IFN-γ production were seen when IL-12 was used in combination with IFN-γ, IFN-β, TNF-α, or IL-4 (data not shown). It was possible that the increased release of IFN-γ in the IL-12 pretreated group was due to differences in cell numbers after TCR stimulation. However, 24 hours after TCR stimulation cells stimulated with or without IL-12 had similar cell numbers, suggesting that IL-12 did not alter proliferation in the timeframe of these experiments. In addition, we examined if IL-12 pretreatment altered the susceptibility to cell death before or after TCR stimulation as assessed by annexin V and PI staining. We found that before and after TCR stimulation both IL-12 pretreated and untreated cells have similar numbers of apoptotic cells. Finally, we explored the status of CD25 on the IL-12 pretreated and untreated cells by flow cytometry. Both IL-12 pretreated and untreated have similar expression of CD25. Together, these data show that prior exposure to IL-12 selectively enhances human activated CD4 T cells to produce more IFN-γ after TCR stimulation. To more fully characterize these effects, the optimal dose of IL-12 that increases TCR-induced IFN-γ production was determined. To test for this, human activated CD4 T cells were pretreated for 6 h with various doses of IL-12 and then stimulated with plate-bound anti-TCR antibodies (anti-CD3 clone OKT3). Pretreatment of cells with doses of 25–100 ng/mL of IL-12 significantly augmented IFN-γ production in comparison to cells with no IL-12 pretreatment. Interestingly, the doses of IL-12 that altered the responses of human activated CD4 T cells to TCR stimulation (25–100 ng/mL) are similar to the doses used in many other in vitro studies but they are above the concentrations previously described in serum of humans. Next, the duration of IL-12 pretreatment required to increase IFN-γ production was examined. As shown in and, exposure to IL-12 for at least 1.5 hours was required to significantly increase the production of IFN-γ in comparison to cells with no IL-12 pretreatment. This finding suggests that the effects of IL-12 are not due to the presence of residual cytokine during TCR stimulation, since alter functions from residual IL-12 would be observed at all pretreatment times. Finally, we examined the duration of the effects after the removal of IL-12. To accomplish this, activated CD4 T cells were rested for various times after treatment with IL-12 before then being stimulated through the TCR (anti-CD3 clone OKT3). As shown in and, the ability of IL-12 to enhance IFN-γ production following TCR stimulation was short-lived and lasted for 3–6 hours after cessation of IL-12 treatment. In previous studies examining murine effector/memory CD8 T cells and secondary effector CD4 T cells, inflammatory cytokines decreased the dose of peptide antigen required to stimulate the maximal T cell response, known as functional avidity. Therefore, we examined if IL-12 pretreatment will have similar effects in human activated CD4 T cells using an antibody based stimulation. To this end, IL-12 pretreated cells were stimulated with titrating doses of plate bound anti-TCR antibodies (anti-CD3 clone OKT3) and then the functional avidity was determined by calculating the dose of stimulatory antibody needed to induce 50% of maximal IFN-γ production (EC₅₀). We found that IL-12 exposure does not alter the functional avidity of human activated CD4 T cells in response to TCR ligation alone. Collectively, these data demonstrate that IL-12 pretreatment transiently increases TCR-induced IFN-γ production after short exposure with IL-12. Furthermore, in an antibody based stimulation system, IL-12 exposure did not alter the functional avidity of human activated CD4 T cells in response to TCR ligation. ## Pretreatment of human activated CD4 T cell with IL-12 primes the TCR-induced release of a range of cytokines CD4 T cells have the capacity to produce multiple cytokines, including IFN-γ, TNF-α, IL-4, IL-13, and IL-10. Numerous studies have established that IL-12 induces naive CD4 T cell differentiation into IFN-γ and TNF-α-producing Th1 subsets. Therefore, we examined whether prior exposure to IL-12 was priming CD4 T cells to selectively produce Th1 cytokines or generally upregulate cytokine production. We observed that TCR engagement led to the production of IFN-γ, TNF-α, IL-4, IL-13, and IL-10 in cells pretreated with media alone. Prior exposure to IL-12 not only increased the TCR-induced production of IFN-γ, but also enhanced the release of TNF-α, IL-4, IL-13 and IL-10 cytokines. This IL-12-mediated potentiation of cytokine production was statistically significant for all cytokines at TCR doses of 6 μg/mL and higher. These data indicate that IL-12 pretreatment in activated human CD4 T cells does not solely increase Th1 cytokines, but instead enhances the production of a range of cytokines following TCR stimulation. To further examine how IL-12 increases the global cytokine release, we examined the cells capable of responding to IL-12 from our heterogenous population of human activated CD4 T cells by determining the expression of IL-12 receptor before IL-12 pretreatment. IL-12 receptor is composed of two subunits IL-12R β1 and IL-12R β2. Co-expression of both subunits is required for the canonical cellular effects of IL-12. We found that IL-12R β1was highly upregulated and IL-12R β2 was minimally increased at the cell surface of activated CD4 T cells. Then, the effect of IL-12 on TCR-mediated (anti-CD3 clone OKT3) cytokine expression was measured using intracellular staining. The TCR concentration dose used in these studies was 6 μg/mL since this was the lowest TCR dose that gave significant differences in the production of all cytokines. Exposure to IL-12 before TCR stimulation for 18h resulted in a significant increase in the frequency of cells producing IFN-γ and a slight, but statistically insignificant, increase in the frequency of cells producing TNF-α and IL-10. Surprisingly, the proportion of cells producing IL-4 and IL-13 remained the same between both IL-12 pretreated and control cells. Furthermore, IL-12 pretreatment resulted in no differences in the MFI of IFN-γ, TNF-α, IL-4, IL-13 and IL-10 compared to untreated cells. Interestingly, only a small fraction of the total cells used in our system (human activated CD4 T cells) responded potentially due to their expression of IL-12 receptor. The fact that IL-12 pretreatment increased the TCR-induced release of TNF-α, IL-4, IL-13 and IL-10 into the culture supernatants but had little to no effect when intracellular levels of these cytokines were measured using an inhibitor of protein transport from the endoplasmic reticulum to the Golgi apparatus (BFA), suggests that IL-12 could mediate its effects by altering the release of cytokines. However because intracellular staining only provides a snapshot of cytokine production during a short time window, we confirmed our results by measuring intracellular cytokines at different time points and in the presence or absence of different secretion inhibitors (BFA and Monensin). We found a similar pattern of cytokine secretion when we repeated the experiments using BFA, Monensin, or no secretion inhibitor. Moreover, this cytokine secretion profile was observed at earlier and later time points of TCR induction (6h and 24h) (data not shown). Therefore, IL-12 increases the number of cells capable of producing IFN-γ, while altering the release of other cytokines via a separate mechanism. ## The IL-12 mediated enhancement of cytokine production following TCR stimulation is mediated by a function of both transcriptional and post-transcriptional effects We next examined whether the IL-12 potentiation of cytokine production was a consequence of increased transcription of cytokines. To explore this possibility, total RNA was isolated from untreated and IL-12 treated human activated CD4 T cells before and after TCR ligation (anti-CD3 clone OKT3). Then the levels of mRNA expression of IFN-γ, TNF-α, IL-4, IL-13 and IL-10 were assessed by quantitative real-time PCR. In the absence of TCR stimulation there were no differences in the mRNA expression of the cytokines between IL-12 treated and untreated cells, showing that IL-12 does not directly stimulate the production of cytokine mRNA. As expected, TCR induction upregulated the mRNA expression of IFN-γ, TNF-α, IL-4, IL-13 and IL-10 when compared to non-TCR stimulated cells. TCR stimulation increased the mRNA expression of IFN-γ and lightly enhanced the mRNA expression of IL-10 in IL-12 pretreated cells in comparison to cells pretreated in media alone. In contrast, the TCR-induced increase in TNF-α, IL-13 or IL-4 mRNA expression was not altered in IL-12-treated cells in comparison to untreated cells. The same pattern was observed after shorter times of TCR induction (6–8 h), indicating that IL-12 pretreatment does not alter the kinetics of TCR-induced cytokine mRNA expression (data not shown). Together with the intracellular staining studies, these results suggest that the IL-12-mediated priming of cytokine release is driven by two separate mechanisms, increased mRNA expression for IFN-γ and increased release of TNF-α, IL-13, IL-4 and IL-10. ## IL-12 pretreatment enhances the activation of select signaling molecules downstream of the TCR To further identify mechanisms by which IL-12 pretreatment potentiates the TCR- mediated production of cytokines, we examined the effects of IL-12 on the expression of surface molecules involved in T cell activation. We found no significant changes in the expression of the TCR, CD4, the costimulatory receptors CD28, ICOS or SLAM or the adhesion receptors CD2, CD49d, and CD11a after pretreatment with IL-12. This indicates that overt changes in surface expression of key receptors were not responsible for the effects of IL-12 on cytokine production. Following IL-12 stimulation, the Janus kinase-STAT signaling pathway is activated and leads to STAT4 phosphorylation. STAT4 is considered to be one of the critical mediators of the canonical IL-12 effects because STAT4-knockout mice have impaired Th1 differentiation and IFN-γ production. Therefore, we examined if the IL-12-mediated increase of cytokine production following TCR stimulation was mediated directly by STAT4. To address this possibility, the total expression and the activation of STAT4 were determined before and after IL-12 pretreatment using quantitative immunoblotting. Treatment with IL-12 significantly increased the phosphorylation of STAT4 (Y693) peaking at 15–45 min following treatment. STAT4 phosphorylation returned back to near basal levels by the end of the six hour IL-12 pretreatment, which is the time when we stimulated the cells through the TCR in most of our experiments (Figs). The total expression levels of STAT4 decreased significantly following IL-12 stimulation and became almost undetectable by the end of the IL-12 pretreatment, suggesting that STAT4 was getting degraded upon exposure to IL-12 signals (Figs). Since STAT4 activation is similar before and after 6 hours of IL-12 pretreatment and STAT4 protein expression is almost undetectable after 6 hours of IL-12 pretreatment, we conclude that the IL-12 mediated increase in cytokine production upon TCR stimulation is likely not mediated directly by STAT4 alone synergizing with TCR stimulation signals. Ligation of the TCR results in the activation of signaling molecules that ultimately promote transcriptional changes at cytokine genes. Therefore, we examined if IL-12 pretreatment altered the activation of signaling molecules downstream of the TCR. In support of this possibility, previous studies have shown that TCR-mediated signaling was enhanced in murine CD4 and CD8 T cells exposed to pathogen-induced inflammation and in human T cells pretreated with IL-7, IL-15 or the TLR5 ligand flagellin. To test this possibility, human activated CD4 T cells were left untreated or exposed to IL-12 and then stimulated using crosslinked anti-TCR/CD4 (anti-CD3 clone OKT3 and anti-CD4 clone RPA-T4) antibodies for various times. We use anti-TCR in combination with anti-CD4 for signaling studies because this provides the best enhancement of proximal and distal TCR signaling. Then changes in the total expression and phosphorylation of TCR-induced signaling molecules were measured using quantitative immunoblotting. The activity of the tyrosine kinases LCK and ZAP-70 is induced early after TCR stimulation. LCK activation is dictated by the phosphorylation of Y394 and Y505, where increases in LCK Y394 and decreases in Y505 phosphorylation are correlated with enhanced LCK activity. To detect changes in LCK Y394 phosphorylation, we used an anti-SRC pY416 antibody, which recognizes all SRC kinases, including LCK and FYN, when they are phosphorylated on their activating sites. LCK is the predominant kinase associated with CD4/plasma membrane and this kinase is the main phosphorylated SRC kinase detected in CD4 T cells. Interestingly, IL-12 pretreatment resulted in a trend towards increased TCR-induced phosphorylation of LCK and FYN in comparison to cells treated in media alone; however this did not reach statistical significance. In addition, IL-12 pretreatment significantly reduced the phosphorylation of LCK at its inhibitory site Y505. In contrast, the phosphorylation of the activating tyrosine on ZAP-70 (pY319) was similar between IL-12 treated and untreated groups; the disconnection between LCK activity and ZAP-70 phosphorylation has been previously observed. ZAP-70 phosphorylates the adaptor protein LAT, resulting in the recruitment and phosphorylation of the phospholipase PLC-γ1 and adaptor protein SLP-76. Consistent with our previous data on ZAP-70, we observed no differences in the phosphorylation of LAT, SLP-76, or PLC-γ between control and IL-12 treated cells. The recruitment and activation of proteins at LAT promotes the induction of the mitogen activated protein kinase (MAPK) and AKT pathways, resulting in the transcription of target genes that regulate cytokine production. We observed that IL-12 pretreated cells had a trend towards increased TCR-induced phosphorylation of AKT (S473) in comparison to control cells that did not reach statistical significance after compiling multiple donors. Furthermore, we observed that the TCR-induced phosphorylation of p38 (pT180/T182) was significantly increased in IL-12 pretreated cells at 2 minutes in comparison to cells treated in media alone. In contrast, there were no changes in the extent or kinetics of TCR-induced ERK1,2 (pT187/pY187) phosphorylation. As shown in, the changes in activation of all the signaling molecules were not due to altered total expression of these proteins. To confirm its use as a loading control, we also examined the expression of GAPDH in both IL-12 pretreated and untreated human activated CD8 T cells in all the samples from the donors used in our studies. After quantifying the data of all the donors used in our studies, we found that both IL-12 pretreated and untreated cells have similar expression of GAPDH. Overall, these findings suggest that prior exposure to IL-12 alters specific TCR-induced signaling pathways in human activated CD4 T cells. ## IL-12 mediated enhancement of cytokine secretion following TCR stimulation is partially regulated by an increase in oxidative metabolism Our previous intracellular staining, mRNA expression, and signaling studies suggest that IL-12 pretreatment increases the release of cytokines *via* increased gene transcription and by an unknown post-transcriptional mechanism. Changes in the cellular function of immune cells, like cytokine secretion, can be post-transcriptionally regulated by alterations in metabolic pathways. This suggests that IL-12 exposure may alter the metabolic state of human activated CD4 T cells to drive the expression of cytokines regulated by a post- translational mechanism. To address this possibility, the metabolic profiles of IL-12 pretreated cells or untreated cells were determined using a metabolic flux analyzer. Mitochondrial respiration and aerobic glycolysis were assessed by measuring the oxygen consumption rate (OCR) and extracellular acidification rate (ECAR), respectively, under basal conditions and after drug-induced mitochondrial stress. Under basal conditions, IL-12 pretreated cells exhibited an increase in basal oxidative respiration compared to untreated cells as measured by the OCR. The capacity of treated cells for oxidative metabolism was examined by measuring OCR after addition of oligomycin (an ATP synthase inhibitor that eliminates OCR due to ATP production), FCCP (a proton ionophore that uncouples ATP synthesis from the electron transport chain (ETC), which reveals the maximum capacity of the mitochondria to use oxidative metabolism), and rotenone plus antimycin A (an inhibitor of complexes I and III that completely shuts down the ETC). IL-12 pretreated cells exhibited a significant increase in the maximal respiratory capacity in comparison to cells treated in media alone. This indicates that IL-12 treatment enhances mitochondrial respiration in human activated CD4 T cells. In contrast, basal and maximal ECAR, were similar in both IL-12 pretreated and untreated cells. Finally, glucose consumption was determined in IL-12 pretreated and untreated cells using a fluorescent glucose analog (2-NBDG) that is used as an indicator for glucose uptake. Consistent with our previous data, we found that both IL-12 pretreated and untreated cells have similar 2-NBDG uptake. Together, these findings suggest that IL-12 pretreated cells have enhanced oxidative metabolism, resulting in increased oxygen consumption rates and increased ability to up-regulate mitochondrial respiration in response to stimulation. On the basis of these findings, we tested if the IL-12-mediated increase of oxidative metabolism was involved in regulating the priming of cytokines that were not transcriptionally regulated. To this end, IL-12 pretreated cells were TCR-stimulated (anti-CD3 clone OKT3) in the presence of an inhibitor of oxidative metabolism (oligomycin) and cytokine secretion was then measured. Consistent with published work, oligomycin was shown to effectively block mitochondrial respiration, as shown by the decrease of OCR in both IL-12 treated and untreated groups. As expected, pretreatment with IL-12 resulted in the increase of IFN-γ, TNF-α, IL-4, IL-13, and IL-10 upon TCR stimulation in comparison to cells treated in media alone. The IL-12 mediated enhancement of IFN-γ, TNF-α, and IL-10 was unaffected by the presence of oligomycin during TCR stimulation. However, unlike IL-12 pretreatment alone, pretreating with IL-12 in the presence of oligomycin did not augment TCR-induced IL-4 and IL-13 production. Notably, the dose of oligomycin used had minimal effects on the production of cytokines by the cells treated in media alone. In addition, we also tested the role of glucose metabolism in the TCR-induced production of cytokines in both untreated and IL-12 pretreated cells. To this end, IL-12 pretreated cells were TCR-stimulated in the presence of an inhibitor of glycolysis (2-DG) and cytokine secretion was then measured. We found that cells treated both with media alone or IL-12, 2-DG significantly decreased the TCR- induced production of IFN-γ, TNF-α, IL-4, and IL-10. These data is consistent with a previous study that examined the effects of 2-DG in the TCR-induced production of cytokines in human T cells and suggest that glycolysis is critical for TCR-mediated cytokine production. These findings suggest that IL-12-mediated increase in oxidative metabolism is involved in regulating the release of IL-4 and IL-13 without altering the release of IFN-γ, TNF-α, and IL-10. # Discussion Previous studies from us and others has demonstrated that prior exposure of activated T cells to inflammatory signals increases the responsiveness of these cells to subsequent TCR stimulation. For example, pretreatment with IL-7, IL-15 or a TLR5 ligand increases the responsiveness of human T cells to TCR stimulation. In addition, murine effector/memory CD8 or secondary effector CD4 T cells exposed to pathogen induced-inflammation have enhanced ability to respond to TCR stimulation. Similarly, murine memory CD8 T cells conditioned with IL-12 and IL-18 *in vitro* have enhanced cytokine production and cytotoxic activity upon TCR re-stimulation. Herein, we demonstrate that prior exposure of human activated CD4 T cells to IL-12 results in increased production of a range of cytokines upon TCR re-stimulation. Mechanistically, we observed that IL-12 mediated its effects through both transcriptional and post-transcriptional mechanisms. The IL-12-driven priming of IFN-γ production was linked to increased mRNA expression, likely mediated by the enhanced activation of certain TCR signaling molecules. In contrast, the IL-12-mediated enhancement of cytokines that are not transcriptionally regulated was partially driven by increased oxidative metabolism. Collectively, these studies highlight underappreciated roles for inflammatory signals in continually shaping TCR-mediated responses via context-specific mechanisms. The canonical role for IL-12 in naïve CD4 T cells is the differentiation of Th1 cells. Based on this model, we expected IL-12 pretreatment to induce primarily IFN-γ and TNF-α production upon TCR stimulation. To our surprise, IL-12 pretreatment of human activated CD4 T cells led to a general upregulation of cytokine production upon TCR stimulation. Interestingly, the effects of IL-12 pretreatment in enhancing the production of IL-4 and IL-13 were not as robust as for other cytokines. However, the levels of IL-4 and IL-13 produced in our studies are similar to what it has been observed in human ex-vivo isolated Th2 cells. After polarization, human CD4 T cells remain flexible and can be re- polarized in different inflammatory environments. Furthermore, emerging literature suggests that full CD4 T cell differentiation occurs at sites of infection and inflammation after they have migrated from the secondary lymphoid organs. Therefore, beyond its role in Th1 polarization, IL-12 may also prepare primed, but not fully differentiated, effector CD4 T cells to release numerous cytokines prior to their final polarization. Similar to our study, the IL-12-mediated production of cytokines other than IFN-γ has been previous reported in T clones, peripheral blood T cells, and tumor reactive T cells. Likewise, several preclinical studies have shown that IL-12 administration increases plasma levels of IFN-γ and other cytokines, such as IL-10. Our data fits well with these findings and demonstrates that the regulation of CD4 T cells responses by IL-12 is more complex than previously appreciated. The effects of IL-12 on human activated CD4 T cells had several interesting features. First, short-term exposure to different doses of IL-12 was sufficient to transiently alter responses to TCR stimulation. The fact that the IL-12 mediated effects on human activated CD4 T cells were short lived suggests that the priming of activated CD4 T cells is tightly regulated to minimize the risk of immunopathology. Next, pretreatment of human activated CD4 T cells with IL-12 did not enhance TCR functional avidity, at least using an antibody based system to stimulate the TCR. Interestingly, previous studies on murine CD8 and CD4 T cells, using an antigen based system found that IL-12 signals from pathogen induced inflammation enhanced the functional avidity of these cells. These distinctions could be attributed to differences between human and mouse T cell responses or caused by experimental variations in our defined IL-12 treatment versus *in vivo* inflammation driven by many cytokines. Finally, the proportion of cells making TNF-α, IL-10, IL-4 and IL-13, and the MFI of all assessed cytokines was not significantly altered by IL-12 pretreatment. We believe that by using an inhibitor of protein transport from the endoplasmic reticulum to the Golgi apparatus, we were incapable of detecting IL-12-driven effects in cytokine release downstream of the inhibitor. Similarly, IL-12 pretreatment altered the gene expression of IFN-γ, but not TNF-α, IL-4, IL-13 or IL-10, following TCR stimulation. Together, these data suggest that IL-12 mediates its effects by at least two mechanisms: increased mRNA expression of IFN-γ resulting in more cells releasing this cytokine and increased ability of cells to release TNF-α, IL-4, IL-13 and IL-10. We have begun to characterize the molecular mechanism by which IL-12 potentiates the TCR-mediated production of a range of cytokines. We found that IL-12 exposure significantly increased the TCR induced activation of p38 and resulted in a trend towards increased activation of AKT without altering the activation of other signaling molecules. IL-12 pretreatment also seems to increase the TCR activation of LCK as shown by the a trend towards increased TCR induced phosphorylation of the activating sites in LCK and FYN and a significant reduced phosphorylation of LCK at its inhibitory site Y505. In contrast, pathogen- induced inflammation increased proximal TCR signaling (ZAP-70, PLC-γ) and ERK1/2 and JNK1/2 without altering the activation of p38 in murine effector/memory CD8 T cells and increased ZAP-70 and ERK1/2 in murine secondary effector CD4 T cells. Interestingly, in human T cells IL-7 and IL-15 mediated their effects by increasing the activation of ERK1/2 following TCR stimulation. Our laboratory also observed that prior activation of T cells with a TLR5 ligand enhances TCR- mediated AKT activation, while simultaneously reducing LCK and LAT phosphorylation. These studies suggest that each inflammatory signal potentiates TCR-mediated signaling via a distinct molecular mechanism. A previous report has shown that p38 plays an important role in the production of IFN-γ by murine Th1 CD4 T cells *in vitro*. In this study, the authors demonstrate that blockade of p38 activity inhibits the gene expression of IFN-γ. Furthermore, AKT activation has been shown to have multiple roles on T cell responses including cytokine release. Unpublished data from our laboratory suggests that the presence of an AKT inhibitor (BML 257) reduces TCR-mediated IL-2 production, indicating the critical role that this kinase plays in cytokine production in human T cells. Similar to our findings, previous literature has reported a crosstalk between IL-12 signals and LCK, AKT, and p38. Visconti and colleagues demonstrated that IL-12 signals activate p38 and AKT pathways without altering ERK or JNK activation in T cells. Furthermore, in human resting and activated NK cells, IL-12 signals were shown to increase LCK activation. Future studies will have to explore how IL-12 signals alter the activation of LCK, AKT, and p38 following TCR stimulation without altering more proximal TCR signals. Together, our results suggest that the IL-12-mediated increase of p38 and AKT could be responsible for the increased gene expression of IFN-γ. This correlation will be further examined in future studies to determine if the IL-12 mediated changes in TCR signaling are in fact responsible for the increased gene expression of IFN-γ. We also found that IL-12 pretreated cells have increased oxidative metabolism which partially regulates the release of IL-4 and IL-13. This suggests that the IL-12 enhancement of cytokines that are not transcriptionally regulated could be partially attributed to enhanced oxidative metabolism. This seems to be a particular effect of IL-12, since exposure to other inflammatory cytokines like TNF-α did not alter metabolic pathways (data not shown). Previous reports have demonstrated that brief exposure to inflammatory signals alters the metabolic state of different immune cells. Furthermore, previous reports have demonstrated that changes in metabolic pathways can regulate immune cellular function like cytokine secretion and proliferation. For T cells, the production of IFN-γ is post-transcriptionally controlled by aerobic glycolysis. Similarly, for DCs, the LPS-induced production of cytokines IL-6, IL-12 and TNF-α is also regulated at the translational level by aerobic glycolysis. However, our studies are the first to demonstrate that the cytokine IL-12 promotes metabolic reprograming towards oxidative metabolism, and that oxidative metabolism is involved in the regulation of cellular functions like cytokine secretion. Our findings highlight a critical question: what is the physiological role of the potentiation of cytokine release by IL-12? Naive CD4 T cells are primed and activated in secondary lymphoid tissues. These cells then migrate through the circulatory or lymphatic system before entering into the sites of infection. During this migration, these cells are exposed to IL-12 and other inflammatory signals. Since human CD4 T cells remain capable of differentiating into multiple CD4 T cell lineages after initial priming and activation, we propose a model in which IL-12 present in blood, infection sites, and/or at inflammatory sites alters human activated CD4 T cells that have not terminally differentiated, rendering them capable of producing a range of cytokines upon TCR activation. This allows the cells to respond faster when they encounter their cognate antigen at sites of infection and be easily polarized into CD4 helper T cell subsets depending upon the cytokine milieu they encounter. This function of IL-12 is the likely reason why clinical trials of IL-12 to treat human cancer have shown limited efficacy. IL-12 treatment will not only drive Th1 T cell differentiation, but also enhance the production of Th2 cytokines and suppressive cytokines such as IL-10. In fact, IL-10 levels are increased in patients treated with IL-12. In conclusion, our studies have uncovered that, beyond its role in Th1 differentiation, IL-12 elevates the responses of activated CD4 T cells for further TCR stimulation by altering TCR signaling pathways and by increasing metabolic respiration. Our findings increase our understanding of the physiologic properties of human CD4 T cells and provide insights into potential avenues to improve the current uses of IL-12 in therapeutics. # Supporting Information We thank members of the Bishop laboratory for valuable comments. We also thank Mahmood Bilal, Mikaela Tremblay, Aline Sandouk, and Michael Zhang for helpful discussions and careful reading of the manuscript. We thank members of the Karandikar laboratory for reagents. We thank Brett A. Wagner and Garry R. Buettner for reagents, help with the Seahorse experiments and their invaluable assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AV NMC JTH MJR JCDH. Performed the experiments: AV NMC. Analyzed the data: AV NMC JTH MJR JCDH. Contributed reagents/materials/analysis tools: AV NMC JTH MJR JCDH. Wrote the paper: AV NMC JTH MJR JCDH. [^3]: Current address: Department of Immunology, St. Jude Children's Research Hospital, Memphis, Tennessee, United States of America [^4]: Current address: Department of Microbiology and Immunology, Microbiome and Disease Tolerance Centre, McGill University, Montreal, Quebec, Canada

# Introduction Changes in sensory function have been identified in many painful conditions of the musculoskeletal system especially among patients with chronic low back pain (cLBP). These changes, described in patients with cLBP, usually affects several physiological functions such as reduced sensory acuity, altered muscle recruitment patterns and reorganisation of the somatosensory regions of the brain cortex. Local muscle vibration is often used to evaluate muscle spindle contribution to movement control. Thus, previous studies on local vibration highlight the importance of muscle spindles in proprioception. During local vibration stimulation, a selective activation of the muscle spindles via the Ia afferent fibers is usually observed resulting in a neuromuscular response referred to the tonic vibration reflex (TVR). Some studies have suggested beneficial effects of local vibration such as significant increases in muscle activity and average mechanical power recorded during arm flexion contractions. Indeed, it has previously been suggested that vibration applied on the muscle belly may alter normal motor unit recruitment patterns, determining an increase in rate coding, synchronization and possibly facilitating the recruitment of faster motor units. Kakigi & Shibasaki, in their study aiming to determine the effects of vibration on pain somatosensory evoked potentials and pain threshold, suggested that stimulation of muscle spindles under vibration influence can lead to increases in the inhibitory mechanisms of painful feeling and, therefore, be used for therapeutic purposes. Trunk proprioception has previously been studied in individuals with low back pain using various protocols such as position sense or pointing task (motor control). Several studies have shown proprioceptive impairments in the lumbar spine of individuals reporting cLBP, but few studies have examined the local vibration effects on the performances of a trunk repositioning task in this population. In 2000, Brumagne et al. suggested that when applied at segmental level L5-S1, a multifidus muscle vibration leads to a significant decrease in pelvis directional error in a sitting position as illustrated by a systematic undershooting of the target position in patients with cLBP. The same authors had also concluded, in a previous study, that further research on the effect of vibration in other postures and other muscle groups was required to clarify the complex mechanism of the lumbosacral neuromuscular function. Also, a recent study conducted by Willigenburg et al. found that patients with cLBP had larger errors in a spiral tracking task requiring circular trunk movements compared to healthy controls. The relative effects of lumbar muscle vibration during the motor control task in their study did not lead to any significant improvements in patients with cLBP. Under stress conditions such as muscle fatigue or mechanical loading, lumbar proprioception may also be highly affected. In a study evaluating the effect of paraspinal muscle fatigue on lumbar spine proprioception, Taimela et al. found that patients with cLBP had significantly poorer ability to sense a change in lumbar position after a fatiguing protocol. It has also been shown that excessive fatigability of lumbar paraspinal muscles is a predictor of a first episode of low back pain and a predictor of long-term back-related disability. Indeed, it was suggested that one mechanism by which fatigue contributes to low back disorders may be spinal instability leading to injury (ie. spinal buckling). To control spinal stability, the central nervous system (CNS) must orchestrate a fine-tuned coordination of trunk muscles involving feedback (reflex) and feedforward control mechanisms. Therefore, the use of vibration stimulation, known to increase spindle discharge and muscle activity, could potentially prevent decrease in muscle spindle firing rate generally observed during muscle fatigue and possibly contribute to the active control of spinal stability. Although a substantial decrease in trunk proprioception under muscle fatigue condition can be expected in patients with cLBP, there is a clear lack of objective studies investigating the influence on local vibration upon back muscle fatigue and the changes in trunk proprioception. Therefore, the primary objective of this study was to determine whether or not local vibration stimulation on *erector spinae* muscles would spontaneously yield changes in control strategy, accuracy and variability of the performance in a trunk isometric force reproduction task in patients with cLBP and healthy participants. It was hypothesized that local muscle vibration would improve trunk sensorimotor acuity in patients with cLBP and would decrease trunk sensorimotor acuity in healthy participants. The second objective of the study was to determine if vibration stimulation applied over fatigued muscles could have short-term benefits on trunk force reproduction parameters. To document the influence of muscle fatigue on trunk sensorimotor control in patients with cLBP is relevant, considering that muscle fatigue limits functionality and may hamper the involvement of cLBP patients in rehabilitation and pain management strategies. The authors tested the hypothesis that muscle vibration would lead to improvements in sensorimotor control of the trunk in patients with cLBP during both the no fatigue and the post-fatigue conditions. # Methods ## Participants The study was conducted at the university’s neuromechanics and motor control laboratory. Sample size was estimated in order to detect a moderate effect size of 0.30 with a significance level of *P* = 0.05 and a desired power of 0.80. A minimum of 12 participants per group was needed considering the abovementioned requirements. Twenty healthy adult participants (7 women and 13 men) without any cLBP history and 20 participants (7 women and 13 men) with non-specific chronic or recurrent low back pain were therefore recruited. Patients and healthy participants were included if they were between 20 and 60 years old. Patients with non-specific cLBP were selected according to previously established criteria for chronic or recurrent low back pain (cLBP: present at least half the days over a 6-month period; recurrent low back pain: present for less than half the days over a 12-month period). Patients presenting any non-mechanical spinal condition, neurologic deficits, and chronic pain syndrome were excluded. Healthy adult participants were recruited based on the following criteria: absence of musculoskeletal or neurological symptoms related to a spine condition. Participants presenting the following conditions were also excluded: ankylosing spondylitis, trunk neuromuscular disease, inflammatory arthritis, scoliosis (15° or more) and previous spinal surgery. Before testing, each participant was informed of all experimental procedures and provided their informed written consent. All procedures were approved by the institutional Research Ethics Committee Involving Human Participants (Comité d'éthique de la recherche avec des êtres humains). ## Clinical outcomes Clinical outcomes included the French validated version of the modified Oswestry Low Back Pain Disability Questionnaire (ODQ) to assess low back pain related disability and the Tampa Scale for Kinesiophobia. Clinical pain intensity of the lower back was assessed using a visual analogue scale (VAS) one week prior to testing and at the moment of testing. ## Preparatory procedures Participants were tested in a neutral standing posture, without trunk flexion nor extension. Force data (torque) was obtained from an isokinetic device (The LIDO Active, Loredan Biomedical, West Sacramento, USA) used only in the isometric testing mode. Participants received personal encouragements from the experimenters as the maximal isometric extension torques of *erector spinae* muscle were first collected. The reference value for maximal voluntary contraction (MVC) was set at the highest torque value obtained in two consecutive 4-second trials. As a warm-up procedure, participants were then instructed to produce a sub-maximal trunk isometric force as quickly as possible. In order to ensure that all participants understood and performed the isometric force reproduction task properly, a familiarization phase was completed before the testing began, in which they were asked to produce a single impulse (shoot and release), but without attempting to correct the force once the contraction was initiated. During this phase, participants received visual accuracy feedback from an oscilloscope located in front of them, which helped them evaluate their performance to correct it for the next trial, if necessary. Participants were asked to produce peak torques set at 60% of their MVC within a 10% margin error of the target goal (ex: 100 ± 10 Nm), while keeping their eyes open for the entire session. The term peak torque, therefore, refers to the highest value of the submaximal extension torque for each trial. This familiarization phase, completed without introducing any form of vibration, was ended after participants completed ten consecutive contractions successfully. For every trial, torque data were recorded at a sampling frequency of 100 Hz. They were digitally filtered with an eighth-order Butterworth filter (10 Hz low- pass cut-off frequency). ## Muscle vibration protocol represents the experiment timeline. Superficial mechanical vibration was applied perpendicularly and bilaterally on *erector spinae* muscle at the third lumbar segment level (L3) using 7.5 cm long eccentric rotating masses (4 cm diameter), weighting about 450 g. As the eccentric rotating masse rotates around the central motor shaft, the mass’ movement can be modelled as a sinusoidal wave. These vibrators, designed with a regulated DC power supply (Zurich Electric RPS-1012 MB, Taipei Hsien, Taiwan), which were held in place with a custom-made Velcro elastic lumbar belt, were placed in a standard position on all participants by the same examiner, to ensure that they were was secured with comparable tension in all tests. Vibration properties of 80 Hz with constant amplitude of 0.85 mm were the same that were used in a recent study conducted on trunk muscle. Participants were asked to perform a set of five trials (force reproduction task at 60% of their MCV) following an auditory signal which was heard every thirty seconds, for each of the vibration conditions (no vibration, 80 Hz). That sequence represented one block of trials and four blocks were completed for a total of twenty trials for each vibration condition. Participants were allowed to rest for 5 minutes between each block in order to limit any fatigue effects. The order of appearance of vibration conditions (no vibration, 80 Hz vibration) differed between block 1 and block 2 (and between block 3 and block 4) to limit any vibration sequence effect. The vibration stimulation (80 Hz) was applied thirty seconds before an auditory signal was activated, and lasted through the torque generation trials, without any rest or delay. Between block 2 and 3, participants completed a trunk extensor muscle fatigue protocol corresponding to a modified version of the Biering-Sorensen test. This test was performed on a 30° Roman chair, with straight upper body, the iliac crest aligned with the chair's limit, and the arms crossed on the chest in a prone position. During the fatigue task, participants were asked to hold a 11.4 kg plate to promptly induce muscular fatigue. The test was completed when the participant could no longer maintain their trunk in a straight horizontal position visually assessed (below the criterion position) or when the participant terminated the test (total exhaustion). Once the fatigue protocol completed, the participants were asked to score their perceived exertion on a 20-point Borg Scale. ## Electromyography Surface electromyography (sEMG) was used to record *erector spinae* muscle activation during the isometric contractions and during the fatigue protocol. Four rigid bipolar electrodes recorded *erector spinae* muscle activation bilaterally at L4-L5 level and L1 lumbar segments. The electrodes were aligned with the muscle fiber direction and placed on the muscle belly. A reference electrode was placed on the T12 spinous process to assess internal vibration damping (essential to sEMG signals processing and noise cancellation techniques) and a ground electrode was placed over the right olecranon of all participants. Prior to application, electrical impedance of the skin at the site of electrode placement was minimized using standard skin preparation techniques. sEMG activity was recorded using a Delsys sEMG sensor sampled at 1000 Hz with a 12-bit A/D converter (PCI 6024E; National Instruments, Austin, TX). ## Data analyses Time to peak torque (TPT), constant error (CE), variable error (VE) as well as absolute error (AE) in peak torque were calculated and compared between experimental conditions. These variables have been successfully used in previous studies and they are described in more detail below. For each trial, the onset of torque and peak torque were determined, wherein onset of torque represents the initial point of the ramp up and peak torque represents the highest torque value reached for one trial. The detection and marking of these two values were made through a visual inspection by the same experimenter. The TPT is defined as the time required to reach the peak torque from the onset of torque generation. The CE represents the positive or negative difference between the peak torque reached and the target torque corresponding to 60% of the MVC; it indicates the amount of direction of error relative to the target torque (measure of bias). A positive CE in trunk extension corresponds to undershooting the target torque, while a negative CE corresponds to overshooting the target torque. The VE measures the inconsistency in movement outcome and represents the variability of the participant’s performances about the mean value; it is calculated by the participant’s peak torque score on each trial and his or her own average score. represents an example of the CE and VE calculations for one participant. Finally, the AE in peak torque represents the average absolute deviation (without regard to torque direction) between the participant’s responses and the target torque, which accounts for both bias and variability. The variable scores (CE, VE, AE and TPT) for one participant represents his average score calculated for each experimental condition. All trials of each participant were analysed and used for the variable calculations. sEMG data were filtered digitally by a 10- to 450 Hz band-pass, zerolag, fourth- order Butterworth filter. The data were collected by LabView (National Instruments) and processed by Matlab (MathWorks, Natick, MA). Because vibration can lead to motion artifacts resulting in spectral components at the approximate modulating frequency and its harmonics, the authors used a simple approach to make the sEMG root mean square (RMS) measurement insensitive to the DC power supply activation burst and vibration motion artefacts. The recorded signals for each participant’s trial were submitted to an intensive frequency domain analysis. On the basis of this analysis, vibration artefacts and related harmonics were excluded using stop-band fourth-order Butterworth filters around the modulating frequency and each harmonic. RMS values during each trunk extension trial were calculated using time-windows corresponding to the onset and cessation of extension torque production. For each participant, mean values across conditions (10 trials) were then calculated and used to establish muscle activation patterns as suggested by previous authors. RMS values were normalized with respect to the trunk extension MVC value. To determine if *erector spinae* muscle exhibited fatigue across the modified Biering-Sorensen test, sEMG median frequency (Fast-Fourier Transform) was calculated from adjacent non-overlapping signal epochs of 1 s. The slope of the median frequency regression and its intercept were used to define the rate of change and median frequency initial values. Normalized slopes were defined as the slopes divided by the corresponding initial values and expressed in percentage per second. Localized muscle fatigue usually cause the sEMG median frequencies to decrease, yielding a negative slope. The sEMG analyses were conducted by averaging normalized sEMG_RMS values from both sides (left and right), as used in previous studies. ## Statistical analyses The Shapiro-Wilk test was used to assess the distribution of the variables. All mean error in peak torque measures (CE, VE and AE) and TPT variables were normally distributed (all *P* \> 0.05). This study was conducted with a counterbalanced measure design. Sampling distribution for each participant was also assessed for outlying observations (standard deviation \> 3). Peak torque data and normalized sEMG_RMS were submitted to a 2 (groups: cLBP and control) × 2 (vibration conditions: no vibration and 80 Hz vibration) × 2 (fatigue conditions: no fatigue and post-fatigue) repeated measure analysis of variance (ANOVA), ran separately for each variable. In order to assess a potential fatigue phenomenon, pre and post-MVC values were submitted to a group × fatigue condition ANOVA with repeated measures on the last factor. Significant interactions or main effects were further analyzed using a post hoc Tukey test. sEMG median time-frequency slopes for each segmental level were submitted to a one-tailed student’s *t* Test to determine if they were significantly different than zero. The significance level was set at *P* \< 0.05 for all analyses. The statistical analysis was performed with Statistica 10 (Statsoft, OK, USA). # Results Demographic and clinical profiles of the participants are presented in. ## Overall results The statistical analyses yielded a significant interaction between groups, vibration conditions and fatigue conditions for the CE variable \[*F*(1, 76) = 6.99, *P* \< 0.01\] and for the AE variable \[*F*(1, 76) = 14.63, *P* \< 0.001\]. No significant interaction effect between groups, vibration conditions and fatigue conditions was found for the VE variable \[*F*(1, 76) = 0.05, *P* \> 0.05\] and the TPT variable \[*F*(1, 76) = 1.28, *P* \> 0.05\]. No significant interaction effect between groups, vibration conditions and fatigue conditions was found for the sEMG_RMS at the L1 level \[*F*(1, 76) = 0.12, *P* \> 0.05\]. A vibration main effect was detected, revealing significantly higher sEMG_RMS activity at L1 \[*F*(1, 76) = 6.81, *P* \< 0.05\] with the 80 Hz vibration when compared to the no vibration condition. At the L4-5 level, no significant interaction effect between groups, vibration conditions and fatigue conditions was found for the sEMG_RMS \[*F*(1, 76) = 0.26, *P* \> 0.05\]. However, a significant fatigue × vibration interaction effect \[*F*(1, 76) = 9.15, *P* \< 0.01\], as well as a significant group × fatigue interaction effect \[*F*(1, 76) = 5.81, *P* \< 0.05\] were found for the sEMG_RMS. A significant vibration main effect was observed, revealing higher sEMG_RMS activity at L4-5 \[*F*(1, 76) = 5.76, *P* \< 0.05\] with the 80 Hz vibration when compared to the no vibration condition. A group main effect \[*F*(1, 76) = 7.23, *P* \< 0.001\] was also detected for the sEMG_RMS at the L4-5 level, showing that patients with cLBP had significantly higher sEMG_RMS activity than healthy controls. ## Vibration effects under the no fatigue condition Post hoc analyses revealed a significantly higher CE scores in patients with cLBP compared to the healthy group during the no vibration condition (*P* \< 0.001). The two groups also responded differently to vibration of the *erector spinae* muscle as patients with cLBP showed a significant decrease in CE during vibration exposure (*P* \< 0.001) while the control group conversely showed a significant increase in CE mean scores (*P* \< 0.05) when compared to the no vibration condition. Post-hoc comparisons also revealed a significantly higher AE scores in patients with cLBP compared to the healthy group during the no vibration condition (*P* \< 0.001). Patients with cLBP showed a significant reduction in AE under vibration exposure as compared to the no vibration condition (*P* \< 0.001). In contrast, the healthy participants during the *erector spinae* muscle vibration showed significantly higher AE scores in comparison with the no vibration condition (*P* \< 0.01). Post-hoc comparisons for group × fatigue interaction showed that patients with cLBP had higher sEMG_RMS activity at L4-5 level than healthy controls for the no fatigue condition (*P* \< 0.001). Post-hoc comparisons for the fatigue × vibration interaction effect revealed a significantly higher sEMG_RMS activity at L4-5 level with the 80 Hz vibration when compared to the no vibration under the no fatigue condition (*P* \< 0.01), independently of the group considered. ## Fatigue effects The presence of back muscle fatigue led patients with cLBP to a significant decrease in sEMG_RMS activity at L4-5 level as compared to the no fatigue condition (*P* \< 0.01), independently of the vibration condition. Statistical analysis yielded a main effect of fatigue for pre and post-MVC values (*P* \< 0.01). The average MVC (± SD) in trunk extension for patients with cLBP was 137.53 ± 73.24 Nm before the fatigue protocol and 123.34 ± 68.44 Nm immediately after the fatigue protocol, indicating a 10.32% decrease in MVC (*P* \< 0.05). The average MVC in trunk extension for healthy participants was 130.79 ± 56.68 Nm before the fatigue protocol and 109.82 ± 48.44 Nm immediately after the fatigue protocol, indicating a 16.03% decrease in MVC (*P* \< 0.05). No significant difference between groups was observed for MVCs before and after the fatigue protocol (all *P* \> 0.05). Important fatigue-related changes were observed during the Biering-Sorensen protocol. sEMG time-frequency analysis at the L1 and L4-5 paravertebral electrode sites for both groups indicated that linear regression slopes were all negatives. Student’s *t* Test revealed no significant difference between both groups for the total holding time, and for the L1 and L4-5 linear equation slopes (all *P* \> 0.05). One-tailed *t* Tests revealed significant muscle fatigue at each segmental level for both groups (all *P* \< 0.05). presents the participant’s perceived exertion during the fatigue protocol. ## Vibration effects under the fatigue condition Following the fatigue protocol, healthy participants showed a significant decrease in CE during muscle vibration when compared to the no vibration condition (*P* \< 0.001). A decrease in CE scores during vibration was also observed in patients with cLBP (*P* \< 0.001). During vibration exposure, a difference was observed between the no- and the post-fatigue condition for the control group (*P* \< 0.05). There were significant differences for both the vibration and no vibration conditions where cLBP patients showed significantly higher AE scores as compared to the healthy participants (all *P* \< 0.01). Following the fatigue protocol, AE scores in patients with cLBP were significantly higher than the no fatigue scores for the vibration (*P* \< 0.01) and the no vibration conditions (*P* \< 0.05). The presence of back muscle fatigue led to a significant decrease in sEMG_RMS activity at L4-5 level for the 80 Hz vibration condition, independently of the group considered. Finally, a main effect of fatigue was found for the TPT variable \[*F*(1, 76) = 16.76, *P* \< 0.001\] where values for both group were significantly higher in the post-fatigue condition regardless of the vibration conditions. Participants from both groups took more time, therefore, to reach the peak torque following the fatigue protocol. # Discussion The aim of the present study was to assess the performance accuracy and variability of trunk reproduction force in conditions with and without *erector spinae* muscle vibration, and to evaluate the influence of muscle fatigue on trunk sensorimotor control in patient with cLBP and healthy participants. The present study included a group of cLBP patients with an average mild pain level score (2.3 ± 1.8) at the moment of testing. Results showed that patients with cLBP had significantly lower trunk isometric force reproduction accuracy than the healthy participants. Higher CE and AE mean scores found in patients with cLBP clearly support this observation. Similar findings have been previously reported by Brumagne et al., who found that patients with cLBP had a less refined lumbosacral position sense than healthy individuals in a sitting position. The results reported by Brumagne et al. provide evidence for reduced trunk neuromuscular control during dynamic contractions in patients with cLBP, and the results of the present study provide evidence for reduced trunk neuromuscular control during isometric contractions in patients with cLBP. It is worth mentioning that measures of error such as CE, AE and VE mean scores reported in the present study and the one conducted by Brumagne et al. are considered as outcome measures and not process measures (see Schmidt & Wrisberg for detailed description). Sensorimotor disturbances of the spine could result from modifications in somatosensory afferent activity, which can be due to trauma or to the modulatory effect of pain and sympathetic activation on muscle spindle sensitivity. Consistent with this explanation, Myers et al. suggested that increased afferent signals sent by pain receptors are believed to override and subsequently decrease proprioception afferents. Reweighting of sensory signals based on location have also been demonstrated in patients with cLBP as they seem to adopt a body and trunk stiffening strategy and rely more on lower limb proprioception . In the present study, it is therefore possible that patients with cLBP, having limited somatosensory information from back muscle, had to reweight sensory information from other segments or muscle groups. This way, distorted afferents from back muscle could have been compensated by other undistorted afferents originating from the pelvic girdle and lower limb muscles leading to lower trunk isometric force reproduction accuracy. ## Vibration effects Vibration of the *erector spinae* muscle induced a significant reduction of the CE and AE mean scores in patients with cLBP. The accuracy with which patients with cLBP reproduced a trunk sub-maximal force was, therefore, improved during vibration stimulation when compared to the no vibration condition. This acute effect of local muscle vibration in patients with cLBP has several possible explanations. Hollins et al., in their study on the ability of vibration to compromise detection of a nociceptive stimulus, found that vibrations ranging from 20 to 230 Hz were able to modulate nociception by reducing the noxious stimulus sensitivity. They also concluded that no mechanoreceptive channel appears to have a privileged role in antinociception. Some authors also suggested that muscle vibration could distort muscle’s primary afferent by introducing a bias signal in a parallel channel, because vibration can modestly active the Pacinian corpuscle and others cutaneous mechanoreceptors. Although the neurophysiological mechanism for such changes remains unclear, it is possible that vibration stimulation in this study may have improved the muscle spindle function in patients with cLBP therefore improving somatosensory information processing. By stimulating Ia afferents and modulating the nociceptive pathways activation, vibration stimulation may lead to a transient sensory reweighting of the *erector spinae* muscle resulting in significant improvements in trunk neuromuscular control. It is well known that mechanical vibration administered to tendons or muscles can induce a reflex response. The study conducted by Nakajima et al. suggests that 80 Hz vibration during muscle contraction can lead to specific TVR increases observed with higher sEMG_RMS activity of the biceps. In the present study, local vibration application led to significant increase in *erector spinae* muscle activity at both L1 and L4-5 segmental levels. These results could partly be explained through the TVR effect. Regardless of the experimental conditions, patients with cLBP in this study also showed higher muscle activity than healthy controls at L4-5 level but not at L1 level. This is consistent with the ‘‘redistribution of activity within muscle” theory that has been described in patients with cLBP. It is difficult to explain the improvements in force reproduction accuracy of the trunk observed in patients with cLBP by the vibration-related increase in muscle activity, since healthy controls also showed significant increases in muscle activity during vibration but with higher CE and AE mean scores. It is suggested, therefore, that clinical efficacy of local vibration during a trunk sensorimotor task in patients with cLBP is not directly linked to the *erector spinae* muscle sEMG activity. It is also possible that the pain alleviation usually reported during vibration might have led to a significant reduction in the fear-avoidance behavior during the force reproduction task. Conversely, vibration stimulation significantly decreased the healthy participants’ accuracy during the force reproduction task. This finding is in accordance with a previous study conducted on healthy participants in which *erector spinae* muscle vibration interfered with torque generation sequence by distorting proprioceptive information resulting in muscle lengthening illusion. ## Muscle fatigue and vibration Previous studies have shown that back muscle fatigue is usually accompanied by a diminished control of trunk movements and an altered coordination of trunk muscle activities, as well as reductions in accuracy when trying to generate a given force. It is also well established that patients with cLPB have excessive fatigability of the back muscles which is probably a consequence of pain, rather than a cause. From a theoretical point of view, these deficits, in association with cLBP, may leave the lumbar spine more susceptible to reinjury. Therefore, the purpose of creating a fatigue condition in the present study was to determine if vibration stimulation applied over fatigued muscles could have short-term benefits on trunk force production parameters. The results of the present study suggest that the back extensors fatigue protocol clearly induced muscle fatigue for both groups. However, the level of muscle fatigue between cLBP and healthy participants did not differ. Under the influence of vibration, patients with cLBP showed significant reduction of the CE values, which suggests accuracy improvements in the force reproduction task. AE values with and without vibration stimulation, however, were higher during the post-fatigue condition. A controversy exists, however, about the use and interpretation of AE. The mathematical properties of AE have been shown to be a complex combination of CE and VE, and it remains difficult to precisely assess the relative contribution of each measurement to AE. Based on CE mean score, these findings suggest that acute *erector spinae* vibration may enhance sensorimotor acuity in patients with cLBP in conditions with and without back muscle fatigue, and could, therefore, be considered as an adjunct to actual rehabilitation strategies. It is suggested that vibration stimulation during rehabilitation exercises could potentially reduce the risks related to back muscle fatigue and participate in preventing lumbar spine reinjury. Interestingly, healthy participants during the post-fatigue condition and under vibration stimulation showed significant decreases in CE when compared to the no fatigue condition. Theoretically, muscle spindle discharge decreases during fatiguing static contraction. Consistent with this change, motor units usually demonstrated a decrease in firing rate during muscle fatigue. Griffin et al., in their study on vibration and motor unit firing rate, found that application of periodic muscle vibration maintained the motor unit firing rate during submaximal fatiguing isometric contractions. These authors also concluded that muscle vibration can enhance the excitatory input from Ia afferents to the motoneuron pool and transiently restore the motor unit discharge rate. In the present study, the significant decrease in sEMG activity at L4-5 level observed following the fatigue protocol could not, however, be prevented by local vibration stimulation. Although the hypothesis that fatigue-related modulation of motor unit firing rate during submaximal contractions could be prevented by vibration in healthy controls received initial support, the present findings do not provide any evidence to confirm this assumption. Even though it is not of clinical importance, alternative theories are required to explain the improvement in sensorimotor control under vibration stimulation following a fatigue protocol in healthy controls. Higher TPT scores were found in all post-fatigue conditions for both groups, suggesting a modified control strategy developed under exposure to muscle fatigue. This result may be closely related to the neuromuscular effects of fatigue on muscle activation, including reduction in the ability of muscle to produce force and power. ## Clinical considerations A spectrum of clinical interventions has been proposed to retrain motor control in the presence of musculoskeletal pain and disability. The present findings suggest that 80 Hz vibration stimulation in patients with cLBP is likely to influence motor adaptation of the sensorimotor system, leading to improved accuracy in isometric force production parameters. Bosco, Cardinale, in their study on local vibration on mechanical power and sEMG activity, concluded that vibration stimulation was able to stimulate the neuromuscular system more than others treatments used to improve neuromuscular properties. *Erector spinae* muscle vibration could, therefore, be used as an additional stimulation to the sensorimotor system during rehabilitation exercises. The sensory integration process may increase the contribution on *erector spinae* sensory afferents during motor control exercises performed under vibration stimulation (sensory reweighting). The results of the present study also showed that vibration effects on trunk sensorimotor control (increases in force production accuracy) may operate similarly with and without back extensors muscle fatigue. These findings suggest that vibration may lead to beneficial effects at multiple stages of the rehabilitation process where patients with cLBP present different levels of muscle fatigue. Even if *erector spinae* vibration stimulation suggests short-term benefits on neuromuscular control, potential long-term benefits involving primary outcomes such as spine loading, movement and motor variability need to be further investigated. ## Limitations A potential limitation of the present study is that results may suggest rapid but transient adaptations to muscle vibration, resulting in changes in isometric force production parameters. Although such adaptation is mainly driven by changes in muscle spindle discharge, changes in sensorimotor integration or motor planning remain to be determined. The experimental design of this study, however, cannot provide specific information with regard to the contribution of central pathways and mechanisms aiming at identifying the complete proper neural mechanisms. A second limitation is the sEMG frequency components and the harmonics related to vibration that were excluded from the sEMG_RMS calculation. Excluding vibration-related frequencies may have removed physiological signals from the muscle. Further research should focus on the optimal dose relationship of vibration duration on neuromuscular and performance aspects for cLBP populations. In addition, more studies are needed to determine whether these responses correlate with long-term clinical outcomes. # Conclusion The findings of this study suggest that patients with cLBP have a less accurate force generation sense than healthy individuals, presumably because of altered *erector spinae* muscle spindle afferents. Local muscle vibration led to significant trunk neuromuscular control improvements in the cLBP patients, which suggests that the weighting of proprioceptive feedback from *erector spinae* muscle spindles differs between groups. The significant increase in neuromuscular control observed for the patients in the non-fatigued condition supports the clinical evidence of improved trunk function during vibration application, and contributes to the efficacy of this approach in the management of patients with cLBP. Local muscle vibration also led to significant trunk neuromuscular control improvements in the cLBP patients following a muscle fatigue protocol. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JAB JA MCN MD. Performed the experiments: JAB JA. Analyzed the data: JAB JA FN MD. Contributed reagents/materials/analysis tools: JAB FN MCN MD. Wrote the paper: JAB JA FN MCN MD. [^3]: ‡ These authors also contributed equally to this work.

# Introduction Glioblastoma (GBM), arising from glial cells, is the most frequent and most malignant primary brain tumor in adults. The median survival (MS) for newly diagnosed GBM patients treated with the current standard of care, including surgery, radiation, and temozolomide chemotherapy, is 15 to 18 months. Conversely, brain metastases occur in 5–7% of patients with melanoma and breast cancer. The MS for melanoma and breast cancer patients with brain metastases with the current standard of care, including surgery, radiation, and systemic immunotherapy or chemotherapy is 29 and 2 to 25 months, respectively. These poor outcomes mandate a need for the development of improved therapeutic options. Tumor-targeted therapy is highly desirable due to its high specificity and potency in multiple cancer types. Among the targeted therapies under development, immunotoxins (ITs) have emerged as a class of promising therapeutic candidates. ITs are produced by genetically fusing single-chain variable-region antibody fragments (scFvs) to a toxin molecule, such as the 38 kDa truncated mutant form of *Pseudomonas* exotoxin A (PE38). An improved PE38 variant (PE38KDEL), was designed with a C-terminal KDEL addition to increase the intracellular retention and cytotoxicity of the ITs. ITs bind to cell surface antigens via the scFv portion. Upon antigen binding, they are internalized into endosomes, and the PE38KDEL moiety is cleaved by furin. The catalytically active C-terminal fragment then translocates to the cytosol via the endoplasmic reticulum (ER), where it inactivates elongation factor 2 (EF2) by ADP- ribosylation of the EF2 diphthamide residue, leading to protein synthesis inhibition and apoptosis. The utility and specificity of ITs are dictated by the targeted tumor antigen. Numerous cell-surface tumor antigens have been investigated as therapeutic targets in brain tumors. The melanoma- and glioma-associated antigen chondroitin sulfate proteoglycan 4 (CSPG4) is overexpressed in 90% of melanomas and gliomas as well as in breast cancer. CSPG4 is expressed on ‘cancer stem-like cell populations’ and promotes radiation resistance in GBM. CSPG4 is recognized by monoclonal antibodies (mAbs) Mel-14 and 9.2.27. Mel-14 and 9.2.27 mAbs target different regions of the extracellular domain on CSPG4 and the anti-tumor efficacy of one mAb over the other is unknown. To determine the utility of Mel-14 and 9.2.27 mAbs for targeted brain tumor therapy, we developed two CSPG4-targeting scFv based immunotoxins, Mel-14-PE38KDEL and 9.2.27-PE38KDEL. Compared to other targeted therapeutic approaches, ITs are attractive due to their excellent safety profile and potent cytotoxicity in the pM-fM range. Despite these advantages, cancer cells usually do not exhibit homogeneous susceptibility to ITs. Small molecule Bcl-2 inhibitors (ABT-737, ABT-263, and ABT-199 \[Venetoclax\]) were proposed to combat inherent tumor cell resistance to ITs by studies in cervical adenocarcinoma, pancreatic cancer, and small cell lung cancer. ABT-737, along with its analogs ABT-263 and ABT-199, are BH-3 mimetics that target anti-apoptotic pathways by binding to and neutralizing the pro-survival members of the Bcl-2 family proteins, such as Bcl-2, Bcl-xL, and Bcl-w, thereby promoting apoptosis via the release of Bax and Bak proteins. In this study, we evaluated the synergistic effect of ITs (Mel-14-PE38KDEL or 9.2.27-PE38KDEL) targeting CSPG4 and one of the five small molecule Bcl-2 family inhibitors, ABT-737 and ABT-263 (specific for Bcl-2, Bcl-xL, and Bcl-w), ABT-199 (specific for Bcl-2), A-1155463 (specific for Bcl-xL), or S63845 (specific for Mcl-1) in GBM xenograft cells, and melanoma and breast cancer cell lines. We determined the *in vitro* cytotoxicity of the Bcl-2 family inhibitors/IT monotherapies and combination therapies using a colorimetric cell proliferation assay (WST-1). Further, we established the *in vivo* efficacy of the ABT-737 and 9.2.27-PE38KDEL combination in intracranial GBM patient-derived xenograft (PDX) and melanoma brain metastasis models. Mechanistic studies revealed the factors contributing to the efficacy of ABT-737/9.2.27-PE38KDEL combination therapy in GBM, melanoma, and breast cancer cells. # Materials and methods ## Xenografts and cell lines Human GBM xenografts (D-10-0021 MG, D-245 MG, and D-08-0326 MG) were established from patient tumors obtained through informed consent. The subcutaneous xenografts were harvested when the tumors reached an average volume of 500 mm³ to 750 mm³; tumor harvest date was between 30–45 days post-implantation. None of the subcutaneous tumors succumbed to ulceration or blistering. The maximum tumor volume threshold for the subcutaneous xenografts was set at 1500 mm³ per our approved animal protocol A049-17-02. Human melanoma cell line (H350) were maintained in our laboratory. Human melanoma cell lines (DM440 and DM443) were kindly provided by Dr. Douglas S. Tyler at Duke University. Human breast cancer cell lines (SUM159, SUM159-R113, and Hs 578T) were kindly provided by Dr. Robin Bachelder at Duke University. All cells were cultured in an incubator at 37°C, 5% CO₂, and passaged when reached confluence with Accutase Cell Detachment Solution (BD Biosciences \#561527). All GBM xenografts and the melanoma cell line H350 were maintained in 1x MEM ZINC Option media (Gibco \#05-0009DJ) supplied with 10% fetal bovine serum (FBS). DM440 and DM443 cells were cultured in DMEM media (Gibco \#11995) supplied with 10% FBS. SUM159 and SUM159-R113 cells were grown in Ham’s F-12 base media (Gibco \#11765) with the addition of 5% FBS, 1 μg/ml hydrocortisone, and 5 μg/ml human insulin. Hs 578T cells were maintained in DMEM media (Gibco \#11995) with the addition of 10% FBS and 10 μg/ml human insulin. ## Western analysis ### Preparation of total cell lysates Cells were harvested and solubilized in polysomal lysis buffer (100 mM KCl, 5 mM MgCl₂, 10 mM HEPES pH7.0, and 0.5% NP-40) supplemented with Halt Protease & Phosphatase inhibitor cocktail (Thermo Scientific \#78440). Protein concentrations were determined using the Bradford protein assay (Bio-Rad \#5000001). Equal amounts (5–10 μg) of total protein were loaded onto NuPAGE 4%-12% Bis-Tris gels (Invitrogen \#NP0321) and transferred to nitrocellulose membranes (GE \#10600000) for staining. ### Preparation of cell fractionations Cells were harvested and resuspended in a cytosol extraction buffer \[110 mM KOAc, 25 mM K-HEPES (pH 7.5), 25 μM MgCl₂\] supplemented with 1 mM DDT, 150 μg/ml digitonin (Sigma \#D141) and protease inhibitor cocktail. Cells were incubated on ice for 15 min, then the supernatant containing the cytosolic fraction was harvested by centrifugation at 10000 g for 10 min. The permeabilized cells were washed and centrifuged at 500 g for 2 min at 4°C. The pellets were resuspended in an ER-solubilization buffer \[200 mM KCl, 25 mM K-HEPES (pH 7.5), 10 mM MgCl2\] supplemented with 1 mM DDT, 20 mg/ml n-dodecyl- b-D- maltoside (Sigma \# D4641) and protease inhibitor cocktail. After a 30-min incubation on ice, the mixture was centrifuged at 10000 g for 10 min to remove nuclei, mitochondria, and incompletely solubilized cellular debris. ### Global translation inhibition assay Cells were treated with 5 μM puromycin (Tocris \#4089) for 15 min, then washed twice and immediately lysed in polysomal lysis buffer supplemented with protease inhibitor cocktail. ### Drugs and antibodies ABT-737 was prepared using the same method as described in the WST-1 assay. Primary antibodies against the following targets were purchased from Cell Signaling Technology: p-FAK(#8556), FAK (#7143), p-AKT (#2965), AKT (#9272), p-PKCα (#9375), PKCα (#2056), GAPDH (#2118), Bcl-2 (#4223), Bcl-xL (#2764), Mcl-1 (#5453), Bim (#2933), Bax (#5023), and PARP (#9542). Antibodies against puromycin (Millipore \#MABE343) and furin (ProteinTech \#18413-1-AP) were also obtained. An anti-PE38 antibody capable of binding to both intact and cleaved PE38 toxins was kindly provided by Dr. Ira Pastan’s laboratory at National Institutes of Health (NIH). All experiments were repeated at least 2 times. ## Animal work approval All animal work described in the manuscript have been reviewed and approved by the following animal research ethics committee of Duke University Medical Center: Duke Occupational and Environmental Safety Office, Duke Office of Animal Welfare Assurance, Duke Veterinary Committee, and finally the Duke Institute of Animal Care and Use Committee. The protocol number for the approved animal work is A049-17-02. ## Anesthesia Isoflurane anesthesia (Absolute Anesthesia Inc.) via inhalation was used during surgical procedures as it stabilizes the mice during surgery and allows for efficient post-surgical recovery. ## Euthanasia CO₂, inhalation followed by decapitation was used for euthanasia. All animals were euthanized in accordance with the established humane endpoints listed in our approved animal protocol A049-17-02. For subcutaneous tumors, the humane endpoint is reached when tumor burden reaches 1500 mm³. For intracranial tumors, the humane endpoint is reached using neurological symptoms and body weight loss of 15% or the animals’ inability/desire to move (the mouse does not move forward two steps when prompted gently). ## Intracranial tumor models All experiments were done in accordance with the Institutional Animal Care and Use Committee of Duke University Medical Center (A049-17-02). Animals were group-housed, maintained in a barrier facility, under pathogen-free conditions according to NIH guidelines. Nude mice (≈22–30 g, 6–8 weeks, female:male = 1:1, Duke University, Division of Laboratory Animal Resources) were anesthetized by isoflurane inhalation and mounted onto the stereotactic frame (Stoelting Co.). The anterior cranial region was shaved, and an incision ≈1 cm in length was made in the skin over the skull. A 25-gauge needle attached to a 25-μl Hamilton syringe was used to pierce through the skull at coordinates 2.0 mm left lateral of the sagittal and 0.5 mm anterior to the bregma. The needle was inserted vertically to a depth of 2.5 mm from the dura mater. A total of 1x10⁵ D-10-0021 MG GBM cells freshly dissociated from a subcutaneous mouse xenograft, or 1x10⁵ DM440 melanoma cells harvested from culture, were injected in 5 μl of 1xPBS containing 2% methylcellulose (Sigma \#M0512). Five days post-tumor implantation, the animals were randomized into four treatment groups, with 9–10 mice per group: vehicle control, 9.2.27-PE38KDEL, ABT-737, and 9.2.27-PE38KDEL/ABT-737 combination therapy, randomized according to initial weight. For each mouse in the study, a brain infusion cannula (Alzet \#0008851) attached to a subcutaneously implanted mini-osmotic pump (Alzet \#1007D) was inserted directly into the intracranial tumor site for intratumoral delivery of the vehicle solution or the drugs, at a rate of 0.5 μl/h for 72 hours. The vehicle control group received the PBS-based solution containing 5% Captisol and 2% mouse serum albumin. The IT monotherapy group received a total dose of 0.1 μg of 9.2.27-PE38KDEL diluted in the vehicle solution. The ABT-737 group received a total dose of 2.93 μg of ABT-737 (equivalent to 36 μl of the 100 μM of ABT-737 diluted in the vehicle control solution). The combination group received both 9.2.27-PE38KDEL and ABT-737 at the doses indicated above. During the surgery, animals in different groups were treated in a random order. Animals were observed twice daily for signs of distress. Survival endpoint is defined as the onset of neurologic symptoms (lethargy, seizure, repetitive circling, difficulty breathing, and hunched posture), greater than 15% loss of body weight, or death, whichever comes first. When endpoint was reached, mice were euthanized as described earlier. Kaplan-Meier survival curves were plotted and compared using log-rank test. Each study was repeated at least twice. # Results ## Expression of CSPG4 on tumor cells and *in vitro* cytotoxicity of IT, ABT, A-1155463, and S63845 monotherapies Flow cytometry analysis (FACS) revealed the presence of CSPG4 on all GBM, breast cancer, and melanoma cell lines tested, albeit at varying levels. Q-FACS analysis showed that the surface density of CSPG4 was lowest in SUM159-R113 cells (≈100,000 molecules per cell) and highest in DM440 cells (≈700,000 molecules per cell). Despite abundant CSPG4 cell surface expression, *in vitro* cytotoxicity assays demonstrated that all GBM xenografts, all breast cancer cell lines, and the DM443 melanoma cell line were highly resistant to both Mel-14-PE38KDEL and 9.2.27-PE38KDEL ITs (IC₅₀ \>100 ng/ml) (Tables). Only the H350 melanoma cells were susceptible to the cytotoxicity of 9.2.27-PE38KDEL (IC₅₀ of 11.67 ng/ml;). Both H350 (IC₅₀ = 42.50 ng/ml) and DM440 (IC₅₀ = 68.30 ng/ml) melanoma cells showed weak cytotoxicity upon Mel14-PE38KDEL treatment alone. These data suggest substantial inherent resistance to CSGP4-targeting ITs in the majority of cancer cells expressing the target. To explore the utility of pro- apoptotic enhancers in countering this resistance, we first evaluated the sensitivity of the cell lines in our panel to ABT-737, ABT-263, ABT-199, A-1155463, or S63845 at a concentration of 5–20 μM (see, Figs for the IC₅₀ values). The absorbance of ABT/A-1155463/S63845-treated (Test) vs. vehicle-treated cells (0.5% DMSO in media) (Control) was used to calculate cell viability. A concentration of ABT/A-1155463/S63845 compound that yielded at least 70% cell viability (sub-therapeutic dose) in all three cell lines from GBM, melanoma, or breast cancer was selected for combination therapy. Thus, the concentration of ABT-737 chosen for combination with the ITs was 20 μM (GBM, melanoma), and 10 μM for breast cancer cell lines. For ABT-263 combinations, we chose concentrations of 5 μM (GBMs, breast cancer), and 10 μM for the melanoma cell lines. For ABT-199, the concentrations were 10 μM (GBMs, melanoma), and 5 μM for the breast cancer models. Finally, a concentration of 10 μM (GBM, breast cancer) and 20 μM (melanoma) were chosen for A-1155463 while 5 μM (GBM, breast cancer) and 10 μM were chosen for S63845 (melanoma). ## *In vitro* cytotoxicity of IT and ABT combination therapies The cytotoxicity (IC₅₀) of IT monotherapies or Mel-14-PE38KDEL/9.2.27-PE38KDEL+ABT or 9.2.27-PE38KDEL+A-1155463/S63845 combination therapies are summarized in Tables and, and. In all cell lines tested, except for Hs 578T, 9.2.27-PE38KDEL produced stronger synergy with all three ABT compounds than Mel-14-PE38KDEL (Tables). With the 9.2.27-PE38KDEL+A-1155463/S63845 combinations, overall a lower efficacy (DM440-IC₅₀ = 8–23 ng/ml and D-10-0021 MG-IC₅₀ = 160–200 ng/ml) or no efficacy (SUM159-R113-IC₅₀ = \>100 ng/ml), compared to the 9.2.27-PE38KDEL+ABT combinations was observed. Of the three ABT compounds tested, ABT-737 produced the most significant improvement in IC₅₀ values with both ITs (Tables). As such, 9.2.27-PE38KDEL+ABT-737 combination induced varying degrees of synergy among the different tumor cell lines: for breast cancer models, the 9.2.27-PE38KDEL+ABT-737 combination synergy was most robust in SUM159-R113 (IC₅₀ = 5.75 ng/ml), representing a 17.4-fold improvement of cytotoxicity compared to IT monotherapy. Within the melanomas, the 9.2.27-PE38KDEL+ABT-737 combination worked best in DM440 (IC₅₀ = 0.52 ng/ml), enhancing the IT monotherapy cytotoxicity by at least 192-fold. The most potent synergy among all 12 tumor cells tested was observed with the GBM xenograft D-10-0021 MG (IC₅₀ = 0.045 ng/ml), yielding a \>1,000-fold improvement of the IC_50. Furthermore, Combination Index analysis in D-10-0021 MG showed that even at the lowest IT concentration tested (0.1 pg/ml), the 9.2.27-PE38KDEL+ABT-737 combination demonstrated strong synergistic activity (CI = 0.135). Moderate synergy (CI = 0.622 to 0.719) was observed in DM440 cells with the 9.2.27-PE38KDEL+ABT-737 combination at concentrations of ≤0.1 ng/ml. For the SUM159-R113 cells, synergy with the combination therapy was observed only at concentrations of ≥10 ng/ml (CI = 0.0172). ## Basal expression of furin and selected Bcl-2 family proteins in GBM, melanoma, and breast cancer cells To mechanistically unravel the observed synergistic effects, the basal expression levels of furin, Bcl-2, Bcl-xL, and Mcl-1 were analyzed in our panel of GBM, melanoma, and breast cancer cells. Compared to both GBM and breast cancer cells, the melanoma cell lines DM440 and DM443 demonstrated higher expression of furin. Pro-survival Bcl-2 family proteins Bcl-2 and Bcl-xL are targets of both ABT-737 and ABT-263 while ABT-199 exhibits high specificity to Bcl-2. Expression of Bcl-2 was only detected in DM440. Varying levels of Bcl-xL expression was detected in all cell lines tested while higher expression levels were obvious in SUM159 and DM440 cell lines. Mcl-1 is a pro-survival Bcl-2 family protein with a short half-life. While Mcl-1 is not a direct target of any of the ABT compounds, its expression level exhausts rapidly upon treatment with agents that induce protein synthesis inhibition, such as the PE38 ITs. Mcl-1 expression was noted in all cell lines tested with higher levels seen in DM440, DM443, and D-10-0021 MG. ## Assessment of prosurvival and proapoptotic Bcl-2 family proteins following 9.2.27-PE38KDEL/ABT-737 mono or combination therapy The difference in the susceptibility of D-10-0021 MG, DM440, and SUM159-R113 cells to the 9.2.27-PE38KDEL, ABT combinations impelled us to examine the role of selected prosurvival (Bcl-xL and Mcl-1) and proapoptotic (Bim and Bax) Bcl-2 members in potentiating tumor cell death. Our analysis indicated no significant difference either in the basal or post-treatment levels of all four Bcl-2 family members in all three cell lines tested. A notable decrease in the levels of all four proteins corresponding with PARP cleavage and induction of apoptosis was observed in D-10-0021 MG, DM440, and SUM159-R113 cells. Thus, 9.2.27-PE38KDEL+ABT-737 combination therapy resulted in protein synthesis inhibition followed by degradation of Bcl-xL, Mcl-1, Bim, and Bax in all three cell lines thereby excluding the participation of these proteins in 9.2.27-PE38KDEL+ABT-737 mediated cell death. ## Internalization of the 9.2.27-PE38KDEL alone or in combination with ABT-737 Next we investigated the role of CSPG4 internalization to account for the difference in the sensitivity of tumor cells to the 9.2.27-PE38KDEL, ABT combinations. Flow cytometry analysis revealed that D-10-0021 MG, DM440, and SUM159-R113 cells internalized 95.5%, 74%, and 46% of the surface-bound 9.2.27-PE38KDEL by 4 h, respectively. 9.2.27-PE38KDEL internalization was the least efficient in Hs 578T cells, with only 11.5% of IT internalized at 4 h. Internalization of 9.2.27-PE38KDEL was examined by western analysis of D-10-0021 MG cells treated with 9.2.27-PE38KDEL, ABT-737, or the combination at different intervals. One hour post-treatment, 41% and 61% of intact IT was internalized with the 9.2.27-PE38KDEL and 9.2.27-PE38KDEL+ABT-737 therapies, respectively. The maximal level of intact IT internalization with either monotherapy or the ABT-737 combination was reached at 4 h post-treatment (the relative band intensity of which was arbitrarily set as 100% for comparison). Notably, when compared to IT monotherapy, ABT-737 addition did not significantly affect the maximum level of intracellular, intact IT at 2–4 h post-treatment. An antibody against the ADP-ribosylating catalytic domain of *Pseudomonas* exotoxin was used to determine intact IT internalization and the rate of intracellular IT cleavage by furin. The 38 kDa cleaved IT fragment, was detected 1 h post-treatment with 9.2.27-PE38KDEL alone or in combination with ABT-737. A gradual increase in the accumulation of the cleaved exotoxin was observed up to 6 h of treatment. However, only 9–12% of the maximal level of intact, internalized IT was cleaved after 4 h of treatment. Since the maximal level of intact 9.2.27-PE38KDEL was detected intracellularly at 4 h post-treatment, we extended the time course study to 24 h for examining the accumulation of intact/cleaved IT. The results confirmed that for D-10-0021 MG cells treated with either the IT monotherapy or the combination, the intracellular level of intact IT gradually declined after 4 h post treatment. Accumulation of cleaved exotoxin continued to increase till 8 h post-treatment, after which it began to decline, possibly due to proteasomal or/and caspase- dependent degradation. ## Inhibition of global translation and poly ADP ribose polymerase (PARP) cleavage by mono and combination therapies Since IT-mediated cytotoxicity is through protein synthesis inhibition followed by apoptosis, puromycylation assays were conducted to assess the rate of global protein synthesis inhibition upon IT treatment in D-10-0021 MG, DM440, and SUM159-R113 cells. At 8 h post-treatment of D-10-0021 MG, the 9.2.27-PE38KDEL/ABT-737 combination inhibited global translation by 69%, compared to 36% or 13% with 9.2.27-PE38KDEL and ABT-737 monotherapies, respectively. Similarly, DM440 (8 h post-treatment) and SUM159-R113 cells (16 h post-treatment), exhibited global protein synthesis reduction to a much greater extent in the combination therapy group compared to either of the monotherapies. Caspase-dependent PARP cleavage, which is often associated with apoptosis, was observed as early as 16 h post combination treatment in D-10-0021 MG, where the majority of intact PARP was cleaved. On the other hand, neither monotherapy was able to induce detectable PARP cleavage, even at 24 h post-treatment. Likewise, in both DM440 and SUM159-R113 cell lines, a significantly greater extent of PARP cleavage was observed in the combination therapy compared to the monotherapies. However, near-complete cleavage of PARP (around 90%) was only observed 24 h post combination therapy in DM440, and 30 h post combination therapy in SUM159-R113. Complete PARP cleavage with the 9.2.27-PE38KDEL/ABT-737 combination at earlier time points in the D-10-0021 MG GBM cells compared to DM440 and SUM159-R113 cell lines, corresponds to elevated *in vitro* cytotoxicity noted with the combination therapy in this model. Time course analysis of global translation inhibition and total intact PARP levels (100% intact PARP levels were observed at ≈70–80% translation inhibition) in D-10-0021 MG, DM440, and SUM159-R113 cells confirmed that inhibition of protein synthesis by processed IT resulted in PARP cleavage and apoptosis. ## Relative amount of cleaved exotoxin following the combination treatment of 9.2.27-PE38KDEL and ABT-737 Upon receptor binding and internalization, ITs are activated by furin cleavage. To determine IT activation following 9.2.27-PE38KDEL (750 ng/ml) and ABT-737 (10 μM) combination, relative amounts of the intact and cleaved exotoxin were analyzed in D-10-0021 MG, DM440, and SUM159-R113 cells. At 4 h post treatment, D-10-0021 MG demonstrated 2-fold and 7.5-fold more intact intracellular IT than DM440 and SUM159-R113, respectively. Cleaved exotoxin in D-10-0021 MG cells was detected at 2 h post-treatment and reached maximum levels at 8 h post-treatment. In contrast, significantly lower levels (5–20%) or no cleaved exotoxin was observed in DM440 and SUM159-R113, respectively. Again, at 16 h post combination therapy, near-complete PARP cleavage was detected in D-10-0021 MG cells, while there was minimal PARP cleavage in both DM440 and SUM159-R113 cell lines. Bcl-xL levels declined at the 16 h time point in D-10-0021 MG cells, which could be due to caspase-dependent degradation (in correlation with PARP cleavage). No significant change in Bcl-xL levels were observed in DM440 and SUM159-R113 cells. Total loss of Mcl-1 was noted at 16 h post combination therapy after the appearance of cleaved exotoxin in D-10-0021 MG cells. Since furin-processed exotoxin needs to translocate from the ER to cytosol to inhibit protein synthesis, the relative amounts of intact and cleaved exotoxin in the cytosol and ER fractions of D-10-0021 MG and DM440 cells were investigated. Cell lysates were fractionated into cytosolic, ER, and total fractions. Immunoblot analysis of marker protein distributions confirmed the presence of tubulin only in cytosolic fractions, and the ER-membrane protein Mcl-1 only in the ER fraction, thus establishing proper separation of the different cellular fractions. At 4 h post-treatment for both D-10-0021 MG and DM440 cells, intact ITs were predominantly located in the ER fraction. In D-10-0021 MG, cleaved exotoxins were detected in both ER and cytosolic fractions, whereas in DM440 cells, they were observed only in the ER fraction. Our data indicate that both internalization and furin-dependent cleavage of IT was ineffective for SUM159-R113 cell line. In DM440 cell line, internalization of IT was efficient, but cleavage and translocation of cleaved exotoxin from ER to cytosol was significantly lower than that of D-10-0021 MG cells. A minimal increase (≈1.6 fold) observed in the cleaved exotoxin in cytosolic fractions of D-10-0021 MG cells following 9.2.27-PE38KDEL+ABT-737 treatment, might contribute to the higher sensitivity of the GBM cells to combination therapy. Thus, in the GBM PDX D-10-0021 MG, rapid internalization (1–2 h) of IT, efficient cleavage (2–8 h) and translocation of cleaved (active) exotoxin from ER to cytosol, complete PARP cleavage, and Bcl-xL and Mcl-1 degradation resulted in robust response to 9.2.27-PE38KDEL+ABT-737 therapy. Taken together the mechanistic studies show that the levels of intracellular IT, processed exotoxin, and PARP cleavage determine the sensitivity of tumor cells to the combination treatment. Our study provides a rationale for the difference in the *in vitro* efficacy of GBM, melanoma, and breast cancer cells to the 9.2.27-PE38KDEL+ABT-737 combination. ## *In Vitro* analysis of 9.2.27-PE38KDEL+ABT-737 therapy on CSPG4 mediated signaling pathways Since CSPG4 activated signaling pathways are known to be involved in tumor cell growth, adhesion, migration, and chemoresistance we analyzed the *in vitro* effect of 9.2.27-PE38KDEL/ABT-737 mono or combination therapy on FAK, PKCα, and AKT pathways in D-10-0021 MG, DM440, and SUM159-R113 cells. An initial increase of p-FAK at earlier time points specifically in 9.2.27-PE38KDEL and 9.2.27-PE38KDEL+ABT-737 therapy groups in D-10-0021 MG and DM440 cells was followed by a decrease at later time points. However, these changes were not significant. The levels of p-FAK did not change after treatment in SUM159-R113 cell line. A significant drop in p-AKT was observed as early as 1 h in D-10-0021 MG GBM cells and by 4–8 h in DM440 and SUM159-R113 cell lines. We did not observe any change in the levels of p-PKCα or PKCα in D-10-0021 MG and SUM159-R113 cells while the levels were extremely low for detection in DM440 levels. Our data indicate no difference in the CSPG4 downstream signaling pathways in D-10-0021 MG, DM440, and SUM159-R113 cells post combination therapy. ## *In vivo* efficacy of 9.2.27-PE38KDEL and ABT-737 combination therapy Orthotopic mouse model of primary brain tumors were established using GBM patient-derived xenograft D-10-0021 MG cells. Histological analysis of brains five days post-tumor implantation demonstrated the presence of tumor mass in D-10-0021 MG model. Thus, day five post-implantation was chosen for treatment initiation. Convection-enhanced delivery (CED), utilizing osmotic pumps, has been successfully used to bypass the blood-brain barrier and to deliver ITs directly into brain tumors. The toxicity of 9.2.27-PE38KDEL was evaluated by CED at a total dose of 0.1 μg or 0.3 μg in mice bearing DM440 intracranial tumors. Toxicity events at an occurrence rate of 1–2 mice per every ten mice were observed only in the 0.3 μg 9.2.27-PE38KDEL group. A total dose of 100 μM (2.93 μg) of ABT-737 by CED was found to be safe for the mice without causing any toxicity. Therefore, a total dose of 0.1 μg of 9.2.27-PE38KDEL and 100 μM of ABT-737 were chosen for the *in vivo* combination studies. In the D-10-0021 MG model, orthotopic delivery of 100 μM of ABT-737 did not improve MS compared to the vehicle control (p = 0.2871). The 9.2.27-PE38KDEL (0.1 μg) monotherapy exhibited a modest 10.7% increase in MS compared to the control but failed to reach statistical significance (p = 0.0716). Compared to the vehicle control, the 9.2.27-PE38KDEL+ABT-737 combination therapy prolonged the MS by 60.7% (p = 0.0012). The survival benefit of the combination was also statistically significant when compared to the 9.2.27-PE38KDEL (p = 0.0100) or ABT-737 (p = 0.0198) monotherapies. More importantly, 2/9 (22.2%) mice in the combination therapy group were tumor-free, as confirmed by the H&E staining of their brains at the termination of the study (Day 81). Orthotopic mouse model of metastatic brain tumor was established using DM440 melanoma cells. Histological analysis of DM440 brains five days post-tumor implantation demonstrated the presence of tumor mass. In the DM440 model, neither 9.2.27-PE38KDEL (p = 0.1667) nor ABT-737 monotherapies improved MS (p = 0.1822). However, compared to the vehicle control, there was an 18.2% increase in MS for the combination group (p\<0.0001). The survival benefit with the 9.2.27-PE38KDEL+ABT-737 combination was also statistically significant compared to the ABT-737 (p = 0.0004) or the 9.2.27-PE38KDEL (p = 0.0002) monotherapies. # Discussion Brain tumors are composed of highly heterogeneous tumor cell populations that frequently harbor inherent resistance to therapeutic agents, resulting in incomplete eradication of tumor cells and tumor recurrence. Thus, sensitizing tumor cells is critical for improving the efficacy of IT-based therapies. By combining CSPG4 targeting ITs with ABT-737, targeting Bcl-2 family members, we were able to overcome widespread, inherent IT resistance in a panel of GBM, melanoma, and breast cancer cell lines. The IT+ABT-737 combination improved *in vitro* therapeutic efficacy up to \>1,000-fold (Tables). The ability of ABT-737 to sensitize cancer cells to 9.2.27-PE38KDEL-mediated cytotoxicity was mechanistically confirmed through western analysis, where the combination therapy, but not the monotherapies, enhanced exotoxin cleavage, induced global translation inhibition, and PARP cleavage. Accordingly, only the combination therapy generated significant improvement in MS, and in some cases even cures, in mouse models of primary or metastatic brain tumors (Figs). Our *in vitro* studies demonstrated variability in the sensitivity of individual cell lines/xenografts to 9.2.27-PE38KDEL+ABT-737 combination, i.e., cytotoxicity (IC₅₀) improvement of \>1000-fold (D-10-0021 MG), \>200-fold (DM440), and \>17-fold (SUM159-R113). Interestingly, while Hs 578T and D-10-0021 MG cells expressed similar levels of CSPG4, Hs 578T failed to show improvement in cytotoxicity with all of the combinations tested (Tables), demonstrating that the surface CSPG4 density is not a critical determinant of tumor cell sensitivity to combination therapy. Examination of selected prosurvival and proapoptotic Bcl-2 family proteins at basal level and following 9.2.27-PE38KDEL/ABT-737 mono or combination therapy failed to correlate directly with the sensitivity of tumor cells to the combination therapy ( and Figs). Our *in vitro* studies also excluded the participation of CSPG4 signaling pathways in sensitizing the tumor cells to the combination therapy. However, flow cytometry assay revealed that the internalization rate of 9.2.27-PE38KDEL is a major determinant of tumor cell sensitivity to combination therapy. Once internalized, ITs are processed by furin in the endosome and translocated from ER to the cytosol for ADP ribosylation of EF2 and protein synthesis inhibition. Western analysis revealed that at 8 h post combination treatment there was a 12-fold increase in cleaved exotoxin in D-10-0021 MG compared to DM440. Moreover, cell fractionation studies revealed that at 4 h post-treatment (9.2.27-PE38KDEL or 9.2.27-PE38KDEL+ABT-737), cleaved exotoxins were found in both cytosolic and ER fractions of D-10-0021 MG cells, while they were detected only in the ER fractions of DM440. Thus, increase in cytosolic levels of cleaved exotoxin in D-10-0021 MG compared to DM440 cells corresponds to their enhanced sensitivity to 9.2.27-PE38KDEL and 9.2.27-PE38KDEL+ABT-737 therapies. In D-10-0021 MG cells, compared to 9.2.27-PE38KDEL monotherapy there was ≈1.6 fold increase in cleaved exotoxin levels post 9.2.27-PE38KDEL+ABT-737 combination. Thus, in contrast to previous studies, our results in the GBM xenograft D-10-0021 MG demonstrate that the addition of ABT-737 does not significantly alter the translocation of cleaved exotoxin from ER to cytosol. However, we observed PARP cleavage in all three tumor models (D-10-0021 MG, DM440, and SUM159-R113) post combination therapy, but not with 9.2.27-PE38KDEL or ABT-737 monotherapies. The efficiency of PARP cleavage correlated well with *in vitro* and *in vivo* efficacy of the combination therapy. In the D-10-0021 MG model, compared to vehicle control, 9.2.27-PE38KDEL+ABT-737 combination therapy increased MS by 60.7% (p = 0.0012). The efficacy of combination therapy was highly significant when compared to ABT-737 (p = 0.0198) and 9.2.27-PE38KDEL monotherapies (p = 0.0100;). Furthermore, the combination therapy generated cures in 2/9 mice, as verified by histological staining. Consistent with the lower sensitivity of DM440 cell line to the combination therapy *in vitro*, a modest increase in survival post combination was observed *in vivo*. Collectively, these data confirmed that the 9.2.27-PE38KDEL+ABT-737 combination therapy was able to overcome tumor cell resistance to IT monotherapy *in vivo*, delayed tumor growth, and, in some cases, generated cures. While several groups have tried to gain mechanistic insights into various immunotoxin+Bcl-2 inhibitor anticancer therapies, only two studies have shown modest antitumor efficacy (improvement in survival by several days without cures), in *in vivo* subcutaneous small cell lung cancer and melanoma tumor models. To achieve this modest improvement in survival, the authors utilized a total of eight doses of 50 mg/kg of ABT-737 + 0.4 mg/kg immunotoxin or a total of two doses of 50 mg/kg of ABT-737 and 0.031 mg/kg immunotoxin. Importantly, in our current intracranial study, we were able to generate 22% cures and improvement in survival utilizing a single dose of 0.147 mg/kg of ABT-737 and 0.005 mg/kg immunotoxin against the aggressive glioblastoma tumor. The ABT-737 and immunotoxin doses in the current study are ≈680–2,700-fold and ≈12–6,400-fold lower, respectively than the previous studies. In conclusion, our *in vitro* studies, using a panel of GBM, melanoma, and breast cancer cell lines showed that Bcl-2 inhibitor ABT-737 reversed the resistance of tumor cells to IT treatment. We further confirmed the ability of 9.2.27-PE38KDEL+ABT-737 combination to overcome tumor resistance in orthotopic models of brain tumors. Mechanistic studies using 9.2.27-PE38KDEL and ABT-737 mono- and combination therapy revealed that that increased levels of intracellular IT, cleaved exotoxin, and PARP cleavage were determinants of enhanced sensitivity of tumor cells to the combination. Addition of ABT-737 had little effect on the rate of IT internalization but generated a small increase in the translocation of cleaved exotoxin from ER to cytosol. Our study provides insights into employing IT+ABT combinations for overcoming therapy resistance in primary and metastatic brain tumors. # Supporting information We thank Steven Clayton for his assistance in cell culture and processing the xenograft tumors. [^1]: The authors have declared that no competing interests exist.

# Introduction Ankylosing spondylitis (AS) is a chronic inflammatory disease characterized by inflammation in the sacroiliac joints and spine which causes joint and bone erosion and even ankylosis. Most AS patients develop first symptoms before they are 30 years old. Radiographic progression during the first 10 years of disease is an important prognostic indicator of disease severity; some recent studies suggest early presentation of structural damage can be considered as a good predictor of further damage. The severity of AS is largely genetically determined. Most genetic studies focus on disease susceptibility; however, the published literature lacks candidate gene association studies for disease severity of AS. The goal of this study is to examine several previously identified gene polymorphisms, and their influence on susceptibility to AS and severity of disease, all with the intention of earlier intervention leading to better outcome. The genes studied in the present report include *JARID1A, JMY*, and *PTGER4*. *JARID1A* (jumonji, AT-rich interactive domain 1A) is also known as *KDM5A* (lysine-specific demethylase 5A) and RBP2 (retinoblastoma-binding protein 2). *JARID1A* encodes the JARID1A protein and regulates gene expression involved in numerous cellular functions, including tumorigenesis. JARID1A increases H3K4me3 (tri-methylated histone H3 at Lysine 4) which is recognized by plant homeodomain (PHD), leading to the altered programs of gene expression and progression of tumor. The dysregulation of PHD finger has been implicated in a variety of human diseases including immune disorders and cancer. Additionally, JARID1A interacts with estrogen receptor alpha, and much published literature has studied the relationship between JARID1A and breast cancer. *JMY* encodes for JMY (junction-mediating and regulatory protein), This protein is a transcription co-factor. It was originally identified as a p300-binding protein, and it can augments the p53 tumor suppressor response. *PTGER4* encodes EP4R (prostaglandin E receptor 4) which is the antagonism to inhibit cell growth, proliferation, and metastasis of breast cancer cells; in particular, *PTGER4* regulates the aggressive phenotypes of inflammatory breast cancer cells. Thus, higher level of *PTGER4* and EP4R may lead tumor cell to proliferate. A recent study supports the hypothesis that PTGER4 receptor antagonists may be an alternative approach to prevent tumor metastasis. In genome-wide association studies, rs11062385 in *JARID1A*, rs16876657 in *JMY*, and rs10440635 in *PTGER4* are related to AS susceptibility in patients of western European descent. We hypothesize that the relationship between *JARID1A, JMY, PTGER4* and AS exists in other populations; additionally, particular single nucleotide polymorphisms (SNPs) of these genes may predict the severity of AS. This is a replication study, Chinese Han population is more than one billion. It is reasonable to support the western results using Chinese Han population. In this study, we examined *JARID1A, JMY*, and *PTGER4* genes in patients of the Chinese ethnic majority Han population. # Methods ## 1: Study population In this study, 396 AS patients and 404 unrelated healthy controls who are age and sex-matched are recruited. All AS patients and normal controls are Han Chinese. The Han ethnic group makes up 92% of the population in China and 20% of the global population, making it the largest ethnic group in the world. These samples are collected from PLA general hospital from 2010 to 2013. All AS patients are HLA-B27 positive and they are treated by non-steroidal anti- inflammatory drug routinely; no other treatments are used for patients. In the patient group, 354 male (89.4%) and 42 female (10.6%) are recruited; the average age is 29.6 years (range 16 to 60 years). In the control group, 370 male (91.6%) and 34 female (8.4%) are recruited; the average age is 30.0 years (range 16 to 60 years). Neither sex nor age distributions show significant differences between AS patients and control (p=0.291, 0.670 respectively). The average duration since AS diagnosis is 11.5 years (range 8 to 18 years). The diagnosis of AS has been made by experienced rheumatologists according to the modified New York criteria. The diagnosis was reconfirmed by different rheumatologists. The diagnosis was made before the genetic information was genotyped. In another word, the phenotype was recorded by the rheumatologists blinded to the genetic information. Subjects with inflammatory bowel disease, psoriasis, rheumatoid arthritis, or other autoimmune diseases are excluded from both the AS and the control group. ## 2: Basic data acquisition The Bath AS function index (BASFI) and Bath AS disease activity index (BASDAI) are administered to the patients using questionnaires; these indexes are the most widely used tools for the assessment of AS functional status and activity. The modified Stokes AS Spine Score (mSASSS) is a validated scoring system for spinal structural changes. The lateral views of standard radiographs of the cervical and lumbar spine are used to derive a mSASSS score for each patient. Three of the authors separately assigned the mSASSS scores, and we used the average. ## 3: Severity classification How to classify AS severity is still a field of discussion.. In this study, we define severe type of AS as the disease form in those patients within first ten years of diagnosis who satisfied the indications of surgery, which include inability to stand upright, inability to look straight ahead, or compression of the viscera due to kyphosis that manifests as pain. Patients with the normal type of AS exhibit inflammation of sacroiliac joints, but their spine and other joints were relatively spared; these patients have required only medical treatment. By this definition, 82 AS patients were the severe type, and 314 AS patients were the normal type. ## 4: SNPs selection The SNPs in this study included four in *JARID1A*, four in *JMY* and five in *PTGER4*. *JARID1A* is on chromosome 12, and *JMY* and *PTGER4* are both on chromosome 5. These SNPs are selected to serve as multi-marker tagging algorithm with criteria of r² more than 0.8 and for all SNPs with minor allele frequency more than 20%; population is set as CHB (Chinese Han Beijing). We use the data download from hapmap to select the tagSNPs randomly. Haploview 4.2 software (Broad Institute, Cambridge, Massachusetts, USA) is used to in this procedure. shows the positions of each tagSNP. ## 5: DNA extraction and genotyping analysis We use AxyPrep Blood Genomic DNA Miniprep kit (Axygen Biosciences, Union City, CA, USA) to isolate DNA from 2ml whole blood samples. We employ the chip-based matrix-assisted laser desorption ionization time-of-flight (MALDI-TOF) mass spectrometry technology to detect SNPs. MassARRAY system is used in this procedure (Sequenom, San Diego, CA, USA). All SNPs in the control groups are successfully genotyped; rs7134353 and rs2284336 in *JARID1A* and rs10440635 in *PTGER4* in AS groups are 100% genotyped. The other SNPs are genotyped in 394 of 396 patients. ## 6: Statistical analysis The Hardy-Weinberg equilibrium is tested for all 13 tagSNPs. We use chi-squared test and independent-samples t-test to compare the differences in age and gender between cases and controls. Comparisons of the distributions of the genotype, allele and haplotype frequencies are carried out using the chi-squared test. The relative risks are estimated as an odds ratio (OR) with a 95% confidence interval (CI). The p-values, OR, and 95% CI indicate in the text are used to estimate the significance of the contribution of corresponding genotype to disease risk. The p-values of genotypes indicate in the result tables are used to estimate the significance of the distribution of genotype between cases and controls. Different subgroups in cases are compared to controls separately. Bonferroni correction is needed. Due to the number of SNPs selected in each gene, p-value less than 0.01 is considered statistically significant after Bonferroni correction. The last genotype of each SNP is the major genotype and the last allele is the major allele. They are the reference groups. All three genotypes of each SNP are compared, p-value for individual genotypes are shown only if significant at 0.05 level (details are shown in,). We compared the severe AS group to the entirety of the control group and then normal AS group to the entirety control group. Pearson’s chi-squared test is used to compare the constructed haplotypes. The SNPs which show significant differences between AS patients and controls are considered to be related to susceptibility to AS. The SNPs which p-value are less than 0.01 both in severe AS groups and normal AS groups are considered related to severity of AS. Statistical analyses are carried out with SPSS v.17.0 software package (IBM, Armonk, New York, USA). All three genotypes of each SNP are compared, and p-values are shown in the first line of each SNP. Additionally we compared the first two genotypes with the third one, showing the p-values if only they are significant (p\<0.05). We compare the severe AS group to the entirety of the control group and then normal AS group to the entirety of the control group. ## 7: Ethics statement The blood samples of both AS patients and controls used in this study are part of samples taken for diagnostic tests. During the collection and use of DNA samples, clinical data guidelines, regulations of the local Ethics Committee and the Helsinki Declaration in 1975 are followed. Written informed consents were obtained from all the patients and subjects (or their parents in the case of two patients less than 18 years old). The study procedure is approved by our Institutional Review Board. The full name of the IRB is ethics committee of Chinese PLA general hospital. # Results ## 1: Clinical features Among the 396 AS patients, the mean BASFI is 3.94±1.45 (mean±standard deviation). The mean BASDAI is 5.60±1.25. The mean mSASSS is 13.7±15.0. When comparing the severe AS to the normal AS patient groups, there is no significant difference in sex (p=0.220), age (p=0.097), and duration since diagnosis (p= 0.290). The BASFI is higher in severe AS group (6.07±2.00) than normal AS group (3.38±0.37) (p-value\<0.001), reflecting poorer function of patients in the severe AS group. The BASDAI is similarly higher in the severe AS group (6.28±1.34) than normal AS group (5.42±1.16) (p-value\<0.001), reflecting higher disease activity. The pattern holds for mSASSS (36.4±20.7 versus 7.71±1.86, p-value\<0.001), signifying more radiographic changes in the severe AS patients. ## 2: Genotype and allele The statistically significant SNPs of these three genes related to susceptibility to AS and severity of AS are summarized in. The details of genotype and allele distributions for *JARID1A, JMY* and *PTGER4* are summarized in, and respectively. The genotype frequencies of these 13 tagSNPs are in Hardy- Weinberg equilibrium case groups and control groups. SNPs in *JARID1A* are compared between all AS patients, severe AS patients, and normal AS patients versus the control subjects. The rs7134353 SNP shows significant difference when comparing severe AS patients to controls, with AA genotype higher in severe AS than in controls (p=2.241×10^-4). The rs2284336 SNP shows significant difference when comparing all AS patients to controls, with CT genotype lower in all AS than in controls (p=1.858×10^-4); this SNP also shows significant difference when comparing severe AS patients to controls, with CT genotype lower in severe AS patients than in controls (p=1.217×10^-6), and T allele lower in severe AS than in controls (p=1.144×10^-5). The rs11062357 SNP shows significant difference when comparing severe AS patients to controls, with CC genotype higher in severe AS than in controls (p=1.888×10^-9) and C allele higher in severe AS than in controls (p=3.456×10^-9); this SNP also shows significant difference when comparing normal AS to controls, with CC genotype lower in normal AS than in controls (p=0.002). The SNPs in *JMY* are compared between all AS patients, severe AS patients, and normal AS patients versus the control subjects. The rs2607142 SNP shows significant difference when comparing severe AS patients to controls, with AG genotype lower in severe AS than in controls (p=1.809×10^-4) and A allele lower in severe AS than in controls (p=0.001); This SNP also show significant difference when comparing normal AS to controls, with AA genotype higher in normal AS than in controls (p=0.003) and A allele higher in normal AS than in controls (p=0.007). The rs16876619 SNP shows significant difference when comparing all AS patients to controls, with TT genotype higher in all AS than in controls (p=0.005); this SNP also shows significant difference when comparing severe AS patients to controls, with CT genotype lower in severe AS than in controls (p=0.005), T allele lower in severe AS than in controls(p=1.172×10^-16) ; and this SNP shows significant difference when comparing normal AS patients to controls, with TT genotype higher in normal AS than in controls (p=0.001). Additionally CT genotype is lower than TT genotype (p=0.001). The rs4704556 SNP shows significant difference when comparing severe AS patients to controls, with CC genotype higher in severe AS than in controls (p=5.844×10^-7), C allele is higher in severe AS than in controls (p=2.249×10^-7). The rs16876657 SNP shows significant difference when comparing all AS patients to controls, with AG genotype lower in all AS than in controls (p=0.009); this SNP also shows significant difference when comparing severe AS patients to controls, with AG genotype lower in severe AS than in controls (p=0.006), G allele lower in severe AS than in controls (p=0.009) The SNPs in *PTGER4* are compared between all AS patients, severe AS patients, and normal AS patients versus the control subjects. The rs10440635 SNP shows significant difference when comparing severe AS patients to controls, with AA genotype higher in severe AS than in controls (p=1.126×10^-6), A allele higher in severe AS than in controls (p=1.667×10^-6); this SNP also shows significant difference when comparing normal AS to controls, with AA genotype lower in normal AS than in controls (p=1.763×10^-5), A allele lower in normal AS than in controls (p=0.002). The rs4957341 SNP shows significant difference when comparing severe AS to controls, with AA genotype higher in severe AS than in controls (p=0.003). Referring to the data from Hapmap: in *JARID1A*, rs2284336 is in 100% LD with rs11062385; in *JMY*, rs16876657 is related to susceptibility of AS. These results can support the former researches in western descendent. *PTGER4* shows no association with susceptibility of AS in Han Chinese. The rs11062357 SNP in *JARID1A*, the rs2607142 SNP in *JMY* and rs10440635 in *PTGER4* are related to severity of AS. ## 3: Haplotype Linkage disequilibrium (LD) maps of the 13 tagSNPs of *JARID1A, JMY* and *PTGER4* comparing all AS patients, severe AS patients, and normal AS patients to controls subjects are shown in, and, respectively. These figures have only a little difference, only is shown in the text comparing all AS patients to controls. and are shown in the appendices comparing severe patients and normal patients to controls separately. Analyses of constructed haplotypes are shown in, and in appendices. When comparing all AS to controls, the rs4133101,rs4546432 and rs4383756 SNPs in *PTGER4* show significant difference, the haplotype CTT frequency is higher than controls (p=6.266×10^-8). When comparing severe AS to controls, the rs7134353 and rs4980880 SNPs in *JARID1A* show significant difference, the haplotype TT is lower than controls (p=4.136×10^-4). The rs16876619 and rs4704556 SNPs in *JMY* show significant difference, the haplotype CC is higher than controls (p=2.682×10-7); the haplotype CT is lower than controls (p=4.660×10^-5). When comparing normal AS to controls, the rs16876619, rs4704556 and rs16876657 SNPs in *JMY* shows marginal significant difference, the haplotype TTA is marginal significant higher than controls but cannot pass Bonferroni correction (p=0.028). In conclusion, *PTGER4* is related to susceptibility to AS; *JARID1A* and *JMY* are related to severity of AS. # Discussion Three genes studied include *JARID1A, JMY*, and *PTGER4*. *JARID1A* regulates gene expression and is involved in tumorigenesis; it has been best studied in association with breast cancer. *JMY* encodes a transcription co-factor that augments the p53 tumor suppressor response. *PTGER4* encodes a prostaglandin receptor, and its down-regulation halts certain cell proliferation. *JARID1A, JMY*, and *PTGER4* have been linked to AS in GWAS in patients of western European descents; we focus on particular SNPs of these genes in the Chinese Han population. In comparing 396 AS patients and 404 healthy controls, we find that Frequencies of different genotypes and alleles are analyzed among the different severity AS patients and the controls. The rs2284336 SNP in *JARID1A*, the rs16876619 and rs16876657 SNPs in *JMY* are associated with susceptibility of AS. The rs11062357 SNP in *JARID1A*, the rs2607142 SNP in *JMY* and rs10440635 in *PTGER4* are related to severity of AS. Haplotype analyses indicate *PTGER4* is related to susceptibility to AS; *JARID1A* and *JMY* are related to severity of AS. *JARID1A* is recently found to interact physically and functionally with the Polycomb complex. This protein can influence the differentiation of CD4+ T-cells. Other research supports that *JARID1A* plays an important role in regulation of immune cells such as CD56+ NK cells, CD8+ T cells, dendritic cells and CD34+ cells. In addition to susceptibility to AS, the rs2284336 SNP in *JARID1A* may have association with other autoimmune diseases. With its influence on p53, *JMY* can affects apoptosis during the DNA damage response. *PTGER4*-encoded EP4R signaling mediates ultraviolet induced immunosuppression through modulation of regulatory T cells and RANKL expression; furthermore, EP4R can restrict the survival of immature B cells. The exact mechanisms how JMY and *PTGER4-*encoded EP4R’s effects on the immune system can influence the AS disease processes remain to be elucidated. Histone demethylase JARID1A is found to be related to susceptibility to AS in the western descendent. This mechanism should be investigated in Chinese Han population. The severe AS patient’s subgroup has 82 patients. This may be a low power data. However, these patients are in the nature course of the disease with only non-steroidal anti-inflammatory drugs treatments and have severe deformity, which may be impossible to be found in the western countries due to their regular treatments. We are the first to divide the AS patients into subgroups due to severity. Severity is related to prognosis of AS which is important for patients and therapeutic method choice. In conclusion, in Chinese patients, *JARID1A*, *JMY* and *PTGER4* are related to susceptibility to AS; *JARID1A* and *JMY* are related to severity of AS. These findings may lead to full understanding of the genetic and molecular pathogenesis of AS. Of clinical relevance, the specific SNPs in these genes can be used to guide genetic analysis and counseling, medical and surgical treatment options, and ultimate prognosis. Further studies are needed to elucidate the molecular roles these genes play in AS. # Supporting Information The authors wish to thank all the patients and families that participated in this study, and all clinical doctors helped us in Chinese PLA general hospital. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: WC ZJL. Performed the experiments: WC ZJL CC. Analyzed the data: WC ZJL JYL YW. Contributed reagents/materials/analysis tools: WC CC. Wrote the manuscript: WC ZJL LLS YW. Constructed the haplotype: JYL.

# Introduction Attention Deficit Hyperactivity Disorder (ADHD) is a widespread disorder with approximately 4.8% of children in the United States receiving medication for ADHD symptoms. Despite its prevalence, the long-term impact of medications commonly used to treat ADHD in humans–*e*.*g*., amphetamines (Adderall) and methylphenidate (Ritalin)–on the developing brain is not well known, and many studies demonstrate a high potential for abuse. An improved understanding of the biological basis for ADHD is crucial for finding new targeted drugs and treatments that would alleviate the common symptoms of ADHD, such as hyperactivity and impulsivity, while minimizing unwanted side effects. Twin and adoption studies indicate that the disorder is highly heritable, with heritability estimates ranging from 70–90%, suggesting an outsized role for genetics in the etiology of ADHD. Several genes have been identified from genome wide association studies (GWAS), but together they account for only a small proportion of the heritability. Current molecular and candidate gene-association studies of ADHD have focused mainly on catecholamine systems, in part because of the hypothesized targets of amphetamine and methylphenidate on dopaminergic and noradrenergic systems. An alternate possibility is that catecholamine system dysregulation is a symptom of a different mechanistic process. Human MRI and fMRI studies have repeatedly identified striatum as displaying reduced gray matter volume and function in ADHD patients. This alteration is apparent in both dorsal (caudate-putamen) and ventral (nucleus accumbens) portions of the striatum when comparing morphology and function between ADHD and control subjects. As compared to the other implicated brain regions such as the cerebellum and anterior cingulate cortex, the striatum receives the vast majority of dopaminergic projections from ventral midbrain nuclei including the substantia nigra and ventral tegmental area, supporting the catecholamine hypothesis. However, the connection between the specific genes implicated in ADHD and the re-wiring of the striatum towards an ADHD-like phenotype remains poorly understood. For preclinical exploration and hypothesis generation, we developed a polygenic mouse model of ADHD using a selective breeding approach. Our model includes a High-Active line, selectively bred for 17 generations for increased home cage activity and a Control, unselected line, maintained over the same period. Features of the model relevant for exploring ADHD etiology include polygenicity of the ADHD-like phenotypes, hyperactivity in a habituated environment, motor impulsivity, and symptoms ameliorated with therapeutic doses of amphetamines. The goal of the present study was to document differences in gene expression between the High-Active and Control mice in the striatum and in response to amphetamines, in order to develop a more coherent hypothesis for how patterns of differentially expressed genes in the striatum might contribute to ADHD-like behavior. Recent reports have implicated genes such as latrophilin 3 (*Lphn3*) and fibronectin leucine rich transmembrane protein 3 (*Flrt3*), which are involved in synaptic strength and structural maintenance, in the pathophysiology of ADHD.. *Lphn3* is a member of the latrophilin subfamily of G-protein coupled receptors involved in adhesion between pre- and post-synaptic neuronal membranes. Interactions of Lphn3 with Flrt3 proteins are required for synapse development and function, specifically excitatory synapses. Therefore, we hypothesized that *Lphn3*, *Flrt3* and other related genes that are involved in the development and maintenance of synapses may be downregulated, and possibly lead to decreased striatal excitatory synaptic density. However, beyond the genes and pathways previously implicated in ADHD, we had no prior hypotheses about which subset of genes might display such genotype-by-amphetamine interactions. Given the paradoxical effect of therapeutic doses of amphetamines on the symptoms reported previously, and the dopamine pharmacology of amphetamines, we expected to find genes that were oppositely regulated by amphetamine in the High-Active vs. Control genotypes. # Materials and methods ## Experimental subjects Young adult male mice (average age 6.5 months ± 0.80 SD) from generation 17 (Experiment 1) and generation 22 (Experiment 2) of a long term selective breeding experiment for increased home cage activity were used. In brief, a starting population derived from a systematic inter-cross of 8 genetically variable inbred strains was used as the founder strain for two mouse lines: High-Active and Control. In each generation mice were placed in custom cages which allow video tracking from above. Total distance traveled over days 5 and 6 of a 6-day test was used as the selection criterion. In the High-Active line, each generation, the top male and female from within each family were randomly bred with the only rule that sibling breeding was not allowed. In the Control line, a male and a female was randomly chosen to represent each family in the subsequent generation with the same rule avoiding sibling breeding. In experiment 1, a total of n = 10 High-Active, and n = 10 Control mice, each from a different family, were used. In experiment 2, a total of n = 6 High-Active and n = 6 Control mice were used. # General husbandry Rooms were kept at a constant temperature (21 ± 1 °C) and in a reversed 12:12 light-dark cycle with lights on at 7:30 pm and off at 7:30 am. Food and water were provided ad libitum (Harlan Teklad 7012). Corncob bedding (Harland Teklad 7097, Madison, Wisconsin, USA) was provided in all cages. All procedures were approved by the University of Illinois Institutional Animal Care and Use Committee and adhered to NIH guidelines. The Beckman Institute Animal Facility, where the experiments took place, is AAALAC approved. ## Experiment 1: RNA-seq Mice were phenotyped as per previous work for home cage activity when they were approximately 3 months old. At approximately 6.5 months of age, all the mice were re-phenotyped for home cage activity using the same procedure, 6 days of continuous video-tracking. On the 7^th day, 2 hours after the lights shut off in the animal facility, mice received an intraperitoneal (i.p.) injection (10 ml/kg) of either vehicle (0.9% saline) or amphetamine (0.25 mg/kg d-amphetamine sulfate in 0.9% saline, catalog number A-5880, Sigma-Aldrich, St. Louis, MO), and returned to their home cage for video tracking. Mice were euthanized exactly 2 hours following the injection by live decapitation. Acute response to single dose was used instead of chronic dosing, because effects of amphetamine on ameliorating symptoms of ADHD are not thought to result from neuroadaptations from chronic use, but rather direct acute psychoactive effects. ### Striatum dissection Following, all surfaces and instruments were cleaned thoroughly with RNAse away. Brains were rapidly dissected and immediately placed on an aluminum platform on wet ice with the ventral surface exposed. Using the olfactory tubercles and optic chiasm as landmarks, a razor blade was used to cut a coronal section approximately 1.7 mm thick containing the striatum. The section was then flipped onto its caudal aspect and then the entire bilateral striatum (dorsal and ventral) was carefully dissected from the cortex, medial septum, and olfactory tubercles. The striatum was then immediately placed in a centrifuge tube on dry ice and then stored at -80°C. The entire brain extraction and dissection process occurred over approximately 2 minutes. ### RNA extraction and purification Bilateral striatal tissue was homogenized with an RNase-free disposable pellet pestle (Thermo Fisher Scientific, Waltham, MA) and RNA was extracted using the commercially available RNeasy^® Lipid Tissue Mini Kit (Qiagen Inc., Valencia, CA). Purification of the isolated RNA included treatment with DNase I (Qiagen Inc, Valencia, CA), accordingly to the manufacturer’s instructions. For assessing total RNA yield, aliquot samples were measured with the Qubit^® 2.0 (Life Technologies). Each sample showed a 260/280 ratio over 2.0 and yielded over 14 μg of RNA. Quality and integrity of isolated RNA samples were determined by 28S/18S rRNA analysis with the Agilent 2100 Bioanalyzer (Santa Clara, CA). All samples scored an RNA Integrity Number (RIN) over 8, indicating no signs of degradation. Quality was spot checked by choosing random samples to run on a gel, which confirmed the Bioanalyzer report. ### RNA sequencing RNA-sequencing was performed by the Roy J. Carver Biotechnology Center. RNA-Seq libraries were prepared using the TruSeq Stranded RNA Sample Prep kit with an average fragment length of 200 bp. Libraries were quantified using qPCR, pooled in a pool of 20, and multiplexed across a total of 5 lanes. The libraries were sequenced on an Illumina HiSeq 2500 for 101 cycles in 100 bp paired-end format using a TruSeq SBS sequencing reagent kit, v. 4. Read depth ranged between 41,684,484 reads and 70,449,689 reads with a median read depth of 54,516,050 reads. FASTQ files were generated from the raw sequencing runs using CASAVA v. 1.8.2. Reads in FASTQ format were aligned to the Mus musculus Ensembl GRCm38 genome with Ensembl GRCm38.75 annotation using TopHat2 v. 2.0.10 22. Since downstream applications used a counts-based method, we counted reads mapping to genes in each sample using htseq-count from the HTSeq Python framework, v. 0.6.1 24. Raw RNA-seq data files have been deposited in the Gene Expression Omnibus (<https://www.ncbi.nlm.nih.gov/geo/>) under accession GSE116752. ### Statistical analysis **Behavior:** SAS version 9.2 was used. Total distance traveled in the home cage on days 5 and 6 of the 6 day test during re-phenotyping was analyzed using a t-test. Distance traveled within the 2-hour period following the saline or amphetamine injection was analyzed using analysis of covariance, with the distance traveled on days 5 and 6 entered as a covariate to account for individual variation in activity levels, followed by line (High-Active or Control), and treatment (saline or amphetamine) entered as factors, and their interaction. The following criteria was used to establish normality of residual distributions: skewness between -1 and 1 and kurtosis between -2 and 2, and visual inspection of the histogram. **Differentially expressed genes:** Differential expression analysis used the edgeR software v. 3.8.5 for R v. 3.1.2. Briefly, we filtered out any genes whose expression was not greater than 1 count per million (CPM) in at least 3 samples. The expression matrix was TMM-normalized. Using the GLM functionality of edgeR, we fitted a model corresponding to two-way ANOVA where genotype (High-Active vs. Control), treatment (saline vs. amphetamine), and their interaction were entered as factors. Dispersion was estimated using the edgeR robust methodology, which uses an admixture of the tagwise and trended dispersion estimates while calculating observational weights to be used in downstream modeling. False discovery rates (FDRs) were calculated using Benjamini-Hochberg method, and genes at FDR \<0.10 were considered differentially expressed. ### DAVID enrichment analysis To test for overrepresentation of biological systems in our DEGs we used the NIAID bioinformatics annotation tool DAVID v. 6.8. The lists of upregulated and downregulated DEGs for each of the main effects and the interaction were queried. DAVID functional annotation clustering was performed using default annotation sources and a medium stringency. We used an enrichment score cutoff of \>1.3 as our criterion for significant enrichment as it approximately corresponds to a p-value of \<0.05. ### Weighted gene co-expression network analysis (WGCNA) Modules of correlated genes were identified using the WGCNA software v. 1.34 in R v. 3.0.2. Prior to coexpression analysis, we log₂-transformed data and filtered all genes displaying zero variance. Transformation was performed using the voom+limma function in limma v. 3.22.4. We then performed WGCNA on the transformed dataset using Pearson correlation coefficients with a soft threshold power of 8, which was determined because it had good scale-free topology (R²\>0.8), with good median connectivity (median k = 369.6), and relatively large module sizes, as done previously. We ran WGCNA in signed mode using Pearson’s r for the correlation function with minimum module size set to 30, the deepSplit parameter for the cutreeDynamic function set to 2 and the mergeCutHeight parameter for the mergeCloseModules function set to 0.15. Module eigengene values were calculated for each of these modules and a two-way ANOVA parallel to the one used for the DEG analysis was used to analyze eigengene relationships genotype, amphetamine treatment, and their interaction. ## Experiment 2: Immunohistochemical detection of synaptophysin Stereological analysis of tissue stained for synaptophysin has been used a reliable measure of synaptic numbers within the central nervous system. When mice were approximately 3 months old, they were euthanized by transcardial perfusion with ice-cold saline, followed by chilled 4% paraformaldehyde. Brains were removed, postfixed overnight in 4% paraformaldehyde at 4°C, and then transferred to 30% sucrose solution in phosphate buffered solution (PBS) until sectioning. Brains were sectioned using a cryostat into 40 micron sections which were stored in cryoprotectant in 24-well plates and stored at −20°C. ### Immunohistochemistry A 1-in-6 series of sections throughout the entire rostro-caudal extent of the striatum was stained for synaptophysin using diaminobenzidine (DAB) as the chromogen to estimate relative number of synapses. All sections from all animals were treated at the same time in the same reagents using custom made immunohistochemistry trays where one tray is filled with reagents and all wells are placed in the tray together for consistent staining. Free floating sections were washed in phosphate-buffering solution (PBS) and then treated with 0.6% hydrogen peroxide in PBS for 20 min. Sections were then blocked with a solution of 0.02% Triton-X and 6% normal goat serum (NGS) in PBS (PBS-X) for 1 h, followed by incubation in primary antibody, a rabbit anti-synaptophysin (sc-17750; Santa Cruz Biotechnology, Santa Cruz, CA) at a dilution of 1:5000 in PBS-X mixed with 3% (NGS) for 24 h at 4 °C. Sections were then washed in a mixture of PBS-X and 3% NGS, then incubated in secondary antibody against rabbit made in goat (sc-2030; Santa Cruz Biotechnology, Santa Cruz, CA) at 1:200 dilution in PBS-X for 90 min at room temperature. After a wash in PBS-X without NGS sections were then treated with the ABC system (Vector Laboratories, Burlingame, CA) for 1 h, followed by a wash in a PBS-X without NGS wash buffer. Sections were then stained using a DAB kit (Sigma, St. Louis, MO). To confirm the specificity of the primary antibody, a small number of sections were processed the same way as above except omitting the primary antibody step. No immuno-labeling was detected in these negative controls. ### Stereology **Volume estimation:** The two dimensional area of the striatum was outlined in each of the stained sections using StereoInvestigator software. The thickness of the mounted sections was determined by measuring the depth of the focal length from the top to the bottom of the tissue using a motorized stage. Total volume of the striatum was estimated as the product of the total area of the sections by the thickness of the sections by the distances between sections. **Density of synaptophysin particle estimation:** Density of labeled synaptophysin particles within the striatum was determined using the optical disector in Stereoinvestigator. Two sections per animal corresponding to 21 and 38 from the mouse brain atlas were analyzed the following way. A grid (125 x 125 μm) was placed across the sections to orient a counting frame (4 x 4 μm) in the corner of each grid square. The counting frame perimeter is made up of two “acceptance” edges and two “exclusion” edges. If a synaptophysin bouton falls entirely within the counting frame or it touches the “acceptance” edge, it is counted. All particles that contact the “exclusion” edge were omitted. The number of counted particles was divided by the total volume of the counting sites to obtain the estimate of synaptic density. Density was then multiplied by the previously calculated volume of the striatum to determine the total number of synaptophysin immunoreactive particles. ### Statistical analysis SAS version 9.2 was used. Total volume of the striatum, density of synaptophysin particles, and total number of particles (density multiplied by volume) were analyzed using un-paired t-test between the two groups, High-Active versus Control. The same criteria as described above was used to establish normality of residual distributions. # Results ## Mouse behavior As predicted, mice from the High-Active line traveled approximately 4-fold greater distances than Controls during re-phenotyping when they were 6.5 months of age, at the time when the striatal transcriptome was analyzed (t₁₈ = 3.8, p = 0.001). High-Active mice traveled an average of 1.2 km/day ± 0.24 SEM, whereas Control mice traveled an average of 0.3 ± 0.03 in their cages. Analysis of distances traveled within the 2 hours following injections of either saline or 0.25 mg/kg amphetamines, indicated a significant effect of the baseline activity covariate (F_1,12 = 139.6, p\<0.0001) and significant effect of line after correcting for baseline activity (F_1,12 = 11.8, p = 0.0049). Consistent with our previous reports, amphetamines tended to reduce activity in High-Active mice by 40%, while increasing activity in Control mice by approximately 2-fold, though the interaction was not statistically significant (F_1,12 = 2.5, p = 0.14) given the small sample size of n = 5 per group intended for RNA-seq comparisons. ## Differentially expressed genes A total of 14,472 genes passed the threshold of 1 CPM in at least 3 samples for sufficient expression for statistical comparisons between groups. shows ANOVA statistics and WGCNA module membership for these genes. shows ANOVA statistics for the WGCNA modules. ### Genotype effect A histogram of the p-values comparing High-Active versus Controls shows a concentration of p-values below 0.05, suggesting that the striatum went through large scale molecular and/or physiological changes as a result of selection and/or genetic drift. A total of 262 genes were differentially expressed between Control and High-Active lines at FDR \< 0.10 (p\< 7.6 x 10⁻⁴), of which 141 genes were upregulated and 121 were downregulated in the High-Active relative to Control line. The top differentially expressed upregulated gene (p\<3.1 x 10⁻³³) was *Gm6180*, which is a non-muscle pseudogene for n-cofilin (<https://rgd.mcw.edu/rgdweb/report/gene/main.html?id=735417>). Expression of the *Gm6180* transcript was low in Control mice and highly expressed in High-Active mice, with an approximately 20-fold difference. These results suggest that pseudogene *GM6180* may participate in the downregulation of its parent gene *Cfl1*. Although, *Cfl1* was not significantly differentially expressed by the FDR \<0.10 cut off, it was significantly reduced in High-Active relative to Controls by t-test at p\<0.001 level. Latrophilin 3 (*Lphn3*) was significantly downregulated (p\< 0.0004) in High- Active relative to Control. The ligand of Lphn3, fibronectin leucine rich transmembrane protein 3 (*Flrt3*) was not significantly differentially expressed at FDR\<0.10, but was significantly reduced in High Active relative to Control by t-test at p\<0.04 level. The trans-synaptic interaction between Lphn3 and Flrt3 is critical in modulating synaptic strength and density; decreased levels of *Lphn3* and *Flrt3* suggest that synaptic strength and density is decreased in High-Active striatum. Four canonical Wnt signaling-associated genes—dihydropyrimidinase-like 5 (*Dpysl5*), Ras association domain family member 8 (*Rassf8*), β-catenin interacting protein 1 (*Ctnnbip1*), and angiomotin-like 2 (*Amotl2*) were upregulated (all p\<0.0002) in High-Active striatum relative to Control. The protein products of all four genes are known to interact with glycogen synthase kinase 3β (*Gsk-3β)* and β-catenin (*Ctnnb1)* thus modulating the canonical Wnt signaling pathway. Another Wnt signaling gene, transmembrane protein 44 (*Tmem44*) was also significantly upregulated in High-Active mice (p\<1.4 x 10⁻¹²) (data not shown). Expression of canonical Wnt interacting partners, *Ctnnb1* and *Gsk-3β* were similar between the lines. ### Amphetamine and genotype-by-amphetamine interactions A histogram of the p-values for the main effect of amphetamines and interaction of genotype-by-amphetamines were uniform, suggesting subtle, if any variation in gene expression was attributed to these factors. For the main effect of amphetamine, 53 genes passed the FDR \< 0.10 threshold; 22 were upregulated, and 31 were downregulated. 17 genes showed a differential response to amphetamine depending on the genotype (*i*.*e*., genotype-by amphetamine interaction) at FDR \< 0.10. Of the 17 interacting genes, glyoxalase 1 (*Glo1; p\<1*.*6 x 10*^*−5*) is of particular interest due to recent association with anxiety-like behaviors in mice. Expression of *Glo1* was significantly higher in High-Active than Controls under both conditions, but amphetamines tended to reduce *Glo1* expression in High-Active whereas it increased expression in Controls. ## DAVID enrichment analysis Among the 121 genes significantly downregulated in the High-Active line, functional categories related to lipid biosynthesis, oxidoreductase activity, and PPAR signaling were significantly enriched with scores of 1.67, 1.59, and 1.59 respectively. Among the 141 genes significantly upregulated in the High- Active line, categories associated with ATP-grasp fold and prenylation were significantly enriched with scores of 1.94 and 1.43, respectively. ## WGCNA Following Saul et al. the analysis of all genomes in the genome annotation which had non-zero variance (26,114) in WGCNA yielded 49 modules with a maximum module size of 2536 genes and a minimum module size of 84 genes. WGCNA was unable to cluster 2463 genes, which were clustered into module 0. Three modules (numbered 8, 14, and 44) showed significant differences between the genotypes at FDR\<0.10 (p\<0.002). Module 8 included *Flrt3* and other genes overrepresented for cytoskeletal proteins (DAVID enrichment score: 2.42), mitochondrial activity (DAVID enrichment score: 4.05), DNA damage repair (DAVID enrichment score: 3.13). Module 14 included genes for overrepresented ontologies such as pyridoxal phosphate activity (DAVID enrichment score: 1.80) and proteolysis (DAVID enrichment score: 1.65). Module 44 included genes overrepresented for actin- myosin complex formation (DAVID enrichment score: 1.95), Golgi apparatus (DAVID enrichment score: 1.76), and ATP binding (DAVID enrichment score: 1.40). None of the modules displayed significant main effect or interaction with amphetamines. Results of the DAVID analysis on the three significant modules are shown in. ## Immunohistochemical detection of synaptophysin The density of synaptophysin positive particles was significantly reduced in High-Active relative to Control mice, by approximately 44% (t₁₀ = 8.8, p\<0.0001;). The volume of the striatum did not differ and was estimated to be 15.1 cubic mm (± 1.11 SE) and 14.5 cubic mm (± 0.97 SE), in Control and High-Active mice, respectively. The total number of synaptophysin particles (the product of density and volume) was also significantly reduced in High-Active relative to Control by approximately 46% (t₁₀ = 9.8, p\<0.0001). In High-Active mice the total number of synaptophysin particles was estimated to be 4.0 x 10⁸ (± 0.11 SE) particles per cubic mm, whereas in Control mice the estimate was 7.4 x 10⁸ (± 0.18 SE). # Discussion The main finding of the study is the robust molecular signature of ADHD-like behavior in the striatum. Rather than the usual suspects for the catecholamine hypothesis of ADHD, such as monoamine transporters and receptors, results identify new candidate genes and confirm others. Three converging pathways, synaptic actin regulation, Latrophilin 3, and canonical Wnt all suggested a downregulation in maintenance and regulation of excitatory synapses. This molecular prediction was confirmed by immunohistochemical detection of synaptophysin, a synaptic vesicle protein marker widely used as a marker of synapses. High active mice displayed approximately 46% reduction in number of synapses compared to Control. Taken together with recent findings from animal models and human GWAS studies, results support the hypothesis that ADHD-like behavior is associated with changes in multiple genes and pathways that may ultimately reduce or impair excitatory synapses in the striatum. ## Synaptic actin regulation Pseudogenes often negatively regulate their parent genes. Therefore, it is possible that the upregulation of *Gm6180* exerted a negative influence on *Cfl1* that resulted in *Cfl1* being downregulated in High-Active mice. However, this hypothesis requires additional molecular work before it can be confirmed. A role for *Cfl1* in synaptic remodeling of the striatum and predisposition for ADHD-like behavior was recently demonstrated in mutant mice with *Cfl1* and actin-depolymerizing factor (*ADF)* knocked out specifically in the striatum. *Cfl1* was deleted in the striatum by controlling expression through a CaMKII- Cre transgene (*ADF*-/*Cfl1*- mice). These mice displayed abnormal synaptic structure and decreased excitatory synaptic density. In parallel with the synaptic alterations, the mice displayed ADHD-like symptoms of hyperactivity and impulsivity which were ameliorated with methylphenidate. The *ADF*-/*Cfl1*- mice and our High-Active mice share remarkable convergence in striatal *Cfl1* expression and behavioral phenotypes related to ADHD. If the *Gm6180* interaction with *Cfl1* is indeed true, then *Gm6180*-like interactions may represent novel targets for modulating synapse strength in the striatum with specificity for ameliorating ADHD-like symptoms. ## Latrophilin 3 Decreased *Lphn3* and *Flrt3* expression in High-Active versus Control striatum provides yet further evidence in support of the hypothesis that reduced excitatory synapses in the striatum predispose ADHD-like behavior in the model. The trans-synaptic interaction between the protein products of *Lphn3* and *Flrt3* is of particular relevance due to the fact that genetic variants in *Lphn3* have been associated with ADHD in GWAS studies of large human cohorts. In particular, SNP marker rs6551665 is associated both with susceptibility to ADHD and with response to stimulant medication such as amphetamine and methylphenidate. Further, *Lphn3* null mutant mice display ADHD-like phenotypes such as hyperactivity, differential sensitivity to cocaine, and alterations in whole-brain levels of dopamine and serotonin-related genes and amino acid levels. Our results support our hypothesis that downregulation of *Lphn3* along with the endogenous ligand, *Flrt3*, contribute to fewer or abnormal excitatory synapses in the striatum, which contributes to the ADHD-like symptomology. ## Canonical Wnt pathway Of the 262 genes differentially expressed between lines, 5 are associated with the canonical Wnt pathway. Three of these genes, *Ctnnbip1*, *Amotl2*, and *Rassf8* encode proteins that inhibit the function of the Ctnnb1 protein, suggesting that the canonical Wnt pathway could be disrupted in the striatum of High-Active mice. *Dpysl5* (also known as *Crmp5*) and *Tmem44* were also upregulated in High-Active mice and have known interactions with the canonical Wnt pathway, though the functional directionality of the interactions is not as clear. The over-representation of canonical Wnt pathway genes that are significantly differentially expressed in a manner consistent with downregulation of the Wnt pathway suggest that modulation of Wnt signaling in the striatum is central to the High-Active phenotype. Recent evidence has shown that variation in Wnt signaling is a hallmark of other developmental psychiatric disorders, particularly schizophrenia, bipolar disorder, and autism spectrum disorder. Specifically, key Wnt signaling downregulators have been associated with populations with schizophrenia and bipolar disorder. Drugs that are used to treat both disorders are also known to raise Ctnnb1, a Wnt signaling biomarker, in the brains of adult rats. Amphetamine doses that are known to induce psychosis also decrease levels of Ctnnb1, but the effect of therapeutic doses of amphetamines on Wnt signaling in an ADHD-like brain, to the best of our knowledge has not been explored. The canonical Wnt pathway is crucially involved in synapse formation during brain development and maintenance through cell-to-cell adhesion processes, neuronal formation, and proliferation/migration. In the adult brain, Wnt signaling functions in synaptic cytoskeleton stabilization. Downregulation of Wnt signaling could have many implications, ranging from potential decreases in synaptic density, abnormal synapse formation, and/or impaired cell-to-cell adhesion. *Dpysl5*, a member of the Wnt pathway upregulated in our results, is specifically implicated in controlling neurite outgrowth. Coincidentally, altered actin dynamics due to decreased levels of Cfl1 and trans-synaptic binding molecules described above were predicted to cause similar effects on the synapse. Taken together, these results suggest multiple molecular pathways are at play in the High-Active mouse striatum which converge on synapse formation and maintenance. ## Glyoxalase 1 The vehicle-treated High-Active mice displayed approximately three-fold higher expression of glyoxylase1 (*Glo1)* relative to Control, while amphetamine reduced expression. This response was opposite to Control mice, where amphetamine increased expression. As reviewed by Distler & Palmer (2012), *Glo1* is implicated in anxiety-related behavior in mice. GLO1 is an enzyme involved in cellular digestion of glucose, metabolizing and reducing cellular levels of the byproduct methylglyoxal (MG), primarily in astrocytes as opposed to neurons. MG is a GABA-A receptor agonist and infusion into mice reduces anxiety-related behaviors. This is interesting because the majority of striatal neurons are GABAergic projecting medium spiny neurons. It is possible that the high baseline levels of *Glo1* in the High Active line cause low MG levels which reduce GABAergic signaling in the striatum, but how this causes anxiety, and how an anxiety phenotype might be related to increased physical activity and impulsivity in our lines remains to be determined. ## DAVID The most significantly enriched category among genes downregulated in High- Active mice involves lipid biosynthesis. Most of the brain is composed of lipids, and lipids serve many different functions in the brain, and so it is difficult to interpret this generic result. Low serum levels of fatty acids have been observed in populations with ADHD regardless of diet. It is possible that lipid synthesis dysregulation occurs in the brain as well as other organs but connections to ADHD symptomology remains unclear. The peroxisome proliferator-activated receptor (PPAR) pathway is associated with lipid metabolism, and is also significantly enriched in genes downregulated in High-Active mice. PPARγ inhibits Gsk3β, and Gsk3β inhibits Ctnnb1. Hence, decreased PPARγ inhibition of Gsk3β in High-Active striatum would result in greater inhibition of Ctnnb1, consistent with the hypothesized decrease in Wnt signaling in High-Active striatum. ATP-grasp fold enzymes and prenylation associated genes are enriched in the upregulated genes in High-Active mice. ATP-grasp fold enzymes are involved in fatty acid synthesis. We speculate that ATP-grasp fold enzymes and related pathways may be upregulated in High-Active mice as a compensatory mechanism to counterbalance decreased lipid metabolism and biosynthesis. Post-translational prenylation plays a role in protein localization and interactions, and thus could be involved in a variety of processes that have not been elucidated in this model yet. ## WGCNA Module 8 was most significantly different between the lines, and over-represents genes associated with cytoskeleton proteins, including *Flrt3*. The implication is that in addition to *Gm6180-Cfl1* and *Lphn3*-*Flrt3*, there are many other process that likely contribute to altered cellular and synaptic structure in the High-Active striatum. Module 14 contains *Tmem44 and Glo1* as well as other genes associated with pyridoxal phosphate activity and proteolysis. We know of no current consensus about the involvement of pyridoxal phosphate and proteolysis in ADHD or other developmental disorders. Module 44 over-represents genes associated with the actin-myosin complex and ATP binding. We would expect genes associated with other processes involving actin, such as the actin-myosin complex, to be dysregulated, in line with our synaptic actin hypothesis. The overrepresentation of genes associated with ATP binding also relates to the DAVID result involving ATP-grasp fold genes being upregulated in High-Active mice. ## Reduced synaptophysin labeling in High-Active mice The molecular prediction that synapse number would be reduced in the striatum of High-Active mice was confirmed in a separate set of animals processed for immunohistochemical detection of synaptophysin. The difference was striking, strongly supporting the hypothesis that the molecular changes which promote hyperactivity and impulsivity in the High-Active line do so, in part, by decreasing the number of synapses in the striatum. All the sections from all the animals were treated using the same reagents at the same time, thus the large difference in staining is a real biological difference rather than a batch effect from the immunohistochemistry. It is important to point out that synaptophysin gene expression was similar between High-Active versus Control in the striatum, suggesting that synapses were reduced not by decreasing their rate of formation but rather by altering their regulation, maintenance and stability. However, as synaptophysin does not distinguish between excitatory or inhibitory synapses, further work is needed to identify the specificity in the types of synapses that are reduced in High-Active mice. Previous work done by Zimmerman et al. showed a 16% decrease in excitatory synapses in mice engineered to display reduced Cfl1 in the striatum, and the High-Active mice displayed reduced Cfl1. Human ADHD has been associated with genetic variants in synapse-related genes including synaptophysin. The genetic association between *Lphn3* and ADHD in human populations has been suggested to result from the allele’s effects on synapse stabilization. Lphn3 is thought to play a role cell adhesion and trans- synaptic scaffolding and *Lphn3* was reduced in the High-Active mice. Reduced excitatory synapses is consistent with reduced gray matter and hypo-activation of the striatum observed in ADHD. Future work with the High-Active mice could reveal molecular pathways connecting genetic association between synapse development-related genes such as *Lphn3*, synapse maintenance and ADHD-like behavior. ## Limitations RNA was extracted from a large portion of the dorsal and ventral striatum which includes a mixture of multiple different cell types (e.g., neurons, glial cells and endothelial cells). Hence, differences in gene expression between groups in our study could be driven by changes occurring in one or multiple of these cell types, as we did not differentiate the pathways at the cellular level. The converging evidence points to synapse maintenance and regulation, so it is likely that neurons and astrocytes were the main cell types influencing the gene expression results emphasized. In addition, because we used both dorsal and ventral striatum, we could have missed genes significantly expressed in the dorsal striatum but not the ventral and vice versa, by homogenizing these two distinct regions. Future analyses of samples containing isolated cell types is required for confirmation. The gene expression differences between the High-Active and Control lines in our study could be due to selection for the ADHD phenotype, but could also be due to random genetic drift. The two lines were reproductively isolated for 17 generations, with a relatively small effective population size (approximately 10 families per line). One way to remove this confound in future studies is to develop and analyze additional replicate lines of mice. It is alternatively possible to estimate rates of genetic drift without replicate lines if certain assumptions are made about heritability, effective population size, and inbreeding coefficients. Because the High-Active line was generated through selective breeding, it can safely be assumed that the observed differences in behavior and gene expression relative to the Control line can ultimately be attributed to specific allelic differences that occur at multiple places in the genome between the lines. However, no QTL or gene mapping data are available for these lines, and the RNA sequence data obtained from this dataset is not adequate for statistically establishing genetic associations. In the present study, some of the DEGs between the lines were transcription factors (*e*.*g*., *Preb*, *Nkx6-2*, *Hopx*), which presumably contributed to orchestrating the gene expression signatures of the lines. Future work employing such techniques as chromosome immuno-precipitation sequencing (ChIP-seq) could identify DNA sequence variation at the site of interaction between the DNA binding molecules and the DNA, that ultimately could explain the gene expression regulation. ## Conclusions Results establish substantial molecular reorganization of the striatum in response to 17 generations of selection for hyperactivity. Three pathways were identified that together converge on a story of abnormal synaptic structure and maintenance. As of yet, no known interactions between the three main identified pathways, canonical Wnt, actin depolymerization, and *Lphn3* have been established. These may represent independent mechanisms or complement and support each other. The mechanism underlying the negative association between *Gm6180* and *Cfl1* expression could provide a useful molecular target if it proves capable of altering synapses in a subtle and specific way in the striatum to predispose ADHD. In addition, DAVID analysis suggests a broader dysregulation of lipid biosynthesis in High-Active mice, in line with current human ADHD literature. Future plans include quantifying excitatory synaptic density, dendritic spines, and arborization in High-Active vs. Control striatum to directly test and refine the reduced excitatory synapse hypothesis motivated by the gene expression and synaptophysin results highlighted herein. We hope by unraveling the biology, from the molecular regulators to the brain-circuit level physiological outcomes, it may be possible to one day identify better targets at the root of the pathophysiology of ADHD. # Supporting information This work was supported in total by NIH RO1 MH083807 and DA027487 to J.S.R., and NIH P30 DA018310 and USDA NIFAILLU-538-909 to S.R-Z. There was no additional external funding received for this study. We also wish to thank Dr. Laura Bolton, Dr. Russell Bolton, the Haferkamp family, and the school of Molecular and Cellular Biology at the University of Illinois at Urbana-Champaign for funding for A.S. We wish to thank Alvaro Hernandez at the Roy J. Carver Biotechnology Center for performing the deep sequencing, and the Beckman Institute animal care staff. The authors report no conflicts of interest. [^1]: The authors have declared that no competing interests exist.

# Introduction Cave sediments are important source of information for understanding the history of terrestrial geoecosystems. Due to their concave morphology and solid walls, many caves serve as sediment traps with the ability to continuously collect sediments for millennia. Preserved sequences record long time intervals and are able to survive destructive geomorphological events, such as ice sheet advance, floods or deflation. Although the stratigraphic resolution of clastic and/or biogenic cave sediments is usually much lower than in subaqueous deposits, such as lacustrine or peat bog sediments, caves are among the rare landforms that record subaerial terrestrial conditions. In particular, data on past terrestrial fauna are well protected in caves. This outcome is not only because caves serve as shelters for animals and are natural places of deposition of animal remains but also because a calcareous environment favors preservation of fossil bones and shells. Caves have also been used by humans in prehistoric, historic and recent times as living places, temporary shelters, storerooms or sacral objects. Therefore, they are important archaeological sites, often with multiple cultural levels preserved in superposition and delivering data on the local or regional development of human beings and their culture. For this reason, it is crucial not only to recognize the stratigraphy of cave sediments in particular sites but also to detect regional lithological variability and similarities between sites. This would help to formulate more general stratigraphic schemes that can be used for further inter- and intraregional correlations. Although numerous karst sites with abundant faunal remains have been reported from Central Europe, few of these sequences are stratified and go back as far as the final part of the Last Glaciation. Accumulations of over a meter thick packets of Late Glacial and Holocene clastic cave deposits are rather unique in the region and occur only at single sites, e.g., Veľká Drienčanská Cave in Slovakia, Barová Cave in Czech Republic and Petény Cave in Hungary. At the northern side of Carpathian Mountains, most of cave sites studied until now are situated in Kraków-Częstochowa Upland in southern Poland. This is a karstic region with a very large number of caves and rockshelters. Most of these features contain clastic sediments, in some cases studied by archaeologists, geologists and paleontologists. The lithology and stratigraphy of Pleistocene sediments were recognized in a number of sites in the region\[–\]. In opposition to the Pleistocene series, the Holocene sediments were poorly studied. The main reason is the focus of the scientific community on the Pleistocene and Paleolithic issues. An additional reason is that the internal stratification of the Holocene series in caves of the region is usually poor or completely lacking, and at the most sites, the entire Holocene sequence is limited to only one stratum. Most of these sequences are only a few-dozens- of-centimetres thick and represent only the Middle to Upper Holocene, or just the uppermost Upper Holocene\[–\]. This stratum has usually been described as a dark loam or humus layer. Explanation of this faint stratification is beyond the scope of this paper; however, the most likely reason is past human activity, which was especially intensive during the Medieval and post-Medieval periods. To date, only a few sites with complex Holocene stratigraphy have been discovered in Kraków-Częstochowa Upland: Cave above the Słupska Gate; Żytnia Skała Rockshelters; Nad Mosurem Starym Duża Cave; Rockshelter in Zalas; Zawalona Cave; Żarska Cave, which was excavated recently; most likely Zegar Cave; Perspektywiczna Cave, which is still under excavation; and Shelter in Smoleń III, which is the subject of this paper. Despite this knowledge, there are still many uncertainties in our understanding of the stratigraphy of Holocene cave sediments in the region and their usefulness for paleoenvironmental reconstructions. The most important reason is a lack of solid chronological frameworks for most of the studied sites. Their chronostratigraphy is usually based on singular radiocarbon dates and/or archaeological dating , which in many cases provides rather wide chronological windows. Another factor is lack of Lower Holocene or Lower-to-Middle Holocene at many sites. In some cases there is a hiatus between Upper Pleistocene and Middle or Upper Holocene, in others the entire sequence is restricted to the upper part of Holocene (e.g.,). With only several long sequences in hand we cannot be sure which of them are representative for the region, and which reflect only the local conditions. Therefore, new data are awaited to build a comprehensive reconstruction of paleoenvironment in the region, its changes through time and its spatial variability. Moreover, even the longest and the most complete sequences are usually biased by different degree of preservation and/or accumulation of fossil fauna. Due to this, the direct intra-site correlation between independent paleoenvironmental proxies (such as fauna of rodents and malacofauna) are usually impossible or constrained. Shelter in Smoleń III bears a complex stratified series of Late Glacial and Holocene cave sediments, which was found during an archaeological excavation from 2012–2013. Rich fossil material of snails and vertebrates (rodents and bats) preserved at this site allows reconstructing the paleoecological and paleoclimatic conditions for the recorded span of time. The aim of this study is to explore this rich material and to use the geological and paleontological data from Shelter in Smoleń III as the proxies to build a comprehensive litho- bioclimato-stratigraphic scheme, which can serve as a regional stratotype. ## The studied site Shelter in Smoleń III (ShSIII) is situated near Smoleń village (Pilica comm., Zawiercie dist., southern Poland), in the central part of Kraków-Częstochowa Upland, in the microregion called Ryczów Upland. The rockshelter is situated on the right side of the currently dry Wodąca Valley, 445 m a.s.l., E 19°40′27″, N 50°26′02″. Two entrances are situated next to one another; they are exposed to the NE and open to a small plateau in front of the rockshelter. The rockshelter is most likely a remnant of a larger collapsed cave system. Its total length is 8.8 m, including the branches, and the surface of sedimentary fill is approximately 11 m². The entire interior is controlled by external weather conditions and exposed to daylight, except for the rearmost area. The rockshelter was named *Schronisko w Smoleniu III* (Shelter in Smoleń III) and initially listed under the inventory number 392, then IV.C.12, and finally J.Cz.IV-04.44. The last number and the name are listed in the governmental database of the caves of Poland (<http://jaskiniepolski.pgi.gov.pl/>). ## Geological setting The region is characterized by a variable topography with steep high limestone tors and cliffs and deeply cut valleys. The core of the microregion are the Upper Jurassic massive limestones, partially covered by red desert sands and loams dated to Paleogene/Neogene and sandy deposits attributed to the fluvio-periglacial facies of the Middle Pleistocene. Near the site, the sands are covered by loess deposits, which become thicker and tend to form a continuous coverage toward the east. The karstic cavities (caves and rockshelters) in the microregion are filled with the Upper Pleistocene and Holocene sediments. The most complex stratigraphic sequence has been found in the nearby Biśnik Cave, situated 1.1 km to southwest, where pre-Pleistocene, Middle-to-Upper Pleistocene and Holocene sediments have been preserved. However, the Holocene sediments of Biśnik Cave exhibit reduced thickness and weak lithological variability. ## Archaeological site The rockshelter was recognized as an archaeological site in 2012, during the research project focused on the ancient settlement pattern in the region. The test pitting started in 2012, the excavation continued in 2013 and revealed a multi-episode sequence of human activity\[–\]. It is noteworthy that only minor attention has been given to the Holocene archaeological sites in caves of Kraków-Częstochowa Upland until last years, especially to the stratigraphic situation of the archaeological finds. Therefore, the thorough investigation of ShSIII, including the modern excavation standards, makes this site an important record of the Holocene human activity in the region. # Material and methods ## Excavation and sampling Permission for archaeological excavation was granted by the Regional Heritage Office (Polish: Wojewódzki Urząd Ochrony Zabytków w Katowicach; document No: K-AR.5161.61.2012.JP, ID: 3110, permission No 74/2012), and permission for the field works was granted by the Polish National Forests, regional office in Olkusz (Polish: Lasy Państwowe–Nadleśnictwo Olkusz) who is a land owner and forest manager (permission No: ZG5/503-1/2012 and ZG5/503-1/2013). The site was excavated from 2012–2013 according to the archaeological standards of the test pitting. Excavation covered 10 squares of the archaeological grid (approximately 7 m² of surface in total) and reached the bedrock, revealing a maximum 2.6-m-thick sequence of clastic sediments. The deposits were divided into archaeological layers according to differences in the color, consistency and amount of limestone debris observed in the field. These layers well reflect the geological stratification, and in some cases, they represent soil horizons. The excavation was performed by removing 10-cm-thick intervals of sediment (or 5-cm-thick where the concentration of artifacts or fossils was high) from the squares. The arrangement of layers was recorded by drawing and photographing for each interval. For 3D recording of our work, we used a north-oriented meter grid for horizontal location and depth below the site stratum (an arbitrarily selected point on the wall) for vertical location. The following finds and samples were collected: - Archaeological artifacts–each find was collected and its position recorded in 3D standard. - Macrofossils–each bone, shell and charcoal detected at the site was collected and its position recorded in 3D standard. - Paleontological samples–approximately 5-liter samples of the fine fraction of sediments (macrofossils collected separately, larger clasts removed by hands) were collected from each square, separately for each layer from each 5- or 10-cm interval. The samples were then wet-sieved through a 0.5-mm mesh, and any paleontological material was picked by hand. - Geological samples– 36 samples of sediment with undisturbed grain size composition (i.e., fine fraction collected together with associating pebbles, cobbles and boulders), approximately 7 liters each, were taken from the walls of archaeological trenches according to methodology by Madeyska-Niklewska. At least one sample was taken from each layer if possible, and in the case of layers with greater thickness, several samples were taken from one layer. The details on sample locations are provided in. - Luminescence dating samples–four 0.5-liter samples of fine fraction were taken under dark conditions from selected layers with low amounts of limestone clasts and secured in opaque containers. The collections of archaeological material and paleontological remains are stored at the Institute of Archaeology, Nicolaus Copernicus University in Toruń (Poland). Any paleontological finds from sieved samples (i.e., most of mollusk, rodent and bat remains) are stored under the inventory numbers of the samples (and Files). Archaeological finds and macrofossils are stored under their individual repository numbers. ## Lithological and geochemical analyses Geological samples were dried and separated into fractions with a set of sieves (2, 4, 10, 20, 40 and 80 mm meshes). Macroscopic archaeological or paleontological material was removed by hand. The coarse fraction (\>2 mm), composed almost exclusively of limestone clasts, was soaked in sodium phosphate solution for 24 h to dissolve clay aggregates and then washed with a shower on a 2-mm sieve. After drying, the material was weighted to calculate the amount of each fraction. The fine fraction (\<2 mm) was homogenized by mixing and divided into smaller samples by quartering. Approximately 5-g samples of fine fractions were analyzed on a Malvern Mastersizer 2000 laser diffractometer with a HydroG adapter (analysis subcontracted to the Department of Geoecology and Palaeogeography, Maria Curie-Skłodowska University in Lublin). The procedure followed the widely accepted methodology. Morphological classification of limestone clasts was performed on the \>20 mm fraction, according to published methodology. Four morphological classes were distinguished: A–sharp-edged (or angular) clasts; Ba–slightly smoothed (subangular) clasts with blunt edges; Bb–smoothed clasts with rounded edges (no edges recognizable), but with still-identifiable flat facets; C–highly smoothed (rounded-like) clasts with no recognizable edges or flat facets. The chemical composition of the sediments was determined using approximately 1-g samples of homogenized \<2 mm fraction. The analytical method was inductively coupled plasma mass spectrometry (ICP-MS) performed in the Bureau Veritas Minerals Laboratory (Vancouver, Canada) according to the published procedure. The content of organic matter was determined using another approximately 1-g aliquot of \<2 mm fraction via oxygenation in 30% H₂O₂ for 24 h on a shaking plate. If the reaction was still ongoing (bubbling or dark color), an additional portion of H₂O₂ was added for the next 24 h. The content of organic matter was calculated as the difference between the weight of the dry sample and that of the dry residue after the reaction. ## Chronometric dating ### Radiocarbon dating AMS radiocarbon dating was conducted in the Poznań Radiocarbon Laboratory (Poland). The dating material included snail shells, human and animal bones, charcoals and one example of organic char residue preserved on the internal surface of the ceramic pot. In the case of snails, each dating was performed on one identifiable shell of *Isognomostoma isognomostomos*. Shells of *Discus ruderatus* were also dated, but due to their small size, several shells from the same layer, square meter and depth were mixed together. A small aliquot of each shell sample was checked with X-ray diffraction (at the Faculty of Geology, University of Warsaw, Poland) prior to dating to exclude any material with negative effect of aragonite-to-calcite recrystallization. Only samples evidencing pure aragonite were dated. Chemical pretreatment followed Brock et al.. In the case of bones, the dated fraction was collagen extracted according to Goslar et al.. Small aliquots of extracted collagen were analyzed with a CHNS elemental analyzer, and atomic C:N ratios was calculated to check the collagen quality. The range of 2.9–3.6 was assumed acceptable according to the literature. For two samples, the collagen yield was too low for both elemental analysis and radiocarbon dating, and these samples were dated without the C:N ratio being confirmed. In the case of charcoals, cellulose was extracted according to AAA method. All dates were calibrated using the OxCal v. 4.3 software package versus the IntCal’13 radiocarbon calibration curve. Calibrated dates are presented as ky BP (i.e., thousand years before 1950), as the ranges within a 94.5% probability interval. The OxCal v. 4.3 sequence model was applied to estimate the chronology of layer boundaries. It followed the procedure used by Krajcarz et al., based on the widely accepted methodology. ### Luminescence dating The luminescence dating of samples was subcontracted to the Department of Geoecology and Palaeogeography, Maria Curie-Skłodowska University (Lublin). Post-IR IRSL₂₉₀ dating was performed on the polymineral fine grains (4–11 μm). The extraction procedure followed the accepted methodology. Equivalent doses were determined using a post-IR IRSL protocol. ## Paleozoological analyses Mollusks, rodents and bats are the most numerous groups of fossil animals in ShSIII and therefore were chosen for the paleozoological analysis to provide supportive paleoecological and paleoclimatic data for stratigraphy. In addition to these groups, the fossil material also comprises the remains of amphibians, reptiles, birds, insectivores and large mammals including humans. For the purpose of paleoecological analysis, layer 3 has been divided into two parts (3 lower and 3 upper) due to its great thickness and abundant rodent and mollusk remains. ### Identification of mollusk remains Standard methods\[–\] were applied for mollusk analysis. The shells collected during sieving were carefully cleaned. All completely preserved shells and their identifiable fragments were identified under a binocular microscope using taxonomical keys\[–\] and counted applying the schemes for broken individuals. The number of individuals was counted separately for each layer. For some damaged individuals, only the genus or family levels were determined, and the calcareous plates of slugs were counted together under the heading of Limacidae. ### Identification of rodent remains The rodent fossil assemblage consists of single bones and isolated teeth. Specimens were identified in terms of skeletal elements and species when possible. The taxonomical attribution of remains was based on the following diagnostic elements: first lower molars (Arvicolinae); mandibles, maxillae and isolated teeth (Muridae, Gliridae and Dipodidae); mandibles, maxillae, isolated teeth and diagnostic postcranial bones (*Cricetus cricetus*, *Sciurus vulgaris*). Remains were identified under a binocular microscope with use of the comparative collection of recent rodents and following the general criteria given in the identification keys. The separation of *Microtus arvalis* and *M*. *agrestis* was based on the method given by Nadachowski. The taxonomic classification followed Wilson and Reeder for all taxa except for *Clethrionomys glareolus* (former *Myodes glareolus*) and *Lasiopodomys gregalis* (former *Microtus gregalis*), which were classified after later publications. For each species, the number of identified specimens (NISP) and minimum number of individuals (MNI) were counted according to accepted method. The MNI was counted separately for each layer. ### Identification of bat remains Specimens were identified, when possible up to the species level, on the basis of diagnostic traits of the hemimandible and/or the skull. In particular, the attribution of the taxonomic group was based on several different characteristics: the number of teeth, the shape of the hemimandible, the traits of the fourth lower and upper premolars and those of the upper and lower molars. Remains were identified and measured through a binocular microscope, comparing them with specimens belonging to a collection of recent bats. The general criteria given in the identification keys\[–\] were fulfilled, and the remains were classified taxonomically according to Wilson and Reeder. The taphonomic parameters NISP and MNI were calculated according to the same method as for rodents. ### Paleoecological reconstruction Paleoecological inference was based on the environmental preferences of all mollusk, rodent and bat species found in the sediments. The structure of taphocoenoses and succession were presented on frequency bar diagrams. The diagrams were composed of both the absolute numbers of shells (or the MNI in the case of rodents) for samples containing less than 50 individuals and the percentages of the total sum for those with more than 50 individuals. In the case of bats, only the MNI values were shown due to low number of remains. Mollusk, rodent and bat assemblages were distinguished on the basis of their structure and composition, supported by statistical clustering analyses. The clustering was done for samples (layers) with more than 10 shells or with MNI greater than 10 by applying a paired group algorithm and Horn’s overlap index for abundance data using the PAST software package, version 3.22. Both the character and succession of the assemblages were used in the paleoecological and paleoclimatic reconstructions. To reconstruct the changes in ecological diversity in the stratigraphic sequence, the assemblages were analyzed via application of the richness and the Simpson index of evenness. In the case of rodents, specimens identified as *M*. *agrestis*/*arvalis* were excluded, while *Apodemus sylvaticus*/*flavicollis* and *Apodemus* sp. were counted as one taxon (genus *Apodemus*). The number of taxa per sample (layer) is a measure of richness. The greater the number of taxa present in a sample, the 'richer' the sample. The Simpson 1-D index was calculated as 1 − Σ(n_i/n)², where n_i is the MNI of the particular taxon in an ‘i-sample’ (a layer) and n is the total number of MNIs in a sample (a layer). The index was calculated using the PAST software package, version 3.22. The index represents the probability that two individuals randomly selected from a sample will belong to different taxa. To present the changes in the environment or landscape around the site, we used the environmental preferences of the species. Each mollusk taxon was assigned to one of four ecological groups: F–shade-loving species; O–open-country species; M–mesophilous species; H–hygrophilous species. In the case of rodents, which usually inhabit a range of environments, we used the method of habitat weightings. Each taxon was assigned to the habitat(s) where it can be found at present, with percentage weight(s) attributed for each of its habitat(s) (details are provided). The rodent fauna of ShSIII was assigned to eight habitat types based on a habitat identification mode provided in the literature\[–\], with modifications for habitats typical for Central and Northern Europe. The used habitats were as follows: Tu–tundra; OA–open anthropogenic cultivated areas; OD–open dry (steppes); OH–open humid (meadows); OW–open woodland (shrubs, forest margins, steppe-forests); FT–temperate forest (broadleaf forests, mixed forests); FB–boreal forests (taiga); Wa–water bodies. Data on the species distribution and their preferred habitats were taken from Mitchell-Jones et al. and the International Union for Conservation of Nature (IUCN) Red List maps\[,–,–\]. Bats with their flight ability are more mobile animals and this restricts their usefulness as the indicators of local paleoenvironment. However, some species which are sedentary and use old trees as summer roosts (i.e., *Barbastella barbastellus*, *Myotis bechsteinii* and *Plecotus auritus*) clearly indicate the presence of old-growth forests in the vicinity\[–\]. Bats hunting near water (*M*. *dasycneme* and *M*. *daubentonii*) indicate the presence of open water bodies. # Results ## Lithostratigraphic sequence The lithostratigraphic sequence comprises fifteen layers, some of them being lateral variants. A characteristic feature of ShSIII is the lithological dissimilarity of sediments between different parts of the site, despite its small size (Figs). Three lithostratigraphic zones could be distinguished on the basis of these dissimilarities: entrance, central and rear. They were recognized at the squares of archaeological grid as follows: entrance zone–C/4, D/4, E/4, and partially C/5; central zone–the remaining area of C/5 and entire D/5, E/5, D/6, E/6, F/6 and F/7; rear zone–C/8 and C/9 (see also). Detail lithological and geochemical parameters are provided in. ### Entrance zone The lower parts of the sedimentary sequence are similar at the entrance and central zones. The sediments of the upper part in the entrance zone are darker and more enriched in humus, as a result of the greater influence of external conditions and pedogenic processes. The total thickness of the sequence reaches 260 cm. **Layer 9** is a limestone debris composed of coarse sharp-edged clasts. The matrix is a yellowish brown to brown (Munsell color: dry 10YR 5/4-6/4, moist 10YR 5/3-5/4) silty loam to sandy silt. The limestone bedrock below the layer is cracked into angular blocks, and this regolith passes into layer 9 gradually, which is marked by decreasing size of blocks and occurrence of matrix. The Ca content is very high in the fine fraction, particularly in the lower part of the layer, where it reaches greater than 30%. This strongly suggests that the fine fraction is mostly composed of physically disintegrated limestone. **Layer 7** is a pale brown to brown (dry 10YR 6/3, moist 10YR 4/3) sandy silt. It is a massive (structureless) loess-like sediment with angular limestone clasts. Layer 9 passes indistinctly into layer 7, and the transition between the layers is marked by a gradual color change and decreasing amount of limestone clasts. The content of Ca in the fine fraction is very low, which indicates the allogenic source of material. At squares C/5 and partially D/5, a lens of dark silt (called layer 8) occurs in the upper part of layer 7. This intercalation enables dividing the layer locally into sublayers 7a (above layer 8) and 7b (below layer 8). **Layer 8** is a grayish brown to dark grayish brown (dry 2.5Y 5/2, moist 10YR 4/2) sandy silt. The texture is similar to that of layer 7, except for the dark color and the presence of charcoals inside layer 8. **Layer 3a** is a grayish brown to black (dry 10YR 5/2, moist 10YR 2/1) silt. The amount of limestone clasts is greater than in layer 7. Angular blocks predominate, but smoothed clasts also occur. The concentration of P increases upward starting from layer 3a, indicating the increasing impact of zoogenic activity on the accumulation. The lower boundary is blurred, and the layer passes downward gradually into layer 7. Close to the walls of the rockshelter, the layer exhibits greater thickness, and its bottom is situated at lower elevation. **Layer 3** is a dark brown to very dark grayish brown (dry 10YR 3/2-4/3, moist 10YR 2/2-3/3) sandy silt. The amount and size of limestone clasts is lower here, and smoothed clasts occur in greater number. However, outside of the dripline (square meters C/4 and partially D/4), the large limestone blocks occur in the layer. Their surfaces and edges are smoothed. Indistinct trough cross- stratification is marked in the bottom part at square E/4. **Layer 1** is a very dark gray to black (dry 10YR 3/1, moist 2.5Y 2/1) silty sand. Limestone debris becomes less abundant and finer upward, and angular clasts are almost lacking, while smoothed, rounded-like clasts are relatively abundant. This layer exhibits unusually high concentrations of Pb and other toxic metals (Zn, Cd, Ag, Sb, Bi, Sn, see), suggesting an increased impact of human activity. Another characteristic of the layer is a high sand fraction. ### Central zone The lithological variability is the greatest in this area, and the stratigraphy is the most complex, with up to eight distinct strata preserved in superposition. The lower part of the sequence shares lithological characteristics with the entrance zone, but in the upper part, the sediments are different. The sedimentary fill achieves a thickness similar to that of the sediments in the entrance zone (230 cm at maximum). The sequence starts with a limestone debris containing very pale brown to yellowish brown (dry 10YR 8/3, moist 10YR 5/6) silty to sandy matrix (called **layer 9**), which is a continuation of layer 9 from the entrance zone. The clasts are finer here, but angular shapes still predominate. The layer reaches the greatest thickness (110 cm) at squares E/6, F/6 and F/7, where it fills the narrow and deep depression of the bedrock, probably a remnant of a vadose canyon. The overlying sediment (called **layer 7**) is a continuation of layer 7 from the entrance zone. It is a yellowish brown to pale brown and brown (dry 10YR 6/3, moist 10YR 4/3–5/4) loess-like sandy silt with limestone clasts. The clasts are mostly angular and become less abundant and finer upward. The concentration of Ca in the fine fraction is very low, which suggests an allochthonous source of fine material. The lower boundary of this layer is indistinct, and layer 9 passes gradually into layer 7. Similar to the entrance zone, locally (at square D/5, where layer 8 occurs), the layer can be divided into sublayers 7a (above layer 8) and 7b (below). Sublayer 7a has a lighter, more yellowish color (dry 10YR 6/3, moist 10YR 5/4) and lower amount of limestone clasts compared with sublayer 7b (color: dry 10YR 6/3, moist 10YR 4/3). The lower boundary of sublayer 7a is not always readable, but locally, especially in contact with layer 8, it is sharp and erosional. **Layer 8** is probably a continuation of layer 8 from the entrance zone. It is a grayish brown to dark grayish brown (dry 2.5Y 5/2, moist 10YR 4/2) sandy silt. It occurs in restricted area, limited to square D/5. The layer seems to fill the shallow depressions in sublayer 7b. The remains of a hearth are preserved in the lower part of layer 8 and labeled as sublayer 8a. **Layer 6** is a light brownish gray to dark grayish brown (dry 10YR 6/2-6/3, moist 10YR 4/2–2.5Y 5/2) sandy silt. The amount, size and morphology of limestone clasts are similar to layer 7. The concentrations of Ca and P increase from layer 6 upward, recording the increasing importance of autogenic deposition and zoogenic activity. The lower boundary of layer 6 is sharp and undulating. Internal sedimentary structures in a form of trough cross-stratification are weakly marked inside the layer. **Layer 5** is a grayish brown to dark grayish brown (dry 10YR 5/2-5/3, moist 10YR 4/2-4/3) silty loam to sandy silt. The amount of limestone debris is greater than in layer 6. The clasts are smoother, and even rounded-like ones occur. The lower boundary is discordant, sharp and undulating. The trough cross- stratification is locally distinct, with bedding inclined toward the entrance. **Layer 5a** is a whitish to grayish brown and dark grayish brown (dry 10YR 5/2-8/2, moist 10YR 4/3-6/3) massive silty loam to clay silt. The sediment is clearly lighter in color than those of the lower strata. Its lower boundary is sharp, and the layer lies discordantly on lower sediments. Locally, layers 5 and 6 are not preserved, and layer 5a lies directly on layer 7. The amount of limestone clasts is relatively low, and the clasts are fine and smoothed. The concentrations of Ca and P are among the highest in the entire sequence. The layer is not continuous and is lacking at square meters D/5, D/6 and E/6. **Layer 4** is a very pale brown to brown (dry 10YR 7/3, moist 10YR 5/3) massive silty loam. The amount of limestone clasts is low; their morphology is similar to that in layer 5a. The clay content increases in the upper part of layer 5a and stays high in layer 4. The concentration of Ca is similar to that in layer 5a, whereas the concentration of P is slightly lower. The lower boundary is indistinct and gradual. **Layer 2** is similar to layer 4 in terms of the grain size composition and limestone clast morphology but differs in color, being more grayish (dry 2.5Y 7/2, moist 2.5Y 4/2). The concentrations of Ca and P are still relatively high but lower than in layer 4. The boundary between layers 2 and 4 is indistinct and gradual. The sequence is topped with **layer 1a**, which is a lateral variant of layer 1 from the entrance zone. This sediment is more brownish than layer 1 (dry 2.5Y 3/2, moist 2.5Y 2/2). The thickness of layer 1-1a decreases toward the rockshelter interior, such that the maximum thickness of layer 1 is 60 cm and that of layer 1a is up to approximately 30 cm. Upper part of layer 1a covers the archaeological feature (layer A), and due to lithological similarities of these two units, the boundaries between them are difficult to follow. ### Rear zone The sequence in the rear zone has a reduced thickness, achieving 120 cm at maximum. Sediments of this zone were recognized in a limited area of approximately 1 m². The rockshelter is shaded, and sunlight does not reach this area. The corridor declines downward to the south and becomes narrower, down to the width not accessible for penetration by an adult human. The concentration of K is low in this zone, which indicates lower infiltration from the top strata, possibly as a result of low rainwater supply in the deeper part of the rockshelter. The sequence starts with **layer 11**, which is a complex stratum of bright reddish-brown color. The average grain size composition indicates a silty loam texture, but the layer is composed of numerous diapirs of brown (dry 7.5YR 5/4, moist 7.5YR 5/4) clay and lenses of reddish brown to strong brown (dry 7.5YR 6/6, moist 7.5YR 5/6) silty sand. This sandy component is most likely an admixture from layer 10. Limestone clasts are absent. The sediment is strongly enriched in Fe and depleted in Ca , what indicates an allogenic origin. High concentrations of Fe and other geochemically related metals (Mn, Co, Zn, Pb, Cu, Mo, Th, Cd, Sb, and REE) are characteristic features of this sediment and are responsible for its reddish color. The thickness is greater than 30 cm. **Layer 10** is a yellow to brownish yellow (dry 10YR 7/5, moist 10YR 6/7) silty sand. Sandy grains are composed mostly of quartz. Such a mineral composition, together with the lack of limestone clasts and very low concentration of Ca, indicates an allogenic origin. The sediment is preserved only in pocket-like or tongue-like structures immersed in layer 11. The contact between the two layers, although not in a primary position, is sharp, what indicates the erosional boundary at the bottom of layer 10. The original thickness of the layer is difficult to establish due to disturbances, but it can be estimated to be greater than 5 cm. **Layer 9** is a 15-cm-thick horizon of coarse subangular limestone blocks, with yellowish brown to dark yellowish brown (dry 10YR 5/3, moist 10YR 4/4) silty loam filling the space between blocks. The lower boundary is sharp and erosional. The vertical orientation typical for layers 11 and 10 is not visible here. High concentrations of trace metals, similar to those exhibited by layer 11, indicate that layer 11 served as important source of material. This layer most likely forms the lateral continuation of layer 9 from the entrance and central zones. That layer passes upward gradually into **layer 4**. This is a very pale brown to yellowish brown (dry 10YR 7/3-7/4, moist 10YR 5/4-5/5) silt or clay silt, passing upward into silty loam. This layer is a lateral continuation of layer 4 from the central zone. The concentrations of Ca and P are high and increase toward the top of the layer. **Layers 2 and 1a** are lateral continuations of layers 2 and 1a, respectively, from the entrance zone and share lithological characteristics with those strata. ### Anthropogenic layers Two anthropogenic strata were recognized between the natural strata and were labeled as sublayer 8a of layer 8 and layer A. **Sublayer 8a** is a black (dry 10YR 4/1, moist 10YR 2/1) sandy silt. It occurs in the lower part of layer 8 and represents its darker variant. The sublayer has a form of a thin lens with a thickness of approximately 10 cm and lies at the bottom of a shallow depression dug into layer 8 and reaching sublayer 7b. The presence of charcoals indicates that this deposit is a remnant of a hearth. **Layer A** is a sediment of intermediate characteristics between layers 1, 1a, 2 and 3. It is a dark brown to very dark grayish brown (dry 10YR 3/1-4/3, moist 10YR 2/2-3/2) silty sand. It forms a backfill of an anthropogenic feature. Its lower boundary is sharp and distinct. Stratigraphically, the feature is situated inside layer 1-1a, as its walls cut layer 1 or 1a, and at some square meters, it is covered by the upper part of layer 1a, with an indistinct boundary between the layers. The bottom of the feature reaches layer 2 or 3 (at different square meters) and occasionally layers 4 and 5a. ## Chronometric data A total of 29 radiocarbon dates are accessible for the ShSIII. The dates cover the interval from 14.1 to 0.6 ky BP, i.e., almost the entire Holocene and Late Glacial of the Last Glaciation. In case of bones, the atomic C:N ratio in extracted collagen is in the widely accepted 2.9–3.6 range for all tested samples (only two samples yielded too low amounts of collagen to check the C:N ratio), confirming that the analytical fraction was pure. Four TL dates were also obtained for layers 7b, 6 and 5a. To build the sequence model in OxCal, we used only the radiocarbon dates from the central zone, to achieve the data for a single stratigraphic column and avoid problems due to lateral correlations. Moreover, the central zone delivered most of the dates, while only two radiocarbon dates were obtained from material from the entrance zone. The archaeological feature (layer A) was excluded because its backfill includes redeposited material from the lower layers and may disturb the model of deposition. A date from layer 1a (Poz-53301) was also excluded, as this date was obtained from a human bone, which was most likely redeposited from the grave, i.e. from layer A. The model including all dates from the central zone did not run until the chronologically discontinuous dates were removed. The dates in disorder were mostly those from layers 4, 5a and 5. There are several possibilities for selecting dates for removal. The least- invasive method is to remove three dates, Poz-61237, Poz-61305 and Poz-53303 (marked in as “intrusive material?”), and this method was applied to build the model. The dates from layer A (backfill of archaeological feature) and from a human bone from layer 1a (Poz-53301) were used to build a separate sequence model to determine the boundaries of the phases of human activity. The dates were attributed in an arbitral manner to three phases of human activity (PHA): PHA 2 –Bronze Age; PHA 3 –pre-Roman to Roman Period; and PHA 4 –Late Middle Ages (PHA 1 was reserved for the Late Paleolithic phase). The model for PHA boundaries gave very good overall agreement index A_overall = 96.4%. The model for the boundaries of the natural layers gave weaker agreement index, A_overall = 78.8%, yet still in the acceptable range which is \>60%. This weaker agreement is a consequence of partial overlapping between the extreme dates from layers 2 and 4; 4 and 5a; and especially 5 and 6. The outlier with the lowest agreement index is a date for layer 5 (Poz-61238, agreement index A = 39.7%). The obtained modeled boundaries are provided in. ## Malacological succession The molluskan fauna from ShSIII is composed of 61 taxa of land snails, represented altogether by 10,295 individuals. Fifty-two species are accompanied by 7 taxa identified to the genus level, one to the family level (Clausiliidae), and slug plates represent a collective group of Limacidae. Malacofauna represents four ecological groups (according to widely accepted subdivision), with 30 shade-loving species (group F), 8 open-country species (group O), 13 mesophilous species (group M) and 1 hygrophilous species *Carychium minimum* (group H). There are only slight differences in malacological composition between the central and entrance zones, but they differ in proportions between taxa and structure of the assemblages. No gastropods were found in the lower part of sedimentary sequence (in layers 11, 10 and 9). In the above-lying layers, the mollusk community is bipartite. Only few taxa and up to 60 individuals altogether are noted in layers 7b, 8 and 7a, whereas the higher part of the sequence is characterized by rich malacocoenoses comprising up to 46 taxa and 2,181 individuals in a single layer. Based on both the cluster analysis and the fauna composition, four mollusk assemblages were distinguished in ShSIII. The nominal species are not the most abundant but rather the most characteristic of the population, which enables wider comparisons on the regional scale\[,–\]. ### Assemblage Vt The sequence starts with the assemblage with *Vallonia tenuilabris* (Vt), including layers 7b, 8, and 7a in the central zone and layer 8 in the entrance zone; possibly also layers 7a and 7b in the entrance zone belong to the same assemblage. This poor malacocoenosis contains from 5 to 60 shells per layer, mostly of open-country gastropods, which constitute 35–73% of all individuals. *Vallonia tenuilabris*, a glacial relic that occupies the open areas of so- called “cold-steppes”, the alpine grasslands, taiga, hemiboreal forest and wooded fens\[,–\], is the most numerous in the central zone, being accompanied by the shade-loving species *Discus ruderatus* and *Semilimax kotulae*, both characteristic of a cool and continental climate. In the entrance zone, *D*. *ruderatus* slightly outnumber *V*. *tenuilabris*. ### Assemblage DrVc In layers 6, 5 and 3a, another assemblage dominated by *Discus ruderatus* and *Vallonia costata* (DrVc) can be distinguished. *D*. *ruderatus* represents taiga-type coniferous forests, whereas *V*. *costata* is typical of dry and open habitats and predominates in the sequence. In the central zone (layers 6 and 5) glacial relics *V*. *tenuilabris* and *S*. *kotulae* are still important components of the assemblage, but they disappear in the entrance zone (layer 3a). Moreover, this assemblage records a first period of predominance of shade- demanding species in the sequence. Mesophilous snails are mostly represented by *Carychium tridentatum*, associated with permanent humid conditions. ### Assemblage DroAp A rich assemblage with *Discus rotundatus* and *Aegopinella pura* (DroAp) can be distinguished in layers 5a and 4 of the central zone and the lower part of layer 3 in the entrance zone. This assemblage is represented mostly by forest gastropods (52–75% of all individuals) typical of warm and humid conditions, namely, *D*. *rotundatus*, *A*. *pura*, *Discus perspectivus*, *Balea biplicata*, *Helicigona faustina* and *Isognomostoma isognomostomos*. *V*. *costata* gradually disappears, and the open-country snails become outnumbered by mesophilous species, mainly *Laciniaria plicata*. The occurrence of *V*. *elata*, unknown in the contemporary fauna of the region, is worth noting. ### Assemblage Lp The topmost part of the sequence (the upper part of layer 3 and layers 2, 1a and 1) is occupied by forest and mesophilous taxa, with the most numerous being *Laciniaria plicata* (Lp). Gastropods of open habitats disappear in the central zone and are of secondary importance in the entrance zone. The forest fauna still comprise a large fraction of the total fauna (45–61% of all individuals) but are gradually replaced by snails of wide ecological tolerance, which are the most abundant in layers 1a and 1 (up to 55% of all individuals). *L*. *plicata*, which is typical of humid rocks in open environments, is accompanied by *Helicigona lapicida*, which often hides in deep crevices in shady rocky substrate; *H*. *faustina*, which is common in forest slopes and humid shady rocks; and *I*. *isognomostomos*, which is characteristic of humid mountain forests. ## Rodent succession The fossil assemblage includes 1,716 identified specimens, corresponding to a minimum of 901 individuals (sum of MNIs for particular layers) and representing at least 19 taxa, including 17 taxa identified to the species level. There are no taxonomical differences in the rodent composition between the central and entrance zones. The sediments of the rear zone appeared as very poor in rodent remains and provided only 45 identified specimens in total. For this reason, the usefulness of fauna from the rear zone for statistical processing is not plausible. No rodent remains were collected from the lowermost layers (11 and 10). Very poor collection (3 specimens) was found in layer 9. In the upper sequence, two main parts are distinct, which vary in terms of both taxonomic composition and number of specimens. The lower part (layers 7 and 8) is characterized by a low abundance of rodent remains, low richness and poor diversity. Much richer taphocenosis is preserved in the upper part of the sequence. The number of individuals and number of taxa reach their maxima in layers 5, 5a, 4 and 3 (its lower part). The habitat representation recorded by rodent remains varies clearly between the lower and upper parts of the sequence. The representation of forest environments is low in layers 8 and 7, where the indications of an open tundra environment with the presence of water bodies predominate. The forest component increases in layer 6 (51%) and continues up to reach over 70% in the upper layers (with maximum in layers 4, 2 and the upper part of layer 3). The cluster analysis of rodent assemblage from ShSIII reveals four distinct groups, which reflect the distinct faunal assemblages formed by the changes in the composition and structure of the species. These groups were named after the most abundant taxa. ### Assemblage Dt This assemblage groups layer 8 and sublayer 7b from the central zone. Sublayer 7a from the central zone and layer 7b from the entrance zone possibly also belong to this assemblage. From a quantitative point of view, this is a very poor collection, giving altogether 38 remains (at least 29 individuals). The dominant taxon here is a collared lemming *Dicrostonyx torquatus* (30.8–100% of MNI for sublayers 7a and 7b), followed by the tundra vole *Microtus oeconomus* and European water vole *Arvicola amphibius*. Collared lemming is an inhabitant of cold climatic areas, while the two other voles prefer humid habitats with water bodies. The relatively high representation of the red vole *Clethrionomys glareolus* in layer 8, which is linked with woodlands and prefers densely vegetated clearings, suggests a possible temporal climatic improvement with development of woodland vegetation. Inhabitants of open dry environments (steppes), i.e., *Cricetus cricetus* and *Sicista subtilis*, were recorded in sublayer 7b. ### Assemblage CgLg Layer 6 forms a separate assemblage. The number of rodent remains is low (51 remains, at least 25 individuals). There is a well-distributed diversity, even if the number of species is low. The assemblage is dominated by *C*. *glareolus* (28% of MNI). The second-most-abundant species is *Lasiopodomys gregalis* (16%), a representative of open and cool habitats. Rodents adapted to cold or boggy environments (*D*. *torquatus*, *A*. *amphibius* and *M*. *oeconomus*) are still present. Thus, the layer records a sensible increase in the woodland component (*Apodemus* and *C*. *glareolus* constitute 30% of the MNI). Layer 4 from the rear zone is grouped by cluster analysis with this assemblage, however, the number of specimens is low (15 remains, at least 11 individuals) and this clustering is disputable. ### Assemblage CgMs A rich assemblage dominated by *C*. *glareolus* and *M*. *subterraneus* is recorded in layers 3a, 3 and 1 in the entrance zone, layers 5, 5a and 4 in the central zone. Layer 2 from the rear zone is poor in rodent remains but possibly also represents this assemblage. The species of cold habitats disappear. There is a visible rise in the representation of taxa associated with woodland environments, with domination of the red vole, *C*. *glareolus*. This is the most abundant taxon in the whole upper part of the sequence. Starting from layer 5 in the central zone and 3 in the entrance zone upward, the red vole is followed by *M*. *subterraneus*, also an inhabitant of woodlands, preferring densely vegetated clearings. Mice of genus *Apodemus* also constitute a significant portion of the assemblage. The development of deciduous or mixed forest habitats is testified by the presence of three species of the Gliridae family, which continues in the upper layers of the sequence. ### Assemblage CgMsGg The layers of the uppermost part of the sequence in the central zone (2 and 1a) are characterized by another rodent assemblage. The species representation is similar to the lower assemblage CgMs, and the record of habitat conditions did not change significantly. The visible change is recorded by the higher participation of the edible dormouse *Glis glis*. In layer 1 of the entrance zone and layers 2 and 1a of the rear zone the presence of this species is also marked, which suggests their correlation with CgMsGg assemblage. However, the amount of edible dormouse remains is not as high there as in layers 2 and 1a in the central zone, and therefore the cluster analysis joins these layers with other assemblages. ## Chiropteran succession The bat fossil assemblage includes 369 remains corresponding to a minimum of 246 individuals (sum of MNIs of different layers) and representing at least 24 taxa, including 13 species (for the specimens that were identified to the species level) and 6 groups (remains identified to the genus level or belonging to cryptic groups); the latter were not considered during the statistical analysis. For some specimens, identification of the species was not possible, since the morphological and dimensional characteristics required for a precise assessment were missing. The groups affected by this issues were *Plecotus auritus/austriacus*, *Pipistrellus pipistrellus/pygmaeus*, *Myotis mystacinus/brandtii/alcathoe/daubentonii*. Some remains (mainly fragmented bones) were identified just to the order level (e.g., Chiroptera). No bat remains were found in the rear zone. There are differences in the taxonomical composition of bat assemblages between the central and entrance zones. The sediments of the entrance zone are poor in bat remains, with 41 bat specimens representing 10 taxa. The chiropteran remains are much more abundant in the sediments of the central zone, with 144 specimens belonging to 11 taxa. The presence of *Myotis dasycneme* characterizes the entrance zone exclusively, while *M*. *brandtii* occurs in the central zone instead. No bat remains were collected from the lowermost layers (11 to 7). Layers 5a, 4 and 2 yielded the highest number of bat remains. This increase in the presence of bats could be explained by the advance in the forest environment in the area. The cluster analysis of the bat assemblage revealed six distinct groups. ### Assemblage PaurMbra The group comprised by layers 4 and 5a contains the richest assemblage in terms of number of remains, even if the number of species does not differ from the overlying layer 2. We found 79 individuals belonging to 9 species, for a total of 130 bone remains. ### Assemblage MbecPaurMbra Layer 2 in the central zone seems to be one of the richest layers in terms of a number of remains (NISP = 61, represented by 38 individuals belonging to 9 species). The only fragment attributable to *P*. *pipistrellus/pygmaeus* (the identification of the species is impossible due to a lack of the distinctive characters) was found in this layer. ### Assemblage MdauMbra This assemblage represents layer 1a from the central zone and is comparable to the one found in layer 1 from the entrance zone (i.e., the PaurMdas assemblage). A small number of remains (16 fragments) were found. They belong to 12 individuals of 6 different species. ### Assemblage BbarMdas The lower part of layer 3 from the entrance zone delivered few fragments (16) belonging to 12 individuals of 8 species. This assemblage does not differ considerably from the other layers of the entrance zone. ### Assemblage PaurMdauMdas This assemblage, belonging to the upper part of layer 3 in the entrance zone, includes a small number of specimens (16 bone fragments attributable to 13 individuals of 6 different species). ### Assemblage PaurMdas This assemblage represents layer 1 from the entrance zone. From a quantitative point of view, this is a very poor collection, with only 16 remains belonging to at least 15 individuals. A predominant taxon cannot be defined. # Discussion ## Deposition of sediments The entire sedimentary sequence in ShSIII can be divided into six sedimentary series. The differences between series are marked in sedimentary structures (erosional surfaces and lamination), texture (mostly an abundance of limestone clasts) and geochemistry (mostly the amount of organic matter, Ca and P). ### Series I–reddish clays The oldest sediments are preserved in the rear zone as layer 11. This sediment shares similarities with reddish clays known from the lowermost parts of the sequences from caves such as Biśnik Cave (layers 20–23), Nietoperzowa Cave (layer 17), Tunel Wielki Cave (lower part of layer 1), Cave in Dziadowa Skała (lower part of layer 1) and Żabia Cave (lower part of sequence). These types of sediments were linked with Lower Pleistocene or Pliocene *terra rosa* soils, redeposited to the caves from the surface. ### Series II–sands Layer 10 has also been preserved only in the rear zone. The bright color of sand suggests a correlation with reddish or yellowish sands known from other caves of the region, where they also form a bottom part of the sequences. Such sandy deposits are known from caves such as Nietoperzowa Cave (layer 17), Koziarnia Cave (layer 21), Tunel Wielki Cave (upper part of layer 1), Perspektywiczna Cave (layer 11), Cave in Dziadowa Skała (upper part of layer 1) and Wylotne Rockshelter (layer 8). The accumulation of sands may be connected with Middle Pleistocene glaciations, when the area of Kraków- Częstochowa Upland entered the fluvioglacial regime, or with later colluvial redeposition of fluvioglacial sediment. Both layers 11 and 10 fill the narrow depression of the bedrock, a form which was called a “bottom rill” and which is most likely the remnant of a vadose canyon. This depression was also noticed in the central zone, but here, the sediments resembling layers 10 or 11 have not been preserved. The depression is filled with limestone debris of layer 9, which also occurs in the rear zone, but with limited thickness and at higher elevation. This result indicates that erosional event happened between accumulation of layers 10 and 9, which was responsible for almost total removal of layers 10 and 11, except for the rear zone. ### Series III–limestone debris and loess Layer 9 records the cold climatic conditions. A high amount of angular limestone clasts with Ca-rich silty matrix indicates that physical weathering of the bedrock and consequential rockfall was the most important depositional factor, followed by physical weathering of the rockfall debris. An absolute lack of faunal remains additionally suggests very cold periglacial conditions. This sediment may be correlated with layer III from the nearby Shelter above the Zegar Cave, described as loess with limestone debris, with very sparse faunal remains. The reddish color of layer 9 suggests that layer 11 (and possibly also layer 10) served as an additional source of material for layer 9. Most probably, some remnants of layers 11 and 10 survived the erosion adhered to the walls or inside the niches and dropped down during the deposition of layer 9. A similar situation concerning the Lower Pleistocene red clays redeposited into the Upper Pleistocene series was described for Tunel Wielki Cave. Debris of layer 9 filled the deep vadose canyon cleaned by the previous erosion and achieved over 1 m thickness inside the depression, but it reached only limited thickness on the elevations, such as in the rear zone. Layer 7 was deposited on the uneven surface of the rockfall debris and filled the depressions. The layer achieves the greatest thickness near the entrance and becomes thinner toward the cave interior. It disappears at a distance of approximately 300 cm from the dripline. It is a loess-like sediment or “cave loess”. Respecting the lithostratigraphic scheme of those authors, the lower part of layer 7 (with limestone debris) correlates with their unit A, and its upper part (less debris, more silt) with unit C. Analogous sediments occur in a number of caves in the region, including the closest ones: Biśnik Cave and Shelter above the Zegar Cave. Indistinct lower boundary with layer 9 indicates gradual change in sedimentary environment from physical disintegration of the bedrock related to cold and humid climate to eolian accumulation typical for cold and dry conditions. Intercalation of dark-colored layer 8 allows dividing layer 7 into sublayers 7a and 7b, but this possibility occurs on a limited area. Sublayer 8a in the lower part of layer 8 represents an anthropogenic activity. This deposit was formed by firing a hearth in a shallow pit dug in the loess. Few faunal remains have been found in layer 8. These include both the bones of larger mammals (hare and elk), which may be related to human hunting activity but also the remains of rodents and snails, likely deposited by natural agents. Presence of fauna and charcoals records a temporal climatic warming, replaced later again by periglacial conditions as marked by the overlying loess of layer 7a. Trough cross- stratification visible in sublayer 7a also indicates a role of colluvial processes in the formation of this stratum. ### Series IV–colluvial sediments Holocene sequence starts from layer 6, as indicated by radiocarbon dating and explosion of the abundance and diversity of faunal remains. Trough cross- stratification and lamination in layers 6 and 5 are distinct indicators of erosion and colluvial redeposition of sediments by washing and/or mudflows. Inclination of sedimentary structures indicates that the main direction of a geological transport was oriented from the rear zone of the rockshelter to the entrance and probably outside. Redeposited material differs from any lower sediment by the higher amount of organic matter, Ca and P. This result means that before the redeposition, but after the sedimentation of layer 7, another type of sediment was accumulated in the rockshelter. The original strata did not survive the denudation, but their material has been preserved in a form of layers 6 and 5. The source of this material was most likely situated in the rear zone. This series represents a period of geomorphological instability, when episodes of erosion and colluvial redeposition alternated several times in the rockshelter. The reason for these processes was likely high water supply, possibly linked with humid climate or degradation of the Pleistocene permafrost. The series is topped with an unconformity, which may be regarded as the last record of erosional-colluvial activity. Records of analogous processes were recognized in other cave sites in the region and linked with the lithostratigraphic unit E of the loess-like cave deposits. ### Series V–massive loams Massive (structureless) silts and loams with limestone debris of layers 5a, 4, 3a, 3 and 2 represent another series. No sedimentary macrostructures except for interlayer boundaries are visible, and even these boundaries are indistinct and gradual. This indicates a low sedimentation rate and slow or almost ceased geological transport. High amount of limestone clasts and high Ca content in the fine fraction support a hypothesis of an autochthonous source of the clastic material. A high concentration of P indicates intense zoogenic accumulation of excrements of cave dwelling animals and/or carcasses of their prey. The limestone clasts are smoothed and even rounded-like. Madeyska linked the smoothing process with both mechanical erosion and chemical dissolution by dripwater and a generally humid and warm climate. Layers 5a and 4 are particularly distinct due to their bright color and higher amount of clay. Increased content of clay and yellowish color indicate chemical weathering of limestone, i.e., the dissolution and removal of carbonates, accompanied by deposition of water-insoluble clay residuum enriched in Fe oxides. The bright color and higher amount of clay in cave sediments are regarded as an indicator of warm climate and allow correlating of these layers with the Holocene climatic optimum. In the entrance zone, the stratigraphy is less distinct due to pedogenesis (see the section “Pedogenesis” below). Noteworthy is the presence of large limestone blocks in layer 3. Some of them reach more than 90 cm in diameter. These blocks represent a massive rockfall event. Other large blocks (some of them longer than 1.5 m) protrude from the ground at the plateau in front of the rockshelter and possibly represent the same episode. As the excavation area uncovered only a limited part of this accumulation of blocks, it is difficult to conclude regarding its origin. It could represent the climatic conditions favoring physical weathering and disintegration of rock, i.e., a cold climate, but this is in contradiction with the record of a climatic optimum within the series. The tectonics can be excluded, as there are no signs of recent tectonic activity in the rockshelter, nor in the region. An alternative explanation is a catastrophic event, such as an accidental fall of a heavy block from the upper part of the hill, which could cause a massive avalanche. ### Series VI–dark humus-rich top sediments The uppermost layers 1 and 1a differ from sediments of the underlying series V. The most impressive feature is their dark color, related to increased humus content. The sand fraction is clearly higher. The unusually high concentrations of Pb and other toxic elements (Ag, Bi, Cd, Sb, Sn, Zn) suggest an increased impact of human activity. However, no archaeological artifacts made from these metals were found in the rockshelter. This suggests indirect pollution from the outside. It may be noteworthy that lead, zinc and silver mining in the southern part of Kraków-Częstochowa Upland has been active at least since Middle Ages and is still ongoing. This medieval and modern industry could have polluted the region, e.g., via an eolian activity. Sediments of this series contain detrital material, i.e., bones with quite early radiocarbon dates. These bones likely come from the backfill of archaeological feature, layer A, and testify later disturbance of the grave, possibly by burrowing animals. ## Postsedimentary disturbances The sediments of ShSIII are relatively intact. Cryoturbation and bioturbation have had limited importance and affected only narrow stratigraphic horizons. The most extensive disturbances are those produced by pedogenesis and human activity. ### Cryoturbation Although the thick series of sediments were connected with cold climate (layers 7 and 9), the traces of cryoturbation are restricted in the sedimentary profile. The only distinct disturbances caused by frost action are visible in layers 10 and 11. A complex of these layers together with frost structures are cut by the unconformity and covered by layer 9. This indicates that the cryoturbation event took place before the deposition of layer 9 and the preceding erosion. ### Bioturbation Larger fossil burrows were found inside layers 5a and 5. The size of burrows suggests a constructor of a badger or fox size. These structures were filled with a sediment resembling layer 4. One of badger bones found inside one burrow gave a radiocarbon age of 5.0–4.8 ky BP (Poz-53303). Some further dates obtained for layers 5a and 5 (Poz-61237 and Poz-61305) are consistent with dating of layer 4. These data allow correlating the burrowing activity with the deposition of layer 4 or a period soon after. The presence of a brown bear skeleton suggests that the rockshelter was used by bears as a hibernation den. It cannot be excluded that bears dug the hibernation pits and disturbed the sediments. However, the bones of a bear were found in a secondary position inside the backfill of an anthropogenic feature (layer A). The direct relationships between the bear remains and the sediments cannot be reconstructed. An interesting example of the recent bioturbation are funnel-shape pits in the modern ground surface, dug in the layer 1a and filled with excrements, possibly of a badger. Two such structures were found during excavation, each approximately 20 cm deep, approximately 20–30 cm wide in the upper diameter and narrowing downward. The total volume of excavated excrements was approximately 10–12 liters. These pits can be interpreted as animal latrines. Production and use of such communal latrines is a known behavior of badgers. Such badger activity might be responsible for the lateral scattering of some material from the backfill of the grave (layer A) throughout the rockshelter, and its incorporation into layer 1a. ### Anthropogenic disturbances The first sediment-disturbing human activity in the rockshelter was related to PHA 1 and happened during the Late Paleolithic. It affected the loess layers 7b and 8 in the area close to the rockshelter entrance. The Paleolithic visitors dug an oval pit that served as a fireplace. This trough-like object had a diameter of 60 cm and a depth of 30 cm and must have been filled with loess sediment (layer 8a) soon after humans had left the site. The second distinctive phase of the anthropogenic disturbances is related to the burial of a woman in the late Middle Ages (PHA 4). The grave was constructed in the central zone and covered an area of approximately 2 m² (at squares D/5, D/6, E/5, E/6 and E/7). The length along NE-SW line is approximately 200 cm, whereas the width along NW-SE line is approximately 100 cm, and the greatest depth reaches 60 cm. The object contacts the limestone walls of the rockshelter (the bedrock) from the west and south-west, while several larger stones that can be interpreted as a relic of a purposefully built wall have been found in the entrance zone, at the north side of the grave. The boundaries of the grave cut primarily layers 1, 1a, 2 and 3 and partially layers 4 and 5a in the southern part. The object is filled with sediments named layer A (field numeration “1a/2” and such name was used in a previous paper). Inside the entire backfill of the object, charcoal dust and fine charcoals were dispersed. Lithologically, the backfill resembles layers 1, 1a and 2, whose material was probably used to fill the grave. The backfill does not exhibit any visible stratification. This suggests a quick and single-event deposition of the sediment. Toward the bottom, the sediment becomes lighter in color and more clayey, similar to layer 4. This is probably a result of using the redeposited sediment of layer 4 during the initial filling. Archaeological material within the object is associated with different phases of human activity (PHA 2, PHA 3, and PHA 4, see). It is mixed up and does not exhibit any systematic layout, except of human bones linked to PHA 4. This indicates that medieval people destroyed the earlier objects and used their backfills while constructing the grave. It seems probable that other features (perhaps graves or pits) associated with PHA 2 and PHA 3 were present in the central zone before the Middle Ages. However, the construction of the medieval grave completely consumed these older objects. ### Pedogenesis In the entrance zone, the stratigraphy of the upper part of the sequence is obscured by dark coloration, related to a high content of organic matter. The dark color becomes gradually lighter downward from the top of layer 1 to the bottom of layer 3, which is also marked by the gradual decrease in organic matter. This indicates that the organic matter was dispersed in profile by illuviation. Layer 1 represents an A-horizon (topsoil) and layer 3 represents a B-horizon (cambic) of a cambisol. The development of the soil was (and still is) possible here due to localization outside of the cave, as layers 1 and 3 occur only in the entrance zone, close to the dripline or outside of it. Recently, this area has been covered by forest litter composed mostly of beech, maple and oak leaves. Layer 3a possibly represents a separate, earlier phase of pedogenesis. This stage of soil development was possibly halted by the erosion recorded in the bottom of series V and/or the accumulation of large limestone blocks visible now inside layer 3 (see the section “Series V–massive loams”). Some strata situated inside the rockshelter, namely, layers 1a and A and to a lesser extent layer 2, are also dark-colored and enriched in organic matter. This enrichment can be an effect of occasional input of litter to the interior of the rockshelter or redeposition of the humiferous sediment of layer 1 from the entrance zone (or topsoil material from outside the rockshelter) by wind, sheet wash and/or walking animals and humans. ## Age model ### Chronological frame of the sequence The radiocarbon and IRSL dates indicate the Late Pleistocene and Holocene chronology of the sediments in ShSIII. Dates arranged by layers show a consequent chronological order. The only inconsistency that prevented the OxCal sequence model from running (two dates were removed from the model; see the section “Chronometric data”) is connected with layers 4-5a-5. It is noteworthy that these layers were a subject of bioturbation, as burrows filled with sediments of layer 4 were detected in layers 5a and 5 (see the section “Bioturbation”). That burrowing could be responsible for mixing of the subfossil material and observed chronological disorder. Two dates removed from the model (Poz-61237 and Poz-61305) were achieved on snail shells excavated from layers 5a and 5 but fit well to the set of dates for layer 4. These two dates likely represent intrusive material according to the terminology of Bronk Ramsey, in this case material from layer 4 that had been redeposited to the lower layers through animal burrows. ### Pleistocene sediments No dates are available for the lowermost layers 9, 10 and 11. The chronology of these layers can be approximated as being older than the earliest date accessible for the upper strata, which is 28.9 ± 1.7 ky BP. Layers 7b and 8 clearly represent the Late Pleistocene. The radiocarbon dates fall between 14.5 and 12.5 cal ky BP for both layers. Bayesian modeling estimates the age of lower boundary of this series to be between 16.2 and 13.8 ky cal BP with 95.4% probability (see and the section “Chronometric data”). The IRSL dates indicate much older age of sublayer 7b, approximately 28–21 ky cal BP. These dates likely represent a periglacial conditions of loess accumulation, when biogenic material was not deposited. Loess accumulation in the region, including its eolian deposition within the caves, has been dated to 27–11 ky BP. No dates are available for layer 7a. The modeled statistical age of this layer is between 12.7 and 10.2 ky cal BP (with 95.4% probability; see and the section “Chronometric data”) and correlates with the Younger Dryas stadial and/or Preboreal phase of the Early Holocene. ### Holocene sediments Radiocarbon dates indicate almost continuous accumulation of biogenic material through entire Holocene. The greatest gap in the probability distribution of radiocarbon age falls before 10.5 ky cal BP, i.e., in the beginning of Holocene. Series IV (layers 5 and 6) includes the oldest Holocene sediments. The series was deposited during relatively short time interval according to radiocarbon dates, which lasted for several hundred up to a maximum of three thousand years. The IRSL date achieved for layer 6 shows a much older age of clastic material than radiocarbon dating of animal remains and corresponds with the age of layer 7b. Regarding the colluvial origin of the series, this date can likely indicate a colluvial redeposition of the material of layer 7b, which has been incorporated into layer 6. A relatively early age is also indicated by the IRSL date for layer 5a. This age is earlier than any other date obtained for layer 7b and may possibly record another source of material (layer 9 or sediments occurring outside the rockshelter?). It is difficult to conclude on the sedimentary process responsible for accumulation of this material. No macroscopic traces of colluvial transport were identified. The important role of zoogenic deposition (see the section “Series V–massive loams”) suggests that deposition of dust or mud by walking and trampling animals could be one of the accumulation processes. The longest time interval is recorded by layer 4. Its deposition lasted for 4–5 thousand years according to the Bayesian model. This interval corresponds to the late part of the Early Holocene and the entire Middle Holocene according to the formal stratigraphic subdivision of the Holocene. It also corresponds to the entire Atlantic climatic chronozone. Layer 3a from the entrance zone correlates with layer 4 according to the dating results and similarities in fossil rodent fauna, while the correlation between layers 3 and 4 is suggested by fossil rodent and mollusk fauna (see the sections “Malacological succession” and “Rodent succession”). Layer 2 corresponds with Late Holocene. Poor chronometric data are available for layers 1-1a. However, assuming that the archaeological feature corresponding to PHA 4 cuts these layers and the lower boundary of PHA 4 was estimated to be 2.0–0.7 ky cal BP, the accumulation of layers 1-1a started before that phase. This dating is in good agreement with the modeled boundary of layers 2 and 1a, which is 2.9–2.4 ky cal BP. ## Paleoecology and reconstruction of paleoenvironment The abundance of fossil material makes the ShSIII an important paleontological site. Although Holocene paleofaunas were reported from other caves of southern Poland and adjacent countries, only few of these sequences go back as far as the final part of the Last Glaciation\[–,\]. Most of the profiles are only a few dozens of centimeters thick and cover only the middle or just the youngest part of the Holocene\[–\]. Despite some disturbances noted in the narrow stratigraphic horizons in the ShSIII, the composition of the mollusk and rodent faunas appears to be environmentally coherent, which makes both groups good paleoenvironmental indicators. ### Reliability of paleoenvironmental reconstruction The mollusk thanatocoenoses in caves represent well the diversity of habitats in the surroundings, comprising shells of species penetrating the cave, of those living on the surface near the cave entrance and the shells dropped from the surrounding rocks during sedimentation. The rather good state of preservation of shells in ShSIII suggests that they did not undergo any intensive transportation but rather were accumulated in sediments in the primary place of deposition after the mollusk death. The rodent remains were deposited mostly from pellets dropped by carnivorous birds, such as owls resting in a cave. Rodents likely could have inhabited a range of habitats surrounding the cave, which had been used by the birds as their hunting territories. Therefore, rodents represent a wider area than mollusks. It is also expected that predatory birds prefer open habitats over canopy for use as hunting grounds, so as a consequence, the abundance of open-habitat rodents can be overrepresented in the fossil record. Bats are unique among the fossil fauna, as they are troglophiles who live inside caves. Some of the recorded taxa are known to hibernate in caves, while the others use caves as their summer roosts. Bats are also migrants and some species move between their winter and summer areas at the distance of over hundred kilometers. Thus, species who used the caves as their hibernacula do not represent the habitats surrounding the cave. This may be the case for *M*. *dasycneme*. This species is known to hibernate in caves, so its presence in ShSIII is possibly an effect of death during winter. The water bodies used by this species as hunting grounds were located in its summer area, which could be far from the site. ### Late pleistocene The oldest assemblages of mollusks and rodents (Vt and Dt, respectively) found in layers 7b-8-7a represent a rather poor fauna. Both assemblages are composed of species of cold climate, known from tundra and steppe biomes. Cold-tolerant snails *Vallonia tenuilabris* and *Discus ruderatus* indicate a cool continental climate and predominance of open habitats around the cave. The same habitats are preferred by lemmings. Dry open environments, i.e., steppe, are also indicated by the presence of *C*. *cricetus* and *S*. *subtilis*. However, some patches of cool, humid and shady habitats favorable for *S*. *kotulae* also occurred, which indicates a mosaic of habitats. The rough and hilly topography of Kraków- Częstochowa Upland certainly was a factor favoriting the diversity of habitats. This is confirmed by regional paleobotanical data, which indicate that during Late Glacial of the Last Glaciation the clusters of trees dominated by pine, spruce and larch occurred beside the rocky slopes occupied by steppe-like communities. The most informative is layer 8 with the highest number of shells. The occurrence of *S*. *kotulae* and *D*. *ruderatus* and lack of species typical of arctic climate, e.g., *Pupilla loessica*, *Vertigo genesii* and *Vertigo geyeri*, may indicate Bølling-Allerød interstadial. Similar malacofaunas of this age are known from calcareous tufas and landslide deposits of southern Poland\[,–\]. The rodent remains, though still low in number, follow the mollusk trend. They document higher abundance of woodland and humid environments during deposition of layer 8 in comparison to layers 7a and 7b. Additionally, the first appearance of a woodland species *C*. *glareolus* is recorded in layer 8. ### Pleistocene-holocene transition Rich and diverse mollusk fauna of DrVc assemblage points to the beginning of the Holocene and amelioration of the climatic conditions compared to the Late Glacial. Radiocarbon dating of layers 6 and 5 from the central zone and 3a from the entrance zone indicate the Preboreal and Boreal phases of Early Holocene, and possibly the very beginning of the Atlantic phase. The mollusk composition resembles the *Ruderatus*-fauna known across Europe from many terrestrial sequences between the Preboreal and the older part of the Atlantic phase\[,–,–\]. Expansion of shade-loving taxa, namely, *Discus ruderatus*, suggests growth of taiga-type coniferous forests, but some grasslands and sunny slopes preferred by *Vallonia costata* also occurred around the site. If shells have not been reworked, the environmental conditions in the beginning of Holocene favored the longer occurrence of glacial relics *Semilimax kotulae* and *Vallonia tenuilabris*. The last one was noted in Preboreal deposits in the nearby Cave above the Słupska Gate. Gradual disappearance of glacial relics may correspond to gradual climate warming and some oceanic influences inferred from the first appearance of *Sphyradium doliolum*, *Discus rotundatus*, *Aegopinella pura* and other species of warm and humid conditions. The rodent assemblage CgLg found in layer 6 is still not abundant; however, it exhibits the highest Simpson index in the sequence. This reflects the high diversity of species and of habitats, as both the species of forest and of open landscape occur. Like in the case of snails, the rodent glacial relics survived, namely, *Dicrostonyx torquatus* and *Lasiopodomys gregalis*. According to Nadachowski, the glacial relics could survive in the region until the beginning of Late Holocene. Layer 6 is marked by the first appearance of *M*. *subterraneu*s and genus *Apodemus* in the sequence. The first one was found in Late Glacial deposits of Shelter above the Niedostępna Cave, which is situated approximately 26 km away. In ShSIII this species is lacking in strata of that age and becomes abundant starting from layer 5 upward. This observation suggests reconsidering the early appearance of *M*. *subterraneus* in Shelter above the Niedostępna Cave, which could be an effect of postdepositional reworking. In another site with well-preserved Holocene deposits, Nad Mosurem Starym Duża Cave, its first appearance is even later, estimated to be ca. 4 ky BP. Rodents of layer 5 record rapid improvement of environmental conditions. This is reflected by the increase of the remain number and higher amount of forest dwellers, such as *C*. *glareolus*, red squirrel *Sciurus vulgaris* and glirids *Eliomys quercinus* and *Muscardinus avellanarius*. Interesting issue is a stratigraphic (and time) shift between the start of dominance of forest fauna in the mollusk and rodent records. The pick of forest- dwelling mollusks begins in layer 6 (or in layer 3a in the entrance zone), while in the case of rodents, only in layer 5 (or 3 in the entrance zone). This situation may be an effect of taphonomic biasing. As the accumulation of rodents prefers the taxa of open landscape (see the section “Reliability of paleoenvironmental reconstruction”), the dominance of forest rodents in the taphocenosis could have been intense since the forested area was virtually the only habitats in the surroundings, although patches of forest inhabited by forest rodents had been abundant earlier. The lack of chiropteran remains in the lower layers and their scarcity in layers 6 and 5 may be an effect of the unfavorable environmental conditions. ### Holocene climatic optimum Predominance of mixed and deciduous forests around the cave during sedimentation of layers 5a, 4 and 3 is evidenced by DroAp mollusk assemblage, CgMs rodent assemblage, PaurMbra chiropteran assemblage and radiocarbon dating (9.5–4.5 ky BP). *Discus ruderatus* was replaced by more demanding *D*. *rotundatus*, *D*. *perspectivus* and other strictly forest species, e.g., *A*. *pura*, *Balea biplicata*, *Ruthenica filograna* and *Isognomostoma isognomostomos*. This records the progressive warming and was likely connected with transition from continental to oceanic circulation across Central Europe and the climatic optimum of the Holocene–the Atlantic phase\[,–\]. These data are in agreement with floristic changes in Central Europe. The climatic optimum is indicated by the presence of *Vestia elata*, which has narrow ecological tolerance and favors warm and humid conditions. This snail has been described from Atlantic phase at other sites in the region and disappeared during the Subboreal phase. Today, only limited relict sites with this species are noted in Poland in the Beskidy, Bieszczady and the Holy Cross Mountains. The composition of rodent taphocenosis in layers 5a, 4 and 3 is similar to that in layer 5. All these layers are characterized by the highest richness of taxa. The important difference in comparison to layer 5 is much higher number of remains. Simpson index of diversity is lower than in most of other layers, which is a result of strong predominance of two species, *C*. *glareolus* and *M*. *subterraneus*, both typical for woodlands. The chiropteran assemblage PaurMbra, found in these layers, is also among of the richest in the sequence, both in terms of number of taxa and number of remains. The thanatocoenosis is characterized by taxa which prefer cold caves as their hibernacula, with temperature below 6°C (which stays in accordance with small size of the ShSIII and testifies that the shelter already had than a similar size to the modern one), and big trees as summer roosts (*B*. *barbastellus*, *M*. *bechsteinii* and *P*. *auritus*). This suggests the presence of old-growth forest with sparse open areas (because of the presence of *Eptesicus serotinus*). The occurrence of *Rhinolophus hipposideros*, a species that typically prefers warmer caves to hibernate, could be limited to the summer period. ### Post-optimum deterioration of habitats Starting from approximately 5 ky BP, the malacofauna of ShSIII is dominated by mesophilous species. The forest malacofauna is still diverse but significantly outnumbered, mostly by *Laciniaria plicata*. The overall tendency of an increasing number of gastropods of wide ecological tolerance in costs of shade- loving taxa suggests a gradual deforestation and exposure of rocky walls around the cave. A similar situation is recorded in other Subboreal and Subatlantic sequences from caves and rockshelters of Central Europe. The upper part of the sequence is characterized by a high abundance of edible dormouse *Glis glis*, which is the third most numerous rodent species in layers 2 and 1a. Its abundance probably coincided with a change in forest cover around the cave and rise in mature beech forest, which is a preferred habitat of this species. Beech forest also constitutes the main type of the recent natural vegetation in the microregion. The expansion of this tree into the region was quite late, the beech forests appeared here approximately 3.5 ky BP. This dating corresponds well with the presumed chronology of the mentioned layers. The uppermost layers 1 and 1a record also the first appearance of *M*. *arvalis* and *Micromys minutus*. The arrival of both these species was most likely of synanthropic character, as they prefer artificial landscapes (agricultural fields and meadows). A similar observation of the appearance of *M*. *arvalis* during the late part of Late Holocene was noticed from other sites in the region. Particularly, the Holocene profile of Nad Mosurem Starym Duża Cave shows the same pattern of rodent succession. The presence of single remains of tundra species (*Dicrostonyx torquatus*) in layer 1a may be an effect of post-depositional disturbances. Chiropteran fauna notes decline in richness and number of specimens, following the trend in mollusks and rodents. There is a clear predominance of taxa foraging in wooded areas (i.e., *B*. *barbastellus*, *M*. *bechsteinii*, *P*. *auratus* and *M*. *brandtii*). The presence of *M*. *myotis* can be related to a lack of undergrowth in the surrounding forest, while *M*. *mystacinus*, *M*. *nattereri*, *M*. *emarginatus* and *P*. *austriacus* suggest the presence of open and/or diversified areas, too. All these data likely indicate deterioration of habitats, most likely of anthropogenic origin. ## Chronology of human activity Four episodes related to human activity have been recorded at ShSIII. The trace of the oldest is the relic of a Late Paleolithic hearth (PHA 1), probably related to the short-term camp. Another episode (PHA 2) is associated with a possible burial containing human and animal remains, which has been fragmentarily preserved in the secondary deposit inside the backfill of the medieval grave, and whose radiometric dating implies the activity of an Early Bronze Age community. The next phase of human stay (PHA 3) happened in the late pre-Roman period. Fragments of the vessel and the animal deposit were all redeposited into the medieval grave. Their context and nature imply the sepulchral character of human activity from 4th to 1st century BC. Typology of the vessel based on stylistics and comparison with the other sites indicate the second half of the 1^st century BC. Radiocarbon dating of the residue collected from the inner surface of the vessel fragment and of the wildcat mandible with cut marks indicate somehow earlier chronology, between the 4^th and 1^st centuries BC. The youngest phase of human activity in the shelter (PHA 4) is a late Middle Age burial. The cultural episodes identified at the rockshelter have numerous analogies in the nearest area. The Early Bronze Age sites are common in the caves of the region. The archaeological material (mainly pottery and flint artifacts associated with the production of tetrahedral axes) indicates that human activity in the other caves in the region was limited primarily to the plateaus outside of the caves or elevated places over the caves\[,–\]. During the late pre-Roman period, the human activity in the caves of the Kraków-Częstochowa Upland increased. Ceramic vessels are the most common artifacts in other sites, but coins, elements of belts, fibulae and toiletry objects have also been found. Typically, the archaeological inventories suggest the economic and utilitarian meaning of caves for those people. It should be emphasized that the nature of deposits in the ShSIII, i.e., a buried vessel and a sepulchral deposit of the European wildcat (*Felis silvestris*), make them unique among the Celtic sites in the region. There are many examples of human activity in the caves of Kraków-Częstochowa Upland during the Middle Ages. It is highly probable that the medieval burial in ShSIII could be related to the medieval castle Smoleń situated on a neighboring hill and the nearby village Pilica. However, the mysterious fact is a human burial inside a cave. This happened several centuries after the Christianization of the region, when cemeteries were commonly used as burial places. Considering the small space of the rockshelter, the multicultural deposits and their sepulchral nature indicate the supermaterial significance of the place from Antiquity to the Middle Ages. It seems that the rockshelter has been visited for sacral or magical purposes. ## Shelter in Smoleń III as a lithostratigraphic stratotype The almost two meters-thick sequence of Late Glacial and Holocene sediments in ShSIII is unique among the caves of Kraków-Częstochowa Upland and the caves of Central Europe in general. Although Upper Holocene cave sediments are quite common in the southern Poland (e.g.,), the Lower Holocene is preserved only on very rare occasions. In case of ShSIII, the entire profile of Lower-Middle-Upper Holocene is preserved, which is an exceptional situation. Taking this into consideration, we propose to regard ShSIII as a regional stratigraphic stratotype of Holocene clastic cave sediments. The synthetic stratigraphy of ShSIII is presented in. The lithology and geochemistry of sediments allow reconstruction of the depositional conditions and can be regarded as a reference for future studies of cave sediments in the region. The main lithostratigraphic similarities shared by ShSIII with other sites in the region include: 1. Loess deposits in the lower part of the sequence (series III). Loess and loess-like sediments are widely known from the near-entrance facies of caves and rockshelters in Central Europe. A synthetic approach to their lithostratigraphy has been presented for Kraków-Częstochowa Upland, and lithostratigraphic units ‘A’, ‘C’ and ‘E’ of that scheme can be identified in ShSIII. 2. A dark-colored, humiferous stratum representing Bølling-Allerød interstadial (i.e., layer 8 here), which forms an intercalation inside a loess packet. Similar strata have been detected in Pekárna, Barová and Nová Drátenická caves in Czech Republic and in Rockshelter in Zalas, however, in the latter the layer occurs within sands. Their age was confirmed by radiocarbon dating (16.2–14.2 in Pekárna Cave; 17.2–13.2 ky BP in Nová Drátenická Cave; and 14.6–12.6 ky BP in Rockshelter in Zalas) and is similar to radiocarbon age of layer 8 in ShSIII. 3. Colluvial activity recorded in Lower Holocene (series IV). Traces of colluvial processes have been identified in similar stratigraphic position (at least, in the lower part of the Holocene series) in Cave above the Słupska Gate, in Zawalona Cave, in Rockshelter in Zalas, and in Nad Mosurem Starym Duża Cave. 4. Rusty-colored stratum with increased amount of clay in the Middle Holocene (here defined as series V). Recorded by layers 4 and 5a in ShSIII, this unit has analogies in Cave above the Słupska Gate, Rockshelter in Zalas and in Pekarna Cave in Czech Republic. 5. Dark-colored humus-rich sediments at the top of a sequence (series VI). Similar strata are known from the most of caves and rockshelters in the region. However, they usually represent an entire Holocene sequence. The case of ShSIII shows that this situation is most likely an effect of removal of older Holocene sediments, followed by the most recent accumulation of series VI. The additional value of this site is that its stratified lithostratigraphic sequence contents an abundant mollusk shells and considerable number of vertebrate remains. Cave sites with long profiles of subfossil fauna of mollusks, rodents and chiropterans are unique for Central Europe and present a great value for reconstructions of paleoecology in the region. The composition and structure of the mollusk and rodent assemblages and to lesser extent also chiropterans, allow reconstructing the changes in paleofauna, which reflected the changes of climate and vegetation around the cave. Several mid- European terrestrial sequences record changes in faunal succession of Pleistocene–Holocene transition and/or the entire Holocene (e.g.,). Among them, those with relatively rich paleontological material are in minority. Long and rich malacological successions are known from Barová Cave in Czech Republic, Veľká Drienčanská Cave in Slovakia and Petény Cave in Hungary. North to Carpathian Mountains, such records are known from Shelter over the Słupska Gate, Zawalona Cave and Duża Cave at Mączna Skała. Long and rich rodent successions are known from Barová Cave in Czech Republic, Nad Mosurem Starym Duża Cave, Rockshelter in Zalas and Zawalona Cave. Despite rather local character of the fauna diversity in those sequences, some general traits in transformation of fauna and their habitats may be distinguished throughout Central Europe. They show some common features with the sequence under study, such as: 1. Poor faunal assemblages of the Late Glacial of the Last Glaciation and Early Holocene. 2. Appearance of warm-adapted fauna in Preboreal and Boreal phases of the Early Holocene, recording the climatic warming and development of coniferous forests with patches of open habitats. 3. Deciduous and mixed forests of the climatic optimum, reflected especially in rich taphocenosis of snails and bats, with species of higher thermal and moisture requirements. 4. Deterioration of natural vegetation during the Late Holocene and increasing anthropopressure, visible in the appearance of species inhabiting open landscapes and synanthropic rodents. Increasing anthropopressure in the region is also marked by the presence of three human activity phases (PHA 2, PHA 3 and PHA 4) from Late Holocene, recorded in this small rockshelter in opposition to lack of any archaeological record from the Early or Middle Holocene. # Conclusions Almost two meters-thick sedimentary sequence of Late Glacial and Holocene sediments in ShSIII is unique among the mid-European caves, and among the caves of its northern part (Carpathian foreland) in particular. Although Holocene sequences are known from other sites in Central Europe, only on rare occasions they have been set in the certain chronological framework. Here we present a long sequence of clastic cave sediments which is also well-dated, with 29 radiocarbon and 4 IRSL dates provided and with Bayesian model of the chronology of stratigraphic units. This makes the ShSIII a key site for the understanding the regional Holocene lithostratigraphy. Based on the lithology and geochemistry of sediments we reconstruct the depositional conditions offering a reference for future studies of cave sediments in the region. We present also an abundant paleontological material, which records significant changes in paleofauna reflecting climate and environmental changes in the cave surroundings. The unusually long faunal succession places ShSIII among the crucial sites providing the paleoenvironmental and paleoecological proxies for the Holocene of the Carpathian foreland. Taking all these facts into consideration, we propose to regard ShSIII as a regional stratigraphic stratotype of Holocene clastic cave sediments. The well-dated lithological sequence of ShSIII will serve as an important reference for future inter-site regional correlations. In particular, this creates an opportunity for archaeological regional studies, especially that archaeological cave sites with Holocene sediments are still excavated in Kraków- Częstochowa Upland\[–\]. Because ShSIII is a small rockshelter, the archaeological excavations consumed the greater part of its sediments. After the excavations, the profiles of unexcavated sections have been protected and the archaeological pits carefully backfilled with stones and previously sieved sediment. This makes the site secured and ready for eventual re-excavation in future, including sampling for further analyses. However, its unique scientific value makes it especially important to elaborate the project of legal protection for this site. The initial actions have already been made, as the surroundings of the site are included in the ‘Natura 2000’ protected area (*Ostoja Środkowojurajska*, site code PLH240009). Despite this, further particular protection of the cave is needed, especially in terms of the future scientific research and access to secured profiles. # Supporting information We are thankful to the officers of the Polish National Forests for their help with the organization of excavations; and to the students of archaeology from Nicolaus Copernicus University in Toruń who participated in the excavation works. 10.1371/journal.pone.0228546.r001 Decision Letter 0 Louys Julien Academic Editor 2020 Julien Louys This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 19 Nov 2019 PONE-D-19-27803 Shelter in Smoleń III – a unique example of stratified Holocene clastic cave sediments in Central Europe, a lithostratigraphic stratotype and a record of regional paleoecology PLOS ONE Dear Dr. Krajcarz, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ============================== While reviewer 1 entered a 'reject' recommendation, this was only because they felt unqualified to examine the bulk of your manuscript. However, they found your study well-written and thorough, an assessment backed up by Reviewer 2's more extensive analysis. Nevertheless, both reviewers indicated that your manuscript lacked some important contextual information that would make it more relevant to general readers. In particular, mention is made of the importance of this site for archaeological and palaeoenvironmental research, but these areas are only minimally addressed in the discussion. They should be more thoroughly introduced at the beginning, with an effort made in the conclusions to tie the major findings of this research within this context.  ============================== We would appreciate receiving your revised manuscript by Jan 03 2020 11:59PM. When you are ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: <http://journals.plos.org/plosone/s/submission- guidelines#loc-laboratory-protocols> Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'. Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out. We look forward to receiving your revised manuscript. Kind regards, Julien Louys Academic Editor PLOS ONE Journal Requirements: 1\. When submitting your revision, we need you to address these additional requirements. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_m ain_body.pdf> and <http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_f ormatting_sample_title_authors_affiliations.pdf> 2\. In your manuscript, please ensure that you have reported specimen numbers and complete repository information, including museum name and geographic location. For more information on PLOS ONE's requirements for paleontology and archaeology research, see <https://journals.plos.org/plosone/s/submission-guidelines#loc- paleontology-and-archaeology-research>. 3\. Thank you for stating the following financial disclosure: The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript. a\) Please provide an amended Funding Statement that declares \*all\* the funding or sources of support received during this specific study (whether external or internal to your organization) as detailed online in our guide for authors at <http://journals.plos.org/plosone/s/submit-now>. b\) Please state what role the funders took in the study.  If any authors received a salary from any of your funders, please state which authors and which funder. If the funders had no role, please state: "The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript." Please include your amended statements within your cover letter; we will change the online submission form on your behalf. 4\. We note that Figures 1 and 3 in your submission contain \[map/satellite\] images which may be copyrighted. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For these reasons, we cannot publish previously copyrighted maps or satellite images created using proprietary data, such as Google software (Google Maps, Street View, and Earth). For more information, see our copyright guidelines: <http://journals.plos.org/plosone/s/licenses-and-copyright>. We require you to either (1) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (2) remove the figures from your submission: 1\. You may seek permission from the original copyright holder of Figures 1 and 3 to publish the content specifically under the CC BY 4.0 license.  We recommend that you contact the original copyright holder with the Content Permission Form (<http://journals.plos.org/plosone/s/file?id=7c09/content- permission-form.pdf>) and the following text: “I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 (<http://creativecommons.org/licenses/by/4.0/>). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.” Please upload the completed Content Permission Form or other proof of granted permissions as an "Other" file with your submission.  In the figure caption of the copyrighted figure, please include the following text: “Reprinted from \[ref\] under a CC BY license, with permission from \[name of publisher\], original copyright \[original copyright year\].”  2. If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only. The following resources for replacing copyrighted map figures may be helpful: USGS National Map Viewer (public domain): <http://viewer.nationalmap.gov/viewer/> The Gateway to Astronaut Photography of Earth (public domain): <http://eol.jsc.nasa.gov/sseop/clickmap/> Maps at the CIA (public domain): <https://www.cia.gov/library/publications/the- world-factbook/index.html> and <https://www.cia.gov/library/publications/cia- maps-publications/index.html> NASA Earth Observatory (public domain): [http://earthobservatory.nasa.gov/](http://earthobservatory.nasa.gov/%20) Landsat: <http://landsat.visibleearth.nasa.gov/> USGS EROS (Earth Resources Observatory and Science (EROS) Center) (public domain): [http://eros.usgs.gov/#](http://eros.usgs.gov/) Natural Earth (public domain): <http://www.naturalearthdata.com/> \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Partly Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: I Don't Know Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: The site seems to be an important one for the thickness of its Late Quaternary deposits, quality and diversity of environmental proxy data, large number of good C14 and other radiometric dates, etc. It is well written (with only minor errors of colloquial English), excellently illustrated and quite adequately supported. But, as an archeologist of the Pleistocene and earliest Holocene, I was disappointed at the virtual absence of archeological (cultural) information or analysis. Nor am I a Central European prehistory specialist. So I am not an appropriate reviewer. My choice by PLOS-1 was not correct. However, I cannot help but notice that this ms. is quite narrowly descriptive; neither the statement of purpose nor the conclusions really tell the reader why this site is important in the overall context of southern Poland or Central Europe generally. What does the site tell us about human activity, adaptations or behavior (especially in a broader context). I imagine that anthropological interpretation was not the authors' goal, but it is a shame that such an interesting site is not more generally contextualized in terms of humans and their cultures in this region throughout the course of the early Holocene. The Editors should seek appropriate reviewers who are relevant natural scientists. Reviewer \#2: The manuscript entitled Shelter in Smoleń III – a unique example of stratified Holocene clastic cave sediments in Central Europe, a lithostratigraphic stratotype and a record of regional paleoecology, deals with the stratigraphy of the archeologically valuable rock shelter. Sediment deposits have been analyzed in detail (morphological classification of limestone clasts, chemical composition...). 29 radiocarbon dating of faunal and human remains and charcoals cover the interval from 14.1 to 0.6 ky BP. Taxonomic composition of fossil faunal assemblages was used to obtain insight to the reconstruction of local paleoenvironment. I would like to compliment the authors for their great effort collecting and analyzing such an important amount of data and effectuating a valuable multidisciplinary research. In cave entrances and especially in rock shelters, where sedimentation rates are high and artifact preservation can be remarkable, many aspects and processes must be considered, and the authors managed that requests very well. The title is appropriate for the content of the text; hence the abstract could give more insight to the findings. Furthermore, the article is well constructed, the sampling and pitting were properly conducted, and analyses used are appropriate and well performed. Methodology and results are well described and accounts for large part of the manuscript. However, there are several issues that should be considered before publishing. \- The manuscript is clearly written. Still, several sentences are confusing in structure or word use. I suggested some changes but it would be good to edited the text by a native English speaker. \- Introduction should give better overview on the topic and the current state of the art in the field. \- The results are described in detail and are well interpreted, but a bit hard to follow. \- Discussion is structured in several sections, it is well explained but there is a clear need of more synthetic interpretations. All sections could be better linked together through time, discussed climatic events, paleo environmental changes and anthropogenic activities. \- Some statements should be more clearly explained and supported by references (for details please see the manuscript) \- I suggest moving the fig. 12 from Conclusion to Discussion, even to separate (sub)section which could be built around it. \- The manuscript would benefit putting the findings in broader context – regional settings (Central Europe, differences and similarities) with previous research. \- If the authors accept this suggestions Conclusion than could be restructured, giving more value to the authors results. Overall, the research itself brings new paleoenvironmental data from a complex climate region, such as Central Europe, and contributes to better understanding of local climate dynamics and anthropogenic activities during the Holocene. However, more synthetic views would substantially increase the value of the presented results and consequently manuscript. Additional comments can be found in the manuscript. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0228546.r002 Author response to Decision Letter 0 14 Jan 2020 Response to reviews We are very thankful to Reviewers for their careful review and many constructive comments. We added the corrections to text according to these suggestions. We also found some minor mistakes (typos) on some figures, and now we submitted the modified versions. Below, we provide our detail responses to Reviewers and explain our corrections. The manuscript with our changes in a review mode is also attached as a separate file. Reviewer \#1: COMMENT: The site seems to be an important one for the thickness of its Late Quaternary deposits, quality and diversity of environmental proxy data, large number of good C14 and other radiometric dates, etc. It is well written (with only minor errors of colloquial English), excellently illustrated and quite adequately supported. But, as an archeologist of the Pleistocene and earliest Holocene, I was disappointed at the virtual absence of archeological (cultural) information or analysis. Nor am I a Central European prehistory specialist. So I am not an appropriate reviewer. My choice by PLOS-1 was not correct. However, I cannot help but notice that this ms. is quite narrowly descriptive; neither the statement of purpose nor the conclusions really tell the reader why this site is important in the overall context of southern Poland or Central Europe generally. What does the site tell us about human activity, adaptations or behavior (especially in a broader context). I imagine that anthropological interpretation was not the authors” goal, but it is a shame that such an interesting site is not more generally contextualized in terms of humans and their cultures in this region throughout the course of the early Holocene. The Editors should seek appropriate reviewers who are relevant natural scientists. ANSWER: We agree that the site has a potential for archaeological interpretations. However, our purpose was to provide in this paper the environmental data (paleoecology and lithostratigraphy), which can be used in future for more detail inter-site correlations. This is also our plan for the future to present such correlations for the entire region of Kraków-Częstochowa Upland, and further, to set archaeological data within this stratigraphic context. Only then we would be able to present a coherent and reliable history of human activity in the area. And that’s our plan, but it needs several steps before, and this paper is thought to be one of them. Now, also to address to the Reviewer’s \#2 comments, we have added new sections to the Introduction and Discussion and re-built the Conclusions, to provide a broader context of our study. Reviewer \#2: COMMENT: The manuscript entitled Shelter in Smoleń III – a unique example of stratified Holocene clastic cave sediments in Central Europe, a lithostratigraphic stratotype and a record of regional paleoecology, deals with the stratigraphy of the archeologically valuable rock shelter. Sediment deposits have been analyzed in detail (morphological classification of limestone clasts, chemical composition...). 29 radiocarbon dating of faunal and human remains and charcoals cover the interval from 14.1 to 0.6 ky BP. Taxonomic composition of fossil faunal assemblages was used to obtain insight to the reconstruction of local paleoenvironment. I would like to compliment the authors for their great effort collecting and analyzing such an important amount of data and effectuating a valuable multidisciplinary research. In cave entrances and especially in rock shelters, where sedimentation rates are high and artifact preservation can be remarkable, many aspects and processes must be considered, and the authors managed that requests very well. The title is appropriate for the content of the text; hence the abstract could give more insight to the findings. Furthermore, the article is well constructed, the sampling and pitting were properly conducted, and analyses used are appropriate and well performed. Methodology and results are well described and accounts for large part of the manuscript. However, there are several issues that should be considered before publishing. \- The manuscript is clearly written. Still, several sentences are confusing in structure or word use. I suggested some changes but it would be good to edited the text by a native English speaker. ANSWER: We are thankful for this advice. However, our manuscript has been sent before the submission to the company (American Journal Experts) which makes professional corrections of scientific texts, and has been extensively corrected by their native speakers who also are the experts in the field of this paper. COMMENT: - Introduction should give better overview on the topic and the current state of the art in the field. ANSWER: Now we have provided a wider background according to this comment. COMMENT: - The results are described in detail and are well interpreted, but a bit hard to follow. ANSWER: Nevertheless, we would like to keep the current structure of the paper. It is quite long, so we believe it will be much easier for readers who are focused only on particular area of interest to find the particular section (for example, data on fossil rodents) if structured like it is now. COMMENT: - Discussion is structured in several sections, it is well explained but there is a clear need of more synthetic interpretations. All sections could be better linked together through time, discussed climatic events, paleo environmental changes and anthropogenic activities. ANSWER: Now we have added new section to the Discussion to provide more synthetic approach. This also fills the requirement from the next comments. COMMENT: - I suggest moving the fig. 12 from Conclusion to Discussion, even to separate (sub)section which could be built around it. ANSWER: We agree with this advice. So now, we rebuilt the Discussion using parts of former Conclusion (including Fig. 12) and this gave also the linkage between different sections of Discussion and we believe this provided more synthetic approach. COMMENT: - The manuscript would benefit putting the findings in broader context –; regional settings (Central Europe, differences and similarities) with previous research. ANSWER: Now our new sections in Introduction and in Discussion address to this constructive suggestion. COMMENT: - If the authors accept this suggestions Conclusion than could be restructured, giving more value to the authors results. ANSWER: Now we have re-built Discussion and Conclusions according to the comments. COMMENT: Overall, the research itself brings new paleoenvironmental data from a complex climate region, such as Central Europe, and contributes to better understanding of local climate dynamics and anthropogenic activities during the Holocene. However, more synthetic views would substantially increase the value of the presented results and consequently manuscript. \- Some statements should be more clearly explained and supported by references (for details please see the manuscript) Done COMMENT: Additional comments can be found in the manuscript. ANSWER: We corrected manuscript according to the advices provided by Reviewer in the attached PDF. According to the Reviewer’s \#2 questions: \- line 867 – here we would like to say generally about the process, without referencing to chronological periods \- line 883 – there are no signs of neotectonics in the rockshelter nor in the area. We inserted the appropriate information into the text. \- line 1069 – we did not identified fossil guano. However, the micromorphological analyses has not been implemented yet into this site, which in future can help to resolve this problem. 10.1371/journal.pone.0228546.r003 Decision Letter 1 Louys Julien Academic Editor 2020 Julien Louys This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 21 Jan 2020 Shelter in Smoleń III – a unique example of stratified Holocene clastic cave sediments in Central Europe, a lithostratigraphic stratotype and a record of regional paleoecology PONE-D-19-27803R1 Dear Dr. Krajcarz, We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements. Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication. Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at <https://www.editorialmanager.com/pone/>, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. With kind regards, Julien Louys Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0228546.r004 Acceptance letter Louys Julien Academic Editor 2020 Julien Louys This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 31 Jan 2020 PONE-D-19-27803R1 Shelter in Smoleń III – a unique example of stratified Holocene clastic cave sediments in Central Europe, a lithostratigraphic stratotype and a record of regional paleoecology Dear Dr. Krajcarz: I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. For any other questions or concerns, please email <plosone@plos.org>. Thank you for submitting your work to PLOS ONE. With kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Julien Louys Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Classical ecological theory posits that changes in the availability of nitrogen, an essential resource, can lead to changes in the species composition of plant communities due to differences in species-specific nutrient requirements. This understanding is well-supported by studies of short-lived plant species for which long-term studies provide data on multiple generations of plant populations (ex. grasslands). Yet, our understanding of the responses of long- lived species such as trees to changes in nitrogen availability is based primarily on studies of the growth response of individual trees or seedlings and saplings to experimental changes in nitrogen availability, that are scaled-up to the population level. Such studies suggest that tree population growth is strongly dependent on available nitrogen and that populations have strong self- reinforcing feedback effects on this resource via differences in plant litter chemistry. However, it remains unclear whether growth responses by individual trees are a good predictor of expected population-scale patterns given changes in the availability of essential nutrients. Given that future climate warming is expected to increase rates of nitrogen cycling, it is essential to assess the effect of climate-induced changes in nitrogen cycling on changes on tree population dynamics over time scales appropriate for long-lived plant species (i.e. 100's to 1,000's of years). Here, we use a mechanistic modelling analysis of palaeoecological data to determine if – during an interval of rapid climate change – changes in nitrogen dynamics lead to vegetation change or if community composition alters nitrogen dynamics. Long-term ecological data from the fossil record yields reliable proxies of plant population dynamics and changes in landscape-scale nitrogen cycling. Recent palaeoecological studies have shown a high level of concurrence between changes in the nitrogen cycle and shifts in vegetation composition. When these changes occur concurrently it is not possible to assess the direction of causation; therefore, it is essential to analyze the mechanistic relationships underlying the changes in vegetation composition and nitrogen cycling. We aimed to infer this mechanistic relationship and to test the stability of it over a threshold change in climate and a change in the dominant tree species on the landscape using a model-fitting and model-selection analysis of long-term ecological data. We used palaeoecological data spanning 8,000 years from Kis-Mohos Tó in Northeast Hungary that was interrupted by a threshold change in climate involving abrupt warming of 10°C over a 60 year period. We tested the goodness of fit of three sets of candidate mechanistic models for explaining the dynamics of two dominant tree populations, *Pinus spp.* (pine) and *Quercus spp.* (oak), with changes in the nitrogen cycle: nitrogen-dependent population growth, nitrogen-dependent population growth with a positive feedback effect and nitrogen independent population growth with a positive feedback effect (see for equations). We determined the relative strength of each mechanistic model in terms of its goodness of fit to the data by using an information criterion ; this form of analysis indicates the most likely mechanistic relationship between each tree population and the nitrogen cycle. Furthermore, the tested the effect of a nonlinear change in the trajectory of climate dynamics on plant-nitrogen cycle interactions. # Results ## Palaeoecological Reconstruction Tree population changes were reconstructed with fossil pollen accumulation rates (PAR) from lake sediments; PAR is a proven indicator of past changes in tree species density. Pine was present in the landscape from 16,000 cal. yr. BP and became dominant between 14,000 and 12,000 cal. yr. BP (see.A). Oak was also present in the landscape in low abundance from 16,000 cal. yr. BP and it had a short-lived period of expansion around 14,000 cal. yr. BP. Oak became dominant on the landscape at around 11,000 cal. yr. BP. This represented an ecosystem shift from coniferous-dominated to deciduous-dominated forest following rapid climate warming at 11,700 cal. yr. BP. Changes in the nitrogen cycle were reconstructed by two different proxies: stable nitrogen isotopes (δ¹⁵N) and elemental concentrations of total nitrogen (%TN) within the sediments. Stable nitrogen isotopes have been shown to correspond with changes in available nitrogen in the terrestrial ecosystem. Total nitrogen primarily represents the amount of nitrogen being transported into the lake from the surrounding catchment, and includes both plant-available and plant-unavailable forms of nitrogen. Coincident with the abrupt change in climate, there was a drop in δ¹⁵N-inferred nitrogen availability that was coincident with a temporary increase in total nitrogen (%TN). This suggests that a short-term increase in the amount of nitrogen entering the lake occurred with the rapid climate change. By 11,500 cal. yr. BP, when %TN reduced to previous levels there was a corresponding increase in δ¹⁵N, which exceeded its pre-12,000 cal. yr. BP level. By this time, oak had become the dominant tree species in the forest and the pine population began a rapid decline. During the time of oak-dominance, δ¹⁵N remained high and %TN increased gradually, yet at a faster rate than under pine-dominance. Oak dominance lasted until after 10,000 cal. yr. BP, when other broadleaved trees such as hazel, lime and elm dominated the ecosystem ; this transition corresponded with rapid and significant increases in %TN and continued high values of δ¹⁵N, which suggest that high amounts of nitrogen were entering the lake system. ## Mechanistic Relationships We used Akaike Information Criterion (AIC) weights to demonstrate the relative amount of support for each mechanistic model of tree population-nitrogen cycle dynamics. Furthermore, to establish whether or not there is evidence of a threshold change in ecosystem functioning coincident with the change in climate at 11,700 cal. yrs. BP, we further separated the coupled nitrogen-tree population models into two types: a threshold model, where the models were fitted separately to the pre- and post- climatic threshold periods; and a non- threshold model where ecosystem functioning was assumed to be constant over the entire series. The AIC-best model for describing the relationship between both the pine and the oak populations and the nitrogen cycle was the nitrogen independent population growth with feedback effects (i.e. plant-driven nitrogen cycle) model, where the interaction between the population dynamics and the nitrogen cycle occurs via declining plant litter. This was the most-likely mechanism by which the populations interacted with the nitrogen cycle in both the threshold and non- threshold cases. Therefore, this form of interaction between the dominant vegetation and the nitrogen cycle held true despite a threshold in climate being passed and a shift from coniferous-dominated ecosystem to a deciduous-dominated one. The ability of the threshold and non-threshold versions of the models to predict the observed relationships between the tree population dynamics and the nitrogen cycle was determined by calculating and comparing the relative AIC weights of the threshold versus non-threshold versions of the plant-nitrogen cycle models. The threshold version of the model provided a better fit to the data than the non-threshold version: AIC weights indicate that threshold models have 100% of support from the data and this held true for both the oak and pine populations. ## Changes in Plant-Nitrogen Relationship Following the threshold change in climate at 11,700 cal. yrs. BP, there were notable changes in the estimated parameters for both tree populations. Maximum likelihood estimated parameters of the AIC-inferred best plant-nitrogen cycle model suggest that climate warming was associated with increased rates of conversion of decaying biomass of both species into available nitrogen, increased background loss rates of available nitrogen but decreased background loss rates of total nitrogen. There was no change in the rate of conversion of decaying biomass of pine on total nitrogen following warming while there was an increase in the rate of conversion of decaying oak biomass into total nitrogen. There were increases in the population growth rates of both species with warming after 11.7k cal. yrs. BP, but contrasting effects on loss rates: loss rates of oak decreased (although these parameters were not found to be significant) while loss rates of pine increased relative to the cool period prior to 11.7k cal. yrs. BP. The net effect of changing biomass on nitrogen cycling, as inferred by the product of the loss rates of each species and the conversion rates of their decaying biomass into available and total nitrogen show that the oak population had a consistently greater positive effect on both proxies of the nitrogen cycle than the pine population. # Discussion Contrary to current thinking, results from our study suggest that the nitrogen cycle was not a driver of secondary succession in this ecosystem during the period of rapid and significant climatic warming at the late-glacial/early postglacial transition. The threshold change in climate was associated with a decrease in the predicted mortality rates of long-lived oak trees, resulting in reduced opportunities for replacement by pine. Given *Quercus spp.*'s greater shade tolerance than *Pinus spp.* we suggest that competition for light may have been the process by which oak replaced pine following the threshold change in warming. Our finding that secondary succession occurred independently of changes in nitrogen availability supports the findings of Menge et al. that forest ecosystems are not limited by available nitrogen over long time scales. These results however are in contrast with the assumptions made in select dynamic global vegetation models (DGVMs) that available nitrogen can limit tree population growth during climate warming. Our results demonstrated stability in ecosystem functioning across the interval of rapid climatic change and despite a change in the dominant tree species. Climate change was found to affect the rates of nitrogen cycling: following the threshold change in climate, our model indicated that there were higher rates of conversion of decaying biomass into available and total nitrogen. The effect of changes in biomass on the nitrogen cycle was modified by concurrent changes in the loss rates of both species; yet, there was a consistently greater effect of increasing oak density on available and total nitrogen, which concurs with contemporary studies of nitrogen concentrations in leaves/needles in *Quercus spp.* and *Pinus spp.* and differences in nitrogen resorption proficiency. Thus, while the mechanism underlying both species' interaction with the nitrogen cycle remained constant, the observed vegetation changes led to increased rates of nitrogen cycling given warming and the shift to oak-dominated landscape even though this feedback effect was not found to have a demonstrable impact on the fitness of either population at the millennial time scale. It is important for future work to test the context-dependence of our findings by validation with other appropriate palaeoecological datasets of coniferous to deciduous forest transitions and changes in nitrogen cycling during the late- glacial warming period. Furthermore, in order for model-fitting and model- selection analysis of palaeoecological data to yield sufficient information to be useful for predictive modelling, this approach needs to be used to analyze a greater variety of transitions in dominant plant functional types. These caveats notwithstanding, results from this study demonstrate how the dynamics associated with ecosystem functioning can remain relatively stable following a major environmental perturbation. # Materials and Methods ## Palaeoecological Analyses The fossil pollen and stable isotope time series were derived from an early postglacial sediment sequence taken from a peat bog, Kis-Mohos Tó (20°24′30″E; 48°24′40″N), in Northeast Hungary. The bottom 2.4 m of an undisturbed 8.86-m sedimentary sequence was collected from Kis-Mohos To′ and analyzed for a range of proxies, including fossil pollen, elemental concentrations and stable isotopes of nitrogen. Seven radiocarbon dates for this section of the core were obtained by a combination of bulk and AMS (Accelerator Mass Spectrometry) analyses. These dates were calibrated using BCal and the 2004 Northern Hemisphere ¹⁴C atmospheric calibration curve and interpolated using linear interpolation. Our time-series (n = 47) included measurements of all proxies (i.e. fossil pollen, elements and staple isotopes) taken from the same time periods. Samples were analyzed at a resolution of two to eight centimeters throughout the 2.4 m sediment core section. The sediment was prepared for pollen analysis following Berglund and Ralska-Jasiewiczowa ; this method involves adding a known quantity of exotic pollen, which allows for the determination of pollen concentration. In order for our pollen data to provide the most appropriate representation of past tree population densities, we converted the pollen concentration values into pollen accumulation rates (PAR), : the pollen concentration (grains/cm³) divided by the sediment accumulation rate (cm²/yr) giving PAR (grains/cm²/yr). Elemental (%TN) and stable nitrogen isotope (δ¹⁵N) analysis followed the methods described in Talbot. Each bulk sample was dried at \<40°C and then ground with a pestle and mortar. About 40 mg of each dried and homogenized sample was transferred to a small plastic vial and sent to the Research Laboratories for Ancient History and Archeology (RLAHA) at the University of Oxford where they were analyzed by mass spectrometry. ## Non-Linear Population Dynamic Models ### Candidate Models There were three groups of candidate models used in this study: nitrogen- dependent population growth, nitrogen-dependent population growth with a positive feedback effect of decaying plant biomass on nitrogen cycling and nitrogen independent population growth with the same feedback effect. These different groups of models were used to test the hypothesis of whether or not the populations were dependent on nitrogen for population growth and whether or not they had a feedback effect on the nitrogen cycle. Within each group we applied different versions of the model to the data in order to detect the best functional form for describing the relationship between the populations and the nitrogen cycle (for the nitrogen-dependent population growth models). In particular, we applied two different plant-resource uptake models to determine which of these provided the best explanation of the joint plant population-nitrogen cycle dynamics: saturating uptake and linear uptake. To test whether or not there was evidence that density-dependent population growth limited the population growth rates, we incorporated the logistic density dependence function in the population growth model and compared the goodness of fit of this model to the fit of the models without this function. In the case of the nitrogen independent population growth models, we varied the functional form of the density-dependent control on population growth between a logistic and an exponential function. ### Model-Fitting The model-fitting approach uses maximum likelihood estimation to find the set of parameters that yield the best fit to the nitrogen and tree species' dynamics data. Model-fitting was achieved by integrating a set of differential equations using a Runge-Kutta 4 numerical integration routine to generate expected values of the nitrogen availability and tree population data over variable time steps (*variable* due to the fact that there are not uniform time differences between observations in palaeoecological data). The model was initialized with the first measurement from the data, and expected values were generated from user-defined initial parameter values. For example, using the model of saturating resource- dependence without a density-dependent effect on growth, the following set of equations would be integrated:to give the expected values of nitrogen (*N(T)*) and vegetation (*x(T)*) at each census point (*t*). *τ* represents the variable integration time-step between census points, which is equivalent to the time lapsed between observations in the palaeoecological record. The expected values change as the set of model parameters, **P** (e.g. **P** = \[*λ, a, b, γ_N, r, γ_x*\] for the above model), are varied. A maximum likelihood method is used to determine the set of model parameters, **P**, that provide the best fit of each model to the data (i.e. that minimizes the difference between the model-generated data and the observed data). The following likelihood function is used to evaluate the goodness-of-fit of each model: This likelihood function assumes that model errors are normally distributed and that stochastic events on each population are primarily driven by external environmental effects. Evaluating the likelihood function involves estimating the set of model parameters (**P**), the standard deviation associated with nitrogen abundance (*σ_N*), the standard deviation associated with tree abundance (*σ_x*) and their covariance (*ρ*). At each time step *j* the expected values are compared to the observed values with the objective of finding the set of model parameters, **P**, that minimize the difference between these observed and expected values. This was accomplished by using a simplex optimization routine implemented in C, which searched for the set of **P** that minimized the negative log-likelihood value. Confidence intervals for the maximum likelihood estimated (MLE) parameters (**P**) were calculated from the likelihood profiles following Morgan. To determine if there was a threshold change in the relationship between tree population density and the nitrogen cycle, a threshold was simulated by splitting the data at the point where rapid climatic warming is known to have occurred (i.e., around 11,700 cal. yr. BP). The ten models were fitted separately to each segment of the split (i.e. threshold) dataset (n₁ = 24, n₂ = 23), which allowed the parameters to vary across this period of climatic change. The likelihood of the threshold models was determined by summing the likelihoods of each segment. The same models were also applied to the whole dataset (n_whole = 47), where the parameters were held constant over the entire series. ### Model Selection Model-selection was used to determine the model of ecosystem functioning that provided the best representation of the tree population density and nitrogen cycling data. The model-fitting step provided a negative log-likelihood value for each model; this was used to calculate an Information Criterion score for each model. The Akaike Information Criterion (AIC) was used, which is a weighted goodness of fit criterion. The AIC score was converted into a normalized indicator of support for each model by calculating AIC weights (*w_i*), where larger values indicate greater evidence in support of the model. # Supporting Information [^1]: Conceived and designed the experiments: ESJ MBB KJW. Performed the experiments: ESJ. Analyzed the data: ESJ. Contributed reagents/materials/analysis tools: MBB. Wrote the paper: ESJ MBB KJW. [^2]: The authors have declared that no competing interests exist.

# Introduction Dengue is caused by one of four distinct dengue virus serotypes (DENV-1–4) belonging to the family of Flaviviridae, genus *Flavivrus*. Dengue infection is transmitted to humans through the bite of infected mosquitoes of the genus *Aedes*, commonly *Aedes aegypti* and *Aedes albopictus*. The clinical picture of the illness may range from asymptomatic, non-specific acute febrile illness (dengue fever) to a fatal illness with hemorrhagic fever and circulatory shock known as dengue haemorrhagic fever/dengue shock syndrome. During the last decades, dengue has evolved rapidly as viruses have spread worldwide expanding from Southeast Asia to the Caribbean and Latin American. Moreover, it is expected to disseminate further in receptive regions where competent vectors are present. This has led to a change in the status of numerous countries, which have evolved from non-endemic (no serotype present) to hyperendemic (continuous co-circulation of multiple virus serotypes), as well as to an increase in the frequency of severe disease forms. Currently, dengue is a pandemic with worrying public health implications throughout the tropics and subtropics. Indeed, DENV is considered as the most important re-emerging virus, directly threatening 2.5 billion people living in tropical/sub-tropical areas and resulting in around 50–100 million cases of dengue fever and 250 000 cases of dengue haemorrhagic fever, together with a mortality rate of 25 000 yearly. Serological evidence of dengue infection has been reported in different countries or islands of the Eastern African costal region. DENV-2 was the first serotype formally identified in the Southwest Indian Ocean Islands during the large outbreak that affected the Seychelles in 1976–1977 and Reunion Island in 1977–1978 with an estimated attack rate of respectively 60% and 35%. Subsequently, circulation of DENV-3 subtype III between the East-Africa coastal region, Southeast Asia and Latin American has been well characterised. In addition, DENV-1 circulation was reported during the outbreaks of dengue both in the Comoros in 1993 (estimated attack rate of nearly 30%) and Reunion in 2004. However, due to the lack of sustainable surveillance systems, available data are sparse in most of the countries of this region either during minor epidemics or during inter-epidemic periods. This is particularly noteworthy in Mayotte, where the only documentation of DENV circulation in the territory dates back to 1943. Since then, DENV circulation has been documented in neighbouring countries or islands which have frequent trade and population contact with Mayotte. These islands share similar conditions that favour the spread of DENV, presence of dengue-competent vectors, comparable environmental and climatic conditions, lack of effective mosquito control, rapid population growth, uncontrolled urbanisation and inadequate water supply and waste management systems. Recently, the dramatic appearance of persistent regional and international outbreaks of Chikungunya virus (CHIKV) fever in 2005–2007 highlighted the vulnerability of the region to arthropod-borne viruses and their capacity for rapid expansion across the region and beyond,. These outbreaks have drawn attention to the need for better understanding of the local epidemiology of arboviruses in general and of DENV in particular. In order to provide reference data on previous DENV circulation in Mayotte, we have performed a serological survey with the goals of estimating the seroprevalence of DENV IgG-antibodies among individuals aged ≥2 years and to identify potential sociodemographic and environmental determinants of seropositivity. # Materials and Methods ## Setting, design and population The study was conducted in Mayotte, an insular French-administered territory located in the Comoros archipelago between the Eastern African coast and Madagascar. With regard to the 2002 census, the estimated population in 2006 was 175 000 with a density of 468 inhabitants per square km. The population originates primarily from Africa. Flow of immigrants from the neighbouring Comoro Islands accounted for approximately 35% of Mayotte's population. The study setting has been described previously. The survey was carried out in conjunction with a CHIKV-related population-based serologic survey in November and December 2006, targeting the island population aged ≥2 years. The survey used a multistage cluster sampling technique. In a first step, 40 of the 400 enumeration areas- EAs (clusters) provided by the 2002 census blocks were sampled randomly with probability proportionate to size, the size being the number of household in the cluster. Secondly, a systematic walk from a random starting point was performed to select 7 to 12 households (defined as groups of persons sleeping or eating together) within each EA. Finally, in each selected household, all adults (aged ≥15 years) and one individual aged 2–14 years (identified by using the next birthday selection method) were invited to participate in the survey. Consequently, individuals belonging to the 2–14 year age group were under-represented in our sample compared to the general population, since 43% of total population of Mayotte is aged 0–14 years. To improve the level of participation, household visits included weekends and holidays. Moreover, when eligible members of the house were absent at the time of the initial visit, surveyors made up to two additional visits. ## Data collection Participation in the survey was proposed on a voluntary basis. A pre-coded and pre-tested structured questionnaire was administered face-to-face at the house by trained surveyors in the local language (Shimaori). Sociodemographic data were collected relating to age, gender, educational level, country of origin and length of residence in Mayotte (for foreigners). Data were also collected on the housing environment, relating to the type of house construction (Makeshift, adobe and stone, concrete) the number of household residents, sanitary conditions and household facilities such as electricity and drinking water source. Data on the housing environment and on asset-ownership were collected with the collaboration of a reference person for the household. A six-point household asset index was calculated as a proxy for socio-economic status for adults. This index was constructed on the basis of availability of electricity (1 point), a flush toilet within the household (1 point), piped water as source of drinking water (1 point), possession of a television set (1 point), radio (1 point), and a refrigerator (1 point). The number of items were summed for each household and the household assigned to one of two grades, namely low economic status (total score\<median threshold) and medium to high socioeconomic status (total score ≥median threshold). ### Laboratory methods A venous blood sample was obtained from each participant. Immediately after puncture, blood samples were stored at 4–8°C and forwarded the same day to the laboratory of the Mayotte Territorial Hospital, which was responsible for all serologic testing. All serum samples were tested for specific IgG DENV antibodies as a marker of past infection using the Focus ELISA kit (Dengue Fever Virus IgG Focus Diagnostics, Cypress CA, USA). A commercial assay for the detection of IgG antibodies to dengue viruses was used (Focus Diagnostics, Cypress, CA, US). The test is an indirect IgG assay using a 96-well microtiter plate coated with equal proportions of inactivated, purified DENV types 1–4 (Type 1: TH-Sman; Type 2: TH-36; Type 3:H87; and Type 4: H241). Briefly, patient sera and controls were diluted 1∶101 in the kit diluent, and 0.1 ml was added to wells. Negative and positive controls were included, as was a kit-supplied calibrator/cutoff control. After incubation for 1 h at 37°C and washing three times with phosphate-buffered saline (PBS), peroxidase-labeled rabbit anti-human IgG conjugate was added to each well for 30 min. After the conjugation step and the final wash series with PBS, tetramethylbenzidine (TMB) was added, and the final reaction product was measured in a spectrophotometer at a wavelength of 450 nm. An index value was obtained for both control and patient samples by dividing the absorbance value of the patients and controls by the absorbance value of the calibrator (cutoff control). The optical density (OD) of the specimen well was compared with the OD of the well containing a calibration cut- off (CO) serum sample provided with the kit. Those samples with OD: CO ratio ≥1 were recorded as positive. The IgG ELISA assay used presented a sensitivity of 100% and a specificity of 88.3%. ## Statistical analysis Data were double entered and corrected for data entry errors using EPIDATA® software version 3.0 (Epidata Association, Odense, Denmark). All analyses were performed with STATA® software version 10.0 (Stata Corporation, College Station, USA). In the interest of representativity as regard to the reference population, each participant was weighted by post stratification according to age, gender, geographic location and household type in order to account for unequal probabilities of selection resulting from sampling cluster design of the study. For this purpose, sampling weights were prepared and provided by the Regional Bureau of the French National Institute for Statistics and Economic Studies using the 2006 population estimation on the basis of the 2002 census. Crude and adjusted prevalence rates are presented by sociodemographic and household variables. Overall and subgroup-specific DENV seropositivity prevalence rates were estimated and differences within sub-groups were compared using binomial survey-adjusted Wald chi-square tests. P values\<0.05 were considered statistically significant. In a next step, explanatory bivariate analyses using population-adjusted weights to account for the survey design were performed to assess potential associations between DENV seropositivity (outcome of interest) and sociodemographic and housing variables. Crude odds ratios (OR) and their 95% confidence interval (95% CI) were estimated to explore the strength of the association between DENV seropositivity and the variable of interest. This analysis was only performed in the adult subgroup (≥15 years). Indeed, educational level and length of residence in Mayotte were not considered applicable for risk factor analysis for individuals aged 2–14 years. Next, the association between variables identified in the bivariate analysis and the outcome of interest were explored by multivariable logistic regression analysis in order to control for confounders. For this purpose, stepwise logistic regression models including sociodemographic and household variables were performed. All variables significantly associated with seropositivity at a probability threshold of ≤0.25 in the bivariate analyses were entered into the initial multivariate models. For this purpose, 36 individuals were excluded due to missing data for at least one of the variables of interest, leaving a total of 852 observations. A backward stepwise procedure was used to select variables retained in each of the final models. Furthermore, in order to retain a final model, a two-by-two interaction terms were checked for variables that remained statistically significant in the pre-final model. The strength of the association between selected variables and DENV seropositivity was estimated by adjusted odds ratios (AOR) with their 95% CI, all AOR excluding 1.0 being considered as significant. In order to reflect adequately association with outcome, confidence intervals for AOR estimates were calculated on the basis of a log transformation with the standard errors computed by the delta method that took into account both clustering of individual observations within households and non-independence of the variables at household and region levels. ## Ethical considerations Publicity and information about the survey was provided through the media, field visits to the local population, and contact with local and national authorities. During visits, surveyors introduced themselves to selected households and explained the objectives and all procedures to the household members concerned. A written consent form was first obtained from adults or guardians of those individuals aged \<18 years prior to the inclusion, unless they were illiterate, in which case signed consent was sought from an appropriate family member chosen by the participant. As the survey was combined with the determination of CHIKV seroprevalence, it is important to note that we solicited the agreement for testing several arboviruses including DENV. This study was reviewed and approved by the ethics committee of the Bordeaux University Hospital Centre, France, in compliance with all French regulations on protection of human subjects. A numbered code was attributed to all participants to track blood samples. All data were handled confidentially and anonymously. # Results ## Characteristics of participants and prevalence of serum DENV antibodies Forty villages (clusters) were sampled and 508 inhabited household visited. Of these, 418 households (82.3%) agreed to participate in the study. The mean household size was 5.7±2.4 individuals \[range: 1 to 13\]. The selection procedure allowed the recruitment of a total sample of 1156 individuals. Two participants in the 2–14 years age group provided insufficient serum to undergo serological evaluation and were thus excluded from the analysis. Comparison of the sociodemographic characteristics of the study sample to the reference demonstrated an adequate agreement. However, the study sample was under-represented in men compared to women (43·2% vs. 50·2% in 2006 population projection) as well as over-represented in individuals belonging to the 15–24-years age-group (25.5% vs. 20.2% in the 2006 population projection). These minor differences were accounted for in the analyses of seroprevalence and risk factors by allocating a weight to each participant. The overall weighted seroprevalence of DENV antibodies in Mayotte for individuals aged ≥2 years was 22.7% (95% CI, 18.2–27.3). The age-specific weighted prevalence increased with age (χ2 for trend = 11.86, P\<0.0006). The peak age-specific prevalence was observed in individuals of the 25–54 year age group while those of the 2–14 age group presented the lowest prevalence rate. When the analysis was restricted to individuals born in Mayotte, the prevalence rate was highest in individuals aged over 35 years (1.6% in the 2–14 age group, 4.6% in the 14–24age group, 6.2% in the 25–34 age group, 15.7% in the 35–44age- group, 22.2% in the 45–54 age group and 23.6% in individuals over 55 years of age). Of the 302 children (age group 2–14), eight had serum DENV antibodies (weighted prevalence: 2.2%); Of these eight children, three were less than 10-years old and born in Mayotte. No difference in the seroprevalence of DENV antibodies was observed according to gender. The highest seroprevalence percentages were observed in the Northern and the North-Eastern part of the island. ## Association between sociodemographic and household variables and DENV seropositivity In the bivariate analysis, DENV seropositivity was significantly associated with a number of demographic and environmental variables. For example, seropositivity was positively associated with being born in another Comoro island (OR: 8.60; 95% CI: 5.56–13.27), whereas it was negatively associated with time at school ≥6 years (OR = 0.42, 95% CI: 0.26–0.66) and a longer duration of residence in Mayotte. Individuals who had resided in Mayotte for ≥15 years were protected from past infection as compared to those who had resided for less than 5 years (OR = 0.18, 95% CI: 0.10–0.34). Moreover, living in a household below the median asset index threshold or with a larger number of household members (3–4 individuals) was significantly associated with DENV seropositivity. In the multivariate logistic regression model, only two demographic variables remained significantly associated with DENV seropositivity, namely increasing age and being born in another Comoros island. Concerning household-related variables, living in a household of three to four members, as well as living in makeshift house and a socioeconomic status index below the median threshold were positively associated with DENV seropositivity. # Discussion This is the first large-scale survey of the prevalence of dengue infection and its potential determinants that has been conducted in Mayotte. The overall weighted prevalence of IgG DENV antibodies was 22.7%, ranging from 2.2% in children to almost 40% for individuals aged 45–54 years. This increasing age- related prevalence of IgG DENV is consistent with previous studies conducted in other endemic settings. We observed that previous DENV infection in adults was independently associated with increasing age, low socio-economic status and household size. In contrast, no difference in seroprevalence was found for gender. Previous reports of potential gender differences in dengue infections have been inconsistent, with some finding a higher prevalence in men, others a higher prevalence in women, and others no gender difference, in line with our findings. This inconsistency may be related to variations in gender-specific exposure according to local cultural practices. The study identified a striking association of IgG DENV prevalence with certain sociodemographic variables. Comorian natives immigrating to Mayotte aged 15 years or over were around ten-fold more affected than individuals of the same age born in Mayotte. This finding may be, at least in part, related to the outbreak of dengue that occurred in the neighbouring Comoros islands in 1993. This epidemic hardly affected Mayotte, which may explain the association observed. Moreover, the age distribution of IgG DENV in the native population of Mayotte reinforces this hypothesis. The peak of IgG DENV prevalence in this group is observed in individuals aged 35 years or more, suggesting a link with the major dengue outbreak that affected the South-Western Indian Ocean islands, in 1976–1978, which was first reported in the Seychelles (1976–1977) and subsequently in Reunion (1977–1978). Although this outbreak was well documented in these two countries, DENV circulation in Mayotte was not reported at the time. However, it can be speculated that the 1976–1978-DENV outbreak had in fact probably spread to Mayotte as well, as did the recent persistent regional outbreak of CHIKV infection, first reported in Reunion. Another important finding is the evidence of past infection among younger individuals, particularly those native to Mayotte and born after the Comoros outbreak of 1993. This may suggest potentially that low level transmission of DENV in Mayotte has continued during the last decade. This is the first suggestion of inter-epidemic transmission that has been described in Mayotte. With respect to household variables, a higher risk for previous DENV infection was found in households with a socio-economic threshold below the median. In addition, there was a positive association between household size and DENV seroprevalence, which was restricted to households with three to four residents. This particular finding may be related to the structure of the questionnaire that assessed the number of inhabitants per household rather than the density of inhabitants per room. The latter variable may be more pertinent for evaluating potential associations between DENV circulation and housing. Our study has a number of limitations that should be considered. Firstly, it was a cross sectional study relying on past exposure, rather than a prospective incidence study. Secondly, specific documentation of DENV serotypes that firmly establish the circulation of different viral variants was not performed due to logistic and financial constraints. On the other hand, this study was conducted on a representative sample of the target population, which is one of the strengths of the study. In particular, it was weighted with respect to age, sex, and housing conditions relevant to the local population, and also took into account local migratory flux. The advantage of using commercial ELISA to perform a high throughput testing for DENV serology screening is well documented. ELISA testing allows detection of all serotypes (DEN-1–4) and has a similar performance as the hemagglutination test, considered to be the gold standard for indirect serologic diagnosis of DENV infection. Besides, ELISA testing for DENV may also detect cross-reactive antibodies to other flaviviruses, such as West Nile virus, whose circulation has been reported in Madagascar. However, it should be noted for the limitation of this assumption that yellow fever has never been reported and that immunization against the disease is neither recommended by French Health authorities nor by the US CDC. This is true for local inhabitants as well as for travellers to the island. Moreover, no epidemic of West Nile virus has yet been notified either in Mayotte or in Reunion Island. This is the first cross-sectional study to demonstrate the circulation of DENV in Mayotte using methods that discriminate between different age groups and between different origins of subjects. Hence DENV infection should be considered in the differential diagnosis of influenza-like illnesses in routine clinical practice in Mayotte. Moreover, routine laboratory surveillance should be enhanced to improve the accuracy of detection of viral circulation trends, including the identification of the specific DENV serotypes involved. In parallel, community-specific habits, customs or behaviours that increase the risk of mosquito-borne infections should be identified to facilitate implementation of prevention campaigns by health services and to educate the general population. A lot has been learnt from previous major outbreaks of arboviral infections in the South-Western Indian Ocean region, notably during the recent chikungunya virus epidemic. Continuous efforts should be put into developing regional cooperation to set up early warning systems and to ensure prompt responsiveness to counter rapidly-spreading vector-borne viruses as due to the presence in Mayotte of two competent vectors *Aedes aegypti* and *Ae. Albopictus*. Finally, from a global health perspective, this region at the crossroads of Asia and Africa should be more thoroughly monitored for dengue virus circulation, due to the increase in intercontinental exchanges and legal and illegal immigration. We thank the inhabitants of Mayotte who kindly accepted to participate to this survey and the staff of the “Centre Hospitalier de Mayotte” and the “Direction des Affaires Sanitaires et Sociales” for their tremendous support. We are also indebted to the members of SEROCHIKMAY Survey Team for the collection of data. We are grateful to Gilles Greneche from the Regional Bureau of the French National Institute for Statistics and Economic Studies in la Reunion Island for extensive support in data weighting. We are also grateful to the copyright holder and original publisher of the West Indian Ocean map represented in. [^1]: Conceived and designed the experiments: DS KE DM. Performed the experiments: DS. Analyzed the data: DS KE DM. Contributed reagents/materials/analysis tools: DS CG AM TL ED FP. Wrote the paper: DS KE DM. Critical revision of the manuscript for important intellectual content: CG AM TL ED FP. Administrative, technical, or material support: CG AM TL ED FP. [^2]: The authors have declared that no competing interests exist.

# Introduction Hepatitis C virus (HCV) causes an infection that can produce both acute and chronic hepatitis. After acute infection, approximately 20–25% of HCV patients experience spontaneous clearance mediated by an immune response. One of the key components of the immune system forming the first line of defense against HCV is cell-mediated immunity, specifically the T lymphocytes and natural killer (NK) cells. The HCV-specific CD4⁺ and CD8⁺ cells are associated with viral clearance via cytolitic or non-cytolitic mechanisms. The NK cells can inhibit virus replication in infected hepatocytes by direct cytotoxicity and the production of inflammatory cytokines, such as interferon gamma. Impaired anti- HCV activity of T- and NK cells therefore could lead to a higher likelihood of developing chronic HCV infection. This could be the reason why HIV-infected patients have lower rates of HCV viral clearance than HCV-monoinfected subjects. The implementation of combination antiretroviral therapy (cART) has considerably improved the life expectancy of patients living with HIV. As a result of achieving sustained virological suppression with cART, the immune system partly recovers and a certain reversal of both the defective T and NK cells can be observed. Consequently, once the immune system recovers, other infections may resolve spontaneously, even in cases of long-term chronic infections such as chronic hepatitis C (CHC). This situation is defined as spontaneous clearance of CHC and only a few cases have been documented, which is why little is known about this rare occurrence. The aim of this study was to evaluate the rate of spontaneous resolution of CHC in a cohort of HIV-infected subjects after effective use of cART. # Material and methods ## Study population Five hundred and nine HIV/HCV co-infected patients in follow-up at two reference hospitals in Andalusia (southern Spain) between January 2000 and September 2016 were analyzed retrospectively for the achievement of spontaneous clearance of CHC. Those patients identified with spontaneous clearance of CHC had to meet the following criteria: i) they had never been treated for HCV infection, and ii) initiation of cART was not before the follow-up period. ## Variable collection and definition The main variable of the study was spontaneous clearance of CHC, defined as a negative HCV RNA measurement, preceded by at least two quantitative HCV RNA tests with a minimum of 12 months between them. Clinical, demographic and analytical variables were collected from patients identified with spontaneous clearance of CHC from diagnosis of HCV infection until the end of the study (September 2016). These variables were: liver fibrosis stage, cART regimen, HIV viral load (copies/mL), CD4⁺ count (cells/mL), HCV viral load (IU/mL), HCV genotype and IL28B genotype. Fibrosis stage was determined by liver biopsy or, after 2007, by transient elastography (FibroScan®, Echosens, Paris). In such cases, a liver stiffness value of ≥14.6 kPa was defined as indicating liver cirrhosis. Plasma HCV RNA loads were measured by quantitative PCR assay (CobasTaqMan, Roche Diagnostic Systems Inc., Pleasanton, CA, USA) with a limit of detection of 15 IU/mL. HCV genotype was determined by hybridization assay (INNO-LiPa HCV, Bayer Corp., Tarrytown, NY, USA). HIV viral load was measured by PCR (CobasTaqMan, Roche Diagnostic Systems Inc., Pleasanton, CA, USA), with limit of detection set at 20 copies/mL. ## Statistical analysis A descriptive analysis was performed. The number of cases of spontaneous clearance of CHC was expressed as a percentage, using a two-sided 95% CI with respect to the overall population calculated on the basis of the exact binomial distribution. GraphPad Prism version 6 (Mac OS X version; GraphPad Software; San Diego, California, USA) was used to calculate percentages and plot graphs. ## Ethical aspect The study was designed and performed according to the Helsinki Declaration and was approved by the ethics committee of the Reina Sofia University Hospital, Cordoba, Spain. The study did not require informed consent because the patients were not interviewed directly and the information was collected from completely anonymous pre-existing records, thus ensuring the protection of personal data in accordance with Law 15/1999 of 13 December on Personal Data Protection. # Results A total of 509 HIV patients with chronic HCV infection were included in the study. Of these, 3 (0.59%; 95% CI: 0.15%-1.6%) patients experienced spontaneous clearance of CHC. The main characteristics of these patients are shown in. The cases are summarized as follows: ## Case 1 A 34 year-old man, an injection drug user (IDU), was diagnosed with HIV infection in 1993. He received neither clinical follow-up nor cART until August 2004. At this point, the patient was asymptomatic, with a CD4⁺ count of 267 cells/mL, an HIV RNA viral load of 86,500 copies/mL and an HCV RNA viral load of 2,280,000 IU/mL. He initiated cART in August 2004 with a regimen of lopinavir boosted with ritonavir (LPV/rtv) plus lamivudine (3TC) and zidovudine (AZT). In June 2005, with a CD4⁺ count of 447 cells/mL, the patient achieved an undetectable HIV RNA viral load. At that point, his HCV viral load remained detectable with an HCV RNA titer of 4,710 IU/mL. In September 2006, the patient was asymptomatic. His HIV and HCV viral loads were undetectable and were still negative at the last visit in May 2016. ## Case 2 A 25 year-old woman was diagnosed with post-transfusion HIV infection in 1987. In April 2000, the patient had a CD4⁺ count of 154 cells/mL with an HIV RNA viral load of 1,610 copies/mL. She initiated cART with efavirenz (EFV) plus AZT and 3TC, but had poor adherence until she voluntarily abandoned cART in February 2002. The patient did not achieve an undetectable HIV viral load during the on-cART period. In August 2001, she was diagnosed with HCV genotype 3 infection, with a viral load of 550,000 IU/mL. Between 2002 and 2006, the patient voluntarily continued interruption of cART and had successive detectable HCV RNA viral loads (between 134,000 and 1,380 IU/mL). In August 2005, a liver biopsy was performed, which was compatible with a diagnosis of cirrhosis. The patient also showed a platelet count of 41,000 cells/mL, total bilirubin of 2.4 mg/dL, albumin of 3.90 g/dL, and an INR of 1.3 (Child-PT score of 6). Ultrasound findings showed splenomegaly (15cm) and portal hypertension (15mm). In March 2006, the patient had a CD4⁺ count of 88 cells/mL, HIV viral load of 808 copies/mL and HCV with a viral load of 1,270 IU/mL. The patient then initiated treatment with SQV/rtv plus emtricitabine (FTC) and tenofovir (TDF). In April 2007, her CD4⁺ count was 127 cells/mL and she achieved undetectable HIV and HCV viral loads. At that point, the liver stiffness measurement using transient elastography was suggestive of liver cirrhosis (21.8 kPa). Her HIV and HCV viral loads remained negative until May 2015, when the patient died of colorectal cancer. ## Case 3 A 22 year-old man with hemophilia was diagnosed with HIV infection in 1992. In the years that followed, the patient was asymptomatic, his CD4⁺ count was over 800 cells/mL and his HIV viral load was below 10,000 copies/mL without use of cART. In June 2001, the patient showed HCV genotype 3 infection with a viral load of 504,000 IU/mL. At that point, the patient had a CD4⁺ count of 855 cells/mL and his HIV viral load was 11,200 copies/mL. Between 2002 and 2005, the patient showed two consecutive detectable HCV viral loads. In June 2005, the patient experienced a significant drop in his CD4⁺ count (up to 148 cells/mL) and his HIV viral load was 100,000 copies/mL. One month later (July 2005), the patient initiated cART with a regimen of efavirenz (EFV) plus 3TC and didanosine (DDI). In March 2007, cART was changed to EFV plus FTC and TDF showing a CD4⁺ count of 366 cells/mL, HIV RNA viral load of 209 copies/mL and HCV RNA viral load of 4,870 IU/mL. In January 2008, the patient was asymptomatic and both HIV and HCV viral loads were undetectable and remained negative until the last visit in July 2016. # Discussion Once chronic HCV is established, the infection is very unlikely to be eradicated without using HCV therapy. However, several cases of spontaneous clearance of CHC have been reported in the setting of HIV/HCV-coinfection, including patients with several years of detectable HCV viral loads. In our cohort, we found three cases of chronic spontaneous HCV clearance after effective use of cART. The rate of this rare event was 0.59%, similar to those found in other studies. The possible mechanisms involved in this event have still not been clarified. Effective treatment with cART suppresses HIV replication and restores the total CD4⁺ count. Immunologic reconstitution as result of successfully implementing cART could therefore be the main factor responsible for the spontaneous clearance of CHC. In our study, *Case 1* (nadir CD4⁺ count, 267 cells/mL) and *Case 3* (nadir CD4⁺, 148 cells/mL) showed a remarkable increase in CD4⁺ cells to 651 and 529 cells/mL, respectively when spontaneous clearance of CHC was achieved. However, in *Case 2*, a cirrhotic patient, CD4⁺ reconstitution after effective use of cART was not observed. Along similar lines, Endo *et al*. reported the case of a man whose CD4+ cell count was stable for 8 years before a change of cART regimen, so that the change in regimen was the only explanation for HCV viral resolution. Nonetheless, in the context of immune reconstitution, almost all reported cases of spontaneous clearance of CHC have linked it to the increase in CD4⁺ cell count after effective use of cART. Most HIV/HCV co-infected patients who initiate cART have immune reconstitution. In our series, however, only 3 (0.59%) of 509 patients with CHC were identified with spontaneous clearance of CHC after initiating cART. It may be speculated therefore that other factors are necessary in order to achieve spontaneous clearance of CHC after effective cART. Factors involved in the immune response, such as IL28B, have been widely associated with spontaneous resolution of HCV acute infection and are also predictive of response to HCV treatment. In the context of spontaneous clearance of CHC, *Vispo* et al. found that all patients who experienced spontaneous clearance of CHC (n = 6) bore the IL28B-CC genotype. Likewise, *Stenkvist* et al. described 3 patients in his series who carried the favorable genotype. As was the case in those studies, all the patients in ours who experienced spontaneous CHC clearance also had the same favorable genotype. This suggests that the IL28B genotype may be an important factor for the spontaneous clearance of CHC. On the other hand, HCV genotype 3 has been associated with a higher rate of acute self-limiting HCV infection, so that spontaneous clearance of chronic HCV infection with genotype 3 is less frequent than in infections caused by other HCV genotypes. In our series, 2 of 3 cases (the remaining case was not available) were infected with genotype 3. In this line, *Grint* et al. also described 9 cases of spontaneous clearance CHC and 5 of them (55.6%) were patients with genotype 3. This suggests that, as was observed for the spontaneous resolution of acute HCV infection, HCV genotype 3 infection could be a favorable factor for the spontaneous clearance of CHC. More studies in this area are needed. Several limitations should be noted. Due to the retrospective design of this study, HCV RNA testing was not determined on a regular basis in all patients. One consequence of this may be that the event was underestimated. At the same time, due to the low number of cases presented in this series, statistical differences among possible associated factors could not be established. In conclusion, the spontaneous clearance of CHC is a rare event in the context of HIV/HCV co-infected patients. These data however could have some impact on the management of HIV/HCV co-infected patients, since treatment-naïve patients, both HCV and HIV, could benefit from the possibility of spontaneous clearance of CHC when cART is administered. We are grateful to Ismael Zafra and Laura Ruiz-Torres for their technical support. The Grupo de Estudio de Hepatitis Virales (HEPAVIR) of the Sociedad Andaluza de Enfermedades Infecciosas (SAEI) is constituted by Juan Macías Sánchez, Francisco Téllez Pérez, Ángela Camacho Espejo, Mario Frias, Pedro Lopez-Lopez, Teresa Brieva, Isabel Machuca, Antonio Ramón Collado Romacho, Julián de la Torre Cisneros, Marcial Delgado Fernández, Elisa Fernández Fuertes, Isabel María Gea Lázaro, José Antonio Girón González, María Amparo Gómez Vidal, José Juan Hernández Burruezo, Luis Fernando López, Cortés, Miguel Ángel López Ruz, Manuel Márquez Solero, Nicolás Merchante Gutiérrez, Dolores Merino Muñoz, Leopoldo Muñoz Medina, Karin Neukam, Mohamed Omar Mohamed-Balghata, Inés Pérez Camacho, Juan Antonio Pineda Vergara, Ana Arizcorreta, Miguel Raffo Márquez, Antonio Rivero Román, Luis Miguel Real, Jesús Rodríguez Baños, María Mancebo, María José Ríos Villegas, Alberto Romero Palacios, José Alberto Terrón, Josefa Ruiz Morales, Antonio Rivero Juárez, Antonio Vergara de Campos, José Hernández Quero, Almudena Torres Cornejo, Ignacio Suarez Lozano, María del Carmen Gálvez Contreras, Montserrat Pérez Pérez, África García Navarrete, Francisco Javier Casas Ciría, Emilio Campos Dávila, José Carlos Roldán Morales, Sandra Lorenzo Moncada, María José Garrido Vega, Guillermo Ojeda Burgos, Javier Borrallo Torrejón, Inés Pérez Camacho, Sandra Lorenzo Moncada. Contact: <ariveror@gmail.com>. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** ARJ FT AR. **Data curation:** MF MPP AC IM SLM PLL AR. **Formal analysis:** MF ARJ FT MPP IM SLM PLL AR. **Funding acquisition:** AC AR. **Investigation:** MF ARJ FT MPP AC SLM AR. **Methodology:** MF ARJ FT AC IM AR. **Project administration:** ARJ AR. **Software:** MF ARJ. **Supervision:** ARJ FT AR. **Validation:** ARJ FT AR. **Writing – original draft:** MF ARJ AR. **Writing – review & editing:** MF ARJ AR FT MPP AC IM SLM PLL AR. [^3]: ¶ Membership of the Grupo de Estudio de Hepatitis Virales (HEPAVIR) of the Sociedad Andaluza de Enfermedades Infecciosas (SAEI) is provided in the Acknowledgments.

# Introduction The relationship between pairs of individuals is an important topic in many areas of population and quantitative genetics. The degree of relationship is measured as the proportion of the genome identical by descent shared by the pair and can be inferred from pedigree information. But there is a variance in the realized, or actual, proportion of genome shared as a consequence of Mendelian sampling and linkage. For instance, two full-sibs can share zero, one or two alleles identical by descent (giving a variance of 1/2 in the number of alleles actually shared), whereas non inbred father and son share exactly one allele (variance of 0). Formulae have been published for the variance of actual relationship for a number of specific types of relatives (see and references therein) but a general formula has not been developed. Deviations of coancestry from the ideal situation of infinite unlinked loci cause linkage disequilibria across pairs of loci. These disequilibria are used extensively nowadays for mapping regions controlling traits (e.g., by genome- wide association studies (GWAS)), genomic selection in crop plants and domestic animal populations, phasing of markers for imputation or quantitative trait locus detection, or control of stratification in GWAS through, for instance, principal component analysis. Therefore, a mathematical formulation of these deviations is critical for the understanding of modern methods of genetic analysis, even if they are based on molecular markers. For instance, Powell et al. suggested a “reconciliation” of identity by state (IBS, critical in GWAS studies) and identity by descent (IBD, used in pedigree analysis) through a notion of base population and the use of Wright’s F fixation indices. However, this assumes ideal populations. In the case of plant and animal breeding populations, pedigree is usually known. For the simplest one locus situation the coancestry between two individuals is the probability that two alleles chosen at random, one from each individual, are identical by descent. The fraternity coefficient is defined as the probability that single locus genotypes (both genes) of two individuals are identical by descent. The purpose of this note is hence to develop the theory to predict the variances and covariances of realized coancestry and fraternity coefficients for pedigreed populations at a single locus. Here we develop a simple expression and algorithm to calculate the variance of these two coefficients and verify it through computer simulation. # Materials and Methods In this section we show how the moments of coancestries can be calculated from the identity coefficients developed by Harris and Gillois, and the generalized kinship coefficients of Karigl. ## The Nine Condensed Identity States In the genealogical analysis we consider ‘virtual’ genes that are all different in the founder population. In this setting, when ignoring the paternal or maternal origin, the relationship between two individuals for one locus can be described exhaustively with the nine condensed identity coefficients. The calculation of these coefficients from pedigrees is fairly well known. shows the nine condensed identity states as they have been presented in the literature, starting from S₁ (the four copies are identical by descent), to S₉ (the four copies are non-identical). The probability of each state S_k is usually denoted as. The nine condensed identity coefficients express the identity given the segregation at the previous generation, for example, two full brothers whose parents were founders can be at the state S₇ if they both received the same copy from the sire and the dam, they can be at state S₈ if they received the same copy from the sire or the dam but not both. Finally they can be at state S₉ if they received different copies from both the sire and the dam. The probabilities of these three states are easy to obtain based on Mendelian segregation rules:, and. Identities of a founder parent- offspring case are even easier, because only has a nonzero probability:. In general, the states and the probabilities define a categorical distribution for each pair of individuals. Each independent locus observed in these two individuals is an independent realized value of this categorical distribution. Using the Iverson brackets notation, the moments of the conventional categorical distribution for the identity states, and are:where “*i = k*” is a probabilistic event meaning that the realized value for a given locus in a given pair of individuals is the state “*k*”, the Iverson variable equals 1 if the event is true and 0 if the event is false. ## The Variance of the Coancestry Coefficient Accounting for Segregation The expected coancestry coefficient between two individuals can be calculated from the condensed identity coefficients using the formula orWhere and. When considering variance due to segregation, each constant has to be replaced by the corresponding Iverson bracket. The variance of the relationship between two individuals due to the segregation can be easily obtained from formula 3 as: or,(note that stands for expected coancestry and for realized or actual coancestry, i.e., given the Mendelian segregation). Replacing formulas (1) and (2) in (4) gives:which has up to 25 nonzero terms. Reordering the terms of formula 5, which can be easily expressed as a function of generalized kinship coefficients. Following Karigl’s notation,where the term is a generalized coefficient of kinship for two pairs of individuals. Covariances between coancestry relationship coefficients can be also derived from the generalized kinship coefficients. The proof is included in the Supplementary Material and the result is the obvious generalization of formula 6. The number of covariances of the coancestry coefficients is very large. If there are different coancestry coefficients, there are different covariances. ## The Variance of the Fraternity Coefficient Accounting for Segregation The fraternity coefficient can be obtained from the condensed states of identity : In this case, formula 5 can also be used but the correct weights must be set, that is. In this case only has up to four nonzero elements.(note that *D* stands for realized fraternity and *d* for expected fraternity). Formula 5, in fact, also holds for any linear aggregate of. For instance, or correspond respectively to inbreeding coefficients of individual *i* and *j*. A formula for the covariance of two fraternity coefficients is given in. # Results In this section, the coancestry of a pair of English Setter dogs is presented in order to illustrate the calculations and to compare the results with MonteCarlo simulations. Afterwards, the coancestries of 11 Latxa sheep were also analyzed in order to test the computational feasibility of the algorithms. ## Example. The Genetic Relationship between the Setters Dash 2nd and Moll 3rd During the XIX century, animal breeders sometimes planned in-and-in pedigrees to keep the blood “pure”. Edward Laverack implemented the matings presented in to achieve “the perfection of form adapted to speed nose and endurance” required for hunting dogs. The intricacy of this ancient pedigree satisfies our testing purposes. Sting and Belle 2^nd, close relatives after 5 generations of full brother matings, were mated with Cora 2^nd and Fred 1^st, collateral relatives. The resulting progeny, Dash 2^nd and Moll 3^rd, are simultaneously half brothers and aunt-nephew and their close relatives are highly inbred. For that reason, the relationship between these two dogs has nine non-null condensed identity coefficients, that is In we present the coancestry coefficient and its variance calculated after formulae (3) and (4), as well as MonteCarlo estimates of coancestries and their the variances, obtained by gene dropping. Both the theoretical values and the MonteCarlo estimates agree perfectly well. It is well known that the inbreeding of an individual is equal to the coancestry between his parents. Nevertheless, it is interesting to note that the variance of the actual inbreeding coefficient of an individual is not equal to the variance of the actual coancestry between his parents. For instance, the expectation of the inbreeding of Dash 2^nd is 0.4297 calculated from the condensed identity coefficients between Dash 2^nd and Moll 3^rd. Although not included in, the coancestry between his parents, Cora 2^nd and Sting, calculated from their shared condensed identity coefficients, is obviously 0.4297. However, the variance of the actual inbreeding of Dash 2^nd is 0.2451, yet the variance of the coancestry between its parents is 0.0922. In general, the variance of the inbreeding will be equal or greater than the corresponding coancestry because it accumulates an extra step of Mendelian segregation. It can be shown algebraically that the variance of the realized inbreeding coefficient F is *f*(1–*f*) (where *f* is the expected inbreeding coefficient) as it should be. In effect, by definition, the realized inbreeding is. Applying formula 6 and the result presented in , i.e., it turns out that.Results presented in lines 3 and 4 of agree with that formula. There is also a good agreement between the theoretical covariance between two coancestries and the values estimated by MonteCarlo simulation. For example, the covariance between the coancestry of the pair Cora 2^nd and Siting and the coancestry of the pair Dash 2^nd and Moll 3^rd has a theoretical value of 0.04188. The covariance between the coancestry of the pair Cora 2^nd and Sting and the coancestry of the pair Fred 1^st and Belle 2^nd is 0.04466. The corresponding MonteCarlo estimates for both covariances are 0.04179 and 0.04436. ## Example. Long Pedigree in Latxa Breed A complex pedigree of 6175 animals of the Latxa sheep was analyzed. We computed the coancestries and variances and covariances of coancestries of the last eleven individuals (in renumbering order). This was an expensive task, for the recursions in Karigl’s method required the computation of \>340,000,000 coefficients. The results for four individuals are in and ; the pedigrees of those four individuals are known for 9–11 generations (often incompletely: the number of equivalent complete generations is respectively 2.85, 4.06, 4.34, 3.28). Individuals 1 and 8 are father and son; individuals 1 and 2 are slightly related. Neither 1 nor 9 are inbred. It can be seen that low relationships (e.g. animals 1 and 2) have proportionally higher variances, as shown by, whereas null relationships have a null variance. Interestingly, there are negative covariances among relationships. Covariances between realized coancestries are most often very small, except the covariances between very close ones, e.g., which is natural. # Discussion In this paper we have shown that variances and covariances of coancestries and inbreeding coefficients can be calculated analytically using the classical condensed identity coefficients. These tasks require using Karigl’s double-pair coancestries. For small pedigrees or pedigrees with many generations, it is better to use tabular algorithms to obtain all double-pair coancestries, but in genealogies with a large number of individuals and a small number of generations, recursive function strategies were implemented to calculate only the coancestries required. An intermediate strategy (i.e., our method) was to store those coefficients that are being calculated for further use. Both approaches are computationally demanding because the number of double pair coancestries is *n*⁴, where *n* is the number of individuals. However, in practice the number of computed coefficients may be much lower: 340×10⁶ relationships were computed in the Latxa example, against a possible total of. The extension to several independent loci is straightforward. The categorical distribution in formulas (1) and (2) has to be replaced by the corresponding multinomial distribution, which basically consists in dividing variances and covariances by the number of loci. However, linkage affects variation in the actual identity coefficients between individuals with the same pedigree, and therefore increases its variance. The treatment of linkage is difficult and has been partially dealt with by several authors (see and references therein). An interesting suggestion by Goddard, is to use the effective number of loci (M_e) defined as the number of loci that provides the same variance of realized relationship as obtained in the more realistic situation. It can be calculated as where is the effective population size and *L* the gamete length in Morgan. Then, we simply divide the variances by this effective number of loci. In our example the variance of coancestry of Dash 2^ndand Moll 3^rd (assuming a genome with 100 linked loci and a recombination fraction of 0.01 among adjacent loci) was 0.0204 (simulation results) and therefore, the effective number of loci is not far from the value obtained using Goddard’s formula (5.5 by using a). Incomplete pedigrees will lead to estimate error. They will tend to bias coancestries and their coancestries downwards and increase their errors. For instance, two individuals with no recorded father will have null coancestries, and their common descendants (if any) will have smaller coancestries and covariances of the coancestries. Methods to deal with unknown paternities include either the use of uncertain paternities, or of pseudo-parents. Rules to derive approximate covariances of coancestries may be obtained from those methods. Instead of pedigree, markers are often used to derive relationships, for instance by computing measures of molecular coancestry and referring them to descent. These relationships are computed after observing the molecular state of the individual (not of their parents), i.e., the Mendelian sampling is somehow “observed”. So, molecular-based relationships do not suffer from sampling due to Mendelian sampling. They suffer, nevertheless, from lack of definition of allelic frequencies (i.e., which base population do we refer to?) and from sampling error due to the finite number of markers and linkage. # Supporting Information We thank Eva Ugarte (NEIKER, Vitoria, Spain), and CONFELAC (Vitoria, Spain) for the Latxa pedigree, and Morris Villarroel (Universidad Politécnica de Madrid, Madrid, Spain) for correcting the English manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Developed the theory: LAGC AL CC MAT. Conceived and designed the experiments: LAGC AL MAT CC. Analyzed the data: LAGC AL. Wrote the paper: LAGC AL MAT CC.

# Introduction While multiple models have been put forth regarding the pathophysiology of schizophrenia, the vast majority of evidence suggests that schizophrenia stems from neurodevelopmental deficits, resulting in disturbance of glutamatergic neurotransmission and especially NMDA receptor-mediated signaling many years later. NMDAR dysfunction models are based upon the observation that psychotomimetic agents such as ketamine and phencyclidine (PCP) induce symptoms of schizophrenia in healthy subjects and provoke relapse in schizophrenics by blocking neurotransmission at NMDA receptors. In rodents, NMDAR antagonists induce schizophrenia-related behavioral abnormalities. While these psychotomimetic effects of NMDAR antagonists have fostered the notion of a hypoglutamatergic state in schizophrenia, recent data suggest that these effects are linked to a loss of NMDAR-mediated GABAergic inhibition, leading to excessive glutamate release and neuronal hyperexcitability in the prefrontal cortex (PFC). In support of this model is the recent demonstration of the antipsychotic efficacy of group II metabotropic glutamate 2/3 (mGlu2/3) agonists, which decrease glutamate release and normalize NMDAR antagonist- induced increases in PFC glutamate. These developments suggest that elevation in the cellular balance of excitation and inhibition within the PFC may be involved in the pathophysiology of schizophrenia. According to the neurodevelopmental model, the etiology of schizophrenia may involve pathologic processes caused by both genetic and environmental factors that begin before the brain approaches its adult anatomical state in adolescence. Multiple lines of evidence from brain pathology, genetics, environmental factors, and gene-environment interactions support this neurodevelopmental model. Numerous reports document the presence of various neuropathological findings in schizophrenia patients, including ventricular enlargement, reduced white and gray matter diffusion anisotropy, and abnormal laminar organization ,. At the perinatal stage, a major risk for schizophrenia is birth complications, especially perinatal hypoxia. Since hypoxia impairs energy-dependent glutamate transport, allowing extracellular glutamate to reach excitotoxic levels, it is possible that increased NMDAR activity caused by excessive glutamate plays a role in the neurodevelopmental deficits of schizophrenia. We recently generated mutant mice in which glutamate receptors are overstimulated by knocking out glutamate transporters GLAST and GLT1, which are essential for maintaining low extracellular glutamate levels. GLAST/GLT1 double- knockout (DKO) mice demonstrate multiple brain defects that are similar to schizophrenia-associated developmental defects, including enlarged lateral ventricles; disorganization of neocortex, hippocampus, and olfactory bulb due to impaired neuronal migration; and defective corticothalamic and thalamocortical axonal projections. All glutamate receptor subunit classes, including NMDA, AMPA, kainite, and metabotropic receptors, are widely expressed throughout the embryonic brain. To confirm the involvement of excess NMDAR signaling in these developmental defects, we generated DKO mice carrying the NMDA receptor 1 subunit (NR1)-null mutation (triple knockout, TKO). NR1 deletion in DKO mice almost completely rescued multiple brain defects including cortical, hippocampal, and olfactory bulb disorganization and defective corticothalamic and thalamocortical axonal projections. # Results NR1 deletion in DKO mice almost completely rescued brain defects in the cerebral cortex, hippocampus, and olfactory bulb at E16.5. In E16.5 WT mice, cerebral cortex is laminated, with the following layers: marginal zone, cortical plate (CP), subplate, intermediate zone (IZ), and ventricular zone. In the DKO cerebral cortex, the CP border on the IZ was obscured. In contrast, this abnormal laminar structure was completely restored in TKO cerebral cortex. Densitometry scans demonstrated apparent border between high optical density (OD) bins in pial side and adjacent low OD bins, corresponding to CP and IZ respectively, in WT and TKO cerebral cortex. In contrast, no apparent border was observed in DKO cerebral cortex. There was a significant difference in the average OD of segment 5 (star), corresponding to the CP border on the IZ, between WT and DKO or TKO and DKO (P\<0.01), but not between WT and TKO. In the DKO hippocampus, the pyramidal cell layer (PCL) was less densely packed. In contrast, abnormal PCL structure was restored in TKO hippocampus. Densitometry scans demonstrated apparent OD peaks corresponding to PCL in WT and TKO hippocampus. In contrast, no OD peak was observed in DKO hippocampus. There was a significant difference in the average OD of segment 5 (star), corresponding to the PCL, between WT and DKO (P\<0.01) or TKO and DKO (P\<0.05), but not between WT and TKO. In the DKO olfactory bulb, the mitral cell layer (MCL) was absent. In contrast, MCL was restored in TKO olfactory bulb. The quantification demonstrated apparent OD peaks corresponding to MCL in WT and TKO olfactory bulbs. In contrast, no OD peak was observed in DKO olfactory bulb. There was a significant difference in the average OD of segment 7 (star), corresponding to the MCL, between WT and DKO (P\<0.01) or TKO and DKO (P\<0.05), but not between WT and TKO. These results suggest brain defects in the cerebral cortex, hippocampus, and olfactory bulb in DKO mice were almost completely rescued by NR1 deletion. Since disturbed laminar and layer organization in DKO mutants resulted from abnormal radial neuronal migration, these results suggested that excess NMDAR activity impaired radial migration. Furthermore, NR1 inactivation rescued subplate neurons, which are vulnerable to early hypoxic-ischemic brain injury and important for the development of thalamus–neocortex connections. Consistent with subplate neuron rescue, severe disruptions to L1-positive corticothalamic and thalamocortical axonal projections were almost completely restored. Recent report suggested that NR1 deletion from GABAergic neurons contributes cortical hyperexcitability and schizophrenia-associated behaviors. We examined the development of GABAergic neurons using anti-GABA and anti-GAD67 antibodies. In DKO mice, the expression levels of GAD67 and GABA were increased, whereas their expression pattern was normal. In TKO mice, the elevated GAD67 and GABA expression returned to normal. These results suggest that overstimulation of NMDAR can accelerate the maturation of GABAergic neurons. Taken together, these results suggest that excess glutamatergic signaling via NMDARs, not AMPA, kainite, or metabotropic glutamate receptors, compromises early brain developmental processes. # Discussion Although NMDAR hypofunction does not impair embryonic brain development, its excessive function appears to be detrimental to the developing brain. Excessive NMDAR activation recapitulated schizophrenia-related pathologies in embryonic mouse brain, including enlarged lateral ventricles, disorganization of neocortex and hippocampus, and defective corticothalamic and thalamocortical axonal projections. These data raise the possibility that molecular abnormalities leading to hyper-NMDAR function in the embryonic brain could be risk factors for schizophrenia. In contrast to glutamatergic neural circuit, excessive NMDAR activation in embryonic brain does not impair GABAergic neural circuit development. Instead, overstimulation of NMDAR can accelerate the maturation of GABAergic neurons. These results agree with the previous finding that chronic NMDA exposure accelerate development of GABAergic inhibition in the superior colliculus and may be consistent with the previous result that genetic NR1 deletion in GABAergic neurons impairs the maturation of GABAergic neurons. Since GABA is excitatory in the embryonic brain, the functional consequence of GABA and GAD67 elevations warrants future research. Fetal hypoxia is a common risk factor for a range of neurological and psychiatric disorders including schizophrenia, autism, and epilepsy. These disorders are associated with disruption of the laminar organization of the cerebral cortex. It is suggested that at least 50% of known susceptibility genes for schizophrenia are more likely than randomly-selected genes to be regulated by hypoxia. Furthermore, GLT1, an “ischemia–hypoxia response gene,” is downregulated in the amygdala of patients with schizophrenia. Since energy failure, as in hypoxic episodes, impairs energy-dependent glutamate transport allowing extracellular glutamate to reach excitotoxic levels, our results suggest that fetal hypoxia may induce neurodevelopmental abnormalities via overstimulation of NMDARs. Recently, chromosomal microdeletions of GLAST and GLT1 were linked to schizophrenia – and Wilms tumor, Aniridia, Genitourinary malformations and mental Retardation (WAGR) syndrome, respectively, and GLT1 (on 11p12-p13) was located near an autism risk locus. DKO mice reproduce important pathophysiological events documented in human mental disorders, including impaired neuronal migration and cortical connections. Thus, DKO mice provide a useful tool for elucidating how embryonic disturbance of glutamate may be associated with the neurodevelopmental defects underlying neuropsychiatric developmental disorders. Complex behavioral phenotypes such as those observed in schizophrenia and autism are hypothesized to arise from elevation in the ratio of cortical cellular excitation to inhibition (cellular E/I balance). This hypothesis could unify diverse genetic and environmental factors under a common pathophysiological principle. Our results indicate that elevated cellular E/I balance in the embryonic brain impairs brain development via hyperfunction of NMDARs. Elevation of cellular E/I balance in the adult brain by treatment with NMDAR antagonists such as ketamine and PCP mimics the symptoms of schizophrenia via the hypofunction of NMDARs in GABAergic interneurons,. Thus, the elevated cellular E/I balance hypothesis could account for disturbance of neurodevelopment and neurotransmission in neuropsychiatric developmental disorders. Our results suggest that offset of excessive NMDAR activity may prevent abnormal brain development due to excess glutamatergic signaling, thereby avoiding later mental disorders. # Materials and Methods ## Animals GLAST, GLT1, and the NMDAR-1 subunit (NR1) mutant mice were described previously. Triple-knockout mice were generated by crossing GLT1(+/−) mice with NR1(+/−) mice to obtain GLT1/NR1 double-heterozygous mice \[GLT1(+/−)/NR1(+/−)\], and then breeding GLAST(−/−) male mice with GLT1(+/−)/NR1(+/−) female mice to obtain GLAST/GLT1/NR1 triple-heterozygous mice \[GLAST(+/−)/GLT1(+/−)/NR1(+/−)\]. Finally, GLAST(+/−)/GLT1(+/−)/NR1(+/−) mice were interbred to obtain GLAST/GLT1/NR1 triple-knockout mice. All mice were of C57BL/6J background. Mice were genotyped with tail genomic DNA by described protocols for GLAST and GLT1. For NR1, a set of NR1 primers for wild-type allele and a set of primers for mutant allele (NR1, 5′-AGGAGTGGAACGGAATGATG-3′; Neo, 5′-CAGAAAGCGAAGGAGCAAAG-3′) were used. The day of vaginal plug detection was designated E0.5. All research and animal care procedures were approved by the Tokyo Medical and Dental University Animal Care and Use Committee (0120292C). ## Histology and Immunohistochemistry As previously described, at E16.5, pregnant mice were killed by cervical dislocation, and embryos were fixed with Bouin’s fixative. Three independent embryos for each genotype were used. Paraffin-embedded sections (4 µm) were prepared. Hematoxylin staining was performed following standard protocols. Images were acquired under the constant exposure condition using a BIOREVO BZ-9000 microscope (Keyence, Osaka, Japan). For immunohistochemistry, deparaffinized sections were placed in an antigen retrieval solution (Immunosaver, Nisshin EM, Tokyo, Japan) for 45 min at 95°C and incubated with primary antibodies for 12 h at 4°C. The following antibodies were used: monoclonal anti-microtubule-associated protein 2 (MAP2) (SMI, Lutherville, MD), polyclonal anti-L1 (a gift from H. Kamiguchi, RIKEN Brain Science Institute), monoclonal anti- glutamic acid decarboxylase 67 (GAD67) (Millipore, Temecula, CA) and polyclonal anti-gamma-aminobutyric acid (GABA) (Sigma, St. Louis, MO). MAP2 was visualized by Alexa Fluor 488-conjugated goat anti-mouse IgG (Molecular Probes, Eugene, OR). L1, GAD67 and GABA were visualized by Envision-Plus Rabbit HRP System (DakoCytomation, Carpinteria, CA) and TSA plus Cyanine 3 system (Perkin-Elmer Life Sciences, Boston, MA) according to manufacturer’s instructions. Images were acquired on an LSM 510 META laser-scanning confocal microscope (Carl Zeiss, Thornwood, NY). ## Normalized Optical Density Measurement of Hematoxylin Stained Sections To quantify the optical density (OD) of the hematoxylin stained sections, boxed regions (100 µm width) shown in, were selected, converted to 8-bit encoded grayscale and inverted to white on black, and then pixel intensity profiles along pia-ventricle axis in each region were measured using ImageJ software (3–5 sections for each embryo). In order to adjust intersectional variability of the absolute level of intensity, minimal OD value within each region was subtracted as background from individual OD values, then, individual OD values were normalized to the maximal OD value on each region. In order to adjust variations of section size, the pia-ventricular extent of the cerebral cortex, hippocampus and olfactory bulb were normalized by dividing those into 100 bins equally spaced along the pia-ventricular axis. Average values of normalized OD in each bin were plotted along the pia-ventricular axis. Three independent embryos for each genotype were used. Data are mean ± s.e.m. Statistical significance was calculated every 10 bins (segment) by one-way ANOVA followed by Scheffé’s post- hoc analysis. ## RT-PCR Total RNA was extracted from E16.5 brain using TRIzol Reagent (Invitrogen, Carlsbad, CA). Complementary DNA (cDNA) was synthesized using PrimeScript RT reagent kit with genomic DNA Eraser (Takara, Shiga, Japan). Reverse transcription polymerase chain reaction (RT-PCR) was performed using the following primer sets: GLAST (forward, 5′-AAGCCATCATGCGATTGG-3′; reverse, 5′-CTTGAAAGTGATGGGTAGGG-3′), GLT1 (forward, 5′-CTGGTGCAAGCCTGTTTCC-3′; reverse, 5′- GCCTGTTCACCCATCTTCC-3′), NR1 (forward, 5′-GCCGATTTAAGGTGAACAGC-3′; reverse, 5′-AATTGTGCTTCTCCATGTGC-3′) and GAPDH (forward, 5′-ACTTTGTCAAGCTCATTTCC-3′; reverse, 5′-TGCAGCGAACTTTATTGATG-3′). The thermal cycler conditions were 5 min at 94°C and then 35 cycles of 20 s at 94°C, 30 s at 60°C, and 45 s at 72°C, followed by 7 min at 72°C and 60 min at 4°C. We thank S. Tonegawa for providing NR1 mutants, H. Kamiguchi for providing the L1 antibody, and T.R. Matsugami, H. Aizawa and Y. Ito for their technical discussions. [^1]: Conceived and designed the experiments: KT. Performed the experiments: TA YI YKT. Analyzed the data: TA YI. Wrote the paper: KT. [^2]: The authors have declared that no competing interests exist.

# Introduction The cornea is composed of the outer stratified squamous epithelium, the intermediate stroma, and the inner endothelium. The stroma consists of type I/V collagen fibers and proteoglycan decorin, lumican, keratocan, and osteoglycin/mimecan. Type III collagen is also present in low proportions but it increases during wound healing and inflammation. The keratocytes, located between the collagen lamellae in the stroma, are a population of quiescent, mesenchymal-derived cells. Despite being sparsely arranged in the stroma, keratocytes form an interconnected cellular network through long dendritic processes. Upon injury, the keratocytes may either undergo apoptosis or transdifferentiate into an activated fibroblastic repair phenotype. This fibroblastic phenotype seen in corneal wound healing resembles the phenotype that is seen under culture conditions *in vitro*. Standard culture conditions for corneal stromal cells (10% Fetal Calf Serum) alter the keratocytes from their *in situ* phenotype. Of the total glycosaminoglycans (GAGs) synthesized by corneas in organ culture, 47% are keratan sulfates. However, stromal cells derived from bovine, rabbit and human corneas, cultured under standard conditions, have been reported to produce moderate (15%), little (3%), or no keratan sulfates, respectively. Serum- cultured corneal stromal cells also express the fibronectin receptor a5b1 (which is not expressed by keratocytes *in situ*), they have a fibroblastic morphology, and their actin cytoskeleton resembles the one of corneal fibroblasts or even corneal myofibroblasts. However, when primary cultures of corneal stromal cells are cultured in serum-free medium, they exhibit a dendritic morphology and extensive dendritic processes, and their appearance is similar to keratocytes *in situ* and distinctly different from the fibroblastic or myofibroblastic appearance of keratocytes grown in serum-containing medium. Stromal wound healing consists of three stages: repair, regeneration (proliferation and migration of keratocytes), and remodeling, and has been shown to involve a complex interplay between cytokines, growth factors, and chemokines. Based on a number of observations, it is likely that also other signal substances are at play such as neuropeptides and other classical neuronal transmitters. To further elucidate the role of such substances in corneal wound healing it is of importance to study their expression profiles in corneal stromal cells *in situ* and *in vitro*. Similarly, the presence of their receptors is also important to delineate, since the influence of neuropeptides/neurotransmitters on primary corneal cells may depend on not only non-neuronal but also neuronal signaling in the cornea. The cornea is one of the most innervated tissues in the body, containing nerve fibers derived from the trigeminal ganglion. Previous studies have demonstrated that the corneal nerve fibers exert important trophic influences and contribute to the homeostatic maintenance of corneal epithelium. However, the potential role of neuropeptides or neurotransmitters in corneal stromal wound healing remains poorly understood. Studies have confirmed the involvement of nerves, especially nerves secreting neuropeptides, in the processes of diabetic/skin wound healing, including inflammation, epithelialization, and fibrogenesis. The neuropeptides substance P (SP) and neurokinin A (NKA) belong to the tachykinin family which have a variety of pharmacological actions both in the central nervous system and in the periphery. SP and its preferred receptor, the neurokinin-1 receptor (NK-1R), have been found to be expressed by a wide range of not only neuronal but also non-neuronal human cells NKA signals through its high affinity receptor, the neurokinin-2 receptor (NK-2R). Acetylcholine (ACh) is regarded as a classical neurotransmitter, and it acts through either nicotinic ACh receptors (nAChRs) or muscarinic ACh receptors (mAChRs). Catecholamines (dopamine, adrenaline and noradrenaline) constitute a class of chemical neurotransmitters and hormones that occupy key positions in regulation of physiological processes. Adrenaline and noradrenaline act through α (α₁ and α₂) and β (β₁ and β₂) adrenoceptors (adrenergic receptors) in target cells whereas dopamine acts through its D₁ (D₁ and D₅) and D₂ (D₂, D₃ and D₄) classes of receptors. Glutamate is a non-essential amino acid, which binds to the N-Methyl-D-aspartic acid receptor (NMDAR) and acts as a major neurotransmitter in the mammalian central nervous system. All these neuropeptides and neurotransmitters are known to exert effects on processes involved in wound healing. Considering the importance of keratocytes in corneal stromal wound healing, it would be of interest to study the expression profiles of neuropeptides and neurotransmitters in human keratocytes. Therefore, as it has not been documented before and as a basis for further functional studies on the role of neuropeptides/neurotransmitters in corneal wound healing, the present work studied the endogenous intracellular and secreted levels of the tachykinins SP and NKA, and of ACh, catecholamines (adrenaline, noradrenaline and dopamine), and glutamate, as well as the expression profiles of their receptors, in human primary keratocytes *in vitro* and in keratocytes of human corneal tissue sections *in situ*. # Materials and Methods ## Collection of human corneas Healthy human corneas from deceased individuals who had chosen, when alive, to donate their corneas post-mortem for transplantation or research, through written consent and according to Swedish law, were kept in a tissue bank at the corneal donation center at the University Hospital of Umeå, Sweden (Hornhinnebanken Vävnadsinrättningen Laboratoriemedicin Norrlands universitetssjukhus, (<http://www.vavnad.se/cms/sites/vavnadsradet/home/hornhinnor/nationellt- arbete-1/umea.html>). If these healthy donated corneas were not used for transplantation after their collection, they were delivered to the laboratory for research purpose. The Regional Ethical Review Board in Umeå reviewed the study and determined it to be exempt from the requirement for approval (2010-373-31M). The study was performed according to the principles of the Declaration of Helsinki. ## Isolation and primary culture of human keratocytes Healthy human corneas were obtained from donated transplantation grafts, as described in the previous section. Samples were scraped using a sterile scalpel to remove any remaining epithelial or endothelial cells before being washed in sterile Hanks’ Balanced Salt Solution (HBSS; Invitrogen, Carlsbad, CA, USA, \# 14170–138). Using a scalpel, the cornea was separated into two parts: the central part (round shaped, middle part of the cornea) and the peripheral part (ring shaped part consisting of limbus and surrounding tissue). Each part was then minced with a scalpel and digested in 2 mg/ml collagenase (Clostridopeptidase A, Sigma, St. Louis, MO, USA, \# C-1030) diluted in Dulbecco's Modified Eagle Medium: Nutrient Mixture F-12 (DMEM / F-12, Gibco, Carlsbad, CA, USA, \# 21331–046) for 30 minutes at 37°C. The samples were then transferred and cultured in DMEM / F-12 medium supplemented with 2% Fetal Bovine Serum (FBS; Gibco, \# 16000), 1% penicillin-streptomycin (Gibco, \# 15410) and 0.2% L-Glutamine (Gibco, \# 25030). The medium was replaced every second to third day until the cells reached confluence. Confluent cells were detached with 0.05% Trypsin-EDTA (Gibco, \# 25300) and split in a 1:3 ratio. Cells from passage 3 to 4 were used for experiments. For the experiments, cells were maintained in DMEM / F-12 supplemented with either 2% or 0% FBS (western blots, EIAs, ELISAs, colorimetric and fluorometric assays), or only 2% FBS (immunocytochemistry and real-time qPCR). Central part of the cornea and peripheral part (consisting of limbus + adjacent tissue) were analyzed and compared in this study. ## Human corneal tissue sectioning Healthy human corneas were fixed in a solution of 4% formaldehyde diluted in 0.1 M Phosphate Buffered Saline (PBS) (pH 7.4) overnight. After three washes in a Tyrode’s solution containing 10% sucrose the sample was mounted on a piece of cardboard with OCT embedding medium (Miles Laboratories, Elkhart, IN, USA) and frozen in liquid nitrogen-chilled propane. The corneas were then cut to 7 μm sections in a cryostat and sections were stored at -24°C until stained. ## Immunocytochemistry 10⁴ keratocytes per well were seeded in 8 well chamber slides (Corning, Corning, NY, USA \# 354118) overnight (in DMEM/F12 supplemented with 2% FBS) before being fixed in 2% paraformaldehyde (PFA) diluted in 0.1 M PBS (pH 7.4) for 10 minutes. Cells were permeabilized with 0.25% Triton X-100 in PBS for 10 minutes. Fixed cells were washed repeatedly in PBS and then blocked with 1:20 diluted normal serum corresponding to the host species of the secondary antibody for 15 minutes. After carefully disposing of the serum, cells were incubated with the primary antibody overnight at 4°C. Washing and blocking were repeated and secondary antibody was added for 30 minutes at 37°C. Cells were then washed before being mounted in Vectashield mounting medium for fluorescence (Vector Laboratories, Burlingame, CA, USA \# H-1500). A control well was also prepared for each secondary antibody by replacing the primary antibody with PBS. A Zeiss Axioskop 2 plus microscope equipped with epifluorescence and an Olympus DP70 digital camera were used for analysis. ## Immunohistochemistry Tissue sections were fixed in 2% PFA and permeabilized with 1% Triton X-100 for 20 minutes. Slides were then washed 3 times in PBS and blocked with 1:20 diluted normal serum for 15 minutes. After carefully disposing of the serum, cells were incubated with the primary antibody overnight at 4°C. The same antibodies were used as for immunocytochemistry. Washing and blocking were repeated and secondary antibody was added for 30 minutes at 37°C. Sections were then washed before being mounted in Vectashield mounting medium for fluorescence (Vector Laboratories, \# H-1500). A control slide was also prepared for each secondary antibody by replacing the primary antibody with PBS. A Zeiss Axioskop 2 plus microscope equipped with epifluorescence and an Olympus DP70 digital camera were used for analysis. ## Measurements of neuropeptides and neurotransmitters 250,000 keratocytes were seeded into 6 wells plates in triplicates in DMEM / F-12 medium supplemented with either 2% or 0% FBS. After 24 h of culture, supernatants were collected and cells were lysed in RIPA lysis buffer. Neuropeptides and neurotransmitters were measured using following kits according to manufacturer’s specifications: Substance P EIA kit (Phoenix Pharmaceuticals, Burlingame, CA, USA, \# EK-061-05), Neurokinin A EIA kit (RayBiotech, Norcross, GA, USA, \# EIA-NEA1), Amplex Acetylcholine/Acetlycholinesterase Assay Kit (Life Technologies, Carlsbad, CA, USA), Glutamate Assay kit (Abcam, \# 83389), and Adrenaline/Noradrenaline/Dopamine 3-CAT ELISA (Labor Diagnostika Nord, Nordhorn, Germany, \# BA E-5600). ## Western blot 250,000 keratocytes were seeded into 6 well plates in DMEM / F-12 medium supplemented with either 2% or 0% FBS. After 24h of culture, cells were washed with PBS and frozen at -80°C overnight. Cells were lysed in RIPA lysis buffer supplemented with 0.5% Proteinase inhibitor cocktail (Sigma, \# P1860) and diluted in Laemmli Sample buffer supplemented with 2-mercaptoethanol. The cell lysates were separated by sodium dodecyl sulfate / polyacrylamide gel electrophoresis and transferred to PVDF membranes. Blots were incubated with primary antibodies listed in. The antigens were detected with horseradish peroxidase-conjugated secondary antibodies listed in. The images were taken by Odyssey Fc imaging system (LI-COR, Lincoln, NE, USA). Densitometry was performed using Image J analysis software (NIH). ## Real-time qPCR 250,000 keratocytes were seeded into 6 wells plates in triplicates in DMEM/F12 supplemented with 2% FBS. After 24 h RNA was extracted using RNeasy Mini Kit (Qiagen, Venlo, The Netherlands, \#74104). 900 ng of RNA was reverse transcribed into cDNA using High Capacity cDNA Reverse Transcription Kit (Life Technologies, Carlsbad, CA, USA, \# 4368814). To determine the gene expression TaqMan Gene Expression Assays (Applied Biosystems, Carlsbad, USA) were used. cDNA obtained from 40 ng of RNA was run in duplicates by ViiA 7 Real-Time PCR system (Applied Biosystems), and analyzed by ViiA 7 Software (Applied Biosystems). Expression of genes of interest was normalized to the expression of 18S housekeeping gene. Analyzed genes are listed in. ## Statistical analysis Data represent mean ± SD of three replicates. Independent samples *t*-test or one-way ANOVA with Bonferroni post hoc test were applied to determine if there was a statistically significant difference between samples. GraphPad Prism 5 software was used for data analysis. Significance was predetermined at p\<0.05. Experiments were performed successfully at least 3 times. Primary cultures of keratocytes obtained from different cornea donors were assessed individually. # Results ## Characterization of isolated keratocytes *in vitro* Immunohistochemistry of the human corneal tissue showed that the cornea expresses keratocan, a cornea-specific keratan sulfate proteoglycan, and lumican which is a keratan sulfate proteoglycan produced by keratocytes. In order to determine the characteristics and phenotype of cultured corneal cells, and how the culturing conditions affect the cultured cells phenotype, four keratocyte markers were used: keratocan, lumican, CD34 (hematopoietic stem cell marker and an adhesion molecule and aldehyde dehydrogenase (ALDH; corneal crystalline which contributes to maintaining corneal transparency). Western blot and densitometry analyses were performed. Two culture conditions were tested. Central and peripheral keratocytes were cultured in either DMEM/F12 medium supplemented with 2% FBS or 0% FBS. Central keratocytes grown in these two conditions, as well as peripheral keratocytes grown in these two conditions, were compared. Moreover, comparison was made between central and peripheral keratocytes. Keratocan was expressed abundantly in cultured cells. Its expression was significantly higher in peripheral keratocytes cultured in 2% FBS than in central keratocytes grown in 2% FBS. Central keratocytes grown in 0% FBS expressed more keratocan than central keratocytes grown in 2% FBS and peripheral keratocytes grown in 0% FBS. However, peripheral keratocytes grown in 0% FBS had significantly lower expression of keratocan when compared to peripheral keratocytes grown in 2% FBS. Lumican was also expressed abundantly in cultured cells. Peripheral keratocytes expressed significantly more lumican than central keratocytes. CD34 was also expressed in cultured keratocytes. Its expression was significantly higher in peripheral keratocytes grown in 2% FBS than in central keratocytes grown in 2% FBS and peripheral keratocytes grown in 0% FBS. CD34 expression was also significantly higher in central keratocytes grown in 2% FBS in comparison to central keratocytes grown in 0% FBS. ALDH was expressed in cultured keratocytes but at very low levels. Peripheral keratocytes grown in 2% FBS expressed significantly more ALDH than central keratocytes grown in 2% FBS and peripheral keratocytes grown in 0% FBS. Moreover, expression of procollagen I and collagen I was assessed. Procollagen I was significantly higher in central keratocytes grown in 2% FBS than in peripheral keratocytes grown in same conditions. Collagen I expression was significantly higher in central keratocytes grown in 2% FBS than in central keratocytes grown in 0% FBS and peripheral keratocytes grown in 2% FBS. Also, collagen I expression was significantly lower in central keratocytes grown in 0% FBS than in peripheral keratocytes grown in same conditions. Additionally, cultured keratocytes did not express CD45, a common marker for bone marrow-derived cells, CD31, a marker of endothelial cells and expressed only small amount of α-SMA, a marker of myofibroblasts (data not shown). ## Human keratocytes express genes coding for neuropeptides, and for catecholamine and glutamate synthesizing enzymes In order to study if cultured keratocytes express genes coding for neuropeptides or for enzymes taking part in the synthesis of catecholamines and glutamate, qPCR was performed. The gene coding for substance P and neurokinin A (TAC1) was present both in central and peripheral keratocytes. The DDC gene, which codes for DOPA decarboxylase, an enzyme involved in catecholamine synthesis, was also present in both central and peripheral keratocytes. The glutaminase gene (GLS), which codes for glutaminase that generates glutamate from glutamine, and the GOT gene coding for aspartate aminotransferase, which plays a role in amino acid metabolism, were both present in central and peripheral keratocytes. ## Human keratocytes express genes coding for receptors of neuropeptides, catecholamines, and acetylcholine Genes coding for receptors of neuropeptides, catecholamines, and ACh were analyzed by qPCR. TACR1 gene, coding for the preferred receptor of SP, the NK-1R, and TACR2 gene, which is the gene for the preferred receptor of neurokinin A (NK-2R), were present in both central and peripheral keratocytes. mRNA for adrenaline and noradrenaline adrenoreceptors α₁ (ADRA1B) and β₂ (ADRB2) were present in central keratocytes and peripheral keratocytes. DRD2 gene, coding for another catecholamine receptor, the dopamine receptor D₂, was also present in both central and peripheral keratocytes. Genes for muscarinic acetylcholine receptors–CHRM1 (for mAChR M₁), CHRM2 (for mAChR M₂), CHRM3 (for mAChR M₃), CHRM4 (for mAChR M₄), and CHRM5 (for mAChR M₅) were found in central and peripheral keratocytes. ## Human keratocytes produce substance P and neurokinin A, and express their preferred receptors Presence and differences in amounts of the neuropeptides SP and NKA in cultures of human keratocytes was measured by EIA. Two culture conditions were tested. Central and peripheral keratocytes were cultured in either DMEM/F12 medium supplemented with 2% FBS or 0% FBS. Central keratocytes grown in these two conditions, as well as peripheral keratocytes grown in these two conditions, were compared. Moreover, comparison was made between central and peripheral keratocytes. SP levels were significantly higher in supernatants of central keratocytes grown in medium supplemented with 2% FBS than in peripheral keratocytes grown in medium supplemented with 2% FBS. SP secretion was significantly higher in central keratocytes grown in 2% FBS than central keratocytes grown in 0% FBS. However, there were no significant differences in SP levels in lysates of keratocytes. NKA secretion was also significantly higher in central keratocytes grown in 2% FBS when compared to central keratocytes grown in 0% FBS. Expression of SP and of its preferred receptor, NK-1R, was furthermore analyzed by immunohistochemistry and immunocytochemistry. Both the keratocytes in tissue sections and the cultured keratocytes expressed SP and a full-length version of NK-1R, composed of 407 amino acid residues. Expression of NKA and of its receptor, NK-2R, was also analyzed by immunohistochemistry and immunocytochemistry. Both the keratocytes in tissue sections and the cultured keratocytes expressed NKA and NK-2R. To quantify differences in expression of NK-1R and NK-2R between the different culturing conditions and central and peripheral keratocytes, western blot and densitometry analyses were performed. Expression of NK-1R was significantly higher in central keratocytes grown in 2% FBS when compared to peripheral keratocytes grown in same conditions. Expression of NK-1R was also significantly higher in central keratocytes grown in 2% FBS than in central keratocytes grown in 0% FBS. On the other hand, expression of NK-2R was significantly higher in central keratocytes grown in 0% FBS than in peripheral keratocytes grown in same conditions. NK-2R expression was significantly higher in central keratocytes grown in 0% FBS than in central keratocytes grown in 2%. ## Human keratocytes produce catecholamines and express their receptors Presence and differences in amounts of catecholamines (adrenaline, noradrenaline, and dopamine) was measured by ELISA. Again, two culture conditions were tested. Central and peripheral keratocytes were cultured in either DMEM/F12 medium supplemented with 2% FBS or 0% FBS. Central keratocytes grown in these two conditions, as well as peripheral keratocytes grown in these two conditions, were compared. Moreover, comparison was made between central and peripheral keratocytes. Adrenaline was present in both culture supernatant and lysates of cultured central and peripheral keratocytes, with significantly higher levels in supernatants collected from peripheral cells. However, in cell lysates levels of intracellular adrenaline were significantly higher in central keratocytes grown in 2% FBS in comparison with peripheral keratocytes grown in 2% FBS. Central keratocytes grown in 2% FBS showed higher levels of adrenaline than central keratocytes grown in 0% FBS. Moreover, peripheral keratocytes grown in 0% FBS had higher levels of adrenaline than central keratocytes grown in same conditions. Noradrenaline was secreted from both central and peripheral keratocytes, and was also present in cell lysates. Peripheral keratocytes grown in 0% FBS secreted significantly higher amounts of noradrenaline than central keratocytes grown in same conditions. Intracellular levels of noradrenaline were significantly higher in peripheral keratocytes grown in 2% FBS than in peripheral keratocytes grown in 0% FBS. Dopamine was secreted from both central and peripheral keratocytes and its levels were significantly higher in lysates of central keratocytes than peripheral keratocytes. Expression of tyrosine hydroxylase (TH), the enzyme responsible for catalyzing the conversion of the amino acid L-tyrosine to L-DOPA (precursor of adrenaline, noradrenaline, and dopamine), was analyzed by immunohistochemistry and immunocytochemistry. The cultured keratocytes expressed TH whereas keratocytes in tissue sections did not. In order to quantify and compare TH expression under different culturing conditions and two types of keratocytes (central vs. peripheral), western blot and densitometry analyses were performed. Both central and peripheral keratocytes grown in 2% FBS expressed more TH than cells grown in 0%. Moreover, central keratocytes grown in 2% FBS expressed more TH than peripheral keratocytes grown in same conditions. Adrenaline and noradrenaline adrenoreceptors α₁ and β₂ (ADRA1, ADRB2, respectively) were not expressed in corneal tissue sections Expression of these receptors in cultured keratocytes was inconclusive. However, dopamine D2 receptor (DRD2) was expressed in both keratocytes in tissue sections and in cultured keratocytes. Western blot and densitometry analyses showed that central keratocytes grown in 2% FBS had significantly higher expression of DRD2 than both central keratocytes grown in 0% FBS and peripheral cells grown in 2% FBS. The expression of TH in cultured cells but not in cells of tissue sections might be a result of culturing conditions. ## Human keratocytes produce acetylcholine and express muscarinic acetylcholine receptors Presence and differences in amount of ACh was measured by fluorometric assay. Two culture conditions were tested. Central and peripheral keratocytes were cultured in either DMEM/F12 medium supplemented with 2% FBS or 0% FBS. Central keratocytes grown in these two conditions, as well as peripheral keratocytes grown in these two conditions, were compared. Moreover, comparison was made between central and peripheral keratocytes. Both central and peripheral keratocytes secreted ACh, and ACh was also present in both central and peripheral keratocyte lysates. Expression of choline acetyltransferase (ChAT)–an enzyme that is crucial for ACh synthesis–was analyzed by immunohistochemistry and immunocytochemistry. Both the keratocytes in tissue sections and the cultured keratocytes expressed ChAT. Muscarinic acetylcholine receptors: mAChR M₁, mAChR M₂, mAChR M₃, mAChR M₄, and mAChR M₅ expression was analyzed by immunohistochemistry and immunocytochemistry. mAChR M₁ was present in cultured cells but not in tissue sections, mAChR M₂ was absent in both cultured cells and tissue sections. The remaining receptor subtypes (mAChR M₃, mAChR M₄, and mAChR M₅) were expressed in both tissue sections and cultured cells. Presence of mAChR M₁ in culture but its absence in tissue sections might, again, be a result of culturing conditions and/or stressed cells. In order to quantify and compare expression levels of muscarinic acetylcholine receptors under different culturing conditions and two types of keratocytes, western blot and densitometry analyses were performed. mAChR M₁ expression was significantly higher in central keratocytes grown in 0% FBS than in central keratocytes grown in 2% FBS, and also significantly higher in peripheral keratocytes grown in 2% FBS than in central keratocytes grown in same conditions, and finally significantly higher in central keratocytes grown in 0% FBS than in peripheral keratocytes grown in same condition. mAChR M₃ expression was significantly higher in central keratocytes grown in 2% FBS than in both central keratocytes grown in 0% FBS and peripheral keratocytes grown in 2% FBS. Central keratocytes grown in 0% FBS expressed significantly more mAChR M₃ than peripheral keratocytes grown in same the condition, and peripheral keratocytes grown in 2% FBS expressed more mAChR M₃ than peripheral cells grown in 0% FBS. Expression of mAChR M₄ was significantly higher in central keratocytes grown in 2% FBS than in peripheral keratocytes grown in same the condition. Expression of mAChR₅ was significantly higher in peripheral keratocytes grown in 2% FBS than in both central keratocytes grown in the same condition and in peripheral keratocytes grown in 0% FBS. mAChR₅ expression was also significantly higher in central keratocytes grown in 0% FBS than in peripheral keratocytes grown in same the condition. ## Human keratocytes produce glutamate and express glutamate receptor NMDAR1 Presence and differences in amount of glutamate was measured by glutamate assay. Two culture conditions were tested. Central and peripheral keratocytes were cultured in either DMEM/F12 medium supplemented with 2% FBS or 0% FBS. Central keratocytes grown in these two conditions, as well as peripheral keratocytes grown in these two conditions, were compared. Moreover, comparison was made between central and peripheral keratocytes. Levels of glutamate were significantly higher in central and peripheral keratocytes culture supernatants collected from cells grown in 0% FBS than in supernatants collected from cells grown in 2% FBS. Levels of glutamate in cell lysates were significantly higher in peripheral keratocytes grown in 2% FBS than in peripheral keratocytes grown in 0% FBS. Expression of glutamate and its receptor NMDAR1 was also analyzed by immunohistochemistry and immunocytochemistry. Glutamate was expressed in both keratocytes in tissue sections and in cultured cells. Glutamate receptor NMDAR1 was expressed in keratocytes in tissue sections and in cultured cells. To quantify and compare NMDAR1 expression under different culturing conditions and two types of keratocytes, western blot and densitometry analyses were performed. NMDAR1 expression was significantly higher in both central and peripheral keratocytes grown in 2% FBS than in central and peripheral keratocytes grown in 0% FBS. Expression of NMDAR1 was also significantly higher in central keratocytes grown in 0% FBS than in peripheral keratocytes grown in the same condition. # Discussion The present study shows that human keratocytes express an array of neuropeptides and neurotransmitters that traditionally have been seen as neuron specific. It also shows that keratocytes express receptors for these substances, making them susceptible to stimulation by neuropeptides/neurotransmitters in an autocrine and/or paracrine manner, in the latter case possibly through secretion of substances from corneal nerves. The possible importance of this expression profile of neuropeptides and neurotransmitters in human corneal stromal cells is here discussed in the context of the known functional properties of these substances and their relevance for processes included in corneal wound healing. ## Methodological aspects: Phenotype of human keratocytes in primary cultures in vitro In the absence of injury or pathology, keratocytes of the corneal stroma are considered as quiescent cells, with a low rate of proliferation and apoptosis. However the culture conditions (including the factors added to the medium), the concentration of cells, and the stiffness of the support can influence the cell phenotype *in vitro*. For example, it has been shown that FBS can induce a fibroblastic phenotype on keratocytes. To determine whether the culture conditions of the present study (2% FBS or 0% FBS) modified the cells’ phenotype and/or function, expression of several keratocyte markers was analyzed and compared between different culturing conditions. Keratocytes isolated from the stroma and cultured in serum free medium should retain their phenotype and maintain the expression of specific markers such as aldehyde dehydrogenase (ALDH) and keratocan. Keratocan, a cornea-specific keratan sulfate proteoglycan was abundantly expressed in cultured cells regardless of the culturing condition (2% and 0% FBS). However, ALDH, a corneal crystalline which helps to maintain the cellular transparency was expressed in low amounts and surprisingly cells cultured in serum free medium expressed lower amounts of this marker. Lumican is a 38 kDa protein belonging to the small leucine-rich proteoglycans and is expressed in the extracellular matrix (ECM) of various tissues including the corneal stroma in which it is produced by keratocytes. It has been well described that during activation of keratocytes to myofibroblasts the cells decrease the expression of keratan sulfate proteoglycans and increase the expression of α-SMA, a marker of myofibroblasts. Under our culture conditions, cultured cells expressed high amounts of lumican, with non-significant differences between cells cultured in 2% and 0% FBS. CD34 is a well-established marker of quiescent keratocytes *in vivo*. Its role in keratocytes has not yet been elucidated but there is evidence that it might play a role in regulation of differentiation, adhesion, and quiescence. Under our culture conditions, cultured cells expressed high amounts of this protein. Cells cultured in serum free medium had lower expression of CD34 than cells cultured in 2% serum medium. Taken together, the cultured human corneal stroma cells of this study express the most important keratocyte markers and are thus deemed to be of a preserved keratocyte phenotype. Serum is able to alternate the cell phenotype, although to a minor degree in the amount of 2% FBS. As the human cornea can be divided into a central and a limbal part (with partly different clinical features), we derived keratocytes from both parts separately to investigate if these two differ in expression patterns of neuropeptides/neurotransmitters. However, as precise separation of these two parts *in vitro* is impossible, we used a term ‘peripheral’ for keratocytes derived from the limbus region and adjacent corneal tissue. Both central and peripheral keratocytes expressed keratocan, which is a cornea-specific marker that has been shown to be expressed in cultured keratocytes. ## Neuropeptides and their receptors in keratocytes The results of the present study show that both central and peripheral keratocytes in culture express the two tachykinins of interest, SP and NKA, as well as their preferred receptors NK-1R and NK-2R, respectively. SP and NKA belong to a family of neuropeptides which have a variety of pharmacological actions both in the central nervous system and in the periphery. SP and its preferred receptor have been found to be expressed in tumor cells, tenocytes and colonic epithelial cells, and also previously in cultured keratocytes from human cornea. NKA expression has been discovered, for example, in monocytes and lymphocytes. These two substances have been linked to proliferative and apoptotic properties, as well as being known to interact with components of extracellular matrix, mediating migration and playing a role in inflammatory responses. As these mechanisms play a role in stromal wound healing, it might be speculated that both SP and NKA have effects in corneal wound healing processes. The fact that keratocytes also expressed the SP and NKA receptors, suggests that keratocytes themselves indeed could be subjected to SP and NKA actions. ## Catecholamines and their receptors in keratocytes Presence of catecholamines (adrenaline, noradrenaline, and dopamine) was confirmed in both central and peripheral keratocytes in culture. It has been shown that catecholamines are synthesized both in the brain and in non-neuronal organs and cells such as adrenal medulla, gut cells, platelets, and lymphocytes. However, when we analyzed expression of TH–an enzyme responsible for catalyzing the conversion of the amino acid L-tyrosine to L-DOPA–we found that cultured cells express it but the keratocytes in tissue sections do not. Moreover, when we analyzed the expression of adrenaline and noradrenaline adrenoreceptors (α1 adrenergic receptor \[ADRA1\] and β2 adrenergic receptor \[ADRB2\]) we found that they are not expressed in keratocytes in tissue sections, and that their expression in the cultured cells was inconclusive as their expression varied among cells obtained from different donors. As catecholamines play a role in stress response it might be that the cultured cells are in stress conditions and that this explains why they sometimes express the adrenaline and noradrenaline receptors. Also, catecholamines have been found to regulate angiogenesis in tumors. Taken together, the expression of catecholamines and the presence of adrenoreceptors in cultured cells but not in keratocytes in tissue sections suggest that they might play a role in corneal stress response, which might be due to injury. ## ACh and muscarinic receptors in keratocytes ACh, a classical neurotransmitter, is thought to be synthesized by all living cells. In the present study it was found that ACh is synthesized by both central and peripheral cultured keratocytes, and that these cells also express muscarinic ACh receptors (subtypes M₁, M₃, M₄, and M₅). Previous studies have shown that corneal epithelium contains high concentrations of ACh and its enzymes. Furthermore, studies on mAChRs in corneal epithelium and endothelium have shown that mAChR M₂, mAChR M₄, and mAChR M₅ are expressed by these cells, but that mAChR M₁ and mAChR M₃ are not. It would be interesting to study why there are differences in the receptor patterns within the cornea and if there are any physiological reasons for that. ACh is known to induce proliferation in human tenocytes, and keratocytes have an anti-apoptotic effect and facilitate wound healing in epithelial cell, as well as induce angiogenesis, all of which are effects that are of importance also in corneal wound healing. ## Glutamate and NMDA1R in keratocytes Glutamate, an important neurotransmitter, and its receptors have been found in peripheral non-excitable cells in tissues such as taste buds, intestine, spleen, skin, and bone, as well as in platelets and lymphocytes. In the present study, we found that cultured keratocytes express and secrete glutamate. Glutamate receptor NMDAR1 was also expressed by the keratocytes in culture. It has been found that glutamate is linked to modulation of tumor cell proliferation and migration, which might imply that it could play a similar role on cells in corneal would healing processes. ## Differences in expression profiles between peripheral and central keratocytes Limbal epithelial stem cells are located at the basal epithelium of the palisades of Vogt, and regulated by their unique niche components. Compared to the central keratocytes, the limbal keratocytes have a close spatial relationship with the limbal epithelial stem cells, suggesting the possibility of cytokine cross-talk between the two cell types. It has been reported that cells of limbal stromal origin contribute to the proliferation of limbal epithelial stem cells through paracrine signaling. Although limited reports describe the interactions between the limbal stem cells and other limbal niche components, the nerves may be involved in the regulation of limbal niche, since a previous study has confirmed that the limbal stroma is heavily innervated. In the present study, we found that there were significant differences in the expression profiles of the neuropeptides/neurotransmitters and their receptors between peripheral and central corneal stromal cells, which might suggest possible differences in their contributions to the maintenance of limbal epithelial stem cells, a phenomenon previously shown. ## Concluding remarks In conclusion, keratocytes in tissue sections and cultured human keratocytes express a broad range of neurotransmitters and neuropeptides as well as their receptors. The expression profiles seem to be changed upon culturing of the cells and show differences between peripheral and central keratocytes. The results, together with known effects of the neuropeptides/neurotransmitters in other tissues, suggest that neuronal and non-neuronal neuropeptides/neurotransmitters may play a role in corneal wound healing, which warrants further functional studies on their effect in different model systems of the cornea. # Supporting Information The authors extend their sincere appreciation to Dr. Ludvig Backman for technical and scientific advice. We also thank Dr. Maria Brohlin and Ms. Randi Elstad for help in providing the donated corneas from the biobank. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MS SLR PB PD QZ. Performed the experiments: MS SLR PB. Analyzed the data: MS PD SLR PB QZ. Contributed reagents/materials/analysis tools: BB. Wrote the paper: MS PD SLR PB QZ BB.

# Introduction Preeclampsia is a severe pregnancy-associated syndrome, leading to morbidity and mortality of both mother and child worldwide.\[–\] The placenta as a specific organ of pregnancy has been considered to be essentially involved in the pathogenesis of preeclampsia. Placentation defects in the very early stages of pregnancy are thought to be the main predisposing factors for preeclampsia, while the severity of the disease is determined by the maternal response to toxic factors of the placenta. Placentation defects in early pregnancy may result in inadequate trophoblast differentiation followed by subsequent impairment of villous maturation and villous trophoblast turnover. The dysregulation of villous trophoblast during preeclampsia can be followed using placental protein 13 (PP13, LGALS13) as a placenta-specific marker released into maternal blood. This complex process and its cascade of events may induce endothelial dysfunction and systemic inflammation of the mother. Endothelial dysfunction during preeclampsia can be visualized using (anti-) angiogenic markers such as sFLT1 and PGF. On the cellular level, cells have developed the ability to maintain their integrity to environmental stress activating the process of autophagy.\[–\] Failure of finalizing the autophagy process can be identified by changes in the ratio of the autophagy markers, microtubule associated protein 1, light chain 3 beta (MAP1LC3B) and beclin-1 (BECN1), (MAP1LC3B/BECN1). If a cell fails to maintain its integrity, the cell tends to enter into the processes of cell death, ranging from apoptosis to necrosis. Hence, the ratio of MAP1LC3B/BECN1 does not only show the failure of autophagy but also gives an idea on the survival capacity of a cell. Nutrition is a main regulator of autophagy. Vitamin D and the associated cellular signaling components, such as the vitamin D receptor (VDR), are molecules that are highly related to the autophagy process. Vitamin D plays a central role in many cellular processes such as cellular proliferation and differentiation, and immunometabolism either via genomic or nongenomic actions. The involvement of vitamin D in basic biological processes such as autophagy has led to a variety of research questions.\[–\] At the same time, recent data provide evidence that in South East Asian countries including Indonesia vitamin D insufficiency is a major issue. One of the reasons discussed was that only about half of the children in Indonesia consumed dairy products on a daily basis. Therefore, the aim of our study was to define the role of vitamin D in processes of cellular survival in the villous trophoblast of the human placenta in pregnancy pathologies such as preeclampsia and fetal growth restriction. # Material and methods This study included pregnant women attending the Cipto Mangunkusumo Hospital and Budi Kemuliaan Hospital in Jakarta, Indonesia from August to October 2015. We assigned the women and their respective samples to four groups: normal gestation with term delivery, late-onset preeclampsia (LoPE), early-onset preeclampsia (EoPE), and intrauterine growth restriction (IUGR). We defined the inclusion criteria to include all pregnant women whose pregnancy was terminated by cesarean section due to early and late-onset preeclampsia, IUGR and normal term pregnancy without complications in the labor process. We excluded all pregnant women who suffered from other diseases, such as heart disease in pregnancy, diabetes mellitus, auto-immune diseases during pregnancy such as lupus erythematosus, pregnant women with congenital abnormalities of the baby, preterm labor, severe infection during pregnancy or women not willing to take part in this study. We applied the mean difference formula in each group to determine the sample size, which was 10 samples for each group. Hence, the total number of subjects in this study was 40 with 10 subjects in the normal gestation group, 11 subjects with late-onset preeclampsia, 9 subjects with early-onset preeclampsia, and 10 subjects in the IUGR group. The subjects were pregnant women whose delivery dates were based on obstetric medical indications. We defined normal gestation as pregnancy until term without any complications and based the definition of preeclampsia on the 2013 ACOG criteria, including blood pressure of ≥140/90 mmHg with proteinuria after 20 weeks of pregnancy in previously normotensive women. Early-onset preeclampsia (EoPE) was defined as preeclampsia diagnosed before 34 weeks, while late-onset preeclampsia (LoPE) was defined as preeclampsia diagnosed at or after 34 weeks of pregnancy. The major difference between early and late-onset preeclampsia is that early-onset preeclampsia is usually complicated by intrauterine growth restriction and/or other organ involvement with severe clinical features. Intrauterine growth restriction (IUGR) was defined as fetal weight lower than the 10^th percentile on the growth chart. In this study we collected samples from maternal serum, cord venous serum and placental tissues. The amount of 10 ml maternal serum was collected at the time of patient admission in the delivery or emergency room. In the labor process after the baby was delivered, 10 ml cord venous serum was collected and followed by collection of placental tissues. We processed blood serum samples immediately by standardized preanalytical procedures and stored them at -80°C, while we processed paraffin blocks of the placental tissues according to a standardized pathology protocol. We examined the factors sFLT1, PGF and placental protein 13 (LGALS13) using maternal serum, while we evaluated 25(OH)vitamin D levels in maternal serum and cord venous serum. For the measurements of the autophagy proteins MAP1LC3B and BECN1 and the vitamin D receptor we used placental tissues. At the Endocrinology Lab Platform at the Medical University of Graz, Austria, we performed the laboratory measurements of angiogenic factors (sFLT1 and PGF), LGALS13 and 25(OH)D levels. We assessed placental tissues at the Department of Cell Biology, Histology and Embryology at the Medical University of Graz, Austria. We measured the angiogenic factors sFLT1 and PGF using the Elecsys equipment (Roche Diagnostics, GmbH, Vienna Austria). Sample volume was 20 μl for sFLT1 and 50 μl for PGF, range 10–85,000 ng/ml for sFLT1 and 3–10,000 mg/ml for PGF. The sensitivity was \<15 pg/ml for sFLT1 and 10 pg/ml for PGF. The intra-assay variation coefficient was \<2% for both sFLT1 and PGF, the inter-assay variation coefficient was 2.6–30% for sFLT1 and 2.0–2.4% for PGF. We measured placental protein 13 (LGALS13) concentrations utilizing a solid- phase sandwich ELISA with a pair of LGALS13 specific monoclonal antibodies (Hylabs, Israel), marked with amplified biotin-extravidin-horseradish-peroxidase complex and developed with tetramethylbenzidene substrate. Using the optical density at 450 nm against a 650-nm background we translated the optical measurements into a quantitative amount using a calibration curve made of recombinant LGALS13 standards (0–500 pg/ml). We assessed serum vitamin D (25(OH)D) levels of maternal and cord blood by automated chemiluminescence immunoassays on the iSYS automated system (Immunodiagnostic Systems, Boldon, UK, with an intraassay and interassay coefficient of variation of 6.2% and 11.6%, respectively). In this study, we detected the autophagy markers MAP1LC3B and BECN1 and the vitamin D receptor (VDR) using immunohistochemistry. We mounted five (5) μm thick formalin-fixed paraffin-embedded tissue sections on Superfrost Plus slides (Menzel/Thermo Fisher Scientific). According to the standard procedures, we deparaffinized sections in Histolab Clear (Sanova) and rehydrated them through a mixture of 100% ethanol/Histolab Clear 1:1, followed by a series of graded alcohol and finally aqua dest. We performed antigen retrieval for 7 min at 120°C in a decloaking chamber (Biocare Medical) in 10 mM sodium citrate buffer pH 6.0 for the anti-VDR antibody (abcam) and the anti-BECN1 antibody (abcam) or Epitope Retrieval Solution pH 8.0 (Novocostra, Leica) for the MAP1LC3B antibody (Novus biologicals). After 20 min cooling at room temperature, we transferred slides in TBS including 0.05% Tween 20 (TBS/T, Merck). We performed staining of sections either manually or using a staining robot (Autostainer 360, Thermo Fisher Scientific). To quench endogenous peroxidase, we incubated slides with hydrogen peroxide block (Lab Vision/Thermo scientific) for 10 min and after washing with TBS/T, we blocked nonspecific background by incubation with Protein Block (Lab Vision/Thermo scientific) for 7 min. We diluted primary antibodies in Antibody Diluent (Dako) as follows; VDR 1:700, BECN1 1:280, MAP1LC3B 1:3000, and incubated sections for 45 min at room temperature. After washing with TBS/T, we incubated slides for 20 min with the UltraVision HRP-labelled polymer and following another washing step, we visualized the polymer complex by incubating the slides with AEC Chromogen Single Solution (Lab Vision/Thermo scientific) for 10 min. Following washing steps with aqua dest, we counterstained sections with hemalaun and mounted them with Kaiser’s glycerin gelatine (Merck) or Aquatex (Merck). We assessed placental tissues on 3 blocks per placenta, taken from central, intermediate and peripheral positions of a placenta. We quantified staining of images with a Leica DM 6000B microscope with an Olympus DP 72 camera and VIS visiopharm software (Visiopharm, Denmark), based on systematic uniform random sampling. For each slide, the software chose 10 pictures randomly and systematically and placed a point grid with 16 times 12 points on each picture (newCAST software, Visiopharm). Each point was assigned to a specific category. The categories for MAP1LC3B and BECN1 were: intervillous space, villous stroma, villous trophoblast and fibrin. Due to the location of VDR within nuclei, the categories for the vitamin D receptor were: intervillous space, villous stroma, trophoblast nuclei and fibrin. Within the categories ‘villous trophoblast’ and ‘trophoblast nuclei’ staining intensity was ranked 3 for the strongest staining, 2 for moderate staining, and 1 for the weakest staining or 0 if the dot was placed on trophoblast without staining. We calculated the ratio between strong/moderate and weak/no staining (also called positivity) by counting the dots with ranks 2 and 3 divided by the sum of the dots with ranks 0 and 1 within villous trophoblast/trophoblast nuclei and called this ratio ‘positivity’. We quantified total staining in the trophoblast as follows (called ‘stained trophoblast’): dots in villous trophoblast/trophoblast nuclei were ranked 0 to 3 and for each protein and image, and the ranking was added and divided by the total number of dots in villous trophoblast/trophoblast nuclei. Finally, we calculated the ratio of MAP1LC3B to BECN1 using the two above approaches to calculate staining intensity parameters. For each parameter, we divided the two values of MAP1LC3B and BECN1 resulting in the ratio of MAP1LC3B/BECN1. This ratio represents one indicator of autophagy limitations in a cell, as it shows the cellular survival mechanism as one of the aspects of autophagy. We tabulated, edited and coded all data based on the purpose of the study and statistically analyzed the data using R Statistics Software version 3.4.0 for Windows operating system. We used the univariate analysis for descriptive characteristics, and assessed differences between autophagy markers and nutritional status between each group by bivariate One-way ANOVA or Kruskal- Wallis test. Correlations between variables were calculated by Spearman’s correlation tests. Before the start of the study, signed informed consent was given to the research subject and her family and ethical clearance was given from The Health Research Ethic Commitee—Faculty of Medicine University of Indonesia and the Cipto Mangunkusumo Hospital (*Komite Etik Fakultas Kedokteran Universitas Indonesia* No:668/UN2.F1/ETIK/2015) as approval. # Results ## Characteristics Forty pregnant women participating in this study had a mean age of 30.4 ± 5.9 years (standard deviation (SD)). Twenty-three women (57.5%) were between 25–34 years, 12 women (30%) aged above 35 years, and 5 women (12.5%) aged under 24 years. The median gestational age of the subjects was 38 weeks with the lowest gestational age of 25 and highest gestational age of 40 weeks. There was a statistically significant difference in gestational age of the mothers particularly in the early-onset preeclampsia group (p\<0.001). Anthropometric measurements showed a mean body mass index (BMI) before pregnancy was 23.4 kg/m² ± 4.7 kg/m². There were no significant differences in anthropometric variables between the four groups. The results are depicted in. ## Defective placentation and survival capacity There was a significant difference of the sFLT1/PGF ratio between normal and pathological pregnancy cases. In addition, a post hoc analysis showed significant differences of LGALS13 levels between the normal pregnancy and the IUGR group (p = 0.022) as shown in. For stained trophoblast, there were significant differences in autophagic survival markers (p = 0.024, one-way ANOVA test), the MAP1LC3B/BECN1 ratio, as assessed by immunohistochemistry of villous trophoblast in the four groups. There were higher mean values of the MAP1LC3/BECN1 ratio in normal pregnancy (1.63±0.34) and late-onset preeclampsia (1.46±0.41) compared to the early-onset preeclampsia (1.13±0.20) and IUGR (1.39±0.34) groups. The early-onset preeclampsia group had the lowest ratio among the groups and showed a significantly lower ratio than the control group (p = 0.008, post-hoc test). Similarly, in the parameter of positivity the normal pregnancy group showed much higher values compared to the other three groups, but the values did not reach significance (p = 0.169, Kruskal-Wallis test). Measurements of the cell autophagic survival markers MAP1LC3B/BECN1 were also performed in relation to gestational age. There was no significant difference of staining level related to gestational age; however, distribution for each subject in all pregnancy groups showed that subjects in the early-onset preeclampsia group and IUGR group had lower number of MAP1LC3B/BECN1 ratios compared to the other two groups. This study cannot tell whether this was due to early-onset preeclampsia/IUGR or gestational age below 35 weeks. ## Vitamin D in pregnancy The analysis of vitamin D status included 25(OH)vitamin D (25(OH)D) levels in maternal blood, umbilical cord blood, and vitamin D receptor (VDR) in placental tissues. There was a significant difference in median 25(OH)D levels between groups. While 25(OH)D vitamin D levels in the normal pregnancy group were in the normal range, the early-onset preeclampsia group had 25(OH)D deficiency. Also the late- onset preeclampsia and IUGR groups showed 25(OH)D insufficient level based on reference values for normal and deficient vitamin D levels in maternal blood of \>32 ng/ml and \<20 ng/ml. Post hoc analysis showed significant differences between normal pregnancy and all disease groups. There was a significant difference of cord blood 25(OH)D level between groups, in which IUGR had the lowest level. Post hoc analysis showed a significant difference between normal pregnancy and the late onset preeclampsia group (p = 0.009) as well as the IUGR group (p = 0.004) as shown in. ## Placental vitamin D receptor (VDR) The vitamin D receptor (VDR) was assessed as part of the cellular vitamin D metabolism pathway. There was a significant difference in mean VDR staining levels between the four groups. The normal pregnancy group had higher mean VDR staining levels compared to the other three groups. Late-onset and early-onset preeclampsia groups had lower mean values of VDR staining compared to the normal pregnancy group. Post hoc tests showed a significant difference between normal pregnancy and IUGR groups (p = 0.029). The results of vitamin D receptor (VDR) in placenta are presented in and Figs and. VDR staining levels did not show any correlation to gestational age as depicted in. Correlation tests of 25(OH)D levels in maternal blood and vitamin D receptor staining levels in the placenta were performed in order to explore the activity of VDR. There was a significant difference between 25(OH)D levels in maternal blood and placental tissue VDR with parameters of stained trophoblasts among the four pregnancy groups with a weak correlation (r = 0.38; p = 0.015) as shown in. ## Trophoblast survival capacity and placental VDR Correlation between the MAP1LC3B/BECN1 ratio as cellular survival marker and intracellular VDR in villous trophoblast in the four study groups was performed. All groups showed a negative correlation, with the normal pregnancy group showing the strongest correlation, followed by the IUGR group, the late-onset preeclampsia group and the early-onset preeclampsia group. Furthermore, there was a strong and significant correlation between the MAP1LC3B/BECN1 ratio as survival marker and VDR levels in the early-onset preeclampsia group. # Discussion ## Conceptual thinking The production of angiogenic factors and placental proteins by the placenta reflects the physiological processes occurring during pregnancy. Both angiogenic factors and LGALS13 are assumed to play roles in trophoblast differentiation and invasion.\[,–\] Defective placentation will thus result in changes and fluctuative amounts of these markers released into the circulation. The quality of trophoblast differentiation is shown by LGALS13 and the ratio of sFLT1/PGF as angiogenic factors. Inadequate trophoblast differentiation will result in a cellular death spectrum ranging from apoptosis to necrosis and will subsequently induce inflammation and immunological responses of the mother. In this context, not only cell death occurs but also the cellular survival capacity, autophagy, takes over a major role. The dual role of autophagy in either cell death or cellular survival has been shown in the trophoblast already.\[–\] The proposed survival capacity marker, the ratio of MAP1LC3B/BECN1 shows the ability of the trophoblast to survive even in the presence of environmental stress. The conceptual thinking of this study is shown in. ## Defective placentation and cellular survival Defective placentation is a condition occurring as a result of inadequate trophoblast differentiation in early pregnancy. Extravillous and villous trophoblast subtypes do not successfully differentiate and mature and thus the transformation of spiral arteries as well as the placenta proper fail to adequately support the healthy continuity of pregnancy. The question arises where in this spectrum of failure diseases and syndromes develop such as preeclampsia and IUGR. Here we hypothesize that the cellular survival capacity based on autophagy plays an important role in cell death and cellular survival in the above syndromes. The results of failure of trophoblast differentiation can be visualized by the altered ratio of sFLT1/PGF and altered LGALS13 release. The angiogenic factors sFLT1 and PGF play roles in angiogenesis, while LGALS13 is released from the syncytiotrophoblast and is involved in regulating the maternal blood pressure. The results of this study showed normal sFLT1/PGF ratios in normal pregnancy and IUGR cases while there were higher values in late and early-onset preeclampsia. This is in line with the data from the PROGNOSIS study where the authors decribed a cut-off value of 38 for the sFLT1/PGF ratio. Interestingly, the data on LGALS13 did not show an increase in cases with preeclampsia, different to what has been described before. The trophoblast survival capacity was shown using the MAP1LC3B/BECN1 ratio. There was a significant difference in the mean values of the MAP1LC3B/BECN1 ratio among the four study groups. The early-onset preeclampsia group had a significantly lower mean value than the control group. This could be interpreted that early-onset preeclampsia had the worst ability to maintain cell survival compared to all other groups. However, it cannot be ruled out that gestational age may have an impact here as directly age-matched controls are missing. ## Vitamin D in pregnancy Vitamin D is a *secosteroid* hormone which is known to have important roles in many physiological processes, such as regulation of cell proliferation and differentiation processes, immunomodulation, vascular biology, and placental metabolism and function. A meta-analysis recently stated that vitamin D deficiency in pregnancy increases the risk of pathological pregnancies, including increased risks for preeclampsia (2.09), preterm labor (1.58), and low birth weight (1.52). The role of vitamin D in immunological processes at the fetal-maternal interface is the reason for studies on vitamin D in preeclampsia. The high number of preeclampsia cases in Indonesia is suspicious due to the known deficiency of vitamin D in the maternal circulation. Regardless of the geographical position over the equator belt with sunlight all year, we assume there any other factors contributing to this deficiency. The main factor that is assumed is the reduced consumption of vitamin D rich nutrients in the daily nutrition of pregnant women. Looking at the 25(OH)D levels in maternal blood, only the control group presented normal levels above 32 ng/ml. All pathological groups showed significantly reduced vitamin D levels as well as vitamin D deficiency. This data can be interpreted as a direct correlation between low vitamin D levels and inflammation and cellular survival of the trophoblast in pregnancy pathologies. The results showed that normal pregnancy had normal reference levels and thus seems to have normal physiological cellular processes mediated by vitamin D. These results were in accordance with previous studies showing the increasing risk of preeclampsia in vitamin D deficiency. ## Vitamin D receptor (VDR) As the placenta is the source of one of the extra-renal hydroxylase enzymes, it has a role in vitamin D metabolism. In addition to this, the placenta also expresses the vitamin D receptor that is involved in vitamin D signaling in trophoblast. Looking at the staining percentage in villous trophoblast, there was a significant difference in placental vitamin D receptor (VDR) staining between the four study groups. The IUGR group showed significantly lower VDR expression compared to the other groups. This is consistent with a previous study showing low VDR levels in IUGR. In preeclampsia cases, VDR staining was higher compared to IUGR but lower compared to normal cases. In the correlation analysis between vitamin D levels of maternal blood and placental VDR staining in villous trophoblast, there was a significant difference between the study groups. The results of this VDR analysis showed the dynamic activity of VDR as response to vitamin D levels in circulation. This suggests a mechanism of up and down- regulation of the receptor to adapt to environment stress. It shows the up- regulation mechanisms in the preeclampsia groups, while in IUGR the down regulation shows the phenotypic changes of the placenta known as compromised placenta in IUGR. ## Placental VDR and trophoblast survival capacity The correlation test between the trophoblast survival marker, the MAP1LC3B/BECN1 ratio, and placental VDR revealed negative correlations in all study groups. This may explain the inadequacy of cellular nutrition for vitamin D, as systematically reflected by the vitamin D receptor (VDR) and thus the low cellular survival in the pathological cases with vitamin D deficiency. Moreover, the early-onset preeclampsia group showed the significantly strongest negative correlation compared to the other study groups. This result may point to the fact that the vitamin D receptor may play a role in the cellular survival capacity, especially in early-onset preeclampsia. The cellular survival capacity can be integrated into the concept of cellular homeodynamics where the survival capacity can be seen as cellular buffer counteracting environmental or developmental stress. This buffer is supported by hormetins such as nutrition including vitamin D. Vitamin D acts on the genomic and non-genomic level to influence cellular processes and thus may an important regulator in the dynamic processes of pregnancy. Vitamin D and its cellular binding partner VDR may thus have a direct impact on the survival capacity of trophoblast as seen in. # Limitations There were several limitations encountered in this study. First, this pilot study was carried out in a cross-sectional design with small numbers of samples. Additionally, only one measurement was perfomed at the time of delivery. Due to observations of this study that there is a crosstalk between autophagy and vitamin D as survival nutrient, which in turn should stimulate a larger longitudinal and experimental study. Second, the assessments were only performed in a static manner using immunohistochemical staining of autophagy proteins MAP1LC3B and BECN1 and vitamin D receptor. Third, placental tissue was collected at the end of pregnancy which did not allow for real time dynamic interpretation of pregnancy and cellular homeodynamics. Hence, the interpretation should be cautious but comprehensive to capture the multiple facets of this dynamic process. # Conclusion The placenta as a pregnancy-specific organ has mechanisms of adaptation and compensation to counteract environmental stress. This becomes obvious looking at the ratio of MAP1LC3B/BECN1 as survival marker, which can be used to describe the survival capacity of villous trophoblast. Defective placentation in preeclampsia is negatively supported by vitamin D deficiency, which should as act as major biological buffer of cellular homeodynamics. Vitamin D and its cellular binding component vitamin D receptor (VDR) might play a role in the trophoblast survival capacity particularly in preeclampsia. # Supporting information The authors would like to thank Monika Siwetz (Gottfried Schatz Research Center) and Nicole Sobitsch, Christian Kaspar and Sandra Triller (Endo Lab Platform) for their excellent technical assistance. [^1]: The authors have declared that no competing interests exist.

# Introduction Atopic dermatitis (AD) is a chronic inflammatory cutaneous disorder that often affects patients’ and parents’ quality of life. AD is also a common public health problem, because of its increasing prevalence throughout the world, and its significant cost to society. Topical corticosteroids (TCS) combined with emollients remain the mainstay of AD treatment. Their efficacy and safety, when appropriately used, has been clearly established. Paradoxically, the fear of using TCS (usually called TCS phobia, TCP) is a frequent concern for patients and their parents (between 40% and 73% depending on the authors),. This fear could be the main cause of poor therapeutic adherence and consequently poor treatment response for many patients (only 32% of AD patients seem to adhere to medical instructions). The evaluation of TCP is thus an essential step in the management of patients with poor adherence. Few studies have addressed the assessment of TCP; most of them used only one question with a yes/no response. Because TCP appears to be a complex phenomenon, its assessment through binary responses is too simplistic, as it cannot detect different types of fear. Thus, a scale, which is capable of identifying the subtleties of this concept, is needed. Although generic psychology scores exist to explore worries, to the best of our knowledge, no such medical tool has been published to assess specific beliefs and worries related to TCS. Herein, we describe the development and statistical validation of a scale named TOPICOP© (topical corticophobia) aimed at assessing AD outpatients’ and their parents’ worries and beliefs about TCS. This scale will help researchers and clinicians to better understand the factors that influence therapeutic adherence. # Materials and Methods ## Study Design This prospective multicenter study was conducted in five regions in France. A convenience sample was created with the help of 9 hospital dermatologists and 53 dermatologists in private practice. They gave a self-administered anonymous questionnaire to their consecutive AD patients consulting between February and May 2009. Patients were asked to complete the questionnaire before the consultation. The first page of the document was a written explanation of the study that had to be signed by the respondent for participation. In all cases, the patient’s dermatologist had confirmed the AD diagnosis. All patients with AD were included, and parents were asked to fill out the questionnaire for their affected children under 15 years old. The local Ethics Committee of Nantes University Hospital approved this study. ## Questionnaire-construction Process The first step of the questionnaire-construction process used focus-group methodology. This qualitative phase, conducted between September and December 2008, involved 12 adult patients, nine parents of children with AD, and 15 health professionals (8 general practitioners and 7 pharmacists) in the collect of data on patients’ behaviors, beliefs, cognitions, sensations and perceptions of TCP. Five focus-group meetings were held and interviews were transcribed. In addition, four telephone interviews sufficed to reach the data saturation threshold. Analysis was completed by an examination of the literature. The qualitative phase was previously reported in details. The second step enabled us to generate a 51-item questionnaire with an additional 18 items concerning sociodemographic characteristics and health status. To ensure content validity, a panel of seven experts (four dermatologists, one psychologist, and two public health doctors) discussed and retained the list of items. To ensure face validity, the questionnaire was given to a panel of 10 patients to explore the level of understanding, acceptability and time required to complete the questionnaire. Five items were modified slightly in response to patients’ comments. Among the 51 items, 32 concerned the different types of worries and beliefs identified. The other items explored the origins of fears and beliefs (lack of oral or consistent information delivered by caregivers, discrepancies concerning treatment among dermatologists, dermatologists and general practitioners, and between practitioners and pharmacists, roles of family circle and the media…), behaviors and therapeutic adherence and, finally, the characteristics of the patients and their AD. Three items were eliminated from the questionnaire for scale construction because they were specific to children. For the 29 remaining items studied in this paper, response choices had a 4-point Likert-scale format (from totally disagree to do not really agree, almost agree or totally agree; or never to sometimes, often or always). ## Scale-construction Process (Construct, Concurrent and Convergent Validity, Reliability) The first statistical analyses used the usual techniques of descriptive statistics (frequency, means ± SD) and Pearson’s correlation coefficients between items two-by-two. The first step consisted of eliminating items with a rate of missing values \>20%, or a floor or ceiling effect \>50%. Then, when two of the remaining items had a Pearson correlation coefficient \>0.60, one of them was selected in accordance with the consensus of the seven experts. The second step was an explanatory principal component analysis using a varimax rotation for the 19 remaining items. The number of dimensions was determined using a scree plot and clinical relevance. Two criteria were used to attribute each item to one of the dimensions: substantial communality (\>0.60) with one principal component and, when an item exhibited communality across several dimensions, it was attributed to the one for which it maximized internal consistency assessed by Cronbach’s α-coefficient (\>0.7). This strategy led to the removal of seven items, for which neither a sufficient communality with principal components nor an adequate Cronbach’s α-coefficient could be obtained, yielding a robust shorter two- dimension solution. Finally, the homogeneity of the dimensions was assessed using convergent validity (correlation of each dimension item with all the other items in the dimension \>0.40), and divergent validity (correlation of each dimension item with all the other items in the other dimension \<0.40). Overlap was corrected. To explore criterion validity, Pearson’s correlation coefficients were calculated between scores obtained and a visual analog scale evaluating the intensity of fears. ## Structural Equation Modeling Using the Linear Structure–relationship Approach Structural equation modeling was performed to confirm factorial structure and unidimensionality of the dimensions. It is a comprehensive statistical approach to test hypotheses about relationships among observed and latent variables (dimensions). Several statistical indices were calculated in order to verify model fit and to select the best-suited model. Parameter estimation used the linear structural–relationships approach developed by Jöreskog. To do so, the three main criteria used were Steiger’s root mean square error of approximation with the fit being considered good when \<0.1 and very good when \<0.05; Bentler and Bonnet’s normed fit index considered good when \>0.9, and the goodness-of- fit index considered good when \>0.9. ## Score Calculation Four response choices were offered, from totally disagree to totally agree, with points attributed to each one (0, 1, 2 or 3), with higher values corresponding to more severe TCP. Individual scores for all patients who responded to at least half the items plus one in a dimension were calculated by summing responses to items and then dividing that value by the number of items completed, yielding a maximal score of 36, expressed as a percentage. The mean score for a dimension was the sum of individual scores divided by the number of respondents. TOPICOP© scores ranged from 0 to 100. All study analyses were computed using R 2.9 and SPAD 5.6. Structural equation modeling used SAS v9.1 (SAS Institute, Cary, NC) and its “PROC CALIS” procedure. # Results ## Characteristics of the Sample A total of 208 patients or parents were enrolled in the five French regions: hospital dermatologists enrolled 114 patients or parents and dermatologists in private practice enrolled 94. Among the 208 respondents, 144 were parents of children with AD. Mean ± standard deviation (SD) age of the respondents was 32.7±7.3 years. Concerning AD severity, 41.1% of patients or parents reported mild, 46.2% severe and 12.7% very severe disease. The majority of patients or parents (81.7%) were currently using or had recently used TCS. For a general item exploring fear about TCS, 80.7% admitted having fears about TCS. ## Scale-construction Process Twenty-nine items exploring different types of worries and beliefs were studied. No item had a missing-value frequency \>20%. A floor effect (percent of ‘totally disagree’ responses exceeding 50%) was observed for four items and one had a totally agree percentage \>50% (ceiling effect): they were excluded. Among the 24 remaining items, five were removed whose Pearson’s correlation coefficients with another item exceeded 0.6. For example, the correlation coefficients of the two following items “I’m afraid of applying too much cream” and “I’m afraid of using the cream for too long” were 0.75. The expert panel decided to retain the former item. Successive explanatory principal component analyses were then performed on the 19 remaining items and led to the identification of two dimensions. Seven items were removed step-by-step because of their low communality with one of the two factors or because they exhibited communality with both. No item maximized Cronbach’s α-coefficient and all the 12 remaining items had communality \>0.60 within their own dimension. The final questionnaire TOPICOP©, a 12 item scale, accounted for 47.3% of the variance, with a first dimension of six items exploring worries (WOR) accounting for 24.4% of the variance, and a second one, also of six items exploring beliefs (BEL) accounting for 22.9% of the variance; their respective Cronbach’s α-coefficients were 0.79 and 0.78. Correlations between items within a given dimension all exceeded 0.40 and correlations between one item and those of the others in the dimension were \<0.40. Inter-dimension correlation was 0.41. Pearson’s correlation coefficient between the WOR dimension and visual analogue scale score exploring fears was 0.60. The statistical parameters are reported in. The final questionnaire is presented in. ## Structural Equation Modeling with the Linear Structure–relationship Approach Modeling confirmed the existence of two latent dimensions WOR and BEL but the best characteristics were obtained with a hierarchical model including the two latent factors and a global latent factor, bringing the 12 items together. Goodness of fit of the data were very good and all the structural coefficients were highly significant (*P*\<0.001). ## TOPICOP© Mean Scores Based on scores ranging from 0 to 100, mean (±SD) scores were 46.4±24.7 for WOR dimension, 41.1±22.1 for BEL dimension and 43.9±19.6 for the TOPICOP© scale. Results are reported in. # Discussion TCP is a common, worldwide phenomenon in patients with atopic dermatitis, leading to poor local treatment adherence and frequent therapeutic failure. Paradoxically, there is a lack of TCP assessment tools available. We previously showed that TCP is a complex phenomenon : some patients who did not admit to being worried about TCS expressed TCS phobia through their behaviour (need for reassurance or reducing doses). Assessing TCP with binary (yes-or-no) responses is, thus, too simplistic and cannot detect different types of fears and their intensity. The need to develop a scale aimed at assessing the multidimensional aspects of TCP (worries, beliefs and behaviours) is proven. TOPICOP© is the first self-administered scale, developed to evaluate AD patients’ or their parents’ TCP, to have undergone initial statistical validation. It contains 12 items; six of them in each of the two dimensions related to worries and beliefs (WOR and BEL), and exhibits excellent psychometric properties. The statistical validation strategy presented herein follows most of the recommendations of ‘good practice’ for score validation. The first original feature of our study was the methodology used to develop the questionnaire. Item generation, based on a qualitative analysis using focus groups, enabled us to design a questionnaire that explored the AD patients’ or parents’ real-life attitudes, worries, beliefs and behaviors. This first step enabled us to ensure good content and face validity. The second step consisted of a quantitative study derived from the conclusions of the former and enabled us to statistically validate the TOPICOP© scale. This step supported high construct, divergent and discriminant validity and reliability. The small numbers of missing values indicate a very good understanding of the questionnaire. Internal consistency was close to 0.80 as recommended. Items had strong communality with the two principal component analysis-identified factors and accounted for \>47% of the variance. To confirm our results, structural equation modeling was performed and strongly upheld the possibility of calculating a global score. As TCP is mainly a parents’ problem in daily practice, our population was made up of two thirds parents and one third adult patients. We cannot exclude a recruitment bias. As we recommended to give the questionnaire to consecutive patients with AD, it might have been distributed to those with more severe AD or with TCS phobia. Finally, to reduce the recruitment bias, parents and adult patients were enlisted from both hospital outpatient- dermatology departments and private dermatology practice. In France, patients consulting at hospital departments might have more severe disease or more TCS worries leading to therapeutic failure. Nevertheless, despite TOPICOP©’s excellent psychometrics properties, a question remains: is it able to represent the reality of the TCP concept? Corticophobia first appeared 25 years ago in the context of asthma, and is now recognized as a very common but poorly understood phenomenon. In dermatology, TCP is characterized by patients’ fear and excessive anxiety about using TCS preparations. Use of the term “phobia” seems to be a little excessive with regard to its psychiatric definition as, according to psychiatrists, a phobia is an intense but unrealistic fear that can interfere with the ability to socialize, work or go about everyday life, brought on by an object, event or situation. A specific phobia is the fear of a particular situation or object, but worries about TCS are not always unfounded. For example, it is true that TCS can pass into the blood stream or can damage the skin by causing permanent thinning and vasoplegia. The question, then, is not to qualify a patient’s worries and beliefs about TCS as being true or false, but rather to understand to what extent those worries and beliefs have an impact on adherence to treatment. Because worries and beliefs are linked, the TOPICOP© scale comprises items from these two facets of TCP in AD. Items selected concerning beliefs (cutaneous side effects; infections; systemic side effects, principally growth retardation, weight gain or inducing asthma) are consistent with those mentioned earlier by numerous authors. Items related to worries (TCS dependency or addiction, loss of efficacy, need for reassurance) have also been reported. Furthermore, many patients in our study said they did not know the side effects of TCS but were still afraid of using them, highlighting that negative beliefs and attitudes concerning TCS were not always based on scientific findings. Their mean TOPICOP© scale scores demonstrated that TCP is an important and frequent problem in AD treatment, with behavioural consequences that lead to poor local treatment adherence in AD patients or parents and therapeutic failures. Even though the question of patient attitudes to corticosteroids is frequently asked in consultation, the origins of these fears are rarely explored. In fact, there can be a variety of reasons for this reticence: personal experience, divergent advise from pharmacists and doctors, divergent advise from friends and family, information found on the internet etc. The TOPICOP scale should thus help clinicians to better identify their patients’ fears and worries relating to TCS in order to personalize their discourse, target specific blockages and develop pertinent arguments to help patients/parents to adhere to prescribed treatment. Nevertheless, our study was not designed to assess the potential impact of using TOPICOP© scale on patients’ adherence. Furthermore, The TOPICOP© scale should help researchers assess and explore TCP in clinical studies. Indeed, as TCP is closely linked to therapeutic adherence in AD, we suggest that it should be systematically assessed in clinical trials as a potential factor influencing outcomes. Moreover, TOPICOP© scale should be useful in future studies to evaluate to what extent patient education could modify patients’ TCP thresholds. TOPICOP© also covers one of the main research topics proposed by the HOME study group. Finally, the items included in TOPICOP© were chosen according to the oral depositions of patients during the focus group meetings. The items are thus simple and easily understood. TOPICOP© was created and tested in France but the dimensions explored by the scale are not limited to French AD patients and further scale validation in other countries and cultures is required in order to facilitate international comparative studies. These international TOPICOP© versions need to be explored in terms of cross-cultural adaptation and necessitate an accurate translation combined with posteriori verification by a medical expert. Nevertheless, an English version of the TOPICOP© scale is already available, translated by a native English-speaking American scientist and reviewed by our panel of experts. We thank the following physicians for having included patients: Laurent Misery, Paul Young, Myriam Chastaing, Nathalie Danou, Frank Boralevi and Jean-Philippe Lacour, from the university hospitals of Brest, Bordeaux, Nice or Toulouse, and private practitioners. We thank Janet Jacobson and Alan Ball for editorial assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: LM HA-W ALR CL J-FS SB. Performed the experiments: HA-W ALR JM-H J-FS SB. Analyzed the data: LM EA. Contributed reagents/materials/analysis tools: EA MD. Wrote the paper: LM HA-W J-FS SB.

# Introduction Head and neck squamous cell carcinoma (HNSCC) is one of the ten most common cancers in Taiwan and worldwide. In general, HNSCC occurs in the oral cavity, oropharynx, hypopharynx, larynx, and paranasal sinuses. Due to the complicated anatomy of the head and neck, head and neck cancer involves one of the most difficult surgical treatments; therefore, multidisciplinary and diverse treatment strategies are needed. Despite advances and improvements in diagnostic and surgical techniques, chemotherapy, and radiotherapy, the prognosis of patients with HNSCC remains unchanged. Metastases and treatment failures are thought to be responsible for most deaths associated with HNSCC. Understanding the mechanisms underlying tumorigenesis, metastases, and treatment failure may help reduce the morbidity and mortality of HNSCC. Thus, a better understanding of the molecular mechanism of HNSCC aggressiveness is urgently needed to promote the development of a more efficient therapeutic target and to identify key pathways mediating disease progression. The tobacco-related carcinogen nitrosamine 4-(methylnitrosamino)-1-(3-pyridyl)-1-butanone, also known as nicotine-derived nitrosamine ketone (NNK), is a major hazard component in cigarette content and has been recognized as its most potent carcinogen. Tobacco smoking with long-term exposure to NNK, as well as heavy arecoline consumption due to habitual betel nut chewing, have been associated with increased risks for tumorigenesis of head and neck cancers, including in the oral cavity, pharynx, larynx, and esophagus. It also appears that smoking and betel nut chewing are the two most common distinguishing risk factors for HNSCC progression and play pivotal roles in increasing cancer cell growth and survival. Arecoline is a predominant psychoactive agent in areca nuts. Some effects of the areca nut are euphoric or anxiolytic, as with NNK. Based on a large-scale analysis of smoking or consumption of betel quid versus the incidence of HNSCC, arecoline and NNK are thought to be associated with poor responses to chemoradiotherapy and shorter overall survival in patients. Nicotine in tobacco is modified by nitrosation to form nitrosamines and the well-known tobacco- specific carcinogen, NNK, which was reported to enhance cancer progression and metastasis through α7-nAChR and to be a hallmark of the epithelial–mesenchymal transition (EMT). NNK binds the β-adrenoceptor (β-AR) and induces cell proliferation and invasion in pancreatic cancer. The β2-adrenergic antagonist was shown to reduce the activation of NF-κB, extracellular signal-regulated kinase, and Akt-related pathways, resulting in cell death. Arecoline exhibits similar carcinogenic and long-term toxic effects as NNK, and both molecules are alkaloids with comparable structures. Arecoline is a full agonist of acetylcholine muscarinic receptors, and its activity is probably mediated by muscarinic M3 receptors found in the smooth muscles of the blood vessels. Areca- nut chewing was popular in many parts of Asia to induce salivation and euphoria. Activation of muscarinic receptors can lead to Akt stimulation, which inhibits apoptosis and promotes cell survival. The expression of several proteins with aberrant regulation has been found in association with oral cancer, including the epidermal growth factor receptor (EGFR), Akt, and GSK3β.\[–\] Chronic exposure to arecoline promotes the acquisition of cancer stemness, EMT, and chemo-resistance. Cancer stem cells (CSCs) have been identified in many solid tumors, including breast, prostate, and pancreatic carcinomas. CSCs show a high capability for tumor initiation, motility, and invasion, with the overexpression of representative markers such as CD24 and CD44 and the activity of aldehyde dehydrogenase 1 (ALDH-1) being associated with stem cell-like properties.\[–\] Persistent cytotoxicity promotes the activation of CSCs, resulting in treatment failure and relapse, and the use of these substances has been previously associated with cancer incidence and cancer progression. The exact mechanism(s) and cross-linked effects of NNK and arecoline underlying tumor progression in HNSCC remain unclear. As we reported previously, long-term NNK exposure increases anti-apoptosis and therapeutic resistance via the Snail-RKIP signaling pathway. Here, we utilized our non-adhesive culture system to investigate the characteristics of HNSCC cells following long-term and combined treatment with NNK and arecoline. The aim of this study, therefore, was to validate the effects of two major risk factors and the associated signaling pathway involved in modulating tumor growth, apoptosis, and stem cell properties. Our current findings provide insight into the molecular mechanism underlying HNSCC and reveal a possible therapeutic strategy for improving HNSCC prognosis. # Materials and methods ## Preparation of cells and subsequent sphere culture Six HNSCC cell lines were initially prepared and tested and details were seen in our previous report. Two representative HNSCC cell lines, SCC25 (catalog number CRL-1628) and FaDu (catalog number HTB-43), were chosen for subsequent study The cells were cultured in 10-cm dishes with a non-adhesive surface. The 10-cm dishes were made to be non-adhesive by coating them with a thin agarose film. Cells were plated and the culture medium was changed every other day until sphere formation occurred, as we described previously. ## Long-term exposure of cells to NNK and arecoline NNK was obtained from ChemSyn Laboratories (Lenexa, KS). Arecoline was purchased from Sigma Chemical Co. (St. Louis, MO, USA) and dissolved in DMSO to a concentration of 100 mM. Parental SCC25 and FaDu cells were plated at a density of 5 × 10⁴ live cells/10-cm dish and treated with arecoline at a final concentration of 0, 10, 50, 100, or 150 nM for 3 months. Then the cells were harvested and cultured with NNK (50 nM) and/or arecoline (130 nM), with the medium being changed every other day. Arecoline (130 nM) and NNK (50 nM) were tested and found to be the most optimal dose for further analysis. Three experimental groups were prepared to examine the effects of NNK and arecoline, alone or in combination. The first and second groups of HNSCC cells were treated with NNK and arecoline separately, and the third group was treated with both NNK and arecoline. ## Cell viability assay The cells were seeded in a 6-well plate at a density of 3 × 10⁴ cells/well, and cell proliferation was measured by the 3-(4,5-dimethyl- thiazol-2-yl)-2,5-diphenyltetrazolium bromide (MTT) reduction method. After incubation with NNK and arecoline at different concentrations for 0, 24, 48, or 72 h or more than 3 months, the cells were incubated with 2.5% MTT solution (5 mg/ml) for cell viability assay, as previously reported. The same test was repeated three times, and the optical density was calculated for statistical analysis. ## Apoptosis assay Apoptotic cells were detected based on annexin V expression (Gene Research, Taipei, Taiwan) according to the manufacturer’s guidelines. After staining, the cells were incubated for 30 min in the dark at room temperature with 5 μl of a fluorescein isothiocyanate (FITC)-conjugated anti-annexin V antibody. The cells were then analyzed with a FACSCalibur instrument (Becton Dickinson, San Diego, CA, USA). ## Migration assay A total of 1 × 10⁵ cells was seeded in the top chamber of a 24-well plate with micropore polycarbonate membrane filters containing 8-μm pores (Becton Dickinson Labware, Lincoln Park, NJ). The bottom chamber was filled with RPMI 1640 medium containing 10% fetal bovine serum as a chemoattractant. The migrated cells were harvested after 24 h and stained with hematoxylin. The migrated cancer cells were then visualized and counted from 5 different visual fields (magnification, 100×) under a light microscope. ## Invasion assay The 24-well plate Transwell^® system with a polycarbonate membrane filter was employed to evaluate the invasion ability of cells and the details were followed by previous protocol. ## Western blot analysis Whole cell lysates (50 μg) were separated by electrophoresis on 12.5% denaturing polyacrylamide gels. The membranes were incubated overnight at 4°C with primary antibodies (0.1 μg/ml) against Oct4, Nanog, β-AR and phospho EGFR were purchased from Abcam Corporation (Abcam, Cambridge, UK), Snail, Twist, Fibronectin, E-cadherin, CD21, CD44, CD133, ALDH-1, phospho AKT, phospho NFκB, α7-nAChR, Bcl-2 were acquired from Santa Cruz Biotechnology (Santa Cruz, CA). Bax was obtained from Becton, Dickinson and Company (BD, Transduction Laboratories TM), MDR-1 and ABCG2 were purchased from Millipore Corporation (Millipore Corporation, Billerica, MA, USA), caspase-3 and PARP were obtained from Invitrogen Corporation (Invitrogen, Camarillo, CA, USA) in Tris-Tween-Buffer- Saline buffer containing 3% non-fat milk. Subsequently, each membrane was washed and incubated for 1 h at 25°C with a secondary anti-mouse, anti-rabbit, or anti- goat antibody conjugated with horseradish peroxidase (1: 1000; Santa Cruz Biotechnology, Inc.), as previously reported. ## Chemosensitivity assay Cells were seeded in 10-cm dishes at a density of 1 × 10⁶ cells/dish. For the chemosensitivity assay, cells were treated with 0 to 100 μM cisplatin (Cis; Sigma–Aldrich) for 48 h. Relative cell survival was determined by the MTS assay using the CellTiter 96 Aqueous One Solution Cell Proliferation Assay Kit (Promega, Madison, WI, USA). ## Flow cytometry A single-cell suspension of 1 × 10⁶ cells containing trypsinized cells and spheres was suspended in 1 mL of phosphate-buffered saline (PBS) and ALDH1 (ALDEFLUOR Assay Kit; Stem Cell Technologies, Durham, NC) as previously reported. ## Statistical analysis All data are shown as the mean ± SD. Differences between groups were calculated using Student’s *t*-test. Dose-dependent effects were measured by linear regression. All statistical analyses were performed using SPSS software, version 15.0 (SPSS, Inc., Chicago, IL, USA). # Results ## Long-term exposure to NNK and arecoline promoted migration and invasion and enhanced the EMT with morphological alterations in HNSCC cells We first determined the effects of NNK and arecoline exposure on cell propagation in three experimental groups of HNSCC cells by performing proliferation assays. Following a 3-month exposure to various concentrations of arecoline, SCC25 and FaDu cells were stimulated with NNK (50 nM), arecoline (130 nM), or a combination of NNK (50 nM) and arecoline (130 nM) for 24 h, 48 h, 72 h, or 3 months, as previously reported. Proliferation assay results showed that NNK and arecoline enhanced the cell growth rate in the short term in both cell lines; however, long-term treatment with NNK and arecoline eventually slowed cell proliferation until growth equilibrium was achieved. This standard condition was used for further experiments. Next, we investigated differences in the effects of NNK, arecoline, and combined treatment on migration and invasion and whether the treatments were involved in the EMT of HNSCC cells. EMT occurs during embryogenesis and is also a sign of tumor cell instability, resulting in the migration to neighborhood tissues. Long-term alkaloid exposure induced EMT in both cell lines, with a morphological conversion from an epithelioid to a mesenchymal appearance. To determine the effect of alkaloids on invasion ability, HNSCC cells were incubated with 50 nM NNK, 130 nM arecoline, or a combination of 50 nM NNK and 130 nM arecoline, after which they were prepared for invasion and migration assays. Combination treatment with NNK and arecoline promoted greater degrees of invasion and migration than treatment with either reagent alone (*p* \< 0.05). Western blot analysis revealed that long-term exposure to NNK, arecoline, and their combination altered the expression of representative EMT markers, including decreased E-cadherin expression, increased fibronectin expression, and increased expression of EMT regulators (Snail, Twist, and NFκB) in both cell lines. A synergistic effect of combination treatment was found, based on the overexpression of all of the aforementioned markers, especially NFκB. ## Long-term NNK and arecoline exposure promoted sphere formation and the overexpression of stem cell markers in HNSCC cells The efficiency of sphere formation was evaluated to examine self-renewal in CSCs following long-term NNK, arecoline, and combined exposure using a non-adhesive culture system. Most cell clusters transformed into spheres after 7 d in culture. In contrast, control cells only formed irregular cell masses that lacked a spheroid appearance. Flow cytometric analysis of the representative CD133 and CD24/CD44⁺ markers showed significant overexpression in long-term alkaloid-treated cells compared with that in parental cells (*p* \< 0.05). Identification of common CSC markers for HNSCC also showed upregulation of Nanog, OCT-4, and ALDH-1 in cells exposed to long-term alkaloid treatment compared with that in the parental cell lines. Again, a synergistic effect involving the overexpression of all aforementioned CSC markers was found in the experimental group exposed to combination treatment. ## Long-term exposure to NNK and arecoline increased drug resistance and upregulated MDR1 and ABCG2 expression The observation that alkaloids enhanced the CSC population suggested their involvement in modulating chemoresistance, an important characteristic of CSCs. To investigate the expression profile of alkaloid-exposed cells and chemoresistance, control and alkaloid-exposed cells were treated with Cis and subjected to MTT assays and western blot analysis. Significant differences in chemosensitivity were found between long-term alkaloid-treated cells and parental cells after exposure to various Cis concentrations for 24 h (*p* \< 0.05). Western blotting revealed that long-term alkaloid exposure resulted in overexpression of the *MDR1* and *ABCG2* genes compared to that in the parental cells, which was enhanced synergistically in cells treated with both NNK and arecoline. ## Long-term exposure to NNK and arecoline enhanced anti-apoptosis signaling Flow cytometric analysis (using annexin V staining) showed that long-term exposure to NNK and arecoline significantly attenuated (*p* \< 0.05) apoptosis in both cell lines following Cis treatment (80 μM), as shown in. Immunoblotting demonstrated that long-term NNK- and arecoline-treated cells showed dramatically increased levels of the anti-apoptotic oncoprotein Bcl-2 in both cell lines. In contrast, the levels of the apoptosis-promoting proteins Bax, cleaved caspase-3 (cl-caspase 3), and cleaved-poly (ADP-ribose) polymerase (Cl-PARP) decreased in both HNSCC cell lines exposed to long-term NNK and arecoline treatment. Again, a synergistic alteration in the expression of apoptotic markers was found in the combination treatment experimental group. ## Long-term exposure to NNK and arecoline induced the activation of EGFR but not the α7 nicotinic acetylcholine receptor (α7-nAChR) and β-AR EGFR, α7-nAChR, and β-AR each play pivotal roles in regulating the alkaloid- induced growth and progression of cancer cells. Here, we determined whether alkaloids could synergistically affect the activation of EGFR following long- term exposure. Western blot analysis showed that α7-nAChR stimulation was mainly induced by NNK and that β-AR expression was slightly induced by arecoline. Nevertheless, in cells treated with a combination of NNK and arecoline, abundant EGFR expression was stimulated in a long-term manner. In addition, treating cells with anti-EGFR antibody to neutralize EGFR activity resulted in downregulation of phospho-EGFR expression induction of the downstream phosphor- AKT and NFκB proteins caused by long-term treatment with NNK and arecoline in SCC25 and FaDu cells. Collectively, these results support our hypothesis that EGFR plays a more critical role is a pivotal receptor in promoting the proliferation of HNSCC cells following long-term treatment with NNK and arecoline. # Discussion Long-term exposure to environmental chemicals such as NNK and arecoline can increase tumor initiation and the aggressiveness of existing tumors, especially in HNSCC. However, the underlying mechanisms and signaling pathways responsible for tumor progression following long-term exposure to NNK and arecoline in combination have not been studied and need further clarification. This study was designed to include experimental groups with long-term (3-month) NNK and/or arecoline exposure to mimic habitual smoking and/or betel-nut chewing in daily life, which represented a longer and higher exposure than those used in previously reported studies conducted over a 2-month period. Long-term exposure to NNK has been reported to be associated with tumor initiation and promotion. Previous data showed the importance of α7nAChR in NNK-induced cell proliferation for lung cancer, breast cancer, and pancreatic cancer. Previous reports also showed that arecoline exerted a partial agonist activity with α7-nAChR in a dose-dependent manner or with habitual use. Areca is of much more significance as a drug of abuse because it is a selective partial agonist of both α4- and α6-nAChR. The silent agonist activity of arecoline for α7-nAChR indicated a mechanism for its effects on immune cells and also revealed its significant potential involvement in the carcinogenic effects of areca use. Accumulating evidence has also demonstrated that β-adrenergic receptors mediate the proliferative and anti-apoptotic effects of NNK in non-lung cancer cells. Furthermore, the mitogenic effects of NNK have been found to be mediated by β-adrenergic receptors, indicating that NNK promotes the growth of gastric cancers via PKC and ERK1/2 phosphorylation in a β-adrenergic receptor- dependent fashion. Our previous results showed that NNK induces HNSCC cell proliferation through a α7-nAChR–EGFR signaling axis to facilitate the growth of HNSCC. Our results from this study demonstrated that the arecoline-induced proliferation of HNSCC cells was mediated by both EGF and β-adrenergic receptors. However, the effects on EGFR and AKT signaling have been reported to be involved in cancer cell growth, up-regulating downstream anti-apoptosis signaling, sphere-forming capability, invasive and migratory potentials, and chemoresistance in numerous cancers.\[–\] In this regard, our results clearly verified that dual activation of EGFR following long-term treatment with NNK and arecoline promoted cancer proliferation more robustly than NNK or arecoline treatment alone. Following long-term exposure to NNK and arecoline, EGFR effector molecules (other than α7-nAChR and β-AR) were prominently induced and activated. Furthermore, we also demonstrated that abolishing EGFR activation by antibody treatment suppressed AKT and NFκB phosphorylation and cancer cell activation in HNSCC. EGFR overexpression in response to AKT phosphorylation and NFκB activation inhibited apoptosis, which has been correlated with low patient survival rates. We successfully confirmed that pAKT and NFκB activity (via the EGFR–AKT signaling pathway) played significant roles in the proliferation of SCC25 and FaDu cells. Previous data have demonstrated that the suppression of apoptotic pathways played an important role in response to long-term stimulation with alkaloids.\[–\] Our data clearly demonstrated that downregulation of apoptotic proteins including CL-PARP and cl-caspase 3 supervened following long- term treatment with NNK and arecoline in both cell lines. The Bcl-2 family of proteins is well-characterized regulators of apoptosis. The pro-apoptotic protein Bax can act as a gateway for caspase-mediated cell death. The Bcl-2:Bax ratio is an important determinant of the susceptibility to apoptosis, which was also confirmed by our results. The EMT is a critical step in the development of metastasis and acquisition of resistance to targeted therapeutics, including tyrosine kinase inhibitors against EGFR. Replacement of the epithelial marker E-cadherin with the mesenchymal marker fibronectin significantly was shown to increase Snail expression and slightly increase Twist expression, which together represent typical features of EMT. EGFR/AKT-mediated signaling is involved in different metastatic cancers, and its role in chemoresistance is well documented. Thus, inhibition of EGFR activation and pAKT as a complement to TKI-based therapy can reverse cellular mechanisms, leading to an attenuation of chemoresistance. A previous study on the alterations of EMT markers in lung cancer demonstrated that decreased E-cadherin or increased fibronectin levels are associated with poor survival outcomes. Inhibition of tumor invasion or migration is one of the goals of suppressing CSC properties, especially in patients who have metastases instead of only a primary lesion. The results of Boyden chamber assays showed the enhancement of the long-term effects of combined NNK and arecoline exposure on SCC25 and FaDu cell motility and migration, which are integral steps in the metastatic cascade of tumor cells, suggesting that long- term exposure to NNK and arecoline can promote invasion and metastasis to distal sites. As described above, mounting evidence also suggests a link between CSCs and the EMT. Herein, we showed that long-term exposure to NNK and arecoline in HNSCC cells induced the development in CSCs linked with the EMT, which is key for tumor progression. From the standpoint of CSC-related gene products, validated stemness gene-related proteins, such as Nanog and OCT4, support the invasive competence of tumor cells, which can self-renew.\[–\] Based on our data, overexpression of ALDH1, Nanog, and OCT4 was also found in long-term NNK- and arecoline-treated cells compared with the levels in control HNSCC cells. In this study, as in other reports, no specific marker of CSC properties in HNSCC was delineated, and the relevance of other CSC-associated markers, such as CD24/CD44⁺ or CD133, is still debated. In summary, we demonstrated that long-term exposure to NNK combined with arecoline activated EGFR/AKT signaling and was involved in anti-apoptosis, CSC properties, and Cis resistance in HNSCC cells. These findings provide new insights into the potential molecular mechanisms of long-term NNK- and arecoline-induced anti-apoptosis and CSC characteristics in head and neck cancer cells. The results presented here suggest that both NNK and arecoline not only are strong carcinogens by themselves but also can synergistically promote the growth and progression of HNSCC cells when administered in combination. The ability of NNK and arecoline to synergistically promote proliferation, adherence-independent growth, EMT, and CSC properties, which could be inhibited by administering an anti-EGFR antibody, suggest that NNK and arecoline might contribute significantly to the growth and metastasis of EGFR-sensitive tumors. Targeting this pivotal receptor when managing tumor aggressiveness and therapeutic resistance might prove beneficial in treating HNSCC, as shown in our proposed model. Our data suggest that understanding these mechanisms is critical in developing effective therapeutic strategies with the potential of CSC inhibition in treating tumors that develop after NNK and arecoline exposure. This study was partly supported by the Tri-Service General Hospital (Grant Numbers TSGH-C106-040 and TSGH-C106-038) and the National Science Council, Republic of China (Grant Numbers NSC105-2320-B-039-028-MY2 and NSC105-2314-B-016-028-MY2). We thank Chun An Ho for her assistance in performing some experiments in this study. [^1]: The authors have declared that no competing interests exist.

# Introduction Enterohemorrhagic *Escherichia coli* (EHEC) O157:H7 was first identified as etiological agent of bloody diarrhea in the early 1980s and has since been recognized worldwide as a cause of food- and waterborne infectious diseases. It may also lead to the development of hemolytic uremic syndrome (HUS), an infection sequelae characterised by hemolysis and renal failure which can lead to long-term kidney damage or fatal outcome. *E. coli* O157:H7 has caused many outbreaks in the past three decades, with a wide range of clinical illness. In 1982, the first outbreak of O157:H7 involved at least 47 cases of diarrhea in Oregon and Michigan, associated with consumption of undercooked beef patties at fast food restaurants. Subsequently there were two multi-state outbreaks in the US. In 1993 an outbreak from hamburgers had 501 cases of diarrhea reported, including 151 hospitalizations (31%), 45 cases of complicated HUS (9%), and three deaths, while in 2006, an outbreak associated with spinach had high rates of both hospitalization (50%) and HUS (16%). Outside the US, a massive outbreak associated with consumption of white radish sprouts in Sakai, Japan in 1996 had about 7,470 school children infected, 1,000 hospitalizations for severe gastrointestinal symptoms, 100 HUS cases and three deaths. A less well known massive outbreak of O157:H7 occurred in Xuzhou, China, in 1999, with 195 hospitalized patients who had clinically diagnosed HUS and 177 deaths, which has only been reported in Chinese journals. Despite that a large number of virulence genes in O157:H7 have been identified, the factors critical for HUS development is poorly understood. The hallmark of the disease is the production of Shiga toxins in the intestine by O157:H7 leading to the damage of the endothelial cells and potential HUS. Two forms of the Shiga toxin, Stx1 and Stx2, are known with the latter being more cytotoxic which may increase the risk of developing HUS. Variants of *stx*2 have been observed in O157:H7 and strains carry one or two *stx2* alleles are more likely to cause HUS. It has also been shown that the Shiga toxin genotype rather than the amount of Shiga toxin or the cytotoxicity of Shiga toxin *in vitro* correlates with the appearance of HUS. A recent study suggests that the 2006 Spinach outbreak strain has evolved to higher virulence, which has increased in frequency in recent years in the US, and is significantly more likely to be associated with HUS. In this study we investigated the genomic and phenotypic properties of the 1999 China outbreak associated O157:H7 strains. We sequenced the genome of the 1999 China outbreak strain Xuzhou21, and compared it with the genomes of the 1996 Japan outbreak strain Sakai, the 1982 and 2006 US outbreak strains EDL933 and TW14359. We show that Xuzhou21 carries a highly inducible *stx2* expression and has the ability to provoke significantly elevated inflammatory response. # Materials and Methods ## Laboratory investigation of the outbreak Isolation of O157:H7 from patient's fecal sample was done using Sorbitol- MacConkey agar. Sorbitol non-fermenting colonies were selected for biochemical and agglutination tests with O157 and H7 antisera. *E. coli* O157:H7 was isolated from other sources by enrichment using anti-O157 coated immunomagnetic beads and plating on Sorbitol-MacConkey agar. ## Bacterial isolates The O157:H7 strains used in this study were isolated from feces of human patients and animals between 1988 and 2005 from Xuzhou City, Jiangsu province (80 strains) and the neighboring Anhui (25 strains), Henan (nine strains) and Shandong (six strains) provinces. Four isolates from other parts of China (Yunnan province), two isolates from USA and eight isolates from Japan were also included, as well as three fully sequenced strains EDL933, Sakai and TW14359. ## Multilocus sequence typing MLST was performed on 15 housekeeping genes (*arcA, aroE, aspC, clpX, cyaA, dnaG, fadD, grpE, icdA, lysP, mdh, mtlD, mutS, rpoS, uidA*). A detailed protocol of the MLST procedure, including allelic type and sequence type (ST) assignment methods, can be found at the EcMLST website (<http://www.shigatox.net/mlst>). MEGA 4.0 was used to construct phylogenetic trees and eBURST was used to cluster STs into clonal complexes which consist of STs differing by one of the fifteen genes typed. ## Cytokines assays Peripheral blood mononuclear cells (PBMCs) were isolated from fresh blood from healthy individuals. PBMCs were diluted sequentially up to four-fold in a 96 well plate. Bacterial cells inactivated by heat at 56°C for one hour were added to the PBMCs and incubated for four hours. IL-6 and IL-8 were assayed by use of Fluorokine MAP Multiplex Human Cytokine Panel A of the Luminex 100 Analyzer with the X-Y Platform (Luminex, USA). ## *stx2* expression assays Cultures were grown in Luria-Bertani medium at 37°C with shaking. For induction of the Stx2 phage, mitomycin C (BBI, American) was added to a final concentration of 0.5 µg/ml and incubated for three hours. Total RNA was extracted with RNeasy Mini Kit (Qiagen, Germany). Real-time PCR determination of gene expression was performed with the Rotor-Gene Q Real-Time PCR system (Qiagen, Germany) using a One Step SYBR® PrimeScript™ RT-PCR kit (TaKaRa, Japan), according to the manufacturer's instructions. *gapA* was used as an endogenous control. The relative level of gene expression was calculated using the 2^−ΔΔCT method. Each sample was run in triplicate. ## Verotoxicity assay One hundred microliter aliquots of MEM-resuspended bacteria at a multiplicity of infection (MOI) of 100 or supernatant were added in triplicate to Vero cells (10⁴ cells per well) in 96-well tissue culture plates, then incubated at 37°C in a 5% CO₂ atmosphere. After 4, 8, 12, 20 hours, the supernatant was collected and the release of the cytoplasmic lactate dehydrogenase (LDH) was evaluated using the Cytotox96 kit (Promega, Madison, Wisconsin, USA) according to the manufacturer's instructions. The percentage of cytotoxicity was expressed as (experimental release−spontaneous release)/(maximum release−spontaneous release)×100. The spontaneous release was the amount of LDH activity in the supernatant of uninfected cells and the maximum release was that when cells were lysed with the lysis buffer. ## Genome sequencing and analysis Chromosomal DNA from Xuzhou21 was sequenced using a 454/Roche FLX machine, according to the manufacturer's protocols, producing 318,830 reads with an average length of 410 bp that were assembled de novo into 159 contigs of at least 1,000 bp, using the 454/Roche Newbler assembly program. Gaps were closed by directed PCR and sequencing with BigDye terminator chemistry on an ABI 3730 capillary sequencer. To avoid sequence miscalls in homopolymeric tracts, all questionable sites and the sites different from other *E. coli* O157 strains were checked by resequencing of PCR products using an ABI 3730. Genome annotation and comparison were done as described previously. The collinear blocks of the O157:H7 genomes were determined using BLASTN. Then the alignment within each of the block was obtained using Mauve. The generation of the final plot, identification of recombination segments, and allocation of SNPs to specific lineages were done similar to the method used in the study of *Vibrio cholerae* by Feng *et al.*. The genome sequence and pO157 sequences of Xuzhou21 have been deposited in the GenBank database under accession numbers CP001925 and CP001926 respectively. # Results ## Epidemiological investigation of the outbreak The O157:H7 outbreak occurred between April and September and peaked in June, 1999 with 195 HUS cases and 177 deaths from 52 villages of seven counties in Jiangsu and neighboring Anhui province. Of the 195 cases, 167 (85.6%) were over 50 years old with only two less than 20 years old and 121 (62.1%) were female. The National Institute for Communicable Disease Control and Prevention, China CDC, commenced the outbreak investigation on June 28, 1999. Three and two strains of O157:H7 were isolated in Xuzhou city from fecal screening of 30 HUS and 25 diarrhea patients respectively. Thirty six sera collected from 42 HUS patients (85.7%) tested positive for IgG against EHEC-hemolysin or O157 lipopolysaccharide. Thus both bacterial and serological data confirmed that the outbreak was caused by O157:H7. The source of the infection was investigated using a case-control sample of 146 HUS patients and 840 healthy individuals, matched in age, sex and residence. No hand-washing before eating, consumption of fruits or vegetables without washing, consumption of leftover foods without heating, no fly-net cover for foods and high density of flies in kitchen were found to be statistically associated with infected patients. Using magnetic beads coated with antibodies against the O antigen, O157:H7 was isolated from six of 67 (9.0%) fly specimens, four of 74 (5.4%) raw meats and three of 83 (3.6%) cooked meats. O157:H7 was also isolated from live animals including 32 of 189 (16.9%) cattle, 50 of 605 (8.3%) pigs, 91 of 590 (15.4%) goats and 52 of 604 (8.6%) chickens raised in courtyards of families with and without HUS patients in the same villages. ## The 1999 China outbreak isolates belonged to a novel clone ST96 The 15-locus multilocus sequence typing (MLST) scheme developed by Whittam and colleagues (<http://www.shigatox.net/ecmlst/>) which offers higher resolution than the standard 7-locus MLST was used to type a selection of 124 O157:H7 isolates including the five isolates from human patients from the 1999 Xuzhou outbreak and 68 isolates from other sources in the outbreak region during the outbreak period. MLST differentiated the 124 isolates into five sequence types (STs) (ST23, ST24, ST96, ST97 and ST98). ST23 was predominant with 83.1% of the isolates. All five Xuzhou outbreak human isolates and five animal isolates from the outbreak region were found to belong to a novel ST, ST96. Two other novel STs were ST97 and ST98. The new STs have been submitted to EcMLST database. The recently fully or partially sequenced O157:H7 strains, fall into existing STs except that three strains have new ST profiles (ST110, ST111and ST112). These STs, together with nine O157 STs (ST23, ST24, ST25, ST31, ST84, ST101, ST110, ST111,ST112) from the EcMLST database, were analyzed by eBURST. Using the definition of one gene difference out of the 15 genes as a clonal complex, eBURST analysis found that ST96 is closer to ST97 and forms a clonal complex with seven other STs including the four genome sequenced strains, EDL933, Sakai, TW14359 and EC4115. ## Genome sequencing of a 1999 China outbreak isolate Xuzhou21 reveals a clonal relationship with Japan outbreak strain Sakai Xuzhou21 isolated from an HUS patient from the 1999 Xuzhou outbreak was selected for genome sequencing. As reported previously, Xuzhou21 shared the same PFGE pattern with all four other human isolates from the same outbreak and seven other animal isolates isolated during the outbreak. The Xuzhou21 genome consists of one chromosome of 5,386,223 bp and two large plasmids (pO157 and pO157_Sal). We compared Xuzhou21 with EDL933, Sakai and TW14359 genomes. The overall relationship based on full genome sequences together with changes along the branches is shown in. Using the method of Feng *et al.*, we differentiated the base changes into mutational and recombinational changes and were also able to allocate mutational differences to specific lineages using O55:H7 strain CB9615 as an outgroup. The majority of the base changes are due to mutations. There are seven recombinational events with six being less than 200 bp while one involving 7,747 bp. The mutational SNPs separated Xuzhou21 from EDL933 and grouped it with Sakai with 17 SNPs supporting this branching order. However two SNPs support Xuzhou21 and TW14359 as a group. Fifty five unique SNPs are carried by the Xuzhou21 genome that separate the Chinese outbreak strain from Sakai, besides the 17 branch supporting SNPs. The genome data also allowed us to assign Xuzhou21 to lineage I of the LSPA lineage typing scheme, and clade 3 of Manning *et al.* SNP based typing scheme. It is interesting to note that Xuzhou21 is in the same clade as EDL933 although it is closer to Sakai based on genome data as none of the 17 Xuzhou21-Sakai branch supporting SNPs was in the set of Manning *et al.* SNPs. A complete genome (EC4115) and 25 incomplete genomes of O157:H7 were recently published by Eppinger *et al.*. EC4115 has been shown to be closely related to TW14359. Although EC4115 was isolated during the 2006 spinach outbreak, it may represent a strain from a separate outbreak (see Eppinger *et al.* 2011). Our comparison of EC4115 and TW14359 revealed only 46 SNPs and thus EC4115 was not used for a detailed comparison with Xuzhou21. Eppinger *et al.* also identified two strains (EC4501 and TW14588) that are closely related to Sakai. We extracted the SNPs from the genome sequences of these two strains and used the SNPs common with the mutational SNPs we identified above to construct a phylogenetic tree. EC4501 and TW14588 remain close to Sakai and Xuzhou21 diverged earlier than the other two strains. Both EC4501 and TW14588 are in Manning *et al.* clade 2 and the genomic relationship is consistent with the clade relationship defined by Manning *et al.*. Note that EC4501 seems to be far more incomplete than TW14588 with over 270 of the 902 SNPs common to the four complete genomes are missing in EC4501, which may affect the branching order of EC4501 and TW14588. In comparison to Sakai, there are five large insertions and seven large deletions of \>100 bp in Xuzhou21 with the majority of these indels located in prophage genomes. Six genes are specific to Xuzhou21, among which, four genes are located in the Stx2 phage genome including *ninE* and two genes encoding putative phage replication functions while the remaining two, both encoding a transposase, located together and are flanked by an integrase gene on both sides. Surprisingly 30 genes were lost in Xuzhou21. Most of the genes lost are phage related. Two EDL933 phages, CP-933R and CP-933T (equivalent to Sp10 and Sp13 in Sakai), are absent in Xuzhou21. On the other hand, Sakai lost only six genes but gained 99 genes. Most of the genes gained are phage related, including phage Sp18. Sixty six genes were gained by the common ancestor of Xuzhou21 and Sakai. All except two genes are in five blocks of two to 15 genes which were likely to have been gained by single events. None of these genes gained were known virulence-related genes. The novel plasmid pO157_Sal in Xuzhou21 has been reported previously by our group, and this plasmid was only observed in ST96, ST23 and ST98 strains. pO157_Sal contains no known virulence genes. Interestingly, a similar plasmid (pEC4115) has also been found in EC4115. Our comparison showed that the two plasmids differ substantially. Only 21 of the 52 pO157_Sal genes are homologous to genes in pEC4115 with amino acid level identity ranging from 28% to 51%. The majority of the homologous genes are type IV secretion system (T4SS) genes and genes associated with plasmid function. We also performed a phylogenetic analysis of the plasmids using TraE which showed that both plasmids belong to the same cluster but are distantly related. It should be noted that several other O157:H7 strains carry a T4SS and were grouped together based on the TraE. The TraE sequences were extracted from partially sequenced genome sequences, and it is not known whether the T4SS in these strains was also plasmid borne. The pO157 plasmid, which is present in all O157:H7 strains, shares high similarity among the four strains. Xuzhou21 has the smallest number of changes with two SNPs in non-coding regions, one single base insertion and two single base deletions. It is interesting to note that all changes occurred singularly in one of the plasmids only. Thus there were no parsimony informative changes to infer relationships among the four plasmids. ## Xuzhou21 induced high levels of cytokine response and *stx2* expression We determined whether there is a difference among the three strains Xuzhou21, EDL933 and Sakai in proinflammatory response as inflammation associated with HUS is marked by the release of cytokines/chemokines,. Using peripheral blood mononuclear cells (PBMCs), we assayed the level of production of two proinflammatory cytokines, IL-6 and IL-8. As shown in, the levels of expression of the two cytokines induced by Xuzhou21 and Sakai were significantly higher than that induced by EDL933 (ANOVA, *P*\<0.01). Xuzhou21 also induced a significantly higher level of IL-8 than Sakai (*P*\<0.01) while both induced similar levels of IL-6 (*P* = 0.36). Previous studies suggest that the production of the Stx2 phage is associated with an increased risk of HUS. We measured the level of the transcription of *stx2* in Xuzhou21, EDL933 and Sakai using real-time quantitative PCR. We first measured the *stx2* expression under non-inducing conditions. *stx2* was expressed constitutively in all three strains and its transcription levels in Sakai and Xuzhou21 were only 3.7 and 7.5 times higher than that in EDL933. As the Stx2 phage can be induced during infection leading to a much higher level of *stx2* expression, we used mitomycin C, a stimulus of the SOS response, for induction to determine levels of induced *stx2* expression *in vitro*. Relative to the constitutively expressed level in EDL933, induced *stx2* expression levels were 4.5, 32.7, 68.6 times higher in EDL933, Sakai and Xuzhou21 respectively and the difference was statistically significant (ANOVA, *P*\<0.01). Interestingly we also observed more complete cell lysis in Xuzhou21 after three-hour Stx2 phage induction. Vero cell cytotoxicity assay was also performed. Xuzhou21 showed higher cytotoxicity after 12-hour incubation than EDL933 and Sakai (data not shown). We tested all other isolates in ST96 for induced expression of *stx2* and found that the isolates produced high levels of *stx2* with an average of 86.7 times higher than EDL933 and ranging from 36.6 times to 174.4 times, suggesting that elevated *stx2* expression is a general characteristic of the ST96 clone. # Discussion The outbreak of O157:H7 in Xuzhou, China, in 1999, led to 195 hospitalized HUS patients and 177 deaths, and is one of the largest O157:H7 outbreaks known, although it is hardly known outside China and was previously only reported in Chinese journals. From the epidemiological investigations, the outbreak was mainly associated with peasants living with animals carrying O157:H7 in the household, including goats, pigs, chickens and cattle. Courtyard animals carrying O157:H7 contaminated the surrounding environment through fecal shedding and persons who had poor personal and kitchen hygiene practice were more likely to be infected. It is well established that farm animals are carriers of O157:H7. Additionally we found that 9% of the flies tested were positive for O157:H7 and thus they are important carriers in this outbreak. Flies may not just be mechanical vectors as O157:H7 can multiply inside the fly's mouth and be excreted through fly fecal matter. Therefore poor hygiene and multiple routes of transmission may be the major contributing factors to the massive outbreak. However, increased transmission would have expected to increase number of infections but not higher number of HUS rate and high mortality rate. Host factors may contribute to higher mortality with a disproportional number of HUS cases and deaths in the older age groups. We showed that the outbreak was caused by a new sequence type, ST96. Genome sequence analysis of Xuzhou21 did not reveal any new virulence genes on its chromosome. However, Xuzhou21 carries an additional novel plasmid which contains a T4SS and may contribute to the virulence of Xuzhou21. A recent study showed that the strain represented by TW14359 that caused the 2006 spinach- associated outbreak caused a much higher rate of HUS (15%) in comparison to other outbreaks in the US. TW14359 has been shown to be on a divergent lineage from Sakai and EDL933. Comparative studies of TW14359 with other O157:H7 strains have revealed many genetic changes, but none of the strain-specific genes definitively contributes to HUS although a number of virulence factors were found. Shiga toxin is a key virulence factor for O157:H7 pathogenesis, and indeed for any STEC infections including the recent O104:H4 outbreak in Germany. The Stx1 and Stx2 prophages are present in all four genome sequenced strains and TW14359 also carries a Stx2c. Like Sakai and EDL933, the Stx1 and Stx2 prophages in Xuzhou21 were inserted in *yehV* and *wrbA* respectively. Kulasekara *et al.* showed that TW14359 produced a 150 fold increase in Stx2 expression under mitomycin C induction and the induction was much higher than that occurred in EDL933. We also observed this differential inducibility between Xuzhou21 (and Sakai) and EDL933. The *stx2* expression in Xuzhou21 was 68.6 times of that under non-inducing conditions, twice of that induced in Sakai (32.7 times) and 15 times of that induced in EDL933 (4.5 times). Therefore, a crucial difference between the strains seems to be the differential inducibility of the *stx2* expression. A recent study compared the *stx2* expression of strains represent four different SNP clades using ciprofloxacin as the inducing agent and found that *stx2* was more highly induced in clade 8 strains including TW14359 than clades 1 to 3 strains. Our finding that Xuzhou21 expresses a higher level of *stx2* than EDL933 is interesting as both Xuzhou21 and EDL933 belong to clade 3. There is clearly a large variation in *stx2* expression even within a clade. This observation also suggests that TW14359 and Xuzhou21 evolved to express higher levels of *stx2* independently. The inducibility of *stx2* expression must lie within the Stx2 prophage genome as the Stx2 phage is inserted at *wrbA* in Xuzhou21 and Sakai, a different site (*argW*) from that in TW14359. We compared the four Stx2 prophage genomes and found many differences among them, making it difficult to pinpoint the genetic determinants that exert the effect on *stx2* expression since both unique phage genes and nucleotide polymorphisms could affect *stx2* expression. Overall Xuzhou21 and Sakai shared more phage genes, consistent with whole genome relationship. The Q antiterminator upstream of *stx2* is crucial for *stx2* expression. However no variation was observed in this region among the four strains. Thus the observed phenotypic differences cannot be genetically linked to polymorphisms in the Q antiterminator genes, consistent with previous findings. Antirepressors play an important role in phage induction. One gene in the Stx2 prophage genome, ECs1199, encoding an antirepressor protein, was found to be absent in EDL933 but present in all three other genomes, in addition to an antirepressor (ECs1214) common to all four Stx2 prophages. We speculate this additional antirepressor regulates *stx2* induction and may contribute to the difference between Xuzhou21/Sakai and EDL933. We also examined other regulatory proteins including CI, CII, CIII, Cro, O and P. Variation in CI and Cro cannot be correlated with *stx2* expression differences observed since only Sakai varies in these two genes. CII, O and P are all polymorphic among the four strains with each strain being different whereas CIII is identical among all except EDL933 which differs by one amino acid. It is possible that variation in these genes differentially affect *stx2* expression. The data of the two proinflammatory cytokines tested, IL-6 and IL-8, are consistent with the difference between EDL933 and Xuzhou21 (and Sakai). The levels of IL-6 and IL-8 of PBMCs induced by Xuzhou21 and Sakai were similar, but almost doubled the amount induced by EDL933. Although systemic inflammatory response does not seem to precede the development of HUS, there is a significant amount of data showing an association between inflammation and the development or the severity of HUS. There is also an intricate direct relationship between cytokines and Shiga toxins. It is known that IL-6 and IL-8 are stimulated by Stx2,. Xuzhou21 and Sakai induced much higher levels of *stx2* expression under mitomycin C induction, confirming this direct relationship of a highly inducible Stx2 phage with vigorous cytokine responses. Stx2 also induces production of other chemokines such as SDF-1α, SDF-1β, and RANTES which simulate platelet function and renal thrombosis associated with HUS. Therefore these interactions further support the key role of *stx2* inducibility in virulence and disease development. In conclusion, the 1999 China O157:H7 outbreak is so far the largest O157:H7 outbreak in terms of the number of HUS cases and deaths. We show in this study that the outbreak was caused by a new sequence type, ST96 and the outbreak strain Xuzhou21 is most closely related to Sakai. We found that Xuzhou21 has the capacity to provoke significantly elevated proinflammatory responses, and carries a highly inducible Stx2. Our analysis of Xuzhou21 provided further insight into the genetic basis of virulence of O157:H7. # Supporting Information [^1]: Conceived and designed the experiments: YX PW RL CY JX. Performed the experiments: YX PW XB ZC XL AZ Yan Wang SZ HS. Analyzed the data: YX PW Yiting Wang RL JX. Contributed reagents/materials/analysis tools: HW JR HJ ZZ LW. Wrote the paper: RL JX. [^2]: The authors have declared that no competing interests exist.

# Introduction Nonalcoholic fatty liver disease (NAFLD) is the most common form of chronic fatty liver disease worldwide. Approximately 25% of the global population is estimated to have some level of NAFLD. The World Health Organization in 2016 reported over 1.9 billion overweight adults, which paralleled the increase in patients diagnosed with NAFLD. The National Health and Nutrition Examination Survey (NHANES) estimated nearly 40% of adults in the US are obese. Obesity severity increases the likelihood of developing NAFLD ranging from 75% in overweight individuals to 90–95% in morbidly obese individuals. NAFLD is associated with metabolic syndrome (MetS) and MetS is linked to obesity, type 2 diabetes mellitus (T2DM), dyslipidemia and hypertension, the top four risk factors for NAFLD. NAFLD is a continuum of fatty liver diseases ranging from hepatic macrosteatosis to nonalcoholic steatohepatitis (NASH, the progressive form of disease), cirrhosis, hepatocellular carcinoma (HCC) and liver failure. NAFLD occurs in children and adults, and both males and females. Factors contributing to the onset and progression of NAFLD include diet, lifestyle, genetics, sex, ethnicity and genetic polymorphisms. Since there are no FDA-approved treatment strategies for NAFLD, current treatment strategies focus on treating the comorbidities associated with NAFLD, such as obesity, type 2 diabetes, insulin resistance and dyslipidemia. According to the National Institute of Diabetes and Digestive and Kidney Disease (NIDDK), “NAFLD is a silent disease with no obvious symptoms” <https://www.niddk.nih.gov/health-information/liver-disease/nafld-nash/symptoms- causes>. As such, NAFLD is typically diagnosed in patients with pre-existing diseases, such as obesity, type 2 diabetes, dyslipidemia and metabolic syndrome (MetS). The current view of NAFLD onset and progression to nonalcoholic steatohepatitis (NASH) involves the accumulation of hepatic lipid (steatosis) including excessive hepatic cholesterol, diacylglycerols, ceramides, and oxidized lipids that promote inflammation, hepatic injury and fibrosis. In addition, the gut-liver axis plays a role in the onset and progression of NAFLD. The liver is the first organ to encounter dietary products absorbed by the gut. Problems occur when the integrity of the gut is compromised, such as leaky gut that can lead to gut luminal products entering the bloodstream. Some of these products are derived from gut microbes and are known to promote systemic inflammation. For example, NAFLD is associated with a low-grade metabolic endotoxemia. Endotoxin (lipopolysaccharide, LPS) is a gram-negative bacterial cell wall component that activates the innate immune system through Toll-like receptors, specifically TLR4. This process likely contributes to hepatic inflammation and dysregulation of hepatic metabolism. Interestingly, germ-free mice do not develop diet-induced fatty liver disease revealing the importance of diet and gut-microbes in the onset and progression of NAFLD. The diet most commonly associated with NAFLD is the western diet (WD). The WD is high in saturated (SFA) and monounsaturated fat (MUFA), cholesterol, simple sugar, and low in fiber. The WD also has an abnormal content of essential fatty acids (EFAs), i.e., linoleic acid (LA, 18:2, ω6) and α-linolenic acid (ALA, 18:3, ω3). The EFAs are either below recommended intake levels or the WD has a high ω6/ω3 polyunsaturated fatty acid (PUFA) ratio. A high ω6/ω3 PUFA ratio is associated with inflammation. PUFA also represent a low percentage (less than10%) of all fatty acids in the WD. Several clinical reports indicate that NAFLD severity, i.e., progression from steatosis to NASH, cirrhosis and HCC, is associated with a decline in hepatic C_18-22 ω3 and ω6 PUFA. Preclinical studies have established that the WD-induced decline in hepatic C_18-22 ω3 and ω6 PUFA is associated with NAFLD onset and progression to NASH. While clinical and histological markers of NAFLD are well-defined, the role of chronic ingestion of a WD-like diet on the onset of NAFLD and its progression to NASH has not been fully elucidated. There is little data in the human or preclinical literature on the antecedent events preceding the abnormal accumulation of lipids in the liver. Our aim in this report is to fill this gap by using an established preclinical model of WD-induced NAFLD to identify early changes in systemic and hepatic markers that precede overt evidence of insulin resistance, obesity and NAFLD. Accordingly, we fed age-matched female and male *Ldlr*^*-/-* mice a purified low-fat diet (LFD) and a western diet (WD) for 40 wks to promote NASH. Our use of the *Ldlr*^*-/-* mouse as a preclinical model of NAFLD is based on the following information. *Ldlr*^*-/-* mice used in this study are on the C57BL/6J background. In contrast to wild type C57BL/6J mice fed the WD, *Ldlr*^*-/-* develop the full spectrum of NASH characteristics in response to the WD. Interestingly, recent studies indicate that excessive hepatic cholesterol content plays a significant role in human NAFLD and NASH. Moreover, cholesterol lowering drugs, like statins and/or ezetimibe, are used in \~52% of diabetes patients to improve dyslipidemia and in NAFLD patients to reduce the risk of HCC. WD-fed *Ldlr*^*-/-* mice become obese and develop multiple markers of NAFLD including insulin resistance (HOMA-IR), dyslipidemia, endotoxemia and elevated markers of hepatic macrosteatosis, inflammation, fibrosis as well as evidence of hepatic injury \[alanine aminotransferase (ALT) and aspartate aminotransferase (AST)\]. In addition, the WD has major effects on the hepatic lipidome, including changes in hepatic oxylipins. Since previous studies documented an inverse relationship between NAFLD severity and the hepatic oxylipidome, we were interested in determining whether WD rapidly alters the hepatic oxylipidome. Finding rapid effects on the hepatic PUFA and the oxylipidome may reveal a potential cause and effect role for specific PUFA and oxylipins in the onset and progression of NAFLD and NASH. Our goals in this study were to: 1) identify early markers of WD-induced disease; 2) determine which lipids might serve as biomarkers of NAFLD to NASH; and 3) determine if male and female mice respond similarly to the WD. Herein, we establish that one week on the WD was sufficient to induce significant changes in multiple systemic and hepatic markers linked to NASH, including dyslipidemia and inflammation, as well as changes in hepatic essential fatty acid (EFA) content. These early responses to the WD likely set the stage for disease progression resulting in significant liver injury and NASH. # Materials and methods ## Animals and diets All procedures for the use and care of animals used in laboratory research were followed and approved by the Institutional Animal Care and Use Committee at Oregon State University (OSU). The study used two-month-old female and male *Ldlr* ^*-/-* \[B6;129S7- *Ldlr* ^*Tm1Her*/J mice, stock# 002207\] purchased from Jackson Laboratories. The study was carried out concurrently with both female and male mice. Upon arrival, mice were housed (5 mice/cage) at the OSU Linus Pauling Science Center vivarium in the same temperature-controlled room and handled by the same personnel throughout the study. Mice were maintained on a 12-hour light/dark cycle. The study design is illustrated in. All mice were fed a purified low-fat diet (LFD) \[Research Diets: D12450K\] for 2 weeks (wks) prior to initiating the feeding trial to acclimate the mice to a purified diet and the vivarium. The 40-week (wk) time course study of female and male mice consisted of two randomized groups for each sex. The Low-Fat Diet (LFD) \[Research Diets: D12450K\] group and the Western Diet (WD) \[Research Diets: D12079B\] group. The purified low-fat diet (LFD) contained 20% energy as protein (casein, cysteine), 70% energy as carbohydrate (corn starch (52%); maltodextrin (14%); sucrose 0.4%)); 10% energy as fat (soybean oil, lard) and cholesterol (0.002 mg/g) of diet. The purified western diet (WD) contained 17% energy as protein (casein, methionine); 43% energy as carbohydrate (sucrose (30%); corn starch (10%); maltodextrin (3%)); 40% energy as fat (butter, corn oil); and cholesterol at 1.5 mg/g of diet. Both diets contained a vitamin and mineral mix and fiber, while the WD contained an additional antioxidant. The energy density of the LFD and WD was 3.82 kcal/gram and 4.67 kcal/gram, respectively. More details of the diets, including the fatty acid composition of the diets, are described in. Female and male mice were maintained on the LFD for 1 and 40 wks (4 mice/sex and time point). The remaining female and male mice were fed the WD (8 mice/sex and time point). Health status and body weight of the mice were assessed twice weekly. Chow remaining from the previous feeding was weighed, discarded and fresh food added. ## Euthanasia: Recovery of blood and liver After 1 and 40 wks on the LFD and 1, 4, 8, 20 and 40 wks on the WD, mice were fasted overnight. The next morning at 8 AM, mice were euthanatized by CO₂ and liver and blood were collected as described above. A portion of the liver (major lobe) was transferred to vials containing PBS-buffered formalin for histology. The remaining liver was frozen in liquid nitrogen and subsequently transferred to a -80°C. freezer for storage. Collected blood was transferred to vials containing EDTA to prevent clotting. Plasma and cells were centrifuged to recover cells and plasma separately and stored at -80°C. Liver and plasma were analyzed as described in the Methods section above. ## Plasma and hepatic assessments Plasma glucose, total and free cholesterol (T Chol; F Chol), and triacylglycerides (TAG) were measured using kits (Wako Diagnostics; Richmond, VA) as previously described. Plasma insulin (Crystal Chem; <sales@crystalchem.com>), aspartate aminotransferase (ALT), and alanine aminotransferase (AST) were measured using kits from Thermo-Fisher as described. Plasma non-esterified fatty acids (NEFA) were measured using a kit from Sigma- Aldrich and β-hydroxybutyrate (β-HB) was measured with a kit from Zen-Bio. Plasma toll-like receptor agonists (TLR2 Ag and TLR4 Ag) were quantified as previously described. The homeostatic model of insulin resistance (Homa-IR) was calculated as described previously. ## Liver histology After euthanasia of the mice, approximately 100 mg of fresh mouse liver from each animal was fixed in buffered formalin, paraffin embedded, sliced and stained with hematoxylin-eosin (H & E) or Picro Sirius red (PCR) (Nationwide Histology, Veradale, WA). Each slide contained 2 to 4 liver slices. Histological analysis and scoring for microsteatosis and macrosteatosis, inflammation (infiltrated leukocytes) and fibrosis were provided by two investigators: Christiane V. Löhr, Dr. Med. Vet., PhD, board-certified by the American College of Veterinary Pathologists; and K. Denise Apperson, DVM, PhD) using a modified Kleiner scoring system established for mouse models of NAFLD as described previously. Histological samples were blinded as to sex, timepoint and diet. Descriptive statistics were performed in Microsoft Excel for Mac 2011 (v 14.7.2; [www.microsoft.com](http://www.microsoft.com/)) and StatPlus for Mac (v6, [www.analystsoft.com/en](http://www.analystsoft.com/en)). Digital images were taken with a Nikon Eclipse 6 microscope and digital camera (Mpixel) and NIS-BR Elements imaging software (v21.1; [www.nikonmetrology.com](http://www.nikonmetrology.com/)). Digital images taken at 400x magnification were used for steatosis scoring (see). An effort was made to place the central vein of a lobule at one of the corners of the image so that each image covered at least one-quarter of that lobule, including all three zones of the hepatic lobule. Steatosis was objectively analyzed as the average percent surface area occupied by vacuoles using the image analysis software, ImageJ (<https://imagej.nih.gov/ij/>). Two images were taken of H&E-stained sections at 400x and the percent of affected surface area was calculated for each. The two values were then averaged. Steatosis was subjectively analyzed as the percent of affected surface area observed on H&E-stained slides at 100x (10x objective) and 400x magnifications. Vacuolization was characterized as macrovesicular, in which vacuoles displace hepatocyte nuclei, or microvesicular. Macrovesicular and microvesicular steatosis was scored separately. Severity was scored using the scale: 0 (0%), 1 (\>5% but \<33%), 2 (\>33% but \<66%), 3 (\>66%). When possible, the distribution of vacuoles was described as pericentral, midzonal, or periportal. Inflammation was defined as intralobular inflammatory foci of at least 5 leukocytes associated with disruption of hepatic plates or increased hepatocellular eosinophilia. Inflammation scores were calculated as the total number of clusters averaged over 5 fields in H&E-stained tissues examined at 100x (total number of clusters in 3.1 mm²). The following scale was used: normal 0 (\<0.5 foci), slight 1 (\<0.5–1 foci), moderate 2 (1–2 foci), severe 3 (\>2 foci). Fibrosis was also objectively quantified as percent surface area occupied by Sirius Red-stained collagen by image analysis using ImageJ. Two images were taken at 100X from the liver section of each mouse and the calculated percent areas were averaged. Fibrosis was subjectively analyzed to determine severity and distribution patterns. Distribution of fibrosis was described as perisinusoidal, periportal, pericentral, or bridging. The following scale was used: absent (0), mild (1), moderate (2), or severe (3). ## Hepatic lipids and metabolites Hepatic lipids were extracted and process for gas chromatography (GC), ultra- high-performance liquid chromatography coupled to tandem mass spectrometry (UPLC/MS/MS) and liquid chromatography coupled to tandem mass spectrometry (LC/MS/MS) as described. Fatty acid standards for the GC analysis were purchased from Nu-Chek Prep and Cayman Chemical Company. Oxylipin standards were purchased from Cayman Chemical Company. Hepatic glutathione (GSH and GSSG) and selected metabolites were fractionated and quantified using the metabolomic approach described previously, while hepatic bile acids were fractionated and quantified as described. ## RNA extraction and gene expression analysis Liver RNA was extracted using Trizol (Life Technologies), quantified and used for qRTPCR as described previously. The selection of quantified transcripts for this report was based on our previous studies. Cyclophilin was used on the reference gene and mRNA abundance was reported as mRNA abundance Delta C_T. Results in heat maps are reported as mean Delta Ct values at each diet and time point. Results in the graphed data are reported as the mean ± SEM for the Delta Ct values for each diet and time point. contains a list of all genes examined and the primers used for qRTPCR quantitation of mRNA abundance. ## Statistical analysis and data presentation All data presented in this report was assessed for statistical significance using the statistical packages in MS-Excel and MetaboAnalyst 5.0 (<http://www.metaboanalyst.ca/MetaboAnalyst/>). The number of female and male mice in each group is displayed in. ANOVA with Tukey’s HSD post-hoc test was used to identify significant differences (p-value \< 0.05) between groups and the false discovery rate (FDR). The FDR and the results of Tukey’s HSD are reported in Supplementary Data: ( All Female data PLoS1 2023**;** All Male data PLoS1 2023**)**. Volcano plots were prepared using MetaboAnalyst. Excel was used to calculate fold-change and T-test for the 1 and 40 wk comparisons between LFD- and WD-fed mice. Data is presented in graphs, heat maps and a table. Graphs were prepared using MS-Excel and the data is represented as mean ± standard error of the mean (SEM). The heat maps were prepared using the following program (<https://software.broadinstitute.org/morpheus/>). The mean value for each item within each group (time and diet) was used to prepare the heat maps. Significant difference (p \<0.05) at 1 and 40 wks between WD and LFD fed mice in the heat maps and graphed data is designated with an asterisk (_ӿ). # Results ## Food consumption and body weight Female and male mice fed the LFD and WD consumed \~2.5 grams of the purified diet per mouse per day over the 40-week feeding period. There was no evidence of WD-induced hyperphagia. The caloric density of the LFD and WD is 3.82 and 4.67 kcal/g, respectively. The cumulative food consumption, expressed as Kcal, over the 40 wk time course study is reported in. Mice fed the WD consumed more calories than mice fed the LFD leading to greater weight gain in WD-fed mice when compared to mice consuming the LFD. Significant differences in body weight between mice fed the WD and LFD fed groups were achieved after 8 wks on the purified diets. Overall, female and male mice fed the WD gained 80.2% and 76.3% more weight than age-matched female and male mice fed the LFD for 40-wks, respectively. Female and male mice were euthanized at the times indicated in after an overnight fast (Methods). Diet effects on the final body weight and liver weight, reported as Liver weight % of Body Weight (LW%BW), are illustrated in. Body weight of both female and male mice increased over time on both diets with males gaining more weight than female mice. As noted in, statistical differences in body weight (WD versus LFD) were detected after 8 wks on the WD. In, statistical differences in body weight between age-matched LFD and WD fed mice are apparent after 40 wks. We previously reported that mice fed the WD have increased liver weight when compared to mice fed the LFD. Liver weight presented as a % of body weight (LW%BW) is illustrated in. Mice fed the LFD maintain LW%BW at \~ 4% throughout the feeding studies. The dashed line in provides a comparison of LW%BW across all groups. LW%BW increased significantly to \~6% in both female and male mice fed the WD for 40 wks. The earliest change in LW%BW in response to WD in both female and male mice was after 20 wks in both female and male mice fed the WD. ## Western diet effects on plasma markers WD effects on plasma markers associated with dyslipidemia \[total and free cholesterol (T Chol, F Chol), triacylglycerides (TAG)\], non-esterified fatty acids (NEFA), β-hydroxybutyrate, (βHB)\] and glucose, insulin & insulin resistance (Homa IR) are presented in. Plasma total cholesterol (T-Chol) and free cholesterol (F-Chol) in female mice trended upward after 4 wks on the WD and increased significantly over the 40 wks of WD feeding. In male mice, T Chol was significantly increased after 1 wk and 40 wks on the WD, whereas F-Chol trended upward after 8 wks on the WD. These results are expected since the WD is moderately high in cholesterol content (0.15% w/w, and *Ldlr*^*-/-* mice have impaired LDL-cholesterol clearance from blood. While plasma TAG, NEFA and βHB changed significantly across the seven groups of mice based on the FDR values (see, **and Tables**) the pattern of change appeared random and did not follow a consistent pattern. Plasma glucose, insulin and the measure of insulin resistance, i.e., homeostatic model assessment of insulin resistance (Homa-IR) changed significantly over the 40 wks of WD feeding. These values were unaffected by the WD after 1 wk, but increased significantly after 4 wks on the WD and remained elevated throughout the WD feeding period, when compared to LFD-fed mice. These results indicate that the WD promotes dyslipidemia within 1 to 4 wks on the WD and insulin resistance within 4 wks on the WD in both female and male mice. The WD also affects plasma markers of inflammation (**)**. Tumor necrosis factor (Tnfα) was significantly higher after 1 wk on the WD in both female and male mice. After 40 wks on the WD Tnfα levels were significantly higher in female, but not male mice when compared to LFD-fed mice. We also measured plasma Toll-like receptor agonists (TLR-Ag). TLR-Ag are pathogen associated molecular patterns (PAMPs) and likely represent microbial products appearing in plasma. For example, we previous reported the presence of endotoxin, a TLR4 agonist, in plasma of WD-fed mice. Plasma TLR2-Ag trended upward after 4 wks on the WD in female and male mice. After 40 wks on the WD, TLR2-Ag were significantly higher (\~10-fold) in female, but not male mice. TLR4-Ag were significantly higher (\~3-fold) in female mice after 1 wk on the WD. While TLR4-Ag trended upward after 4 to 8 wks on the WD in female mice, plasma levels of these agonists decreased after 40 wks on the WD. The WD had little effect on plasma TLR4-Ag prior to 20 wks on the WD in male mice. After 20 and 40 wks on the WD plasma TLR4-Ag increased 2.4- and 9.8-fold, respectively. These results reveal a different timeline for the WD to alter plasma levels of TLR2-Ag and TLR4-Ag in female and male mice. While the WD effects on plasma TNFα are rapid, increasing after 1 wk on the WD, WD effects on plasma TLR2-Ag and TLR4-Ag require 4 wks on the WD to show an increase in plasma levels in female and male mice. ## WD effects on hepatic steatosis Histology is the gold standard for clinical assessment of NAFLD status. Hepatic histology was scored by two veterinary pathologists (**see** **and Figs &**). Both microsteatosis (**MIS**) and macrosteatosis (**MAS**) were scored. MAS is readily apparent at 100X magnification, while visualization of MIS required higher magnification (400x). MAS is defined as large lipid droplets in hepatocytes displacing nuclei, whereas the MIS score represents small lipid droplets that do not displace nuclei. Representative images of MIS and MAS are shown in the Supplement. MAS was not detected in livers of female and male mice fed the LFD for 1 or 40 wks, or in mice fed the WD for 1 to 4 wks. MAS, however, was observed in male and female mice as early as 8 and 20 wks on the WD, respectively. Scoring for MIS revealed changes over the 40 wk WD feeding trial in both female and male mice. Briefly, MIS decreased in livers of WD-fed mice at 4 wks and increased afterward. Whereas MAS increased after 8 wks on the WD in both female and male mice. Since lipid droplets (steatosis) contain neutral lipids, both triglycerides (TAG), cholesterol (CHOL) esters, we quantified hepatic TAG and CHOL in LFD- and WD-fed female and male mice **.** While the relative abundance of TAG and CHOL remained relatively unchanged after 40-wk on the LFD, hepatic TAG and CHOL changed significantly over 40 wks of feeding the WD in both female and male mice. In female mice, there was a progressive decrease in TAG, but not CHOL from 1 to 8 wks on the WD; afterward both lipid classes increased in liver. Total CHOL increased \~2-fold after 1 wk on the WD and this amount of CHOL does not change until 20 and 40 wks of WD feeding where CHOL increased \~5-fold when compared hepatic CHOL in 40 wk LFD-fed mice. Hepatic TAG declined from 1 to 8 wks on the WD and increased after 20 and 40 wks. After 40 wks on the WD hepatic TAG does not reach levels of TAG seen of LFD-fed female mice. Overall, this pattern of change in hepatic TAG and CHOL paralleled changes in MIS and MAS. Male mice show little change in the relative abundance of TAG, but a \~2-fold increase in CHOL over the first 8 wks of WD feeding (**).** After 20 and 40 wks on the WD, TAG increased \~1.7-fold and CHOL increased \~1.5-fold when compared to mice fed the LFD for 40 wks. Overall, this analysis revealed major differences in the type of steatosis (MIS versus MAS) and major neutral lipid classes (TAG and CHOL) during the 40-wk WD feeding period in female and male mice. Admittedly, the decline in hepatic TAG and CHOL in female and male mice after 4 wks on the WD was unexpected. We are unaware of other investigators reporting this phenomenon. This is not likely an artifact since the results were seen in both female and male mice using two different methods, i.e., histological scoring and chemical analysis of neutral lipids. We speculate these changes in hepatic neutral lipid storage may reflect changes in lipid partitioning between liver and adipose tissue over the course of WD feeding. Clearly, mice become insulin resistant by 4 wks on the WD which likely impacts lipid storage and mobilization in liver versus adipose tissue. ## Diet effects on liver injury Evidence of WD-induced liver injury is illustrated in **.** Plasma levels of alanine amino transferase (ALT) and aspartate aminotransferase (AST) increased progressively from 8 to 40 wks on the WD in female mice. Plasma levels of both enzymes were significantly increased after 40 wks on the WD. In male mice, ALT, but not AST was significantly increased by the WD after 20 and 40 wks on the WD. These findings indicate that plasma markers of liver injury change well after WD-induced increases in insulin resistance (HOMA-IR, and systemic inflammation (Tnfα), but parallel the increase hepatic MAS and neutral lipid storage after 20 wks on the WD. ## The western diet induces hepatic fibrosis Hepatic fibrosis develops as a result of liver injury and liver injury increases after 8 wks on the WD. Picro Sirius Red (PSR) staining of liver slices was used to visualize collagen fibers. PSR staining is apparent in areas surrounding major hepatic vessels, including the portal triad composed of the hepatic artery, portal vein, bile duct, and central vein. Our studies show that WD- induced hepatic fibrosis develops in the centri-lobular/sinusoidal regions (**Figs &**) by 20 wks in mice fed the WD and worsens by 40 wks. Fibrosis scoring was carried out by veterinary pathologists as described in the Methods section and presented in. Based on scoring, hepatic fibrosis increased in female and male mice after 8 wks on the WD (**Figs &).** In addition to steatosis and fibrosis, we also quantified hepatic inflammatory infiltrate to obtain the hepatic inflammation score. provides representative images of hepatic inflammation as observed in the H & E stained liver samples. We present the three histology scores (inflammation, steatosis and fibrosis) as stacked column plots to illustrate how these three parameters change over time in response to the WD. The dash line in represents the basal level of inflammatory infiltrate in female and male mice. The inflammation score increased in both female and male mice 1.8- and 1.5-fold, respectively, after 1 wk on the WD. In female mice the inflammation score increased progressively to 8 wks. Afterward the inflammation score decreased to levels seen in mice fed the LFD for 40 wks. In male mice, the elevated inflammation score after 1 wk on the WD was not sustained at 4 wks, but was increased at 8, 20 and 40 wks on the WD. After 40 wks of WD feeding, the inflammation score was increased nearly 2-fold in male mice. Both steatosis and fibrosis are prominent in male and female mice after 20 wks on the WD. In female mice, the increased hepatic inflammation score after 1 wk on the WD precedes the onset of MAS and fibrosis scores. This early inflammatory response of female mice to the WD coincides with the \~2-fold increase in plasma TNFα. In male mice the hepatic inflammation score increased in parallel with the scores for MAS and fibrosis (**)**. ## Volcano plots reveal differential effects of the WD on female and male mice A major goal of this study was to identify early markers that changed significantly in response to the WD that might play a role in NASH onset and progression. The preceding data analysis (**Figs –**) has revealed a time course for onset and progression of major NASH markers, including the onset of insulin resistance, systemic inflammation, hepatic steatosis, inflammation and fibrosis in response to the WD. One outcome from this analysis is that female and male mice appear to have a different time-line for the response to the WD, particularly with respect to WD induced systemic and hepatic inflammation. To gain additional insight into the disease process, we used a statistical approach, i.e., volcano plots, to identify markers that responded rapidly to the WD in female and male mice. Overall, we assessed 295 markers **( and Tables),** including anthropometric, plasma and hepatic markers already discussed, as well as lipidomic (fatty acids, bile acid and oxylipins) and metabolomic (glutathione and NAD) and transcriptomic markers. The volcano and pie plots (**)** represent a numerical assessment of markers responsive to feeding mice the WD for 1 wk and 40 wks. After 1 wk on the LFD and WD, 25 and 16 markers were significantly increased, while 9 and 1 markers were significantly decreased in the WD fed group when compared to the LFD fed group in female and male mice, respectively. After 40 wks on the LFD and WD, 64 and 47 markers were significantly increased in female and male mice, while 27 and 24 markers were significantly decreased in WD versus LFD fed, respectively. These results add further support to the notion that female mice are more responsive to the WD than male mice. Markers significantly induced in female mice after 1 wk on the WD included TLR agonists, hepatic fatty acids, bile acids, hepatic cholesterol, a microbial product (muramic acid, a gram-positive cell wall component) and multiple transcripts. Male mice, in contrast, had fewer markers significantly induced after 1 wk on the WD. These markers included increases in several hepatic fatty acids, a plasma lipid marker and multiple transcripts. Fewer markers were suppressed by the WD after 1 wk on the WD. In female mice, these markers included some hepatic fatty acids, the hepatic glutathione status, i.e., (reduced / oxidized (GSH/GSSG) ratio and several mRNA transcripts. In male mice, only TLR4-Ag were suppressed by the WD within 1 wk. Taken together, these results indicate that female and male mice respond differently to the WD. In **Figs –** we focus our analysis on WD-induced changes in hepatic fatty acids, oxylipins and gene expression markers of NASH. We present the data in both graphs and heat maps. The graphs were constructed using the mean ± SEM, while the heat maps were constructed using the mean value of the marker assessed. Both data presentations used all data for diet (LFD and WD) and all time points (i.e., 1, 4,8, 20 and 40 weeks \[wks\]) in the analysis. Statistical differences (p-value, p \< 0.05) between WD versus LFD at 1 and 40 wks are indicated in the graphs and heat maps by an asterisk (_**ӿ**). ## The WD rapidly alters hepatic fatty acid composition As previously reported, the hepatic fatty acid profile is significantly affected by feeding *Ldlr*^*-/-* mice the WD for several months. In this report we determined how quickly the WD affected hepatic fatty acid content. Accordingly, hepatic lipid was extracted, saponified, methylated and fractionated by gas chromatography as described in Methods. This approach allows for the identification and quantitation of fatty acids in ester linkage in complex lipids, as well non-esterified fatty acids. We present the fatty acid data as stacked column plots to illustrate how the WD affected the mole% of the 4 major fatty acid classes, i.e., saturated fatty acids (SFA), monounsaturated fatty acids (MUFA), ω3 polyunsaturated fatty acids (PUFA) and ω6 PUFA ****. The p-values for the diet effects (WD versus LFD) for SFA, MUFA, ω3 PUFA and ω6 PUFA are presented in. MUFA represent the predominant fatty acid class in liver; the rank order is MUFA \> SFA \> ω6 PUFA \> ω3 PUFA. The WD had minor effects on the overall mole% of SFA, but increased the mole% of hepatic MUFA throughout the 40 wk WD feeding period. In contrast, the mole % of both ω3 PUFA and ω6 PUFA decreased in response to the WD over the same time frame. Thus, feeding mice the WD enriched liver fat with SFA and MUFA and lowered hepatic ω3 and ω6 PUFA. This pattern of change in hepatic fatty acid class in response to the WD was seen in both female and male mice and this outcome confirmed our previous findings on WD effects on liver lipids in mice. The heat maps provide a detailed view of WD effects on the mole% of 23 hepatic fatty acids in the 4 fatty acid classes, SFA, MUFA, ω3 PUFA and ω6 PUFA. In addition, we also included mead acid (MA, 20:3, ω9) in the analysis. The WD significantly affected the hepatic content of nearly all fatty acids in livers of female and male mice over the 40 wk feeding period. Particularly relevant is the decline in the essential fatty acids (EFAs), linoleic acid (LA, 18:2, ω6) and α-linolenic acid, (ALA, 18:3, ω3) within 1 wk on the WD in both the female and male mice. The red arrows (◄) in point to LA and ALA. The significant decline in LA and ALA after 1 wk on the WD was associated with a decline in multiple, but not all, C_18-22 ω6 and ω3 PUFA derived from LA and ALA, respectively. Accompanying the decline in hepatic LA and ALA was an increase in hepatic mead acid (red arrow, ◄) **)**. Mead acid is a well-established marker of essential fatty acid deficiency (EFAD). Thus, consumption of the WD, which is low in EFAs is sufficient to decrease hepatic LA and ALA and increase MA, the EFAD marker. We next examined the effect of the WD hepatic content LA, ALA and palmitic acid (PA, C16:0) and their elongation, desaturation and peroxisomal β-oxidation (pβOX) products. These results are expressed as fatty acid μmoles/g hepatic protein (Mean ± SEM) to provide insight into the hepatic abundance of these fatty acids. Hepatic LA content decreased significantly in response to feeding the EFA-deficient WD and this decline was rapid, \~50% reduction after 1 wk on the WD (**, panel A**). Surprisingly, arachidonic acid (ARA, C20:4, ω6), a desaturation and elongation product of LA, changed little in response to the WD in both female and male mice. The end product of the ω6 PUFA pathway is the C₂₂ ω6 PUFA, docosapentaenoic acid (DPA, 22:5, ω6); and ω6 DPA increased \~3-fold in livers of female mice fed the WD for 1 wk. Hepatic ω6 DPA remained elevated throughout the 40 wk WD feeding period in female mice. In male mice, ω6 DPA was lower in both LFD and WD fed mice and hepatic levels of this fatty acid were not responsive to the WD. The relative abundance of LA: ARA: ω6 DPA in livers of female mice fed the LFD was 400:35:2.5 μmoles/g hepatic protein, respectively. Hepatic ω6 DPA is a minor hepatic fatty acid, its hepatic abundance is affected not only by diet, but also by sex. While LA and ARA are substrates for the synthesis of bioactive oxylipins, some investigators suggest ω6 DPA may improve blood lipids, i.e., LDL-C and HDL-C as well as neural inflammation in an APOE-based Alzheimer’s disease model. The level of ALA in the LFD and WD is 5.14 and 0.45 mole %, respectively. The mole ratio of ω6/ω3 PUFA in the LFD and WD is 9.25 and 11.43, respectively. The decline in hepatic ALA in response to WD feeding parallels the decline in hepatic LA. Fatty acid desaturation (Fads1, Fads2), elongation (Elovl2, Elovl5) and peroxisomal peroxidation (pβOX) converts ALA to EPA and DHA ( **panel B**). Hepatic EPA and DHA levels fall paralleling the decline in ALA. The overall decline in ALA, EPA and DHA after 8 wks on the WD was \~90%, \~80%, \~50%, respectively. In contrast to LA, ALA elongation, desaturation and peroxisomal β-oxidation to form C_20-22 PUFA was not affected by sex; both female and male mice had similar profiles of ALA, EPA and DHA in response to the WD. Since EFAs in the WD are low and we identified a significant effect of the WD on hepatic pathways for LA and ALA conversion to C_20-22 ω3 and ω6 PUFA, we examined WD effects on hepatic mead acid, an EFAD marker. Accordingly, the effect of the WD on hepatic palmitic acid (PA, 16:0), oleic acid (OA, 18:1, ω9) and MA (20:3, ω9) content over the time course of WD feeding is illustrated in. The accumulation of PA and OA in liver is due to both dietary sources and *de novo* lipogenesis (DNL), desaturation (stearoyl CoA desaturase, Scd1) and elongation (Elvol5 and/or Elovl6) to form OA. Further elongation (Elovl5) and desaturation (Fads1, Fads2) converts OA to MA. The pattern of change in PA and OA essentially paralleled the changes in MAS. Note that PA and OA in livers of female mice fed the LFD for 1 and 40 wks is 538 ± 19.4 and 600 ± 10.8 μmoles/g hepatic protein, respectively. MA is a low abundance hepatic fatty acid at 0.13 ± 0.01 and 0.06 ± 0.018 μmoles/g hepatic protein in LFD-fed female and male, respectively. MA is lower in livers of male mice when compared to female mice. In female mice, hepatic MA content increased to 0.82 ± 0.08 μmoles/g hepatic protein after 4 wks on the WD and stayed elevated throughout the duration of the feeding study. In male mice MA increased to 0.29 ± 0.015 μmoles/g hepatic protein after 4 wks of WD feeding. These results reveal a sex and diet effect on hepatic EFA and MA content. The elevation of MA in response to the WD was associated with the onset of insulin resistance and the progression of NAFLD to NASH (**Figs and**). Clearly, low dietary EFA content in the WD induces hepatic levels of the EFAD marker, mead acid. Analysis of plasma fatty acyls revealed low levels of both ω3 and ω6 PUFA in WD- fed mice, but not age-matched the LFD-fed mice. We and others have reported that high fat diets like the WD lower the hepatic abundance of essential fatty acids \[EFAs: LA & ALA. Moreover, several clinical studies report that as NAFLD severity progresses, i.e., NAFLD transition to NASH and cirrhosis, hepatic EFAs and their products decrease. A recent report indicates that key ω6 and ω3 C_18-22 PUFA are low, while mead acid (20:3, ω9) is elevated in plasma of humans with biopsy-confirmed NASH. Explanations for these observations are based on the low levels or an imbalance of EFAs in western-like diets (**)** and the impairment in hepatic desaturation and/or elongation pathways required to convert the C₁₈ EFA precursors to C_20-22 ω3 and ω6 PUFA products. A surprising outcome of our study was the major difference in how the WD affected MA content. Females have a more robust response than males to the WD in terms of the increase in hepatic MA in WD-fed mice. Differences in ω3 PUFA metabolism in male and female rodents have been reported previously. ## WD affects hepatic ω6 and ω3 PUFA-derived oxylipins The decline in hepatic LA and ALA and their desaturation and elongation products described above prompted an analysis of hepatic oxylipins derived from ω3 and ω6 PUFA. We previously reported that feeding male and female *Ldlr*^*-/-* mice the WD not only affects hepatic SFA, MUFA and PUFA composition, but also significantly alters hepatic oxylipin type and abundance. Enzymatically generated oxylipins have the potential to regulate multiple hepatic functions relevant to NASH onset and progression, such as inflammation, vascular compliance (vasoconstriction and relaxation), cell signaling, innate immunity and hepatocyte metabolism through plasma membrane and nuclear receptors. However, these regulatory oxylipins are evanescent; they are short-lived while the inactive metabolites have a longer half-life and are more readily detected. Accordingly, we assessed the effect of the WD on LA, ARA, EPA and DHA-derived oxylipins (**Figs &**). The analysis focused on the hepatic non-esterified fatty acid precursors and oxylipins, since precursor fatty acids are enzymatically excised from membrane lipids by phospholipases and serve as substrates for enzymatic conversion to oxylipins. In addition, we also quantified three non-enzymatically-derived oxylipins, i.e., 8-iso-PGF2α derived from ARA and 18-HEPE & 8-iso-PGF3α derived from EPA. The arrows (◄) in **Figs &** point to non-enzyme-generated oxylipins. These oxylipins form in membranes as a result of non-enzymatic lipid peroxidation and are excised from membranes by phospholipases and appear in the non-esterified fatty acid fraction of cells. As such they are markers of oxidative stress. During the course of these studies, we discovered that the hepatic abundance of several oxylipins differ significantly in the LFD groups fed for 1 versus 40 wks **(Figs &**). We attribute these differences to aging. As such, our analysis will focus on those oxylipins that are not significantly affected by aging in the 1 and 40 w LFD groups. ## ω6 PUFA-derived oxylipins WD effects on hepatic non-esterified LA (18:2, ω6) and ARA (20:4, ω6), as well as non-esterified oxylipins derived from LA and ARA are illustrated in. We quantified hepatic non-esterified LA and 6 oxylipins derived from LA. Hepatic levels of non-esterified LA and LA-derived oxylipins were not significantly by the WD after 1 wk, but were significantly lower after 40 wks on the WD in female and male mice. The suppressive effects of the WD on these fatty acids were apparent after 4 wks on the WD. The affected LA derived oxylipins include: 9(S)-HODE, 13(S)-HODE, 9,10-EpOME, 9,10-DiHOME and 12,13-DiHOME). Changes in hepatic levels of these oxylipins paralleled the decline in non-esterified LA. The WD suppressed hepatic 9(S)-HODE abundance by 40 to 60% after 4 wks on the WD and this effect persisted to 40 wks on the WD in both female and male mice ****. 9(S)-HODE and 13(S)-HODE are bioactive lipoxygenase products, while 9,10- and 12,13-EpOME are bioactive epoxygenase (Cyp2c) products. These oxylipins are ligands for PPAR nuclear receptors and plasma membrane G-protein receptors. Ligand mediated receptor activation alters chemokine and interleukin (IL1β) production as well as neutrophil respiratory burst, mitochondrial function and cell survival. The epoxy-fatty acids (9,10-EpOME and 12,13-EpOME) are converted to inactive dihydroxy fatty acids (DiHOME) by a soluble epoxide hydrolase (Ephx2). Circulating levels of oxidized LA metabolites, like 9(S)-HODE have been reported to impair mitochondrial function and promote apoptosis and NLRP3 activation. In contrast, our study shows that NAFLD progression to NASH was associated with a decline in hepatic LA-derived oxylipins. The results reported in confirmed our previous studies using female and male mice. While total hepatic ARA (20:4, ω6) levels were not significantly affected by the WD, non-esterified ARA showed considerable variability in hepatic abundance over the 40 wk feeding period in both female and male mice. Twenty-two oxylipins derived from ARA were quantified. Nine oxylipins are products of the prostaglandin pathway, either as primary products, like PGE2 or as inactive metabolites, like 6-keto-PGF1α, an oxylipin derived from the prostacyclin, PGI₂. The graph for PGE2 reveals a major decrease in hepatic PGE2 after feeding female and male mice the WD for 4 wks and this WD-mediated suppression persisted for the duration of WD feeding. Other prostaglandins and thromboxane B2 did not reveal a differential response to the WD versus LFD. Hepatic leukotriene B4, a lipoxygenase product, decreased in hepatic content after 1 wk on the WD and this effect persisted to 20 and 40 wks in female and male mice, respectively. Leukotriene B4 regulates multiple leukocyte functions, such as aggregation, release of lysosomal enzymes and nitric oxide production. Other lipoxygenase products include the hydroxy-eicosatrienoic acids (HETE). Hepatic levels of several HETEs were affected by the WD, including 5(S)-HETE, (8(R)-HETE, 8(S)-HETE, 12(S)-HETE & 15(S)-HETE). The graph for 5(S)-HETE in reveals the effect of the WD on this oxylipin. The WD increased 5(S)-HETE \~2-fold after 1 wk on the WD. In female mice, the response of 5(S)HETE to the WD was robust, but biphasic, whereas in male mice the WD response was sustained for the 40 wk feeding period. Other HETEs, e.g., 8(S)-HETE showed opposite responses to the WD in female and male mice suggesting the influence of sex on the hepatic regulation of HETEs. The epoxy-eicosatrienoic acids (EET) are Cyp2C products, while the dihydroxy-eicosatrienoic acids (DiHET) are derived from the EET by the soluble epoxide hydrolase, Ephx2. There was no consistent response of these oxylipins to the WD in either female or male mice. Similarity, there was no effect of the WD on the lipid peroxidation marker, i.e., isoprostane, 8-iso PGF2α (see the red arrow (◄) in). The outcome of the ω6 PUFA-derived oxylipin analysis revealed major effects of the WD on the generation of LA-derived lipoxygenase and epoxygenase products. The WD effect on these products paralleled WD effects on hepatic total and non- esterified LA content (**Figs and**). WD effects on ARA-derived oxylipins, however, were less robust (**Figs &**). *ω3 PUFA-Derived Oxylipins*. Analysis of WD effects on hepatic oxylipins derived from EPA and DHA are reported in. The heat maps include non-esterified EPA (C20:5, ω3) and DHA (C22:6, ω3) as well as twelve oxylipins derived from EPA and 14 oxylipins derived from DHA. Most of the EPA- and DHA-derived oxylipins were products of epoxygenases (Cyp2C and Cyp2J) and an epoxide hydroxylase (Ephx2). Leukotriene B5 is a lipoxygenase product and is implicated in immune modulation. 18-HEPE and 8-iso PGF3α are lipid peroxidation products; see (◄) in the heat maps. As mention in the introduction to the analysis of hepatic oxylipins, there appears to be a significant aging effect in both female and male mice on the hepatic abundance of ω3 PUFA-derived oxylipins. As such, our analysis will focus on oxylipins that have similar levels of hepatic abundance in mice fed the LFD for 1 and 40 wks. Non-esterified EPA was significantly increased in both female and male mice after 1 wk on the WD. Since there is little ALA and no EPA in the WD, we suggest this increase is due to the induction of membrane remodeling and the release of EPA form membrane lipids. The only EPA-derived oxylipin that changed after 1 wk on the WD in was the non-enzymatically derived 18-HEPE. This rapid effect was evident in female, but not male mice. After 40 wks on the WD, EPA, 17,18-DiHETE, 18 HEPE decreased in livers of female mice. While 18-HEPE is reported to be a precursor of pro-resolvin (RvE 1–3), we detected no significant effect of the WD on hepatic RVEe1 abundance in female or male mice. In male mice, EPA and leukotriene B5 decreased while 14,15 DiHETE and the non-enzymatically generated lipid oxidation product, 8-iso PGF3α (◄) increased. Overall, these results indicate that hepatic non-esterified EPA and multiple Cyp2C/Ephx2 products decreased significantly in female and, male after 4 wks on the WD. Moreover, there was also a decline in many of these oxylipins after 40 wks on the LFD, making interpretation of these effects challenging. DHA is the predominant ω3 PUFA accumulating in liver. Of the 16 DHA-derived oxylipins examined, 9 and 8 DHA-derived oxylipins were significantly affected by the WD in female and male mice, respectively. In contrast to the EPA-derived oxylipins, many of the DHA-derived oxylipins in female and male mice were less affected by aging. We illustrate the effect of the WD on the following DHA- derived oxylipins, 10,11-EpDPA, 19,20-EpDPA and 10,20-DiHDPA in. The hepatic level of the epoxy- and dihydroxy-oxylipins was decreased after 4 wks on the WD. Also, no DHA-derived resolvin or protectin was significantly affected by the WD. Overall, the decline in most DHA-derived oxylipins paralleled the decline in non-esterified DHA. The key outcomes of this analysis are that hepatic levels LA- and DHA-derived oxylipins are suppressed after 4 wks on the WD and this suppression was sustained for 40 wks while on the WD. There is also evidence of aging effects on hepatic levels of oxylipins since levels of several oxylipins from mice fed the LFD for 40 wks were lower than oxylipins after 1 wk on the LFD. The WD-regulated LA-derived oxylipins include epoxygenase, epoxide hydrolase and lipoxygenase products, while the DHA-derived oxylipins are epoxygenase and epoxide hydrolase products. While several oxylipins have the potential to play a regulatory role in hepatic function, the changes described above are not consistent with a role in controlling plasma TNFα or the hepatic inflammation score; both increased significantly after 1 wk on the WD. Instead, changes in LA- and DHA-derived oxylipins may impact later events, such as the accumulation of lipid (steatosis) as well as fibrosis that appear in livers after 8–20 wks on the WD (**Figs, &** ). ## WD effects on hepatic gene expression To gain insight into potential mechanisms controlling hepatic lipid levels and various markers associated with NASH, we examined WD effects on hepatic gene expression. The transcripts examined by qRTPCR in this report were based on transcripts quantified in our previous studies. contains DNA sequence data for all primers used for the qRTPCR analysis reported herein. The reference gene for the qRTPCR was cyclophilin. Results in heat maps are the mean value for mRNA abundance Delta C_T, while results presented in graphs are the mean Delta CT value ± SEM. ## Expression of proteins involved in hepatic lipid metabolism We first examined diet effects on hepatic expression of genes involved in lipid metabolism. Of the 33 transcripts examined, 19 and 15 transcripts were significantly affected by the WD in female and male mice, respectively. Long chain acyl CoA synthetase-long (Acsl) plays a key role in the synthesis of fatty acyl CoAs. In livers of female mice, the abundance of transcripts encoding Acsl1, 3 and 4 decreased rapidly (within 1 wk) in response to the WD. In male mice, the WD-mediated decline in acyl CoA synthetase-long chain (Acsl) mRNA abundance was apparent after 4 wks on the WD. Lysophosphatidylcholine acyl transferase (LpCat) subtypes transfer fatty acyl chains to lysophospholipids and likely play a role in membrane remodeling. All three LpCat subtypes (LpCat1, LpCat2 and LpCat4) were expressed at low levels in livers of LFD fed mice and feeding mice the WD increased hepatic mRNA abundance of all 3 LpCats in both female and male mice. The WD effect on LpCat2 (◄) is illustrated in. The hepatic abundance of LpCat2 mRNA trended upward after 1 wk on the WD and showed a progressive increase over the 40 wks on the WD. After 40 wks on the WD, LpCat2 was induced 4.2- and 2.6-fold in female and male mice, respectively. In contrast to the LpCats, expression of peroxisome proliferator activated receptor (PPAR) target genes, i.e., acyl CoA thioesterase-1 (Acot1), Cyp4A10 and acyl CoA oxidase (Aox) declined after 4 wks on the WD and stayed low for the duration of the 40 wk WD feeding study. PPARs are regulated by both fatty acids and oxylipins. As such, the WD-mediated decline in hepatic C_18-22 ω3 and ω6 PUFA and oxylipins derived from these fatty acids (**Figs –**) may account for the decline in expression of these transcripts. The effect of the WD on a representative PPAR target gene (Aox, ◄) is illustrated in. Hepatic Aox mRNA abundance trended downward as early as 1 wk on the WD and showed a significant (\> 60%) decrease when compared to the LFD group after 40 wks on the WD. Hepatic abundance of transcripts encoding PPARα, PPARβ, PPARγ1, PPARγ2 and PPARγ co-activators (Pgc1α andPgc1β) were affected by diet, but the pattern of WD-induced change did not parallel the changes in Acot1, Cyp4A10 or AOX. This outcome suggests that changes in hepatic fatty acids and/or oxylipins levels, rather than changes in PPAR/co-activator expression, may account for the changes in hepatic mRNA abundance of the PPAR-target genes. Sterol regulatory element binding protein-1c (Srebp1c) is a key transcription factor involved in regulating the expression of multiple transcripts encoding proteins involved in *de novo* lipogenesis \[fatty acid synthase (Fasn), fatty acid desaturation, (Fads1, Scd1) and fatty acid elongation (Elovl3)\]. The mRNAs encoding these proteins increased in parallel after 4 wks of WD feeding. A representative transcript for this group (Scd1, ◄) is plotted in. After 4 wks on the WD hepatic Scd1 mRNA was increased 6- to 8-fold in livers of female and male mice and this transcript remained high for 40 wks on the WD. However, Scd1 mRNA increased in LFD-fed male mice after 40 wks revealing an aging effect on Scd1 expression. The other Srebp1c responsive genes (Fasn, Elovl3) were also induced in response to the WD. The time frame for the induction of these transcripts, i.e., from 4 to 40 wks of WD feeding, parallels the time course for WD-induced insulin resistance and the decrease in hepatic C_18-22 ω3 PUFA. Srebp1c is a well-established target for fatty acid regulation. C_20-22 ω3 PUFA suppress Srebp1 expression by affecting the transcription of the Srebp1 gene, Srebp1 mRNA stability and Srebp1c nuclear protein content. The WD-mediated depletion of hepatic C_18-22 ω3 and ω6 PUFA is consistent with the induction of Srebp1c expression, as well as the expression of its target genes (Fasn, Scl1, Elovl3). Diacylglycerol acyl transferases (Dgat1, Dgat2) are key enzymes involved in the terminal stage of TAG synthesis. Hepatic Dgat1 mRNA abundance was rapidly decreased in female mice, but slowly decreased in male mice, in response to the WD. Dgat2 mRNA abundance was not affected by the WD in female mice, but was significantly decreased by the WD after 1 wk in male mice. Hepatic mRNA abundance of two TAG hydrolases (Tgh, Ces1g) decreased in response to the WD and were also affected by aging. While the mRNA encoding perlipin 2 (Plin2), a key lipid droplet protein, was suppressed rapidly (within 1 wk) by the WD in livers of female mice; Plin2 mRNA abundance was significantly suppressed after 40 wks. In male mice, Plin2 was significantly decreased after 40 wks on the WD. Hydroxymethylglutaryl-CoA reductase (HmgCoA Red) and hydroxymethylglutaryl-CoA synthase 1 (HmgCoA Syn1) are key enzymes involved in cholesterol synthesis. Hepatic abundance of these mRNAs was rapidly reduced by the WD in female mice. Expression of these enzymes was not significantly affected by the WD in male mice. Sterol-O-acyltransferases (Soat1, Soat2) are involved in cholesterol ester synthesis. Soat1 mRNA abundance was significantly different after 1 wk on the WD in male mice, while Soat2 mRNA abundance was significantly different after 40 wks in female mice. Finally, Cyp7A1 is a key enzyme involved bile acid synthesis. Its mRNA abundance was not significantly affected by the WD in female and male mice. The outcome of this analysis revealed significant effects of the WD on the expression of multiple genes involved in membrane lipid remodeling (LpCat1, LpDat3, LpCat4), PPAR signaling (Acot1, Aox & Cyp4A10 and Srebp1 signaling (Srebp1, Fasn, Scd1). Some transcripts do not respond similarly to the WD in female and male mice, e.g., HmgCoARed, HmgCoASyn1, Soat1 and Soat2. The lack of consistent effects of WD in male and female mice suggests that changes in the expression of these genes may not play a major role in WD-induced liver pathology. ## Expression of enzymes involved in oxylipin metabolism We previously reported that the WD consistently affected the hepatic expression of prostaglandin synthease-2 (Ptgs2), but had little effect on other enzymes involved in hepatic oxylipin metabolism. Herein, we re-examined the effect of the WD on the expression of these enzymes to determine if sex and duration of WD feeding affected their expression. The WD significantly induced Ptgs1 expression (\~2-fold) in both female and male mice after 1 wk. In female mice, but not male mice, Ptgs1 mRNA abundance increased progressively over the 40 wks feeding period. Ptgs2 mRNA was significantly increased (\~4-fold) after 1 wk of WD feeding in female, but not male mice. However, Ptgs2 mRNA abundance increased progressively in both female and male mice over the 40 wk WD feeding period. Ptgs1 and Ptgs2 are not expressed in hepatocytes, but expressed in other hepatic cells, including cholangiocytes, endothelial, stellate and Kupffer cells as well neutrophils (<https://www.proteinatlas.org/>. The rapid effects of the WD on Ptgs1 and Ptgs2 expression suggests that cells expressing these enzymes are early targets of WD effects on the liver. Despite the rapid effects of WD on Ptgs1 and 2 mRNA abundance, rapid and sustained changes in hepatic Pge2 were not observed. Instead, the hepatic abundance of the inactive prostaglandin metabolite, e.g., 15-keto PGE2 (**, ◄**) increased after 4 wks on the WD. This finding suggests WD-induced PGE2 synthesis, but also the rapid conversion of PGE2 to an inactive metabolite, 15-keto-PGE2. The hepatic abundance of mRNAs encoding epoxygenases (Cyp2c and Cyp2J5) and epoxide hydrolases (Ephx1 and Ephx2) were also affected by the WD, but the response to the WD did not reveal a consistent and significant change over time. The overall outcome of the oxylipin analysis suggests that the WD significantly affects the hepatic abundance of oxylipins derived from LA, ARA and DHA after 1 wk on the WD. This timeline for the WD-induced change in oxylipins coupled with the rapid effects on Ptgs1 and Ptgs2 expression, suggests that LA-, ARA- and DHA-derived oxylipins may be involved in the early regulation of systemic and hepatic inflammation after 1 wk on the WD (**Figs and**). ## Expression of proteins involved in hepatic inflammation Of the 28 transcripts associated with hepatic inflammation examined in **Figs and**, 19 and 22 transcripts were significantly affected by the WD in female and male mice, respectively. Data for several of these transcripts, designated by an arrow in the heat maps (◄), was graphed in to illustrate the effect of sex and time on the response of these transcripts to the WD. Monocyte chemoattractant protein-1 (Mcp1) is expressed in hepatic cholangiocytes, endothelial cells and neutrophils (<https://www.proteinatlas.org/>) and this gene is an early responder to factors promoting inflammation. Hepatic expression of Mcp1 is low in LFD fed mice, but the WD increased Mcp1 mRNA significantly after 1 wk. After 40 wks on the WD, Mcp1 was induced 18-fold in female and male mice. In contrast to male mice, female mice have a more robust induction of Mcp1 after 1, 4 and 8 wks on the WD. CD40 is a member of the TNF receptor super gene family and is expressed in multiple hepatic cells. Hepatic CD40 is expressed at levels lower in male mice than in female mice. CD40 mRNA increased in livers of female and male mice by \~50% after 1 wk on the WD. Overall, CD40 mRNA increased 3-fold after 40 wks on the WD. CD68 is a macrophage marker and its expression was higher in LFD-fed female mice than male mice. After 40 wks on the WD, CD68 was induced 5.7- and 3.1-fold in female and male mice, respectively. F4/80 is a G-protein receptor and a macrophage marker. Its expression in LFD-fed mice was lower in male mice than female mice. F4/80 mRNA was significantly induced in male, but not female mice after 1 wk on the WD. After 40 wks on the WD, F4/80 mRNA levels were increased 3.7- and 2.2-fold in female and male mice respectively. NLR family pyrin domain containing 3 (Nlrp3) is a key protein involved in inflammasome organization and is expressed in hepatic cholangiocytes, endothelial and Kupffer cells and neutrophils. Expression of Nlrp3 was rapidly induced (2.4-fold) by the WD within 1 wk on the WD feeding in female but not male mice. The overall induction of Nlrp3 after 40 wks on the WD was 2.5-fold in both female and male mice. Tumor necrosis factor α (Tnfα) plays a major role in systemic and hepatic inflammation and in the progression of NAFLD to NASH. In liver, Tnfα is expressed in cholangiocytes, Kupffer and T-cells. Hepatic expression of Tnfα was rapidly induced (3.7-fold) in livers of female, but not male mice after 1 wk on the WD. After 40 wks on the WD, Tnfα mRNA was increased 5.8- and 3.3-fold in female and male mice, respectively. Glycoprotein nmb (Gpnmb) is another macrophage product, but also expressed in multiple hepatic cells. Gpnmb is a blood biomarker for NAFLD; and some suggests Gpnmb may have beneficial effects on NAFLD. Gpnmb expression in livers of LFD fed female and male mice was very low. Hepatic abundance of Gpnmb rapidly increased \~7-fold in female, but not male mice in response to 1 wk of WD feeding. The mRNA encoding Gpnmb in female and male mice was induced 32- and 17-fold, respectively, after 40 wks on the WD. Finally, the triggering receptor expressed on myeloid cells (Trem2), like Gpnmb, is another circulating marker of NAFLD. Trem2 plays a protective role in liver by modulating hepatic TLR4 action. In LFD-fed mice hepatic Trem2 is expressed at low levels. After 1 wk on the WD, Trem2 was induced 4.2- and 2.1-fold in livers of female and male mice, respectively. After 40 wks on the WD, Trem2 mRNA was increased 12.4- and 8.6-fold in female and male mice, respectively. The outcome of this analysis has established that 1 wk of WD feeding was sufficient to significantly induced multiple mRNAs linked to hepatic inflammation (**Figs &)**. With the exception of F4/80, levels of transcript induction were greater in female mice than male mice. Moreover, after 40 wks on the WD, female mice had higher levels of transcript abundance than male mice. These results indicate that female mice have a more rapid and robust response to WD feeding than male mice. More important, the onset of change in expression of many of these transcripts precedes any evidence of obesity, insulin resistance, histological evidence of NASH, i.e., steatosis and fibrosis. Instead, the change in expression of these transcripts is inversely associated with the timeline for the WD-mediated decline in hepatic EFAs (**Figs –**). ## WD effects on oxidative stress, apoptosis, fibrosis, extracellular matrix remodeling and cell growth ### Oxidative stress Oxidative stress plays a key role in the onset and progression NAFLD to NASH, cirrhosis and HCC. We quantified transcripts encoding three proteins linked to cellular oxidative stress, including cytochrome B-245 β chain (Cybβ) a component of the NADPH oxidase complex, glutathione S-transferase α1 (Gstα1), and heme oxygenase 1 (Hmox1). All three transcripts trended upward after 4 and 20 wks on the WD in female and male mice, respectively. We also quantified hepatic reduced (GSH) and oxidized glutathione (GSSG) and the GSH/GSSG ratio in the heat map and the. Changes in the GSH/GSSG ratio reflect changes in oxidative stress status within cells where a low GSH/GSSG ratio indicates increased oxidative stress and/or decreased recycling of GSSG to GSH. As illustrated in, LFD-fed female and male mice have a low GSH/GSSG ratio at both 1 and 40 wks. This is expected since the mice were fasted overnight prior to euthanasia (Methods) and fasting is associated with a decline in tissue GSH levels. One wk on the WD does not change this ratio, but afterward, this ratio revealed a biphasic response to the WD. The GSH/GSSG ratio was increased by 4 wks followed by a decline beginning after 20 wks on the WD. Interestingly, this change in the GSH/GSSG ratio paralleled the changes in Homa-IR and MAS (**Figs, &)**. Finding GSH/GSSG values above values measured in LFD fed mice suggests that livers of WD-fed mice likely have sufficient capacity to maintain GSH to prevent robust oxidative stress damage. This finding may explain why we did not find increased formation of the lipid peroxidation markers, derived from LA or EPA, i.e., 8-iso PGE2 and 8-iso-PGE3 (**Figs &**). ### Apoptosis and autophagy We next examined the expression of caspase 1 (Casp1) and two cathepsins (CtsB and CtsS). Casp1 is involved in apoptosis, while the cathepsins are lysosomal proteases involved in autophagy and the turnover of intra- and extra-cellular proteins. Both Casp1 and the cathepsins play a role in NASH pathology. All 3 markers were significantly induced after 40 wks on the WD. We also plotted the qRTPCR data for CtsS. Interestingly, CtsS expression was inversely associated with the hepatic GSH:GSSG ratio. GtsS expression is highest when the hepatic GSH:GSSG ratio is low. This result may reflect a role of oxidative stress in the regulation of CtsS expression. ## Fibrosis and Extracellular Matrix Remodeling (ECMR) Hepatic fibrosis is the hallmark of significant hepatic injury; and injury promoted scarring involves extracellular matrix remodeling (ECMR). As shown in the heat maps and graphs collagen 1A1 (Col1A1), lysyl oxidase (LysOx), matrix metalloproteases (Mmp12, Mmp13) and tissue inhibitor of metalloprotease (Timp1) are well induced after 40 wks on the WD. A closer look at the heat maps suggest that the expression of some of the fibrosis and ECMR markers were affected by the WD after 1 wk on the WD. Moreover, there is a trend for a more rapid increase in these markers in response to the WD in female mice than male mice. Accordingly, we assessed these differences in **,** where we graphed the qRTPCR data for Col1A1, Tgfβ1, Mmp12 and Timp1. Col1A1 is expressed in hepatic stellate cells. In whole liver, Col1A1 is expressed at low levels in LFD-fed mice. Its mRNA increased \~2-fold after 1 wk on the WD in female and male mice. The overall effect of the WD on Col1A1 abundance between female and male mice at 40 wks on the WD versus LFD was 19.6- and 6.5-fold, respectively. The elevated expression of Col1A1 at 20 and 40 wks of WD feeding correlated with the appearance of PSR staining for fibrosis (**Figs &**). Tgfβ1 mRNA displayed a different response to the WD. Tgfβ subtypes are major regulators of hepatic Col1A1 expression and fibrosis. Tgfβ1 is expressed in multiple hepatic cells, but not hepatocytes. The transcript encoding Tgfβ1 was induced 2-fold in female, but not in male mice after 1 wk on the WD. After 40 wks on the WD, Tgfβ1 mRNA was increased 2.7- and 2.1-fold in female mice and male mice, respectively. The induction of hepatic fibrosis is associated with an increase in expression of enzymes involved in ECMR. We examined the effect of the WD on two transcripts encoding proteins involved in ECMR, i.e., matrix metalloprotease-12 (Mmp12) and tissue inhibitor of metalloproteases-1 (Timp1). Like Col1A1, hepatic expression of Mmp12 and Timp1 is very low in livers of LFD-fed mice. Mmp12 was induced 4- and 2.3-fold after 1 wk on the WD in female and male mice, respectively. After 40 wks on the WD, Mmp12 was induced 73- and 25-fold in female and male mice, respectively. Timp1 was induced 2.5- and 1.8-fold in female and male mice after 1 wk on the WD and after 40 wks on the WD, Timp1 was induced 23 and 12-fold, respectively. The outcome of this analysis revealed an apparent coordinate induction of expression of proteins involved in fibrosis and ECMR over the 40 wk WD feeding study. These results also reveal significant changes in hepatic expression of Col1A1, Mmp12 and Timp1 well before the appearance of histological evidence of hepatic fibrosis. ## Hepatic cell growth and differentiation NASH can progress to cirrhosis, hepatocellular carcinoma and liver failure. This transition is associated with significant changes in hepatic cell survival, cell type and cell signaling through the activation of pathways like notch and hedgehog signaling. We previously reported that several gene expression markers linked to these pathways were regulated by the WD in our *Ldlr*^*-/-* mouse model. The heat map and graphs **(Figs &)** describe the time course for WD induced changes in hepatic expression of the following proteins in livers of male and female mice; including: 1) B-cell lymphoma 2 (Bcl2), 2) growth arrest and DNA damage (Gadd45), 3) Notch signaling proteins Hey1, HeyL, 4) Indian hedgehog (Ihh) and hedgehog interacting protein (Hhip), 5) SRY-box containing gene 4 (Sox4), a transcription factor and 5) the transcriptional repressor (Snai1/Snail1). While the WD significantly affected the hepatic expression of most of these markers ( **and Figs**), only Bcl2, Hey1, HeyL, Hhip and Sox4 were significantly changed after 1 and/or 40 wks in female and/or male mice fed the WD. To understand how these proteins function in liver biology, we used the human protein atlas (<https://www.proteinatlas.org/>) to determine which hepatic cells expressed these proteins. Accordingly, hepatic Bcl2 is expressed in cholangiocytes (bile duct), vascular endothelial, stellate, Kupffer and T-cells. Bcl2 regulates apoptosis and cell survival by its effects on mitochondria. Hey1 and HeyL are downstream mediators of Notch signaling. These transcription factors play key roles in vascular development, as well as extracellular matrix remodeling (ECMR) and fibrosis. These proteins are expressed in hepatic endothelial cells, particularly cholangiocytes and vascular endothelial cells. In addition, Hey1 is expressed in hepatic neutrophils, while HeyL is expressed in hepatic stellate cells. The hedgehog markers, Indian hedgehog (Ihh) and hedgehog (Hh) interacting protein (Hhip), are expressed in hepatocytes and stellate cells, respectively. Hedgehog (Hh) signaling is required for liver regeneration, regulation of capillarization, control of the fates of hepatic stellate cells, and promoting liver fibrosis and liver cancers. Hhip interferes with the capacity of Hh ligands to mediate changes in cell function. Sox4 is expressed in hepatic cholangiocytes and vascular endothelial cells. This protein plays a role in multiple cell development pathways, as well as triglyceride metabolism and liver steatosis. Hepatic expression of Bcl2 was significantly increased in liver after 40 wks on the WD. In female and male mice, Bcl2 mRNA trended higher after 4 and 20 wks, respectively, on the WD. While both Hey1 and HeyL were significantly affected by the WD after 40 wks on the WD, only HeyL was significantly increased after 1 wk on the WD in both female and male mice. Ihh is a hedgehog signaling ligand and its mRNA abundance trended upward after 4 wks on the WD in male mice and 8 wks in female mice. Ihh was significantly increased after 40 wks on the WD in both female and male mice fed the WD. Interestingly, hepatic hedgehog interacting protein (Hhip) displayed the opposite response to the WD. Hepatic Hhip mRNA abundance decreased after 4 and 20 wks on the WD in females and males, respectively. In both groups of mice there was a progressive decline in hepatic Hhip mRNA in response to the WD. If cellular Ihh and Hhip protein abundance parallel changes in Ihh and Hhip mRNA, then the inhibitory capacity of Hhip on Hh signaling would be expected to decrease more rapidly in livers of female mice than male mice in response to WD feeding. In female and male mice, Sox4 mRNA trended upward after 4 wks and 20 wks on the WD in female and male mice, respectively. The outcome of this analysis suggests that activation notch signaling (HeyL) is an early event in WD induced NASH, while hedgehog (Ihh and Hhip) and Sox4 signaling are activated after 4–20 wks on the WD. Also, there is a different timeline for the WD regulation of Bcl2, HeyL, Ihh and Sox4 in female and male mice. As such, this analysis reveals a sex- and time-dependent effects of the WD on the regulation of these pathways. Most notable is the rapid effect of the WD on the induction of HeyL mRNA, an event that precedes the appearance of histological evidence of NAFLD and NASH (**Figs, &**). # Discussion The goal of this study was to establish the time course for western diet (WD)-induced NASH in the preclinical *Ldlr*^*-/-* mouse model to assess changes in anthropometric, plasma and hepatic markers of NASH. The analysis included histology, lipid (cholesterol, triglyceride, the fatty acid and oxylipin profile)\] and gene expression assessments. Specifically, our goal was to: 1) identify early markers of WD-induced disease; 2) determine which lipids might serve as biomarkers of NASH; and 3) determine if female and male mice responded similarly to the WD. Our study identified early changes in systemic and hepatic inflammation markers that preceded overt evidence of NAFLD, i.e., obesity, insulin resistance, hepatic steatosis and fibrosis in age-matched female and male *Ldlr*^*-/-* mice. Our study also established that female and male mice do not have identical responses to the WD. Female mice are more responsive to the WD than male mice. This study was carried out concurrently with age-matched female and male mice. The advantages of this study design are: 1) the approach avoids confounding outcomes due to differences in food batch composition, animal age or animal batch from the supplier, personnel handling the animals or vivarium conditions; 2) the approach reveals both similarities and differences in how age-matched female and male mice responded to the WD; 3) the longitudinal design of the study provides insight into the early NADLD markers and the changes in these markers over time 3) the approach increases the power of the analysis by increasing the number of animals in the study and 4) increases confidence in the outcomes when common markers are found to be affected by the WD in both female and male mice. The prevailing view of diet induced NAFLD and its progression to NASH involves the accumulation of lipid (cholesterol, triglycerides, diacylglycerides and ceramides) in liver that promotes hepatic inflammation leading to hepatic injury, inflammation, oxidative stress and fibrosis. The excessive accumulation of hepatic lipid is postulated to be the 1^st hit is a multi-hit pathway leading to NASH. Absent from this model are the antecedent events preceding the abnormal accumulation of hepatic lipid. As mentioned earlier, the NIDDK considers NAFLD to be a “silent disease” with no obvious disease indicators. NAFLD is typically diagnosed in patients with pre-existing diseases, such as obesity, type 2 diabetes, dyslipidemia and MetS. Our study fills this gap and presents an alternative view of the time course for WD-induced NAFLD and its progression to NASH. We list 14 markers of NAFLD/NASH that were quantified in this study. Most markers changed in response to the WD prior to histological evidence of macrosteatosis (MAS), which was apparent after 20 and 8 wks on the WD in female and male mice, respectively. While some of these changes were small, they were the beginning of a long and progressive change over time. For example, histological evidence of fibrosis was first detected after 8 wks on the WD in both female and male mice (**Figs, &).** However, gene expression analysis revealed a small, but significant increase in hepatic mRNAs encoding hepatic proteins involved in fibrosis (Col1A1, Tgfβ1) and ECMR (Mmp12 and Timp1) after 1 wk on the WD. Plasma markers of dyslipidemia (plasma cholesterol), insulin resistance (Homa-IR), and systemic and hepatic inflammation (Tnfα) were significantly increased prior to hepatosteatosis (**, &**) and obesity. Hepatic cholesterol, the new villain in the NAFLD landscape was elevated after 1 wk on the WD in female and male mice and significantly increased after 40 wks on the WD. Metabolic endotoxemia (plasma TLR4-Ag) was elevated in female and male mice after 1 to 4 wks on the WD feeding. This finding indicates rapid changes in the gut-liver axis in response to the WD. Metabolic endotoxemia, i.e., plasma TLR4 agonist and elevated hepatic cholesterol are key factors in the onset and progression of NASH. Additional markers changed in both male and female mice prior to the appearance of MAS (**Figs, and**) including the onset of insulin resistance (Homa-IR), changes in hepatic oxidative stress status (GSH/GSSG) and a significant difference in body weight between LFD and WD-fed mice at 8 wks and LW%BW at 20 wks. Taken together, these findings identify multiple early markers of NASH that precede histological evidence of the NAFLD and NASH. The early plasma markers (Tnfα, TLR4-Ag) may serve as targets for early clinical assessment and therapeutic intervention. Another outcome of our analysis was that female and male mice respond differently to the WD. This is particularly evident in the response of female and male mice after 1 wk on the WD. Female mice have a more robust inflammatory response to the WD when compared to male mice ( and **Figs &**). The relevance of this finding to human NASH, however, is unclear. Hormonal status is known to affect the onset of NAFLD in humans. Premenopausal women are less likely to develop NASH and fibrosis than children, men, and post-menopausal women. Nevertheless, this difference in response of female and male mice to the WD may be relevant in early diagnosis of NAFLD. The physiological basis for this differential response to the WD will require further investigation. In addition to measuring plasma TNFα, clinical assessments of plasma Gpnmb and Trem2 might also serve as early indicators of hepatic and/or systemic inflammation. Our third goal was to determine whether there was a temporal link between the onset of NASH and changes in hepatic EFAs \[linoleic acid (LA, 18:2, ω6 and α-linolenic acid (ALA, 18:3 ω3)\], and their oxylipin metabolites. As indicated earlier, hepatic levels of ω3 and ω6 PUFA in human liver decrease as the severity of hepatosteatosis progresses to NASH and cirrhosis. In humans, the decline in hepatic C_18-22 ω3 and ω6 PUFA is likely due to effects on hepatic PUFA metabolism or changes in hepatic cellular composition. In mice, however, the decline in hepatic C_18-22 ω3 and ω6 PUFA in response to the WD is associated with low blood levels of C_18-22 ω3 and ω6 PUFA reflecting EFA deficiency. EFA deficiency in the *Ldlr*^*-/-* mouse/WD preclinical model is due, at least in part, to the low level of LA and ALA in the WD. Clearly, WD feeding promotes a rapid loss of hepatic EFAs with significant decline in LA and DHA-derived oxylipins (**Figs – &).** While one week on the WD was sufficient to significantly lower LA and ALA in livers of female and male mice (**Figs &**), 4 weeks of WD feeding was required to observe a major decline in LA and DHA-derived oxylipins (**Figs &**). The decline in hepatic LA- and DHA-derived oxylipins occurred as insulin resistance (Homa IR) and the GSH/GSSG ratio increased (**Figs, &**). Whether this observation is a cause & effect relationship or merely coincidence will require additional studies. Our data, however, clearly show that WD suppression of hepatic ω3 and ω6 PUFA and oxylipin content precedes WD-mediated induction of the major hepatic events linked to NASH, such as MAS and fibrosis. In further support of the role EFA deficiency plays in WD NASH, we quantified hepatic levels of mead acid (C20:3, ω9), an EFA deficiency marker. The substrate for the synthesis of mead acid is oleic acid C18:1, ω9; and this fatty acid is elongated and desaturated to form 20:3, ω9. The accumulation of mead acid is likely due to three mechanisms. First, there is increased hepatic MUFA content arising from high MUFA content in the WD (**Figs &**). Second, the WD increases *de novo* lipogenesis increasing the production of palmitate and its subsequent elongation (Elovl3/Elovl5) and desaturation (Scd1) to form oleic acid (18:1, ω9). The third mechanism involves the lack of suppression of Scd1 gene transcription by DHA \> EPA \> LA. Scd1 was one of the first genes to be recognized as a target of C_20-22 ω3 regulation of gene expression. As such, the decline in hepatic DHA is associated with the increase in hepatic expression of Scd1. The mechanism for PUFA-mediated suppression of Scd1 expression involves suppression of Srebp1c nuclear abundance, a key transcription factor controlling expression of genes involved in *de novo* lipogenesis (DNL) and MUFA synthesis. More recent studies have shown that adding C_20-22 ω3 PUFA to the WD restores C_20-22 ω3 PUFA in liver, as well as decreasing Scd1 expression and increasing oxylipins derived from C_20-22 ω3 PUFA. Moreover, these changes in hepatic lipid are associated with attenuation of WD induced NASH as well as promotion of NASH remission, i.e., decreased hepatosteatosis. Taken together with the clinical observations, these findings suggest that dietary EFA sufficiency is an important parameter in the clinical management of patients with NAFLD, from benign steatosis to HCC. EFA deficiency or an imbalance of ω3 and ω6 PUFA, i.e., the ω6/ω3 PUFA ratio, received attention in a recently published longitudinal study. The Raine study used a prospective cohort including 985 participants ranging in age from 14 to 22 years of age. Individuals at increased risk of developing fatty liver disease in early adolescence were associated with high dietary intake of ω6 PUFA and a high dietary ω6/ω3 PUFA ratio. Children who developed NAFLD had increased risk of further complications in adulthood, like NASH, T2DM, cardiovascular and renal disease. Moreover, human clinical trials have established that efforts to lower the ω6/ω3 PUFA ratio through dietary supplementation with C_20-22 ω3 PUFA decreased hepatosteatosis and liver injury. This treatment strategy, however, did not reduce hepatic fibrosis. Our findings, coupled with our previous studies using DHA to mitigate NASH onset and progression support that notion that changes in hepatic C_18-20 ω3 and ω6 PUFA play an important role in liver health. The significant decline in hepatic LA and ALA in response to 1 wk of WD feeding was inversely associated with significant increases in expression of multiple hepatic genes linked to inflammation (**Figs, –**). These hepatic gene expression markers include transcripts encoding Cd40, Cd44, Cd68, F4/80, Gpnmb, Mcp1, Nlrp3, Ptgs1, Ptgs 2, Tnfα, and Trem2 (**Figs and**). These transcripts represent a subset of the transcripts that increased rapidly in livers in response to the WD and persist at elevated levels throughout the 40 wk WD feeding study. This raises the question of whether such changes in hepatic EFAs or their C_20-22 oxylipin products directly affect hepatic abundance of inflammation markers. We addressed this issue in two previous studies, a NASH prevention and a NASH treatment study. Supplementing the WD with DHA attenuated the WD-mediated induction of these and other gene expression markers of inflammation as well as reversing the effects of the WD on hepatic ω3 PUFA-derived oxylipins. These findings suggest that maintaining a level of C_20-22 ω3 PUFA in the liver is critical to attenuate WD-induced hepatic inflammation and progression to NASH. We have established that induction of hepatic inflammation is an early event associated with WD induced NAFLD and this response precedes all other markers associated with NASH, such as obesity, insulin resistance, changes in hepatic GSH/GSSG ratio, and fibrosis. Based on these findings, we recommend attention to dietary EFA intake as a potential target for therapy and the mitigation of NAFLD/NASH severity. The EFAs and their C_18-22 oxylipin derivatives play multiple roles in cell function including: 1) serving as components of membrane lipids, 2) regulators of membrane lipid raft composition and signaling from membranes, and 3) conversion to oxylipins which are ligands for membrane and nuclear receptor signaling. As we show in , the mole% of ω3 and ω6 PUFA decrease while hepatic MUFA content increases. Such changes in hepatic fatty acid type is likely to lower membrane levels of ω3 and ω6 PUFA, which will influence membrane lipid composition and fluidity, signaling from the plasma membrane and the generation of regulatory oxylipins. Because of the pleiotropic role EFAs play in cell function, it is not surprising that a significant decline in their hepatic abundance affects multiple hepatic pathways. Hepatic oxylipins are transient, i.e., short term, regulators of multiple hepatic functions, including inflammation, blood flow and metabolism. There are many hepatic oxylipins affected by the WD. As such, it will be challenging to assign changes in hepatic functions to specific oxylipins. Preclinical and clinical studies report that dietary C_20-22 ω3 PUFA supplementation of mice or patients with pre-existing NAFLD lowers hepatic fat and reduce hepatic injury. However, both preclinical and clinical therapeutic studies reveal that dietary supplementation with C_20-22 ω3 PUFA does not promote total removal of hepatic fibrosis. A recent report indicates that the removal of excess ECM requires the matrix metalloprotease Mmp12. Mmp12 is expressed at low levels in liver, but is induced in response to the WD in livers of WD-fed mice. While DHA supplementation of the WD lowers hepatic expression of Col1A1, Col1A2, it also lowers hepatic expression of several Mmps, including Mmp12. As such, DHA supplementation of the WD may block full removal fibrosis/ECM, as seen in liver histology. This finding represents a significant limitation of C_20-22 ω3 PUFA in NASH therapy. As such, EFAs and/or C_20-22 ω3 PUFA should be considered for preventive use or use in patients who present with hepatosteatosis, but not fibrosis. # Conclusions Time is an important variable in dissecting cause and effect relationships in pathophysiology. Many studies on the effect of diet on the NAFLD and NASH use a single timepoint for their analysis. This report clearly shows that time is a critical factor in defining how dietary factors impact the onset and progression of NAFLD and NASH. Current NAFLD/NASH diagnoses and therapies are based on histological evidence of disease. Our findings indicate that changes in systemic and hepatic makers of inflammation occur rapidly after a dietary challenge using the western diet, and these events precede the development of histological evidence of MAS and fibrosis. Moreover, these silent events worsen over time to impact multiple regulatory pathways that impact the onset of MAS, liver injury, fibrosis and ECMR. Our studies establish that the increase in systemic and hepatic inflammation were inversely associated with a loss of hepatic EFA, implicating hepatic EFA status in maintaining liver health. While we understand how EFAs and their C_20-22 ω3 and ω6 PUFA products regulate hepatic lipid metabolism, less clear is how EFA-derived oxylipins affect liver function. Several reports implicate a role for oxylipins in liver health, but the diversity of the hepatic oxylipin species and the rapid conversion of active oxylipins to inactive compounds complicates an assignment of cause and effect in the management of liver health. There are many dietary and environmental factors contributing to liver health and the onset of fatty and inflamed liver disorders. In cases where NAFLD is attributed to poor diet, such as a diet with high ω6/ω3 PUFA ratio or low EFA content, we suggest correcting EFA status to aide in mitigating NASH severity. # Supporting information [^1]: The authors have declared that no competing interests exist. [^2]: Current address: Arkansas Department of Agriculture, Veterinary Diagnostic Laboratory, Little Rock, AR, United States of America

# Introduction Total knee arthroplasty (TKA) is a well-established modality with a high satisfaction rate for various knee disorders. However, this surgical procedure inevitably perturbs the femoral medullary canal, leading to marrow embolization that reportedly increases the risk of myocardial infarction or cardiac stress postoperatively. Minimizing femoral medullary canal destruction, thus reducing marrow embolism-related morbidities, is an important issue for TKA surgeries. In addition to improving prosthetic alignment, computer-assisted navigation TKAs also contribute to reduced operative blood loss and systemic emboli\[–\]. These observations imply that navigation TKAs may cause less microvascular damage than conventional TKAs. However, the molecular evidence for the differential extent in vascular injury between conventional and navigation TKAs remains elusive. Intercellular adhesion molecule-1 (ICAM-1), vascular cell adhesion molecule-1 (VCAM-1), and platelet endothelial cellular adhesion molecule-1 (PECAM-1) are cell adhesion molecules (CAMs) that contribute to endothelial activation and leukocyte recruitment. They have been employed as markers for endothelial or vascular damage or hemorrhage, including coronary artery disease\[–\]. After total joint surgeries, patients reportedly had higher serum levels of leucocytes and endothelial markers. Therefore, we postulated that serum levels of CAMs in patients receiving navigation TKAs may be different from those receiving conventional TKAs. The purpose of this prospective comparative study was to compare ICAM-1, VCAM-1 and PECAM-1 levels in serum and hemovac drainage of patients receiving navigation and conventional TKAs. We hypothesized that navigation TKAs would lead to less postoperative elevation of serum CAMs compared to preoperative baselines than conventional TKAs. # Materials and Methods ## Patients The protocol for this trial and supporting TREND checklist are available as supporting information. This prospective comparative study was approved by the Institutional Review Board of Kaohsiung Chang Gung Memorial Hospital (IRB 100-0038A3) before recruiting the patients, and was conducted from March 2011 to December 2011. Local officials did not mandate registration before recruiting the patients, but this study was later registered in the ClinicalTrials.gov system (ID: NCT02206321). Our study adhered to the TREND reporting statements. Patients in need of TKA surgery due to degenerative osteoarthritis of the knee visited the outpatient department first and then were scheduled for admission for TKA surgery. They were self-separated into two groups when they visited the outpatient department. Those patients who visited Dr. CJ Wang would undergo conventional TKA, and those who visited Dr. JY Ko would undergo computer navigation TKA. Both senior surgeons had performed more than 1,000 TKAs using the conventional and the computer navigation method, respectively, before the beginning of our study. The patients did not know which surgeon performed conventional or computer navigation TKA before admission. After admission and before the surgery, all of the patients were asked whether they consented to participate in this study. All but three patients consented to participate. The patients knew their allocation as they completed written informed consent, because there were four additional small skin incisions (0.5 cm each) for the insertion of reference arrays for the navigation procedures. No patients shifted to the other group during the study. Those with autoimmune diseases, rheumatoid arthritis, malignancies, previous knee surgery or post-traumatic arthritis were excluded. All surgical procedures, including aseptic dragging and skin preparation were performed with standard protocols in a standard surgical facility. Each patient was given intravenously a single dose of 1 mg prophylactic cefazolin and a pneumatic tourniquet (300 mm Hg), followed by a mid-vastus approach after midline skin incision. Because of the pilot nature of the study, we planned to enroll at least 30 patients for each group after consultation with the statistician. ## Navigation-assisted TKA The bone cuts were extramedullary-guided and mapped using navigation and infrared-based systems (Vector Vision; Brain LAB, Heimstetten, Germany), according to the manufacturer’s instructions. Briefly, two fixed reference arrays with marker spheres were tracked using an infra-red camera, and the marker spheres were fixed to the distal femur (4 mm pins) and proximal tibia (3 mm pins) via two bi-cortical half pins. The hip joint center, distal femur and proximal tibia articulating surfaces, and the medial/lateral malleolus of the ankle were mapped and registered. The femoral component was referenced to the anterior cortex of the distal femur. A multiple-referencing method using epicondylar line, Whiteside line, and posterior condylar line was adopted to determine the appropriate rotation of the femoral component, and the size and position of the femoral prosthesis was optimized by the navigation system. The distal femur cut and chamber cut were guided by the real-time navigation system without reaming and destroying the bone marrow cavity. An extramedullary guide connected to a reference array was employed to determine the tibia cutting level, varus-valgus angle and tibia slope. The rotation of the tibia component was adjusted to fit the femoral component. An extramedullary guiding rod was used to reference the center of the anterior ankle joint, assisting the determination of the tibial component rotation. Bone cut was achieved under real-time navigation. After completing the femur and tibia bone cut, femoral and tibia components (LPS-Flex system, Nexgen; Zimmer, Warsaw, IN, USA) were implanted with antibiotic-loaded cement fixation. A neutral mechanical axis with a deviation of less than 1 degree was obtained after soft tissue balancing, with the assistance of a real-time computer screen. A 1/8-inch hemovac (Zimmer Haemovac; Zimmer, Warsaw, IN, USA) was inserted as a closed drainage system, and then removed 24 hours after surgery. ## Conventional TKA Femoral bone cuts (distal and chamber cuts) were guided by an intramedullary system and guiding instruments. The proper size and positioning of the prosthesis (LPS-Flex system, Nexgen; Zimmer, Warsaw, IN, USA) was based on the surgeon’s experience, and was followed by the implantation of the prosthesis with antibiotic-loaded cement fixation. A 1/8-inch hemovac (Zimmer Haemovac; Zimmer, Warsaw, IN, USA) was inserted as a closed drainage system and removed 24 hours postoperatively. ## Perioperative management Administration of anti-coagulants was halted 1 week before surgery and resumed on the day after surgery. Patients without previous use of anti-coagulants received aspirin for chemical prophylaxis if no contraindications. Intravenous cefazolin 1 g every 8 hours was administered for 3 doses after surgery. The hemovac drainage was kept in place for 24 hours, and then removed. Hemoglobin and hematocrit levels were measured before surgery and 24 hours after the operation. If patients had hemoglobin \< 8 mg/dL or hemoglobin 8–9 mg/dL with unstable vital signs, blood transfusion with packed red blood cells was provided. After removing the hemovac drainage, patients proceeded with continuous passive motion, including active flexion-extension, quadriceps training and walker-aided ambulation with the assistance of physical therapists. Patients with a clean operative wound and stable general condition were discharged if range of motion of the knee exceeded 95 degrees. All complications before manuscript preparation (July 2014) were scrutinized and reported. ## ELISA assessment Five milliliters of drainage from the hemovac was harvested and processed to collect supernatants. Ten milliliters of peripheral blood was drawn from each patient before and 24 hours after TKA and processed to collect sera. All processed supernatants and sera were stored at -80°C till ELISA analysis. Concentrations of all CAMs in sera and drainage supernatants were measured by respective ELISA kits (R & D Systems), according to the manufacturer’s instructions. The technician performing the ELISA analysis was blinded to the patient profile and grouping via the de-labeling process. ## Blood loss assessment Blood loss during the first 24 hours was calculated by the method proposed by Sehat et al. and other authors, which can be summarized by the following formula: blood loss = total blood volume × (Hct_pre-op-Hct_{post- op})/ Hct_mean + volume transfused where total blood volume was calculated based on sex, height and weight, using the method described by Nadler et al (<https://www.easycalculation.com/medical/blood-volume.php>). ## Statistical analyses Data are expressed as median (lower quartiles, upper quartiles). Categorical variables were compared by Chi-square test. As for continuous variables, the Mann-Whitney U test and Wilcoxon signed-rank test were utilized for between- group and within-group comparisons, respectively. The Spearman’s correlation coefficient was used to measure the strength of correlation between the two variables. All statistics were performed by SPSS software, and a p value \< 0.05 was regarded as statistically significant. # Results Ninety patients met the inclusion criteria from March 2011 to December 2011. Three of the 90 patients that underwent TKA surgery declined to participate and were excluded from the study. Finally, 87 eligible patients provided written informed consent and were enrolled. Fifty-four patients underwent computer- assisted TKAs and 33 received conventional TKAs. All 87 patients completed the preoperative and postoperative blood sampling and the collection of drainage from the hemovac. There were no significant differences in gender, affected side, age or BMI between the computer navigation and conventional TKA group. The numbers of patients afflicted with coronary artery disease and coronary artery equivalent disease (including symptomatic carotid artery disease, peripheral arterial disease, abdominal aortic aneurysm, diabetes, and chronic kidney disease) in both groups were not significantly different. The calculated volume of blood loss in the computer navigation group was 955 (772, 1164) mL, significantly lower (p = 0.001) than the 1265 (963, 1475) mL in the conventional group. The skin to skin closure time was 116 (103, 124) minutes in the navigation group and 110 (101. 115) minutes in the conventional group. (p = 0.141) The baseline serum CAMs did not differ between the two groups before surgery. Of interest, postoperative serum ICAM-1, VCAM-1 and PECAM-1 levels in the computer navigation group were 35.5% (p \<0.001), 2.0% (p = 0.037) and 49.3% (p \<0.001) lower, respectively, than those in the conventional group. The extent of postoperative elevation of serum ICAM-1 (p = 0.022) and PECAM-1 (p = 0.003) in the navigation group was significantly milder than that in the conventional group. However, there was no significant difference (p = 0.435) in the postoperative elevation of VCAM-1 between groups (Tables –). The patients in the computer navigation group had lower ICAM-1 levels in hemovac drainage supernatants than patients in the conventional group. The ICAM-1 and VCAM-1 levels in hemovac drainage supernatants were significantly correlated with those in sera. However, PECAM-1 levels in hemovac were not significantly correlated with those in sera. Up to the time of manuscript preparation (July 2014), two patients, an 85-year- old female in the conventional group and a 64-year old male in the computer navigation group, were admitted for angina symptoms, with an interval between surgery and admission of 12 and 14 months, respectively. Coronary angiography revealed no apparent stenosis for the 85-year-old female patient and single- vessel disease for the 64-year-old male patient. Percutaneous angioplasty was performed smoothly for the 64-year-old male. However, no major complications were noted in any of the patients of both groups postoperatively. # Discussion Complications secondary to bone marrow violation are significant concerns after TKA surgeries. Computer navigation TKAs do not involve the reaming of the medullary canal, thus minimizing the destruction of the femoral medullary cavity. While this modality reportedly improves blood loss and prosthesis accuracy, biochemical verification of vascular injury has not been defined\[, \]. In this study, we provided novel evidence that patients had decreased blood loss concomitant with mitigated postoperative elevation of levels of CAMs after computer navigation TKA, which is indicative of its less-invasive nature with regard to the integrity of the femoral medullary cavity. The results of the current study shed new light on the known advantages of computer-aided TKA procedures for patients. Compared to patients with conventional TKAs, patients that underwent computer navigation TKAs had milder postoperative elevation of serum ICAM-1 and PECAM-1 from the preoperative baseline. These molecules are associated with vessel damage, thrombosis and hemorrhage, and reportedly promote atherogenesis of various tissues in pathologic contexts\[, –\]. In the current study, the milder elevation of postoperative serum CAMs from the preoperative baseline reflects the milder vessel-deleterious reactions after computer navigation TKAs. Marrow embolism after conventional TKA might be associated with increased risks of acute cardiac disorders. A retrospective study argued that the increased risk of acute myocardial infarction was attributable to the destruction of the femoral medullary canal in the TKA procedure. Serum ICAM-1 levels are linked to macrovascular disease, and serum VCAM-1 concentrations have been correlated with the occurrence of acute coronary syndrome. Although the current study does not correlate the milder elevation of postoperative serum CAMs from baseline to the lower risk of cardiovascular complications in patients after computer-aided TKA, the timing of our observation overlapped exactly with the “hazardous first two weeks” for myocardial infarction. The significance of differential serum CAMs and the correlation with the incidence of heart disorders after different TKA modalities merits long-term follow-up. We acknowledge the limitation of this study, which is that all participants came from the outpatient department and chose surgeons to perform elective TKA surgery of their own free will. However, the patients did not know which surgeon performed navigation or conventional TKA until the day before the operation, and no patient shifted to the other group during the study. The two physicians were arthroplasty specialists who had performed more than 1,000 TKAs using the method they were familiar with before the study, and neither of them shifted to performing the other technique throughout the course. The shift to a less familiar technique might introduce performance bias, and the choice of a less familiar method would probably prolong the operation time and confound the results. While the current study may not be a double-blinded or truly randomized study design, patient attributes, including gender, age, side in need of TKA, BMI, and heart comorbidities were not significantly different between the groups. The skin to skin closure time for the two groups was similar. All procedures for ELISA analysis were performed using stringent de-coding procedures, and technicians were all blinded to the identity and source of the specimens. Despite all endeavors to minimize possible bias, more rigorously- designed studies, such as randomized double-blind studies, are still necessary to substantiate the preliminary finding. The results of this study cannot be extrapolated to conclude that navigation TKAs lead to fewer postoperative cardiac events and fewer postoperative thromboembolic complications. The correlations between the postoperative CAMs and the incidence of cardiac events or thromboembolic complications warrant a larger study population for further validation. Taken together, our results reveal that conventional TKAs inevitably perturb the femoral medullary canal, leading to the destruction of vascular integrity, and thereby contribute to more apparent elevations of concentrations of CAMs after the surgery. Computer navigation TKA impedes the medullary canal to a lesser extent, which minimizes vessel deterioration and leads to milder elevation of CAMs after the surgery. This study highlights low serum CAMs as emerging biochemical indicators that strengthen the advantage of navigation TKA. The correlation between differential CAM levels and the incidence of heart events deserves further investigation. # Supporting Information We are thankful for the technical support of colleagues in the Fifth Central Lab of the Department of Medical Research. We are also thankful for the invaluable suggestion for the amendment of this manuscript during the revision process by professor Hsien-Yuan Lane, the chairman of the Graduate Institute of Clinical Medical science, China Medical Univeristy, Taichung. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: JYK FSW. Performed the experiments: FSW. Analyzed the data: SJK. Contributed reagents/materials/analysis tools: CJW JYK SHC KKS. Wrote the paper: SJK.

# Introduction Single embryo transfer (SET) is the preferred treatment to limit multiple pregnancies after ART. In order not to compromise the carry home baby rate, the selection of the embryo for transfer in the first cycle becomes even more important. Next to the existing criterion based on morphology, other methods are currently under investigation. The use of quantitative gene expression measurements in cumulus cells (CC), which are in close contact with the oocyte during growth and maturation, seems a promising method. Since the first published study on the subject where CC expression could be related to embryo development, several other studies have investigated this possibility and try to relate CC expression to different endpoints. Examples of endpoints investigated are: embryo development, aneuploidy stage of the oocyte, oocyte nuclear maturity stage and probably the most important from a patient perspective: pregnancy outcome. Confirmation of results between different studies does not seem obvious in the analysis of CC gene expression. In the current literature not many genes were found in common in different studies. For example, hyaluronan synthase 2 (*HAS2*) was higher expressed in good quality embryos compared to low embryo morphology in two studies, but could not be related to embryo morphology in two other studies. Divergences can be due to a different experimental design, with different endpoints, but gene expression can be influenced by known factors as the stimulation protocol of the patients, or not yet assessed factors such as culture media used in the different IVF laboratories. In this study, 47 individual cumulus complexes from 47 intra-cytoplasmic sperm injection (ICSI) patients were retrospectively analyzed with quantitative real- time polymerase chain reaction (QPCR). Using the current sample set, a pregnancy prediction model from a previous study was validated for its predictive power. In a next step, in an attempt to search for new genes with a stronger predictive power, new multivariable models were built considering the 3 genes (ephrin-B2 (*EFNB2*), calcium/calmodulin-dependent protein kinase ID (*CAMK1D*), stanniocalcin 1 (*STC1*)) described earlier and 9 novel genes (glutathione reductase (*GSR*), glutathione peroxidase 3 (*GPX3*), glutathione S-transferase alpha 3 and 4 (*GSTA3* and *GSTA4*), transforming growth factor beta 1(*TGFB1*), progesterone receptor (*PGR*), inositol 1,4,5-trisphosphate receptor type 1 (*ITPR1*), solute carrier family 2 (facilitated glucose transporter) member 1 (*SLC2A1*) and thrombospondin 1 (*THBS1*)) (inter-patient analysis) (see for an overview of all genes). Our patient sample set allowed for an analysis never reported before in literature: CC from oocytes that did not result in pregnancy in the fresh transfer cycle and the CC from their sibling oocytes that resulted in pregnancy after a frozen embryo transfer (FRET) cycle were analyzed (intra-patient analysis). To our knowledge this is the first study to compare pregnant and non- pregnant CC from the same retrieval cycle in a SET setting as would be done in a final clinical application. # Materials and Methods ## Patient population This study was approved by the Ethical Committee of the UZBrussel and the patients written consent was obtained. Forty seven ICSI patients were selected based on the embryo transfer policy (single embryo transfer) and ovarian stimulation protocol prescribed: GnRH antagonist in combination with recombinant follicle stimulating hormone (FSH) (Gonal-f, Merck-Serono, Geneva, Switzerland; *n* = 4 or Puregon, MSD, Oss, The Netherlands; *n* = 43). Causes of infertility were: male factor only (n = 19), female factor only (ovulation disorder n = 3 and tubal infertility n = 2), combination male and female factor (OAT and endometriosis n = 3) and idiopathic (n = 20). Twenty patients had single embryo transfer on day 3 of culture, from these 10 became pregnant. Twenty seven patients had transfer on day 5, from these 9 became pregnant. The day of transfer was decided by the consulting doctor before oocyte retrieval took place taking into account the age of the patient or by the embryologist according to the number of good embryos available. All pregnancies resulted in live births. ## Collection of human cumulus cells and embryo culture Vaginal ultrasound was used to monitor follicular development. The endocrine profile was monitored by analysis of serum 17β-estradiol (E2), progesterone, FSH and LH by electrochemiluminescence on a COBAS 6001 immunoanalyser (Roche, Roche Diagnostics, Mannheim, Germany) using validated assays with respectively sensitivities of 5 ng/l, 0.03 µg/l, \<0.1 IU/l, 0.1 IU/l and total imprecisions (%CV) of respectively \<6, \<7, \<6 and \<6. Final follicular maturation was induced with a dose of 10 000 IU hCG when at least three follicles of 17 mm in diameter were observed by transvaginal ultrasound. Oocyte retrieval was done 36 h later. CC collection was done as described in. Briefly, individual oocyte denudation was performed in 40 µl droplets of HTF-SSS containing 80 IU/ml Cumulase (MediCult, Lyon, France) for not longer than 30 s and washed sequentially in droplets without enzyme. At any time, oocytes were handled individually from this point onwards in order to allow retrospective analysis of the CC per oocyte. After denudation, the CC were plunged directly in liquid nitrogen. ICSI was performed as described previously and embryos were cultured in sequential media of SAGE (CooperSurgical, Leisegang Medical, Berlin). Embryos were vitrified on day 3 or day 5/6 of embryo culture as was described earlier and used in a subsequent FRET cycle. The day 3 embryos were warmed on cycle day 3, cultured overnight and transferred as a day 4 embryo on cycle day 4 ( = synchronized transfer). The day 5 (and day 6) blastocysts were warmed in cycle day 5 (or day 6) in the morning and transferred on the same day. For all 47 patients, only the CC related to those oocytes resulting in embryos selected for transfer were analyzed exception made for 7 of the 28 non-pregnant patients, 1 extra CC sample (except for 1 patient, 2 CC samples) related to a vitrified embryo giving pregnancy after a frozen single-embryo transfer cycle, was analyzed (8 extra CC in total from 7 patients). shows the different samples used for each analysis. ## Gene selection This study is the 3^rd one in a row to evaluate the predictive value of CC gene expression for oocyte quality using QPCR. Over the 3 studies, we followed a precise strategy to choose which genes to analyze regarding to oocyte quality in ICSI patients. The first study identified 4 top genes, 2 predictive for embryo morphology (inositol-trisphosphate 3-kinase A (*ITPKA*) and transient receptor potential cation channel, subfamily M, member 7 (*TRPM7*)) and 2 for pregnancy outcome (syndecan 4 (*SDC4*) and versican (*VCAN*)). It was chosen to include those 4 genes in the next study. *ITPKA* and *TRPM7* were again related to embryo development, but *SDC4* and *VCAN* were not retained this time in the pregnancy models and were replaced by *EFNB2*, *CAMK1D* and *STC1* when using stepwise multiple regression analysis. As several studies showed no relation between the genes predictive for embryo development and pregnancy outcome, it was decided to repeat only the 3 pregnancy related genes in this third study. Our hypothesis is that by this ‘cascade’ testing strategy, the strongest pregnancy predictive genes may be filtered out through the consecutive studies. The other 9 genes analyzed in this study were: *GSTA3, GSTA4, GPX3, GSR, ITPR1, SLC2A1, THBS1, TGFB1* and *PGR*. The 9 new genes were chosen from own unpublished array data comparing CCs related to pregnancy versus CCs related to no pregnancy. ## RNA extraction and cDNA synthesis Total RNA was extracted as described earlier using the RNeasy Micro Kit (Qiagen, Westburg, Leusden, The Netherlands) including a DNase step and addition of 5 ng/µl poly(dA) (Roche Applied Science, Mannheim, Germany) prior to extraction. Extraction was followed by a second DNase treatment (RQ1 RNase-Free DNase, Promega, Leiden, The Netherlands). Reverse transcriptase (RT) was done as previously described with the iScript cDNA Synthesis Kit (Bio-Rad Laboratories, Ghent, Belgium). Negative controls were generated by omitting the enzyme or the RNA in the RT reaction. ## Real-time PCR Primer sequences for *CAMK1D, STC1* and *EFNB2* are listed in Wathlet et al 2012. Primers for *GPX3, GSTA3, GSTA4, PGR, THB1, ITPRA, SLC2A1, GSR* and *TGFb1* can be found in. Both beta-2-microglobulin (*B2M*) and ubiquitin C (*UBC*) were validated and used before as normalization factor. Cycling conditions, negative controls, standard curves and normalization (with *B2M* and *UBC*) are as described earlier, but all PCR reactions were adapted to 10 µl reactions. All values mentioned hereafter are the normalized values to the mean of both *B2M* and *UBC* for each sample. ## Statistics ### Inter-patient analysis In a first analysis, a two-tailed t-test (GraphPad Prism version 4.01 for Windows, GraphPad Software, San Diego California USA) was used to compare cumulus complexes of oocytes resulting in pregnancy or not (19 live birth, 28 non-pregnant). All data were LOG transformed to obtain normal distribution and only *P*-values \<0.0042 were considered significant after Bonferroni correction. In a second analysis, a model was built using stepwise multiple regression analysis as described earlier. Briefly, a linear regression model, with an equation as output ‘y = a + bx + cz + ds +et + fu + gv + hw’, was built with as response variables (x, z, s, t, u, v, w) gene expression and/or patient and cycle characteristics (all are listed in). ‘b–h’ are the respective indexes of the included variables and ‘a’ is the intercept of the equation. A variable was added to the model if the type III *P*-value of the variable was \<0.3 and if the *P*-value of the model was improved. At the end of the model a backwards regression step was performed to exclude redundant variables. Four different models were built to predict pregnancy. In two models only gene expression values were allowed (first the model was restricted to 3 genes, next all genes were allowed into the models as long as they improved the *P*-value of the model). In two other models, the need for correction by patient and cycle characteristics was assessed by allowing all patient and cycle characteristics to the models only composed of genes, when those extra variables could improve the model. By introducing those extra factors, possible inter-patient variability on gene expression could be leveled out and increase the differences related to oocyte quality. For all models, accuracy ((True positive + true negative)/(true positive + false positive + false negative + true negative)) and positive and negative predictive values (PPV =  True positive/(true positive + false negative) and NPV  =  True negative/(true negative + false positive) were calculated. The power of all models was represented by Receiver Operating Characteristic (ROC) and the area under curve (AUC) was calculated. ### Intra-patient analysis Finally, the study allowed comparing 2 (or 3) oocytes originating from one oocyte retrieval cycle with known pregnancy outcome per oocyte, as all embryos were transferred individually in consecutive cycles. For this purpose, from 7 of the 28 patients that were not pregnant in the fresh cycle, the CC of the embryos that were replaced in a subsequent frozen single embryo transfer cycle resulting in pregnancy were compared to those transferred in the fresh cycle. For one patient, two consecutive frozen embryo replacement cycles were analyzed as the first embryo did not end in a pregnancy. Seven genes were chosen for this analysis based on their presence in one of the above mentioned models or their *P*-value of addition, when first added to a pregnancy model. A paired t-test was performed for each gene and the chance to pregnancy was calculated with the earlier defined models from the inter-patient analysis, containing only genes and built on the 47 CC samples (19 live birth), but excluding the 8 CC samples from the frozen cycles. # Results ## Patient population No statistical differences were found between the patient and cycle characteristics of the pregnant and the non-pregnant groups. Mean age and BMI for both groups of patients was low but comparable. Progesterone levels were low in both groups. Mean cycle number attempt was not different in both groups: respectively in the pregnant and non-pregnant group 12 and 17 patients underwent their first cycle, 6 and 10 their second cycle and 1 for both groups their third cycle. Percentages of oocyte maturity and fertilization were normal in both groups, and the percentage of good quality embryos on day 3 was more than 50%. Embryo quality score on moment of transfer was not different for both groups. Nineteen fresh SET cycles resulted in live birth. ## Pregnancy prediction ### Validation of predictive genes (CAMK1D, EFNB2 and STC1) using a model built on a previous sample set A model predicting pregnancy composed of 3 genes, considered in a previous study, was tested in the current, independent patient series. The gene expression values for *CAMK1D, EFNB2* and *STC1* of the CC of the 47 patients were introduced in the equation obtained before and gave a value predicting the chance to pregnancy for each of the 47 oocytes. The obtained PPV and NPV calculated with the 47 samples of this study were 62% and 86%, with an accuracy of 72%. ### Inter-patient analysis: t-test of all 12 genes in the current sample set The cumulus complexes of 28 oocytes not resulting in pregnancy were compared to 19 CC of oocytes that resulted in a live birth. Only *EFNB2* was statistically higher in the pregnant group. *CAMK1D, GSTA4* and *GSR* only showed a trend of higher expression in the pregnant group (respective *P*-values: 0.0068, 0.0123 and 0.0507). Graphs for all genes can be found in. ### Inter-patient analysis: Stepwise multiple regression analysis In a first step to build a pregnancy model, the *P*-value of addition when added as first variable was calculated for all genes and can be found in. First, the model was restricted to the inclusion of 3 of the 12 genes (Model 1). *EFNB2, GSTA4* and *PGR* were retained and gave a model with a *P*-value of 0.0015, a PPV of 68%, a NPV of 79% an accuracy of 73% and an AUC of 0.82. When trying to improve this model by also allowing patient and cycle characteristics (from), no improvement on the previous model was found (Model 1 bis). In a next step, more than 3 genes were allowed into the model if improving the *P*-value (Model 2). Five of the 12 genes were retained in this model (i.e. *EFNB2, GSTA4, PGR, GPX3* and *GSTA3*) which yielded a *P*-value \<0.0001 with a PPV of 78%, a NPV of 83%, an accuracy of 81% and an AUC of 0.93. Adding patient and cycle characteristics improved the model (Model 3). The retained parameters were: days of stimulation, relative E2 and age. The PPV, NPV and accuracy of the extended model all increased to 93% and the AUC to 0.95. Full mathematical models can be found in. ROC curves are shown in. ### Intra-patient pregnancy prediction For seven patients, cryostored CC samples related to cryopreserved embryos which had led to a clinical pregnancy after transfer in a subsequent transfer cycle were analyzed. This material was used to analyze the genes present in the above obtained multivariable models and/or the 5 genes with the smallest *P*-value of addition i.e. *CAMK1D, EFNB2, GPX3, GSR, GSAT4, GSTA3* and *PGR*. As a first explorative method a paired t-test for those seven genes was performed and can be found in. Five genes had an upwards trend in the CC of an oocyte resulting in pregnancy (*P*-value \<0.05; only *P*-values \<0.07 are significant after Bonferroni correction for multiple comparisons): *EFNB2* (only significant difference between pregnant and non-pregnant), *CAMK1D, GSR* and *PGR*. *GSTA4, GPX3* and *GSTA3* had all *P*-values \>0.05. Fold changes between the CC from the same patient were calculated and an average per gene was made (from 1.1 to 2.5). For the five genes with a *P*-value \<0.1 in the paired t-test, the percentage of correctly estimated CC was 71% to 88%, based on the level expected (e.g. higher expression is expected in the pregnant compared to the non-pregnant related CC) from the paired t-test. In a next step the predictive power of the multivariable models (the 3 gene model ( = Model 1) and the 5 gene model ( = Model 2)) obtained in the inter-patient analysis (47 CC samples) was used to rank the CC of each patient for their chance to pregnancy. As all patient and cycle characteristics are identical for oocytes within a single retrieval cycle it makes obviously no sense to use models containing those variables. All CC, except for patient 4, were correctly ranked for their chance to pregnancy. # Discussion The expression of 12 genes in the CC and their capacity to predict the pregnancy potential of the oocyte they surrounded were analyzed in the present study. Three genes (*CAMK1D, EFNB2* and *STC1*) were included from a previous study, where they were coming out as the most predictive ones for pregnancy prediction. *EFNB2* was also significantly up-regulated in the current study in the CC from the oocytes that gave pregnancy and *CAMK1D* showed the same trend (*P* = 0.0068). The model composed of genes only from the previous patient dataset, using *CAMK1D, EFNB2* and *STC1*, was validated in this independent patient group and yielded a similar overall performance of the model, with accuracies of 72% in the current study and 79% in the previous study. Besides the *EFNB2* and *CAMK1D* coming from our previous study, two newly studied genes, *GSTA4* and *GSR*, also showed an up-regulated trend in the CC of the pregnant group. A multivariable approach tested whether it would be possible to predict live birth by using only 3 of the 12 tested genes in the current sample set (inter- patient analysis). The 3 most predictive genes were *GSTA4*, *PGR* and *EFNB2* and resulted in a similar accuracy as the previous model with *EFNB2, CAMK1D* and *STC1* (73% versus 72%). Using three genes, none of the patient or cycle characteristics could improve the model, suggesting that the expression of the three genes was minimally influenced by patient and cycle factors. The live birth model could be improved by including two more genes (*GPX3* and *GSTA3*), which increased the accuracy up to 81% and resulted in an AUC of 0.93. This 5 gene model was improved with three patient and cycle characteristics (age, relative E2 and days of stimulation) and resulted in an optimized model with a PPV, NPV and accuracy of 93%, but similar AUC as the model only containing genes. Ideally, pregnancy prediction based on gene expression should be possible with a limited set of genes to reduce analysis time and cost. Of the 12 genes tested in this patient population, the 2 most recurrent genes are *GSTA4* and *EFNB2*. These 2 genes are present in all 3 pregnancy models and have respectively in 3 out of 3 and in 2 out of 3 predictive models a type-III *P*-value \<0.01. The models also showed that the gene expression values are always more important than the patient and cycle characteristics, as the type- III *P*-values in the models are only significant for the genes and not for the patient and cycle characteristics. Furthermore, the AUC results are comparable between the model containing only genes and the model combining genes and patient and cycle characteristics. For 7 patients, 2 (or 3) CC from oocytes giving embryos with a good morphology and consecutively transferred, were analyzed (intra-patient analysis). For all oocytes, pregnancy outcome was known as all embryo transfers were done in subsequent SET cycles. Having assessed the expression of 7 genes (*EFNB2, CAMK1D, GSR, PGR, GSTA4, GSTA3* and *GPX3*), it was easy to estimate, based on single genes or on a combination of them in a model, which oocyte of the two considered ones would have the highest chance to pregnancy. The best predictions, when looking at each gene individually, were found for *CAMK1D, GSR, EFNB2, PGR* and *GSTA4* with 71% to 86% correct estimations. Combining 3 or 5 genes, according to the earlier models, 6 of the 7 patients had the cumulus complexes correctly predicting the chance to pregnancy. This prediction was the same when using the 3 or 5 gene model. Surprisingly, the multivariable model failed on a different patient than the single gene analysis. The multivariable approach seems stronger than the individual genes as all genes wrongly predicted the pregnancy outcome for the second comparison of patient 6, but the multivariable model predicted successfully, especially as the CC of the oocytes that resulted in pregnancy were not considered for building the predictive model. The failed prediction of the multivariable approach for patient 4, where the individual genes were correct in 4 out of 5, could be due to other factors (e.g. the endometrium status at the moment of transfer). Although this analysis is still limited in patient numbers, these results are encouraging. This is the first study that confirms that some genes predicting pregnancy between patients might also be capable of ranking the quality of oocytes within patients using a multivariable approach and providing a chance to pregnancy for each oocyte. CC gene expression analysis might become a valuable tool in the ART lab but does obviously not take into account the eventual influence of poor sperm quality and out-of-phase endometrium. This study could confirm the use of earlier found predictive genes in a new patient population, with the same stimulation protocol, but with different culture media. This data suggests that *EFNB2* and *CAMK1D*, the only genes that we analyzed in two studies using two different culture media, were not affected by culture media. To test the validity of the current models, a future analyze should involve patients with different stimulation protocols and different culture media. Reasons why the specific genes of this study are important for pregnancy prediction remains speculative. As mentioned previously, *CAMK1D* may, among other pathways, be related to steroidogenesis by its strong correlation with steroid related genes (*CYP11A1, STC2* and *HSD3B1*). In this data set *CAMK1D* also strongly correlated with a steroid related gene i.e. *PGR* next to *EFNB2* and *GSTA4* (all *P*\<0.0001 with Pearson correlation analysis) still leaving the possibilities open for more than one pathway to which *CAMK1D* could be associated with. The presence of *PGR* in the pregnancy models and its predictive power within patients might indicate that steroid-related genes could be helpful in pregnancy prediction. *PGR* has been described before, but no difference between pregnant and non-pregnant could be confirmed with quantitative PCR. *GSTA3* that was present in some of the pregnancy models also has a link with progesterone production reiterating the importance of the steroidogenesis pathway. The function of *EFNB2* in the ovary is not yet known, but B-class ephrins were proposed to be related to the luteinization process and the ephrin B2 receptor was found differently expressed between CC from normal oocytes compared to aneuploid oocytes. Out of the other genes tested, only members of the glutathione family were retained in the models or were significant in the t-test. Glutathione enzymes are important for detoxification actions (of free radicals) in the cells through the use of glutathione. Hypoxia leads to the formation of reactive oxygen species (ROS) which can cause lipid peroxidation, enzyme inactivation and cell damage, resulting in apoptosis not only in CC, but also in the oocyte. Oxidative stress has already been reported by other groups as a possible target to assess oocyte quality. One reason that some transcripts of those pathways are higher in the CC from oocytes giving pregnancy might be that those oocytes are better protected against a stressful environment if needed. Before knowing what are exact role and importance of the considered genes in this study, more experiments need to be performed using animal models to be able to access the antral growth and periovulatory period since in human only 1 time point (i.e. 36h after hCG) is available for analysis. ## Conclusion By testing presumably important genes batch wise in three consecutive studies on different patient groups for which cumulus complexes were frozen individually per oocyte, we retained the most predictive genes for pregnancy and opposed these every time to new candidate genes. This ‘cascade’ strategy attempted to increase the power of pregnancy prediction using CC gene expression as quality marker for oocytes in ART. The strategy proved effective as the model with *EFNB2*, *CAMK1D* and *STC1* from the second study could be confirmed on an independent patient sample set. In an attempt to further improve prediction models for live birth, models were built (inter-patient analysis), still retaining *EFNB2* together with *PGR* and genes related to the glutathione metabolism. The new models proved to be able to rank oocyte for their potential for pregnancy (intra-patient analysis). The validity of our current models, for routine application, still need prospective assessment in a larger and more diverse patient population allowing intra-patient analysis. # Supporting Information The authors thank their colleagues from the Centre for Reproductive Medicine of the UZ Brussel. [^1]: The authors have declared that no competing interests exist. [^2]: Selection of the patients, scored the embryos and revised the manuscript: LVL GV. Supervised the work at the clinical centre: PD. Conceived and designed the experiments: SW TA JS. Performed the experiments: SW. Analyzed the data: SW TA IS WC. Wrote the paper: SW TA JS.

# Introduction The ability to control action prospectively over the first years of life, which involves having priori knowledge of actions effects, constitutes a cornerstone for action development. A powerful mean to examine this feature consists in studying developmental changes in postural control when performing a motor task. Postural control in fact supports motor action by ensuring balance during execution of action. Then, it is necessary that postural adjustments anticipate for the imbalance produced by actions. Several studies have been conducted on the development of postural anticipation when reaching in sitting and draw a twofold conclusion. First, anticipation in sitting postural control emerges as early as the first year of life, with 4- to 12-month-old infants demonstrating some ability to activate the postural (neck and trunk) muscles before the arm muscle that initiates the arm movement. Second, postural anticipation to prepare for the reach becomes consistent only in late infancy, with an increased anticipatory activity of the postural muscles in infants aged over 15 months. Such a developmental trend was further supported by findings from another study concerned with the development of anticipatory postural adjustments during a pulling task while standing. The proportion of pulls involving anticipatory activation of leg muscle to prospectively counteract an expected forward displacement in the body’s center of gravity progressively increased between 10 and 17 months. Interestingly, regrouping data in terms of infants with and without independent walking experience revealed marked increases in the temporal specificity and consistency of anticipatory adjustments as infants gain experience with walking. In the above framework, a study on the use of a contact surface for stabilizing upright posture provided further details on the link between independent walking and the development of anticipatory control. It was demonstrated that infants with no or very little walking experience used contact surface in reaction to their body sway, such like a mechanical support stabilizing posture. Inversely, infants with significant walking experience (∼ 1.5 month post-walking) used it in anticipation to body sway, integrating somatosensory information before swaying to stabilize posture. Walking appeared consequently as a crucial piece for establishing anticipatory strategy to stabilize posture, likely due to the refinement of an internal model for the sensorimotor control of posture that starts occurring at the time infants walk. However, there is evidence that infants within the first months of independent walking demonstrate anticipatory lateral weight shifts before initiating gait, suggesting that some form of anticipatory postural control is already present at independent walking onset. In sum, although postural anticipation is likely already in use before infants start walking, the onset of walking experience seems to bring about significant changes in anticipatory postural control. The purpose of this study, therefore, was to investigate whether independent walking experience plays a facilitative role in the development of anticipatory postural control. To this end, we evaluated whether walking infants show greater sophistication compared to non- walking infants in their adult-like ability to anticipate for postural disturbances induced by a continuously moving platform. Indeed, there is evidence in adult population that anticipatory postural adjustments prevail when the platform moves in a rhythmic fashion, involving the activation of appropriate muscles prior to the turning points of the platform to counteract the destabilization. This comes from the fact that the postural challenge induced by the moving platform is highly predictable and the ongoing sensory inputs can be used by the central nervous system to foretell the mechanical effect of the perturbation. Specifically, the experiment included both infants (some of whom were independent walkers) and adults, who were sitting on a support surface continuously rotating in the frontal plane that caused postural destabilization (i.e., lateral tilt). The subjects’ ability to anticipate for the destabilization induced by the moving platform was examined from the temporal relationship between the muscular activation of neck (i.e., trapezius muscle) and back (i.e., erector spinae muscle) muscles, which function is to counteract lateral postural sway. We predicted that only infants walking independently would activate these muscles in anticipation of the platform rotations to counteract the forthcoming destabilization as adults do. Moreover, we also examined the kinematic behaviour of the subjects to reveal the strategies used to cope with the balance challenge. # Methods ## Subjects Subjects were 10 infants who had not attained independent walking yet (aged 6–13 months, 5 boys), 10 infants who were independent walkers (aged 10–22 months, 7 boys) and 6 young adults (aged 18–30 years, 3 men). The more mature infants within the group of infants at the pre-walking stage were able to stand upright (1 girl, 3 boys) while the less mature ones were still mastering upright posture. Independent walkers had one to six months of walking experience, determined as the time interval from walking onset (i.e., the infant was able to perform at least three independent walking steps) until the testing day. Walking onsets were obtained through parental reports. All subjects were recruited from the community of Marseille and were in the majority from middle-class families. The infants’ caregivers and the young adults gave written informed consent. The local research ethics committee CPP Sud-Méditerranée I approved the experiment. ## Apparatus The subject sat on a software-controlled, electrically driven rotating platform (90 cm×90 cm) that produced sinusoidal roll rotations at a frequency of 0.5 Hz. The peak-to-peak platform rotation was 6.6° and the duration of a trial was 100 s, subjecting the subject to 25 rotations to the right and 25 rotations to the left. The frequency and amplitude characteristics were determined in order for the platform movements to destabilize posture (i.e., lateral tilt) and demand that subjects relied on anticipatory postural adjustments to counteract for the destabilization. Higher frequencies and amplitudes would have not been safely sustainable by the young infants. Further, higher frequency and amplitude oscillations would have produced a too large variability in the response behaviour of the subjects and would have induced neuromuscular fatigue with the potential of confounding the postural adjustments. Conversely, lower frequencies and amplitudes would have not implied enough postural destabilization, leading to small postural adjustments that are difficult to measure. ## Procedures Once arrived in the laboratory, the subject was provided a few minutes to become acclimated to the testing environment and experimenters. The laboratory simulated a common living room intended to provide calm and soothing environment for the subject. Following this period, upper body clothing and pants were removed in order for the experimenters to place electrodes and reflective markers. Bipolar surface electrodes (dimensions 15 mm×15 mm, inter-electrode distance 20 mm) were placed over the surface of lower back and neck muscles, namely the erector spinae (ES) muscle (30 mm lateral to the spinal process of the second lumbar vertebra) and the trapezius (T) muscle (over its cervical part). Reflective markers were placed on several anatomical landmarks of the upper body and head including the head vertex, the mastoid processes, the spinal processes of the seventh cervical vertebra, the sixth and the twelfth thoracic vertebrae, the second lumbar vertebra, the sacrum, and the posterior superior iliac spines. Body segments were subsequently defined from the markers. Two additional markers were also placed on the platform. The subject then sat on the platform in a self-selected position. The experimenters made sure that the subject was positioned along the axis of rotation of the platform with the legs stretched forward and the arms rested on the legs. The subject then underwent 100 s rotations. The platform was located 1.5 m away from a white wall, preventing the visual attention of the subjects from being oriented toward a particular location in the visual field. No instructions were provided to adult subjects as to how they should respond to the platform oscillations in order to place them in similar experimental conditions than infants and measure spontaneous postural responses and kinematic adaptations. Electromyograms (EMGs) were recorded with a TELEMG multi-channel system (BTS, Milan, Italy). The EMG signals were pre-amplified (5,000×), analog bandpass- filtered between 20 and 450 Hz, acquired at a sampling frequency of 1,000 Hz and stored on a PC for off-line analysis. The three-dimensional positions of the markers were acquired at 100 Hz with an ELITE automatic motion analyzer (BTS, Milan, Italy). ## Analysis ### Emg The signals were bandpass filtered from 25 to 250 Hz using a fourth-order bidirectional (i.e., zero-lag) Butterworth filter, bandstop filtered from 57–63 Hz to remove any residual 60 Hz noise, and full-wave rectified. These pre- processed EMG signals were then filtered with a low-pass bidirectional Butterworth filter of order four with a cut-off frequency of 1 Hz, producing an integrated EMG. Note that this filtering outlined the envelope of the EMG pattern while it cancelled small non-recurrent changes in it, which were detrimental to subsequent analysis. Finally, the integrated EMG signals were re- sampled from 1,000 to 100 Hz so that they had the same length as kinematic signals. ### Kinematics The raw marker data were filtered with a fourth-order zero-lag Butterworth filter with a cutoff frequency of 10 Hz. The angular displacements of the segments head, upper- and lower-torso, lumbar and pelvis together with the angular displacement of the platform were afterwards obtained in the frontal plane from the filtered markers’ positions. ### Cross-Correlation The coupling between the periodic activation of the ES and T muscles and the platform rotation was determined using sample cross-correlation estimation : andwhere and *τ* denote the estimate of the correlation coefficient and the time lag between data sets *x* and *y* of length *N*, respectively; *x* representing the integrated EMG and *y* the angular displacement of the platform. Note that cross-correlation was normalized so that in case *x* and *y* would have been similar, the correlation coefficient at lag 0 would have been 1. Technically, Eq. 1 was solved using a fast Fourier transform (FFT) procedure, which was more efficient than solving it directly. The 100-s-long *x* and *y* data sets were first segmented into several overlapping segments that counted 2¹⁵ points (i.e., segments 32.768 s long). Then, 15-bit FFT algorithm was applied to these segments to generate segments’ periodograms that were averaged to obtain an ‘ensemble average’ power spectral density of improved statistical stability. Finally, cross-correlation was obtained by applying the inverse FFT to the cross-power spectral density function of *x* and *y*, which was obtained from the product of the FFT transforms of *x* and *y*. The maximum value of was used as the variable of interest to evaluate the strength of the association between the integrated EMGs and the platform movement, the higher the-value the stronger the association. Another variable of interest was the time lag *τ* at which the maximum-value occurred. A lag of 0 indicates a perfect in-phase relationship between the integrated EMG and the platform; a positive lag indicates that the integrated EMG is delayed with respect to the platform and inversely for a negative lag. Both a negative lag and a lag of 0 were interpreted as expressing postural anticipation. Specifically, each previous variable was averaged over the right- and left-side muscles for each subject. Moreover, cross-correlation estimations were also conducted between angular displacement of body segments and platform movement to determine which segments were most strongly related to the support. In all cases, the accuracy of the time delay calculation was 10 ms, due to the sampling frequency of 100 Hz. ### Marker displacement The displacements of the markers in the medial-lateral direction at the time of maximal rotation of the platform to the right and to the left were averaged over the trial to get information about the strategy of balance control. These displacements were calculated with respect to initial position before the platform start rotating, the lower the averaged displacement the less destabilized the upper body (i.e., a good damping of the perturbation). Given that averaged medial-lateral displacement strongly depended upon the subject’s upper-body height, group data were normalized by removing trends due to height. Specifically, marker data sets for each group of subjects were fitted in a least-squares sense using first-order polynomials. Detrended displacements, were then obtained from original data as:where the index *i* represents any given subject and is the mean value of the original data set. Therefore, the transformed medial-lateral displacements data were no longer correlated with upper-body height and were scaled to a similar range as the original data. ### Angular dispersion Further information on the damping of the perturbation about the body segments was obtained from the averaged angular dispersion (i.e., standard deviation of angular displacement) of the body segments at the time of maximal rotation of the platform. The lower the averaged angular dispersion of a given segment, the more attenuated the perturbation about it. ### Statistics Kruskal-Wallis one-way analysis of variances (ANOVAs) and Friedman repeated measures ANOVAs were used to examine differences in dependant variables between groups and repeated conditions, respectively. When ANOVAs yielded significant results, *post hoc* assessment was performed by Dunn’s multiple comparison tests and multiple comparison correction by a step-down Bonferroni-Holm procedure. One-sample Wilcoxon signed rank tests were also used to evaluate whether time lags *τ* differed from 0. Statistical analysis was performed using SigmaStat statistical software package v.4. The threshold of statistical significance was set at p\<0.05. In case of multiple comparisons, this threshold applied for the whole family of comparisons and the per-comparison significance was evaluated from corrected p-values. # Results ## Coupling between Integrated EMGs and Platform As can be seen on, the lags *τ* between EMGs and platform movement exhibited differences between groups, with a significant group effect revealed by Kruskall Wallis ANOVAs for both the ES muscle, H(2, N = 26) = 12.17, p = 0.002, and the T muscle, H(2, N = 26) = 11.67, p = 0.003. On the one hand, one-sample Wilcoxon signed rank tests indicated that the time lag for the ES muscle in infants at the pre-walking stage (2±10 ms) was not statistically different from zero, Z = 0.38, p\>0.05, while the time lags in walking infants (−68±67 ms) and young adults (−114±180 ms) were significantly lower than zero, Z = −0.72 and Z = −0.78, respectively, p\<0.05. This difference between groups was confirmed using pairwise multiple comparisons, with the time lag in infants at the pre- walking stage significantly greater than the time lags in walking infants and young adults, Q = 2.81, p\<0.01, and Q = 2.97, p\<0.01, respectively. Therefore, the activation bursts for the ES muscle and the turning points of the platform occurred at the same time in infants who had not attained independent walking yet whereas the bursts started before the time of occurrence of the turning points in the other two groups. This result indicated that walking infants and young adults had ability to activate ES muscle in anticipation of platform movements while infants who did not acquire independent walking only demonstrated premises in anticipation. In addition, a group effect was found on the peak correlationof the cross-correlation function, H(2, N = 26) = 9.54, p = 0.008, with a higher correlation in infants at the pre-walking stage (0.61±0.11) as compared to adults (0.38±0.22), Q = 2.97, p\<0.01. Thus, the strength of the covariation between the EMG and the platform movement decreased in adults. On the other hand, a different pattern of findings was observed for the T muscle. The one-sample Wilcoxon signed rank tests indicated that the time lag in infants at the pre-walking stage (87±84 ms) was statistically greater than zero, Z = 0.72, p\<0.05, while the time lag in walking infants (2±119 ms) did not differ from zero, Z = 0.38, p\>0.05, and the adults’ time lag (−127±91 ms) was significantly lower than zero, Z = −0.64, p\<0.05. Further examination of the data using pairwise multiple comparisons indicated that the two groups of infants exhibited significantly higher lags than adults, Q = 3.46, p\<0.01 (pre- walkers vs. adults) and Q = 2.45, p\<0.01 (walkers vs. adults). Thus, there was a clear anticipation of platform movements by the T muscle activity in young adults and a beginning ability in infants who already walked independently. Finally, no group effect was observed on peak correlation. ## Coupling between Segments and Platform Lags *τ* in all three groups were significantly higher than 0, meaning that segmental displacements followed platform rotation. Friedman repeated measures ANOVAs that were conducted on lag *τ* revealed significant differences in temporal coupling between the body segments and the platform in pre-walking infants, χ²(4, *N* = 10) = 35.84, p\<0.001, walking infants, χ²(4, *N* = 10) = 38.08, p\<0.001, and adults, χ²(4, *N* = 10) = 21.01, p\<0.001. Multiple paired comparisons showed that the turning points of the head and the upper-torso oscillations were more delayed than those of the lumbar segment and the pelvis with respect to the turning points of the platform in all three groups of subjects, 2.83\< Q \<5.11, 0.001\<p\<0.01. In addition, Friedman repeated measures ANOVAs that were conducted on the peak correlation revealed significant differences between body segments in pre- walking infants, χ²(4, *N* = 10) = 27.21, p\<0.001, walking infants, χ²(4, *N* = 10) = 33.68, p\<0.001, and adults, χ²(4, *N* = 10) = 15.61, p\<0.004. Specifically, multiple testing revealed that the coupling strength with the platform was larger for the lower body segments (i.e., pelvis and lumbar segment) than for the higher body segments (i.e., head and upper torso) in all groups, 3.11\< Q \<4.95, 0.001\<p\<0.01. Taken together, the above results thus indicated that the higher body segments were more independent from platform movement than the lower body segments. Moreover, Kruskall Wallis ANOVAs revealed a significant group effect for lag *τ* between platform movement and segmental movement for the lower-torso, H(2, N = 26) = 6.58, p = 0.037, the upper torso, H(2, N = 26) = 9.41, p = 0.009, and the head, H(2, N = 26) = 5.75, p = 0.043. Lags were higher in adults (lower- torso: 389±281 ms; upper-torso: 617±472 ms; head: 706±340 ms) than in infants at the pre-walking stage (lower-torso: 285±42 ms; upper-torso: 331±68 ms; head: 389±52 ms) as revealed by multiple comparisons, 2.41\< Q \<3.04, 0.01\<p\<0.05. Therefore, the higher body segments were more independent from platform movement in adults than in non-walking infants. Finally, there was no group effect on peak correlation for all segments. ## Normalized Marker Lateral Displacement Significant differences in displacement across markers were found for both infants at the pre-walking stage, χ²(4, *N* = 10) = 40, p\<0.001, and walking infants, χ²(4, *N* = 10) = 40, p\<0.001. Multiple paired comparisons revealed in both groups an increase of lateral displacement from lower markers to higher markers. Specifically, the lateral displacement of the head marker was found larger than that of the thoracic marker, which was also found larger than that of the pelvis marker, 2.82\< Q \<5.65, 0.001\<p\<0.05. Inversely, the markers of the young adults were all found at the same distance from initial position, with an insignificant Friedman test result, χ²(4, *N* = 6) = 5.86, p = 0.21. Therefore, the young adults remained rather upright, while the infants tilted in congruence with the platform with their upper body behaving like a rigid inverted pendulum. Kruskall-Wallis ANOVAs provided further information on the above difference between adults and infants regarding the ability to remain upright on the moving platform. While no difference between the three groups was observed for the lateral displacement of the sacrum, lumbar, thoracic and cervical markers, a group effect was found for the head marker, H(2, N = 26) = 7.88, p = 0.019. Multiple testing indicated that the lateral displacement of the head marker was significantly lower in adults as compared to infants who did, Q = 2.61, p\<0.01, and did not attained, Q = 2.53, p\<0.01, independent walking. Therefore, this result suggests that the adults adopted a head stabilization in space strategy while the infants did not. ## Segmental Angular Dispersion Within-group analysis of segmental dispersion showed differences between segments in infants at the pre-walking stage, χ²(4, *N* = 10) = 20.56, p\<0.001, and walking infants, χ²(4, *N* = 10) = 33.84, p\<0.001. On the other hand, no difference was revealed in young adults, χ²(4, *N* = 6) = 8.91, p = .07. In infants at the pre- walking stage, paired comparisons revealed that the angular dispersions of the head and upper-torso segments were significantly larger than that of the pelvis, Q = 3.67, p\<0.01, and Q = 3.54, p\<0.01, respectively. In walking infants, the angular dispersions of the head and upper-torso segments were also significantly larger than that of the pelvis, Q = 4.38, p\<0.01, and Q = 4.24, p\<0.01, respectively, and that of the lumbar segment, Q = 3.96, p\<0.01, and Q = 3.81, p\<0.01, respectively. Thus, the perturbations from the platform were attenuated about anatomical segments in young adults while they propagated as going up to higher segments in infants. Between-group analysis of the segmental dispersion indicated a significant group effect for all the segments: pelvis, H(2, N = 26) = 14.97, p\<0.001, lumbar segment, H(2, N = 26) = 14.49, p\<0.001, lower torso, H(2, N = 26) = 15.47, p\<0.001, upper torso, H(2, N = 26) = 14.44, p\<0.001, and head, H(2, N = 26) = 14.32, p\<0.001. Multiple testing indicated that the angular dispersions in infants at the pre-walking stage were significantly larger than those in adults for all segments, 3.92≤ Q ≤3.72, p\<0.01. The angular dispersions in walking infants were also significantly larger than those in adults for the upper segments (i.e., lower torso, upper torso, and head), 2.65≤ Q ≤2.86, p\<0.01. Accordingly, walking infants demonstrated an adult-like ability to attenuate platform perturbations only about lower segments. # Discussion Controlling actions prospectively is a major achievement for infants. Although anticipatory postural adjustments start developing as soon as the first year of life, there has been suggestive evidence that they become more consistent once independent walking is acquired. The present study was therefore intended to provide further evidence that independent walking experience plays a role in the development of anticipatory postural control. Specifically, it was expected that walking infants would show greater sophistication compared to non-walking infants in their adult-like ability to anticipate for postural disturbances induced by the moving platform. ## Anticipatory Postural Adjustments to Platform Rotation: What is Changed with Independent Walking and why does it Change? There were two main results with respect to anticipatory control, as evaluated from muscular adjustments intended to stabilize posture against platform disturbances. On the one hand, infants who did not walk independently demonstrated premises for anticipatory adjustments about the low back (i.e., ES muscle) only. On the other hand, infants who were already walking alone demonstrated ability for anticipatory adjustments about the low back and the neck (i.e., T muscle), including an adult-like ability at the former level and premises for anticipation at the latter level. Therefore, in line with what we had expected, independent walking is not a prerequisite for anticipatory postural control. However, once independent walking has been acquired, a robust, adult-like, ability to anticipate for postural perturbations emerges. These results thus complement previous works that have proposed independent walking as an essential transition point for the development of anticipatory postural control. It is noteworthy that anticipatory adjustments to platform rotation developed in a bottom-up fashion, occurring first at the low back and then developing from the low back to the neck from independent walking onwards. This finding questions the maturational viewpoint that supports a cephalo-caudal gradient in the development of postural responses during the first year of life. Indeed, experiments that have examined postural adjustments when perturbing the support surface or performing reaching movement rather reported that infants by the end of the first year of life prefer to activate neck muscles first, which afterwards disappear for a bottom-up recruitment as observed in sitting children and sitting and standing adults. This outcome difference may come from the fact that instead comparing postural adjustments in infants of different ages, we considered this issue by grouping infants based on the skills they were mastering (i.e., standing and walking). Considering change over time on such basis, Assaiante and Amblard proposed that postural control (i) becomes organized in a bottom-up way from the time upright posture is acquired, and (ii) aims at stabilizing the pelvis to minimize the displacements of the center of gravity from walking onset. As regards the latter point, we did observe a better stabilization (i.e., more attenuation of the perturbation as evaluated from angular dispersion) of the pelvis in walking infants and adults as compared to non-walking infants, supporting the idea that from the moment walking is acquired, postural control (here anticipatory) strongly stabilizes the low back. Thus, it turns out that the present results are in line with an experience- driven explanation for developmental progress, meaning that experience maintaining balance in a variety of postures (standing and walking) is a facilitator to the development of anticipatory postural adjustments. Interestingly, the moderate ability to compensate in advance for the perturbations about the pelvis before independent walking suggests that anticipatory postural control may be also a facilitator to it. In particular, anticipatory muscle activity is required during walking to stabilize pelvis and organize balance control. Therefore, walking may be a facilitator to anticipatory postural control and vice versa, reinforcing each other as proposed by Barela et al.. However, the significance of walking experience on the development of anticipatory postural control should be considered with caution since most of the non-walking infants were younger than the walking infants, so that age was confounded with walking experience in the study. Thus, disentangling the role of age from that of walking experience is critical to definitely establish a link between walking and developmental changes in anticipatory postural control. Previous studies on the role of crawling experience in changes in cognitive, social and emotional development are instructive in this regard (see for a review). These studies have established the experimental designs needed to isolate the role of locomotor experience as a facilitator of development, including for example age-held constant designs (i.e., holding infant age constant while examining variations in walking experience) and lag-sequential designs (i.e., permitting the assessment of the role of age, locomotor experience, and their interaction). Therefore, future studies should use designs of this nature to infer the role of independent walking in the development of anticipatory postural adjustments. ## ‘Balancing’ Behaviour to Platform Rotation: What Sort of Pendulum? An important result was that infants, even the walking ones who demonstrated clear anticipatory adjustments about the low back, did not remain upright, tilting with the platform (i.e., a tight coupling to the platform). This behaviour was reflected in lateral displacements and time lags between the platform and the body segments that progressively increased as going up to the most distal body segment (i.e., the head). Inversely, adults remained more upright and more independent from platform rotations with respect to their distal body segments. Indeed, lateral displacements remained similar for all markers and the time lags between the platform and the lower-torso, the upper- torso, and the head were rather large, reflecting an articulated operation of the head-trunk unit. These findings tend to demonstrate that infants behaved as rigid inverted pendulum with a mass and a spring, its base being attached to the rotating platform, while upper body movements in adults rather resembles that of a non-rigid, non-inverted pendulum rotating about the mass. In other words, while infant were following the platform and were thus counteracting body inertia too late when the platform already went through its turning points, adults learned to counteract body inertia between the two end-points of rotation of the platform to stabilize head in space. This behaviour in adults agrees with previous studies where subjects who were standing on a continuously anterior- posterior translating platform were damping perturbations by stabilizing trunk and head to create a pivot for the passive displacements of the lower limbs, entrained by the platform displacement. However, what are the origins of such difference between infants and adults? Corna et al. reported a transition from a non-rigid, non-inverted pendulum to a rigid inverted pendulum strategy when visual inputs were removed in adults standing on a moving platform, concluding that the central nervous system may use visual information in a feedforward manner to reduce head oscillation in space. Isableu et al. also evidenced that visual field dependant subjects adopted an “en bloc” functioning (i.e., behaved like rigid inverted pendulum) in the absence of adequate visual cues when remaining upright, suggesting that they blocked their joints due to difficulty in using corresponding proprioceptive cues. Although we did observe inverted pendulum behaviour in infants, neither an inability to perceive optical flow nor an inability to use proprioceptive cues or an inability to generate anticipatory motor commands from both sources of information likely explain difference with adult behaviour. First, there is extensive evidence in the literature that infants in the second half of the first year of life are able to perceive optical flow and scale their postural responses to the visual information. Besides, infants should have been also able to interiorise the platform rhythm on the basis of the other sensory inputs, namely the vestibular and the somatosensory inputs. Indeed, it has been established that the vestibular system, which has an important role in balance control, is already very sensitive during the first 6 months of life. With respect to somatosensation, Hadders-Algra et al. demonstrated that compensatory muscle activations, likely triggered by somatosensory signals from the pelvis, already occurred in 5-month-old sitting infants to bring upper body back to vertical once destabilized by a moveable platform. Second, we observed in both pre-walkers and walkers an ability for postural anticipation about muscles, which reflects their ability to integrate multiple inputs (here, visual, somatosensory and vestibular inputs) and use them feedforwardly. Therefore, a remaining explanation for the infants to behave like a rigid inverted pendulum, not being able to stabilize upper-body and head in space, is that priori knowledge of the platform effect on sitting posture was not accurate, so that motor commands sent to the musculature to maintain the upright posture were not appropriate to platform perturbations (i.e., an inappropriate scaling between platform effect and postural command). This latter outcome likely results from immaturity of the central nervous system to update internal model that integrates both body characteristics and external perturbation, as discussed below. ## Building Internal Model Combining Postural Experience and External Perturbation When movement prediction depends on both the individual body and an external object, as here the platform, not only does the central nervous system build an internal model of the object but it also integrates the dynamics of that object into an internal model of the sensorimotor system as a whole. In this view, the anticipatory adjustments observed about the low back muscle in infants having not walked yet indicate that this process already took place before independent walking, although it remained importantly immature. In walking infants, this internal model of the sensorimotor system and platform appeared to be more accurate. Anticipation occurred about both the low back and neck muscles and there was also kinematic adjustment at the low back level that attenuated, in an adult-like way, the platform perturbations. Thus, it is likely that the variety of everyday walking experience enriched the internal model of the sensorimotor system and facilitated integrating the dynamics of the platform within it, although this latter element remained difficult as suggested by the infants’ inability to remain upright. Taken together, results from the present study thus support the idea that postural behaviours underlying motor skills share a similar internal model for postural control. This is in contrast to studies by Adolph, who rather concluded on the absence of sensorimotor learning transfer between the postural milestones (e.g., sitting, standing, crawling, and walking). On the other hand, our suggestion accords with a recent work by Chen et al., which emphasized that any new behavior is built upon some of the basic control elements used in managing the previously learned behavior, enriching the existing internal model. In conclusion, our study indicated that although independent walking is not a prerequisite for postural anticipation, adult-like anticipation ability seems to emerge once this skill is acquired. However, other research designs controlling for age effects are needed to firmly establish independent walking *per se* as a facilitator to the development of anticipatory postural control. With regard to the behaviour adopted by the subjects to maintain balance during the continuous rotation of the platform, infants contrary to adults had difficulties to stabilize upper body and head in space, likely due to an inaccurate estimation of the platform effects on posture. These findings support the idea that postural behaviour relied on an internal model that included the sensorimotor system and the platform, with difficulty in infants for integrating the dynamics of the platform within the model. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: CA. Performed the experiments: FC MZ MV CA. Analyzed the data: FC MZ MV CA. Contributed reagents/materials/analysis tools: FC MZ MV CA. Wrote the paper: FC MZ MV CA.

# Introduction Transcranial ultrasound stimulation (tUS) has gained attention in the past years as a potential new tool for noninvasive brain stimulation (NIBS). tUS has higher spatial precision compared to other NIBS techniques such as transcranial magnetic stimulation (TMS) and transcranial electric stimulation (TES), which presents a possibility of improved targeting. Furthermore, tUS can deliver its energy much deeper while maintaining focal precision—deeper than TMS or TES. Previous studies have demonstrated that ultrasound is capable of stimulating central structures in animals, peripheral nerve pathways in animals and humans, the retina, and intact brain circuits in animals. However, the number of human tUS studies thus far is limited. In primary somatosensory cortex, tUS has been shown to modulate touch discrimination, induce localized somatosensations when targeting the cortical hand representation, and induce changes in intrinsic and evoked EEG dynamics. In primary visual cortex, tUS can induce individual visual phosphenes, percepts of a flash of light, that were accompanied with an evoked potential and blood-oxygenation-level-dependent (BOLD) contrast similar to those seen with photic stimulation. Likewise, the effects of tUS in primary motor cortex (M1) in humans are a focus of active investigation. Much of our understanding of motor cortex stimulation comes from TMS investigations, where it has been the standard for noninvasive M1 perturbation for decades. Suprathreshold single-pulse TMS of M1 induces contraction in the corresponding muscles, and electromyography (EMG) allows for the quantification of these motor evoked potentials (MEPs). Since MEP size increases as a sigmoidal function of TMS intensity above motor threshold, MEP strength is frequently used as an indicator of corticospinal excitability, both in neuromodulatory and behavioral interventions. This is supported by pharmacological evidence that shows motor threshold, the TMS intensity needed to elicit an MEP, is a proxy for the within-subject excitability of the cortico-cortical axons affected by the induced current of TMS pulse. We investigated the neuromodulatory effects of tUS on M1 by analyzing its effect on TMS-evoked MEP. Cortical silent periods (cSP) are a phenomenon of suppressed EMG activity during tonic contraction following single-pulse TMS of the corresponding M1 motor representation. cSPs typically last from 100–300 ms. Importantly, cSPs are considered to be driven predominantly by cortical inhibition from \~50 ms after instigation. Specifically, pharmacological evidence suggests that the cSP effect is mediated by GABA receptor-dependent postsynaptic inhibition. When elicited during tonic contraction of the contralateral hand, cSPs are reported to be observable either following an MEP or without inducing an MEP, at subthreshold TMS intensities. As such, cSP provides a valuable method of investigation of inhibitory mechanism in motor cortex. As a whole, the field is still building its understanding of what, if anything, tUS can affect via M1 stimulation. To date, no tUS study has been able to induce motor contraction through human M1 stimulation. Given that change in motor contraction strength has been the benchmark for M1 modulation studies, the established capabilities of TMS have been leveraged to investigate tUS effects on M1. For example, tUS of the hand area reduces the size of motor evoked potentials (MEPs) evoked by concurrent and concentric TMS. A separate study reported a lasting increase in the size of MEPs after exposure to an ultrasound imaging device. Additionally, tUS of M1 alone was shown to affect reaction times in two motor tasks. Because of its physiological underpinnings, the cSP is in a unique position to be leveraged as an externally detectable phenomenon to better understand tUS effects on M1. Specifically, tUS has been proposed to preferentially affect inhibitory interneurons, feeding well into cSP’s existence as a interneuron- facilitated phenomenon. Additionally, since cSPs have been reported to occur without a preceding MEP, tUS’ apparent inability to instigate an MEP does not preclude its use to attempt induction of cSPs. But as of yet, it is unknown if tUS can engage the necessary inhibitory circuits to instigate a cSP. We addressed this by performing tUS of M1 on participants executing voluntary muscle contraction and analyzing the EMG data from the contracted muscle. # Methods Two experiments were performed in this study. In Experiment 1, we measured how tonically contracted hand muscles respond to single-pulse TMS and single-burst tUS of M1. We performed separate trials using tUS and TMS. In Experiment 2, we measured if cortical excitability changes following tUS exposure. Excitability was gauged using single-pulse TMS before and after tUS exposure. ## Data acquisition ### Participant demographics Research participants were right-handed with no neurological conditions. Due to TMS use, subjects with an increased risk of seizure were excluded. Due to MRI use, subjects with MR-incompatible implants were excluded. Participants were 18 to 42 years old, A subset of Experiment 1 participants (n = 10; mean: 25.9 years) participated in Experiment 2 (n = 8; mean: 26.75 years). Note that subject ID numbers are not sequential since other recruited subjects were used in a different study. Participants provided written consent before participating, and the study protocol was approved by the Institutional Review Board of the University of California, Los Angeles (IRB#17–000958). ### EMG and NIBS placement Electrode sites were cleaned with abrasive skin prep gel (Nuprep) and alcohol wipes. A surface EMG electrode (two 10 x 1 mm contacts; 10 mm spacing) measured the right first dorsal interosseous (FDI) muscle activity, and the signal was amplified (x1000) (Delsys Inc., Boston, MA) and sampled at 5000 Hz. The surface electrode was additionally secured to the finger with medical tape. A wide ground electrode was placed on the back of the hand. EMG was recorded for 1-second epochs around NIBS (both TMS and tUS) onset. A structural MRI (T1-weighted; 0.8 x 0.8 x 0.8 mm voxels) was acquired in a previous visit, and each structural MRI was registered to standard space for NIBS targeting of standard-space coordinates (Montreal Neurological Institute, MNI; 1-mm atlas). Registration was performed in FSL (the FMRIB Software Library) on brain volumes extracted using the optiBET tool for FSL’s BET. NIBS position with respect to the subject’s head was tracked using neuronavigation software (Brainsight, Rogue Research, Montreal, QC) loaded with the subject’s MRI. The neuronavigation software was prepared with pre-determined trajectories, which were the shortest Euclidean distance from the scalp to a set voxel as determined by a custom MATLAB script (Mathworks, Inc., Natick, MA). At the beginning of each NIBS session, five single TMS pulses were given at each target of a 6-target grid over the left motor cortex using MNI space. The grid’s origin was placed at MNI coordinates that correspond to M1_hand as based on a meta-analysis of fMRI motor experiments: x = −39, y = −24, z = 57. This coordinate corresponds morphologically to the cortical ‘hand knob’. The other five targets on the grid were in a 12 voxel-width grid (9.6 mm grid interval) anterior, posterior, and medial, anteromedial, and posteromedial from the M1_hand coordinate in subject space. The targets that elicited the largest, second-largest, and third- largest average MEPs were used as placement points for the NIBS devices. These three positions are referred to below as “TMS target”, “2^nd-best”, and “3^rd-best” targets respectively. For all TMS trials, the TMS coil was oriented with the handle pointed backwards and angled 45° from midline. ### Experiment 1, cSPs Participants moved their index finger laterally during trials to maintain consistent FDI contraction across trials, as monitored by a digital scale (20–40% maximum voluntary contraction, depending on the subject). NIBS was delivered during contraction. Percent maximum contraction varied between subjects so that every subject maintained a comparable level of EMG activity (\~150–200 peaks per second). 20 trials were performed at each of three TMS intensity levels: 90%, 100%, and 110% of % active motor threshold (aMT) (60 TMS trials total per subject). 20 tUS trials were performed for each of the following four parameters: 300-ms burst duration at the TMS target, 300-ms burst duration at the 2^nd-best target, 300-ms burst duration at the 3^rd-best target, and 500-ms burst duration at the TMS target (80 tUS trials total per subject). Subjects were told to use the feedback of the digital scale display to maintain their target FDI contraction force, and subjects were cued to relax between trials to avoid fatigue. Subjects monitoring their contraction level also had the benefit of keeping subject attention constant across trials in Experiment 1, since attention can affect EMG measurements. Trials had a jittered intertrial interval of 10 ± 2 s. The order of tUS vs. TMS- cSP blocks were not counterbalanced since counterbalancing Experiment 1 blocks would have conflicted with acquisition of Experiment 2. ### Experiment 2, Cortical excitability MEPs were measured with the subject’s hand relaxed. TMS was delivered at the same suprathreshold intensity for both “before” and “after” conditions within subjects. TMS was set to 110–120%rMT (percent resting motor threshold), (110%: n = 1; 115%: n = 4; 120% n = 3). %rMT was varied across subjects to assure that each subject had consistent MEP sizes. 20 MEPs were acquired both before and after the tUS exposure protocol. tUS exposure protocol was the same as described in “Experiment 1, cSPs:” (80 total trials: 20 trials each: 300 ms at TMS target, 300 ms at 2^nd-best, 300 ms at 3^rd-best, 500 ms at TMS target). Trials had a jittered intertrial interval of 10 ± 2 s. Participants of Experiment 2 (n = 8) also participated in Experiment 1 (n = 10). For participants of Experiment 2, Experiments 1 and 2 were performed during the same visit. Performing Experiments 1 and 2 at the same visit with the same participant allowed us to minimize the number of volunteers who would be exposed to tUS. ### tUS equipment The tUS device used was a 500-kHz focused piezoelectric transducer (Blatek Industries, Inc., State College, PA). The transducer was manufactured with a face width of 3 cm and a focal point of 3 cm. When measured in water, the transducer produced a focus (-3 dB focus area) centered at 29.6 mm that was 2.8 mm wide and 20 mm long (-6 dB focus: centered at 33.7 mm, width 6 mm, length 33 mm). The transducer was housed in a custom 3D-printed handle, and an infrared tracker was mounted to the housing for neuronavigation. The transducer was driven by 500-kHz sine-wave voltage pulses from a waveform generator (33500B Series, Keysight Technologies, Santa Rosa, CA) and voltage pulses were amplified by a 50-dB radio frequency amplifier (Model 5048, Ophir RF, Los Angeles, CA). A 3-dB fixed attenuator was attached in line following the amplifier. Subjects’ hair was parted at the location of the tUS targets to reduce the hair’s effect on acoustic propagation, and ample ultrasound gel was used to acoustically couple the transducer to the subject’s head. The cooling fans of the server rack-mounted radio frequency amplifier were quite loud, which acted as an auditory mask throughout the duration of tUS session. Cooling fan noise was constant throughout the duration of all tUS sessions. tUS bursts were pulsed with a 1-kHz pulse repetition frequency and a duty cycle of 36%. Each burst lasted either 300 or 500 ms (tUS on for 108 or 180 ms total). Transducer output was confirmed via measurements made via hydrophone in degassed water (1 mm, Precision Acoustics Ltd, Dorchester, UK). Transducer output was set to produce an intensity of 15.48 W/cm² in degassed water (spatial peak, pulse average; I_sppa); spatial peak, temporal average (I_spta): 5.57 W/cm². Using previously measured attenuation of a 500-kHz ultrasound through human skull (-12 dB to -16 dB), these parameters produce an estimated intracranial I_sppa of 3.9 W/cm² (I_spta = 1.4 W/cm²). This is within safe levels (max Mechanical Index: 0.8) and is within intensities of previous human tUS studies. ### TMS equipment Single-trial, monophasic TMS was applied using a figure-eight coil (70 mm diameter) via a Magstim 200² magnetic stimulator (Magstim, Whitland, Dyfed, UK). An infrared tracker was mounted to the TMS coil for neuronavigation. Individual resting and active motor thresholds were determined using simple adaptive PEST (SA-PEST) (Adaptive PEST TMS threshold assessment tool, Brain stimulation laboratory, Department of Psychiatry, Medical University of South Carolina). Trials had a jittered intertrial interval of 10 ± 2 s. ## EMG data analysis ### EMG post-processing EMG traces were high-pass filtered with a 10-Hz cutoff (filter transition: 5–10 Hz), unless noted as unfiltered. This high-pass filter was applied to remove voltage shift and low-frequency noise. All EMG post-processing was performed using MATLAB. ### cSP measurement cSPs were measured using an automated script written in MATLAB, which used a rolling standard deviation (STD) to see when EMG activity quieted below a threshold. The rolling STD window had a width of 3 ms. The cSP threshold was set using the baseline EMG variability—specifically ½ STD of the rolling STD trace the 200-ms period before TMS onset. cSP onset was set to the timepoint the rolling STD first fell below threshold after the MEP peak. cSP offset was set where the rolling STD first rose back above threshold. This method is similar to other published cSP algorithms, albeit using rolling STD (i.e. sliding window STD) instead of the conceptually similar ‘mean consecutive difference’. Additional rules were necessary to handle certain edge cases the algorithm would otherwise not handle as intended. For determining cSP starts, starts were contingent on the raw EMG being near or below zero (specifically, below the same threshold value mentioned above). If no cSP onset point was found within the first 15 ms, the duration was set to zero. For determining cSP offsets, a 15-ms ‘amnesty period’ was included to avoid spuriously classifying MEPs as cSP offsets, since large MEPs could have high STD values in their valleys after the MEP peak. For these cases in which putative cSP duration was below 15 ms, the initially calculated offset was ignored if the EMG signal was quiet immediately following that point (i.e. below threshold 15–30 ms after cSP onset). If not, the initial threshold breach was deemed valid. Determination of cSP onsets or offsets being contingent on EMG being silent for a set period time has been used by other published cSP algorithms as well. ### MEP measurement, resting To quantify the size of MEPs Experiment 2, we used the area of the MEPs. The area under the curve (AUC) of the rectified MEP waveform from 20 to 120 ms after the TMS pulse was estimated via the trapezoidal method in MATLAB. AUC was not used in Experiment 1 (voluntary contractions) because of surrounding EMG activity. ### MEP measurement, voluntary contraction MEP peak-to-peak measurements were measured by the absolute height from the peak to the mean of the two flanking valleys of the peak. Calculation from the mean of both flanking valleys was chosen to reduce the influence of voluntary EMG activity on TMS MEP measurement—especially important in this experiment given we induced small, near-threshold MEPs. To improve accuracy of automated MEP detection during tonic contraction, MEP search was constrained using per-subject exemplar data from trials with overt MEPs. A 10-ms search window was centered around the expected MEP timepoint. Expected MEP timepoint was the median MEP timepoint during resting TMS MEP trials (Experiment 2 TMS MEP data). For the two subjects who did not participate in Experiment 2, trials with visually overt MEPs during tonic contraction were used instead (Experiment 1 TMS MEP data). This search approach was used both for the positive MEP peak and its two flanking negative valleys. Candidate peaks and valleys in the EMG data were found using the *findpeaks* MATLAB function. To avoid minor, extraneous peaks from being selected, only peaks with a prominence and width above the 50^th percentile were eligible. TMS cSP trials were categorized into three labels: *MEP*, *stub*, and *none*. Overt MEPs (*MEP*) occurred within the 10-ms search window and had a peak-to- peak height above 0.5 mV. Potential-but-short MEPs (*stubs*) occurred within the 10-ms search window and had a peak-to-peak height below 0.5 mV. Trials with no peak that met conditions (50^th percentile prominence, width) within the 10-ms search window were labeled *none*. ### EMG characteristics Additional characteristics of the EMG traces were calculated to contrast TMS and tUS effects on the FDI EMG signal during voluntary contraction, as well compare within different periods of tUS trials. First, spectral components were determined by estimating the short-term, time-localized power spectrum of each trial and then taking the mean to get separate average spectrograms for TMS trials and tUS trials. Second, lengths of silences in the EMG signal were calculated with the same sliding window approach used to determine cSP onset and offset (see cSP measurement). Specifically, the cSP algorithm searched for a silence duration from a window centered at 0.001-second intervals from 0.05 to 0.95 s. The first and last 0.05 s were excluded to avoid edge artifacts. The results were then averaged within their respective groups to get mean silence traces. Two additional characteristics were calculated to investigate possible EMG responses time-locked to tUS exposure: the height of the EMG (AUC) and the rate of EMG peaks. AUC was calculated as described above (MEP measurement, Resting) for two 150-ms epochs: from 200 to 50 ms before tUS onset and from 50 to 200 ms after tUS onset. This provides two 150-ms epochs wholly covered by ‘off’ and ‘on’ periods of tUS. Rate of EMG peaks was calculated using *findpeaks* function in MATLAB on each EMG trace, binning peaks by time for the tUS-off or tUS-on periods (i.e. both the pre- and post-tUS periods were included together for tUS- off). All EMG characteristics processing was performed in MATLAB. ## Acoustic simulation ### Skull mask processing The same T1-weighted structural scan used for neuronavigation was used for simulations and skull thickness measurements. Specifically, we used a MPRAGE sequence (slice thickness = 0.8 mm, repetition time = 2500 ms, echo time = 3.6 ms, inversion time = 1000 ms, dimensions = 300 x 320 x 208 voxels, scanner = Siemens Prisma 3T). For acoustic simulations and skull thickness measurements, binary skull masks were produced in BrainSuite using its “Cortical Surface Extraction Sequence”. Skull masks were corrected by hand with the mask brush tool in BrainSuite. In MATLAB, skull masks were linearly interpolated to increase resolution to 0.2-mm- width voxels, and they were rotated such that the tUS trajectory was in line with the computational grid. Masks were also smoothed via morphological image processing both before and after transformation. Masks were cropped to the area of interest, creating a 484 x 484 x 484 volume. Additional skull processing details can be found in documentation for the toolbox, TUSX, made by these authors. ### k-Wave simulations Acoustic simulations were performed using k-Wave, an open-source acoustics toolbox for MATLAB. Each skull mask was imported into k-Wave, providing a computational grid spacing of 0.2 mm. To simulate the transducer, we set a curved disc pressure source (k-Wave function: *makeBowl*) with a curvature radius of 30 mm and aperture of 30 mm to mirror the focal length and width of the real transducer, respectively. The pressure source emitted a 0.5 MHz sine wave, resulting in a grid points per wavelength (PPW) of 14.8. Simulations were performed at a temporal interval of 285 temporal points per period (PPP) for a Courant-Friedreichs-Lewy (CFL) number of 0.0519. Perfectly matched layers (PML) of 14 grid points were added for a total grid size of 512 x 512 x 512. To allow comparison to real-world pressure measurements in the water tank, each tUS trajectory was simulated twice: once to simulate propagation through the skull and once to simulate propagation through water. For skull simulations, the same acoustic properties were given for all points within the skull mask: density of 1732 kg/m³, a sound speed of 2850 m/s, and an alpha coefficient of 8.83 \[dB/(MHz^y cm)\]. The use of homogenous skull acoustic properties has been shown to be effective in simulations within the frequencies used here. All values not within the skull mask were given bulk acoustic values of brain: 1546.3 kg/m³, a sound speed of 1035 m/s, and an alpha coefficient of 0.646 \[dB/(MHz^y cm)\]. Homogenous water simulations were given acoustic properties of water at 20°C: a density of 998 kg/m³, a sound speed of 1482 m/s, and an attenuation constant of 2.88 × 10⁻⁴ \[Np / m\]. An alpha power of 1.43 was used for all simulations. To estimate in-brain pressures experienced by participants for a given tUS trajectory, we used a ratio of pressures from the skull and water simulations. The estimated pressure (*P*_*est*.) at a given location was calculated as $$P_{est.} = \frac{P_{skull\mspace{360mu} sim.}}{P_{water\mspace{360mu} sim.}} \times P_{water,real}$$ Where *P*_{*skull sim*.} is the temporal maximum pressure value at that same location in the skull simulation for the specific subject and trajectory. *P*_{*water sim*.} is the temporal maximum pressure value at the focal point of the water simulation. *P*_{*water*,*real*} is the temporal maximum pressure value at the focal point measured in a water tank of degassed water using the same parameters as used in the experiment (see tUS equipment). To avoid any potential outliers in the simulated data, spatial averaging was performed on *P*_{*skull sim*.} and *P*_{*water sim*.} by taking the mean within a 0.6-mm radius sphere. *P*_{*water*,*real*} was 1.40 MPa for all subjects except one (sbj11), whose *P*_{*water*,*real*} was 1.13 MPa due to the lower waveform generator setting used for that session (user error). Simulations were performed on the Ahmanson-Lovelace Brain Mapping Center computational cluster. Each simulation instance was allocated 24 CPU cores and took approximately 2.5 hours with the C++ implementation of k-Wave (kspaceFirstOrder3D-OMP). ### Target registration NIBS targets and the location of M1_hand were determined via registration to standardized stereotactic space (Montreal Neurological Institute, MNI). Registration was performed with FSL’s FNIRT/FLIRT tools. M1_hand was set to the voxel closest to the MNI coordinates x = −39, y = −24, z = 57. ### Exposure An estimate of cumulative M1_hand exposure was made by multiplying the individual peak pressure at the M1_hand voxel for each of the three tUS trajectories by the time the tUS device was on for that location. Specifically, exposure was defined as $$\sum\limits_{traj = 1}^{n}{P_{traj} \times Time_{traj}}$$ for ***n*** number of tUS trajectories, where ***P***_***traj*** is the pressure at the M1_hand voxel for that trajectory, and ***Time***_***traj*** is time tUS was on for that trajectory. We display these values in the form Pascal-hours (Pa⋅hr). ## Statistics ### Experiment 1, cSPs TMS cSP durations vs. aMT was analyzed with a one-way repeated-measures ANOVA. For post-hoc tests, Welch’s t-tests were performed on the group distribution pairs (90 & 100, 90 & 110, 100 & 110) using the cSP durations demeaned to their respective subject mean (*Duration*–*Subject Mean*). Rate of EMG peaks on vs. off during tUS was analyzed using a paired t-test, with the rates ‘on’ rate and the ‘off’ rate for each trial paired. In other words, the tUS-off portions of the 1000-ms epochs served as the control. ### Experiment 2, Cortical excitability M1_hand excitability was analyzed with a paired ranked non-parametric t-test since the distribution was non-gaussian. We performed this using resampling using a script in R. For null hypothesis testing, permutation was used to create a null distribution of all permutations of before- and after-tUS values swapped within subjects (256 permutations). All 16 medians of each permutation (8 subjects, 2 conditions) were then ranked against one another. The difference of the means of the permuted group ranks for each permutation was used as the values of the null distribution. The value of p equaled the number of permutations in which the absolute difference of the mean ranks was greater than the real absolute difference of the mean ranks. Confidence intervals were calculated using the bootstrap method (10,000 bootstrap samples). Each bootstrap sample was made by sampling with replacement the 16 real median MEP sizes (8 subjects, 2 conditions). The 95% confidence intervals were set as the 2.5^th and 97.5^th percentiles of the bootstrap samples’ differences of the group means. To investigate any association between cortical excitability change and total tUS exposure of M1_hand, an estimate of total tUS exposure for a participant was calculated with the following formula: $$\sum\limits_{traj = 1}^{n}{P_{traj} \times Time_{traj}}$$ Where ***n*** is the number of tUS trajectories used with that participant, ***P***_***traj*** is the pressure (estimated) at M1_hand voxel for that trajectory, and ***Time***_***traj*** is the time the tUS device was on for that trajectory. Determination of ***P***_***traj*** is outlined in k-Wave simulations. The spearman correlation coefficient (r_s) was calculated for these values. ### cSP null distribution To bootstrap a null distribution of cSP lengths our automated cSP algorithm would find if applied to null, non-cSP data, we used a sliding window approach on non-cSP data. This data was real EMG traces collected during tonic contraction by the same subjects and sessions as Experiments 1 and 2—specifically, the one-second tonic contraction trials collected during tUS exposure. tUS trials were deemed valid as null EMG traces since we saw no change in EMG traces between tUS on vs. tUS off (Figs and, – Figs). For every null trial, the cSP algorithm searched for a silence duration from a window centered from each 0.001-second interval from 0.05 to 0.95 s. The first and last 0.05 s were excluded to avoid edge artifacts. All trials, subjects, and sliding-window increments were grouped into a single distribution (686,457 sliding window samples). # Results ## Experiment 1 ### TMS cSPs Single-pulse TMS was performed over left M1_hand during tonic contraction of the FDI muscle (n = 10). TMS was delivered at 90%, 100%, and 110% aMT. cSP duration increased with TMS intensity (ANOVA: F_2,16: 26.31, p \< 0.001; Welch’s t-tests: p \< 0.001, all pairs). The aMT threshold for one subject (sbj11) was set mistakenly low, resulting in TMS intensities lower than intended and therefore elicited very few cSPs. We also examined the size and presence of MEPs preceding cSPs. For trials with overt MEPs, the lengths of the subsequent silent periods were noticeably longer than would be seen by chance. Among trials with a peak within the expected 10-ms time window, trials with a peak smaller than the standard MEP peak-to-peak amplitude threshold of 0.5-mV, henceforth referred to as “stub” trials, mostly showed silences within lengths that would be seen by chance. However, some “stub” trials did show long silence durations on par with those of overt MEP cSPs. Lastly, trials in which there was *no* peak within the 10-ms time window showed silence durations within what would be seen by chance, with only one of these trials showing a silence above the 95^th percentile of the null distribution. ### tUS Single-burst tUS was performed over left M1_hand during tonic contraction of the FDI muscle (n = 10). The 300-ms or 500-ms bursts were delivered at three trajectories per subject, one of which was also the trajectory for TMS. No overt silent periods were visible during single trials of tUS stimulation. To investigate whether tUS caused any suppression of the EMG trace, we investigated the height of the EMG traces (area-under-the-curve, AUC), the lengths of the intermittent contraction silences, the rate of EMG peaks, and the spectral components of the EMG signals. While a drop in signal power of the spectral components occurs due to TMS-induced cSP, no spectral changes are visible during tUS trials. The same disparity is seen comparing the length of silences in the EMG signal, with a clear rise in mean silence period in response to TMS-induced cSP but no change in response to tUS. For comparisons performed within tUS trials, the height of the EMG traces showed no difference directly before versus after tUS onset (150-ms epochs before vs. after tUS). The rate of EMG peaks while tUS was on vs. off also showed no significant difference ( and Figs), with a paired t-test confirming there was only a small but statistically insignificant tUS effect on rate of EMG peaks (Delta: -0.91 Hz; 95% CI: -1.99, 0.16 Hz; p = 0.095). ## Experiment 2 ### Cortical excitability Cortical excitability was gauged before and after exposure to tUS by recording MEPs from single-pulse TMS over M1_hand (n = 8). Both the pre-tUS and post-tUS measurements (1-min post-tUS) consisted of 20 suprathreshold TMS trials with an intertrial interval of 10 ± 2 s. The size of TMS-induced MEPs did not vary between before- and after-tUS conditions, according to a ranked paired non- parametric t-test (Delta: -0.64 mV-ms; 95% CI: -2.39, 0.84 mV-ms; p = 0.51). ### Exposure & excitability To investigate the variability that was present among cortical excitability responses, we compared subjects’ cortical excitability change to total estimated tUS exposure in the session. tUS exposure estimates were made using acoustic simulations in models that matched each experimental setup, with skull data computed from structural MRI of the tUS participants. These data showed no obvious correlation between M1_hand exposure and cortical excitability change (n = 8), with a spearman correlation coefficient of -0.21. ## Acoustic simulation Acoustic simulation results suggest we very accurately ‘hit’ targets we were aiming at. tUS produced pressures in an ellipsoid focus, with a mean FWHM with of 4.5 mm. # Discussion ## Experiment 1 ### No tUS-MEPs We were unable to elicit tUS-induced MEPs at safe intensities. This was the expected outcome, given the lack of MEPs in previous human tUS studies. Some recent animal model work suggests that motor activation via ultrasound stimulation of motor cortex may not be feasible, proposing that previously reported motor contractions in anesthetized animals possibly relied on auditory mechanisms; however, others suggests motor activation via ultrasound stimulation may still be possible. We did not investigate for potential resting tUS-MEPs during rest beyond a single pilot subject. No tUS-induced MEPs appeared during active contraction trials either. ### Cortical silent period, TMS Our data found no cSPs that occurred without a preceding TMS-evoked MEP , despite structuring the study to facilitate a high prevalence of near-threshold MEPs. As such, this contradicts claims in the literature that TMS-induced cSPs can occur without an MEP. One explanation for this difference is that a small “stub” MEP could precede reported “MEP-less” cSPs, with its amplitude not surpassing the amplitude of tonic contraction. This is supported in these data by the consistent appearance of an EMG peak within the latency window expected for TMS-evoked MEPs. If this MEP-cSP dependency is true, this could suggest that cortical silent period is dependent on the recruitment of M1 motor units. TMS preferentially depolarizes axons. While still early in investigation, tUS conversely has often been proposed to preferentially affect inhibitory neurons—though this is certainly not universal. If these assumptions are true, this could explain why tUS struggled to silence corticospinal output in these data. To be clear regarding the notion of “stub” trials within our TMS data, we do not believe all TMS trials classified as a “stub” by the algorithm are MEPs. Rather, we believe there are two underlying distributions that fall under the “stub” designation. The first: trials in which there is a TMS-evoked MEP that is shorter than the standard threshold (0.5 mV). The second: trials in which there was an EMG peak produced *by chance*—created by a peak in tonic muscle EMG activity that fell within the expected time window. We must also note: For cSP length determination, we took a conservative approach on brief EMG activity flanked by periods of silence, referred to in the literature as late excitatory potentials (LEPs). Of the two silent periods flanking an LEP, we included only the first silent period when measuring cSP duration. Since these LEPs appear heuristically as short EMG disruptions of a longer cSP, a visual inspection of the data suggests that ignoring these LEPs would have resulted in less variable cSP durations within blocks. For comparison, these LEPs have at times been ignored in past by-hand cSP measurements. ### Cortical silent period, tUS Single-burst tUS of the hand area in the left motor cortex did not affect tonic muscle contraction of the FDI muscle. Specifically, there were no deviations in gap duration between tonic muscle spikes, spectral components, or prevalence of EMG peaks ( and Figs). This is in sharp contrast to the lengthy silent periods from single-pulse TMS. This is also in contrast to a growing number of studies showing inhibitory effects by tUS. While these data revealed no time-locked tUS effects, recent work using spatially overlapping TMS and tUS devices have found measurable effects in certain conditions by using TMS as an instantaneous probe. Specifically, TMS- induced MEPs were suppressed by concurrent tUS bursts using spatially targeting the same area of M1. Among these studies, *Fomenko*, *Chen et al*., *2020* showed a unique finding that provides insight into our results: while tUS had robust suppression of MEPs amplitudes evoked while the subject’s hand muscles were at rest (i.e. resting MEP), tUS had no effect on MEP amplitudes evoked while the subject’s hand muscles were contracting (i.e. active MEP). In other words, the presence of the target muscle tonically contracting was tied to the loss of the suppressive effects of tUS. If ongoing voluntary motor activity precludes, or at least greatly impedes, measurable tUS effects in M1, this could explain why we failed to find time-locked effects of tUS on the EMG signal. This leaves the question of why voluntary motor activity would prevent tUS effects. One hypothesis we propose is that there could be a floor or ceiling effect involving inhibitory interneurons, which have been proposed to be preferentially affected by tUS. Indeed, electrophysiology evidence from optogenetic tagging shows at least some M1 interneurons increase activity during voluntary motor behavior. Therefore, some of the cells that tUS would have preferentially affected (inhibitory interneurons,) are already being actively modulated at the time tUS energies enter the tissue, thus making further modulation of ongoing activity more difficult. Assuming tUS acts on neurons by opening ion channels, we hypothesize this difficulty in modulation could emerge due to the interneurons in question being in a state of high membrane conductance at the time of tUS (i.e. their tUS-sensitive ion channels are already open). Along with tUS not affecting active MEP, the cSPs that follow tUS-targeted active MEPs were not affected either. This too is interesting considering the cSP phenomenon is dependent on cortical interneurons. This matches our results in that tUS did not increase periods of EMG silence. That said, our approaches do differ. *Fomenko*, *Chen et al*. used TMS to instigate an MEP while we did not. Additionally, the approach to tUS timing differed. In *Fomenko*, *Chen et al*., tUS was turned off 10 ms after TMS onset (mean cSP length = 138 ms, *Fomenko*, *Chen et al*.), while sonication overlapped the full target epoch in our approach. Having noted the differences, the fact that both studies showed no change in EMG activity when the target muscle was being voluntarily contracted is quite intriguing. Another factor that could be at play is a theory proposed by *Fomenko*, *Chen*, *et al*.: that tUS may preferentially modulate GABA_A activity relative to that of GABA_B. Their proposal is based on the fact that tUS increased short-interval intracortical inhibition (SICI) in their study, which is considered a GABA_A-mediated phenomenon. Conversely, they also did not see changes in GABA_B-mediated effects tested: the paired-pulse paradigm long-interval intracortical inhibition and cSP. Considering that our tUS during tonic contraction paradigm is indeed similar to a cSP paradigm of TMS during tonic contraction, it is possible that our paradigm would be sensitive to GABA_B-related activity levels. However, it is the *late* part of the cSP that is believed to be GABA_B-mediated, and we did not even see short increases of EMG silence. As such, we are not sure whether or not our data would add to the GABA_B discussion regarding tUS. But, overall the notion of heterogeneity of tUS sensitivity among GABA_A channels and GABA_B-linked channels is certainly intriguing and warrants further investigation—potentially in combination with the notion of whether tonic activity impedes the ability of tUS to affect M1 circuitry. Though we were looking for effects on EMG signal, we know from the TMS literature that it *is* possible to induce detectable excitation of motor cortex without a measurable peripheral effect. Specifically, electroencephalography (EEG) recordings during subthreshold single-pulse TMS show significant TMS- evoked potentials, despite a lack of a peripheral MEP. Given this, it is possible there were tUS effects that did not interact with corticospinal projections—making them undetectable by EMG and therefore undetectable by our experiment. While EEG is not necessarily a more sensitive readout to study all mechanisms, such as for certain corticospinal excitability experiments, adding methods like EEG and fMRI that record cortical effects directly could provide a more complete picture of time-locked tUS effects in future investigations. ## Experiment 2 ### Cortical excitability Our data showed no group difference in M1 excitability in response to tUS, as indicated by no change in MEPs evoked by TMS at the same M1 trajectory. Our data differ from data by Gibson and colleagues that showed increased excitability of M1 after ultrasound exposure. There are study design differences that could have driven this disparity. The first is the tUS device used. While we used a 500-kHz single-element focused transducer, *Gibson et*. *Al 2018* used an imaging ultrasound device, which consisted of an array of 80 transducer elements emitting frequencies in a range between 1.53 and 3.13 MHz. This frequency range is noteworthy because acoustic attenuation increases as a function of acoustic frequency. *Gibson et al*. *2018* cited papers that used ultrasound imaging devices to image through the skull as evidence of the device’s validity for use over M1. However, all studies they cited placed the device over the temporal window—an area of the skull that is significantly thinner than that over M1 (temporal window: \~3 mm; parietal bone: \~6 mm). In fact, measurement of ultrasound propagation through ex-vivo human parietal skull shows that little to no energy is transmitted at frequencies above \~1.5 MHz \[, –\]. But importantly, our stimulation protocol may not have been ideal for inducing offline effects. We delivered separate bursts of ultrasound (duration: 300–500 ms each) with long gaps between bursts (8–12 s inter-burst interval). Our \~10-second inter-burst interval is very slow compared to repetitive TMS protocols used to modulate cortical excitability, and it ventures into the intertrial interval range suggested for use to avoid central habituation effect in sensory stimulation studies. It is also starkly different from *Gibson et al*., which delivered constant exposure to an ultrasound imaging protocol for 2 minutes. Acquisition of Experiment 1 and Experiment 2 data were integrated, i.e. the same tUS trials were used to both investigate time-locked (Experiment 1) and offline effects (Experiment 2), so changing the tUS protocol was not an option for this study. As such, future tUS studies may want to use more compressed protocols with shorter inter-burst intervals, if neuromodulation is the aim. A compressed approach was shown to successfully affect tUS targets in non-human primates, as shown by the reduction of resting-state fMRI connectivity following a 40-s tUS protocol (pulse repetition frequency: 10 Hz; pulse length: 30 ms). These non-human primate studies showed effects lasting up to two hours after stimulation. However, they also used tUS intensities, 24.1–31.7 W/cm², that were significantly higher than the levels used here or any other human tUS study (human max.: 4.9 W/cm² in *Legon et al*. *2014*). While histological examination in these studies revealed no microstructure damage, the protocol in question still corresponds to a mechanical index of \~3.6—higher than the 1.9 maximum allowed by the FDA for diagnostic imaging. Given that the most robust tUS effects seem to occur at high intensity levels, future studies will need to carefully explore whether consistent, behaviorally relevant tUS effects are feasible at intensities safe for human exposure. Replication of repetitive tUS protocols, at lower intensities, will likely be the first step. Indeed, recently published work showed an increase in motor excitability in humans up to 30 minutes following a condensed tUS protocol (pulse repetition frequency: 5 Hz; burst duration: 500 ms; inter-burst interval: 1.1 s) at intensities safe for human use. Considering the device used by *Zeng et al*. was very similar to the one used here, this reinforces our interpretation that the stimulation protocol used here was not ideal for inducing offline effects. ### Exposure & excitability With this small sample size (n = 8), we saw no correlation between M1_hand exposure and cortical excitability change, though the calculation of this correlation was also low-powered. This conclusion is to be expected since this study was not designed to investigate such a correlation, with M1_hand exposure effectively stratified into two levels depending on whether the M1_hand target was used once or twice (i.e. whether it was the primary NIBS target, “TMS target”). Studies that wish to investigate potential correlation effects of exposure levels would need to expose participants at a variety of different levels and have a larger sample size than used here to increase statistical power compared. While we performed acoustic simulation on 907 cm³ volumes (\~300–350 cm³ of which were grey or white matter), we still chose to tabulate cumulative exposure for a single location: M1_hand. M1_hand was chosen because it is the most reasonable small-volume, easily identifiable cortical area that is known to play a direct role in hand muscle contraction. It is, however, possible that directly targeting M1_hand may not be the ideal choice for modulating voluntary muscle contraction with tUS. This notion of placement is especially important given the spatially precise nature of tUS compared to a TMS or transcranial electrical stimulation—especially with a small, focused ultrasound transducer as used here. ## tUS and M1 While there could be multiple causes, one potential explanation could lie in cytoarchitectural differences between brain regions. Crucially, motor cortex has significantly lower neuronal density compared to somatosensory and visual cortex. As such, for a tUS pressure field of a given size, the number of individual neurons that fall within the focus would be lower in M1 compared to primary somatosensory (S1) or primary visual cortex (V1). This disparity could leave M1 neurons at a relative disadvantage for reaching thresholds to create detectable systems-level effects from tUS exposure. Additionally, the inherent cytoarchitectural and circuitry differences between ‘output’ cortical regions, like M1, compared to ‘input’ cortical regions, like S1 and V1, could likely have a significant role. Alternatively, the fact that we performed tUS during ongoing motor activity may have been a factor, since recent data showed that tUS-mediated suppression of TMS MEPs was not present during tonic motor contraction of the target muscle. Assuming the theory that tUS acts on neurons by opening ion channels, ultrasonic opening of channels should be less impactful if those same channels are already in a state of high conductance. # Conclusion We performed neuronavigated tUS and TMS of primary motor cortex (M1) in healthy volunteers. We found no concurrent change in finger EMG activity from tUS of M1 during voluntary muscle contraction. We also did not find any consistent effect of tUS M1 exposure on motor cortex excitability, as measured by single-pulse TMS of M1. We performed acoustic simulations using structural MRI of the study participants to estimate the degree and location of ultrasound intracranially. Using these simulations, we were unable to find any correlation between cumulative ultrasound exposure of the M1 hand area and M1 excitability change with the uncondensed tUS protocol used here. Within the TMS-only data, our data suggest that cortical silent periods (cSP) may be contingent on a motor evoked potential (MEP) occurring at cSP onset, though at times the MEP may elude visual detection due to a small amplitude that does not rise above the level of tonic muscle activity. This finding questions previous reports of cSPs without MEPs. While the negative tUS results reported here mirror struggles some other investigators have shown when attempting to elicit measurable modulation of M1 by tUS, this was also a pilot study with small sample sizes (n = 8; n = 10). As such, clearer results may emerge with larger datasets or changes in methodology. Lastly, given that both we and others failed to show tUS effects during tonic motor contraction, futures investigation may be warranted to consider how the state of channel conductance may affect tUS outcomes. # Supporting information 10.1371/journal.pone.0267268.r001 Decision Letter 0 Chen Robert Academic Editor 2022 Robert Chen This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 15 Nov 2021 PONE-D-21-31823Ultrasound stimulation of the motor cortex during tonic muscle contractionPLOS ONE Dear Dr. Heimbuch, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. ============================== Both reviewers provided detailed comments that need to be addressed. There should be more throughout discussion of relevant published papers (two were pointed out by the reviewers). The rationale for Experiment 2 as to why the chosen parameters are expected to produce offline effects should be more clearly stated. There is significant limitation with the low number of subjects which should be acknowledged or increase the number of subjects. ============================== Please submit your revised manuscript by Dec 30 2021 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript: A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Robert Chen Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf> 2\. In your Data Availability statement, you have not specified where the minimal data set underlying the results described in your manuscript can be found. PLOS defines a study's minimal data set as the underlying data used to reach the conclusions drawn in the manuscript and any additional data required to replicate the reported study findings in their entirety. All PLOS journals require that the minimal data set be made fully available. For more information about our data policy, please see <http://journals.plos.org/plosone/s/data- availability>. "Upon re-submitting your revised manuscript, please upload your study’s minimal underlying data set as either Supporting Information files or to a stable, public repository and include the relevant URLs, DOIs, or accession numbers within your revised cover letter. For a list of acceptable repositories, please see <http://journals.plos.org/plosone/s/data-availability#loc-recommended- repositories>. Any potentially identifying patient information must be fully anonymized. Important: If there are ethical or legal restrictions to sharing your data publicly, please explain these restrictions in detail. Please see our guidelines for more information on what we consider unacceptable restrictions to publicly sharing data: <http://journals.plos.org/plosone/s/data-availability#loc- unacceptable-data-access-restrictions>. Note that it is not acceptable for the authors to be the sole named individuals responsible for ensuring data access. We will update your Data Availability statement to reflect the information you provide in your cover letter. 3\. We note that you have included the phrase “data not shown” in your manuscript. Unfortunately, this does not meet our data sharing requirements. PLOS does not permit references to inaccessible data. We require that authors provide all relevant data within the paper, Supporting Information files, or in an acceptable, public repository. Please add a citation to support this phrase or upload the data that corresponds with these findings to a stable repository (such as Figshare or Dryad) and provide and URLs, DOIs, or accession numbers that may be used to access these data. Or, if the data are not a core part of the research being presented in your study, we ask that you remove the phrase that refers to these data. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Partly \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Please see attached document for my review. Reviewer \#2: Heimbuch et al., examine the effect of TUS of M1 on corticospinal excitability and inhibition as measured by EMG. In Experiment 1 TUS was applied during (i.e., “online”) voluntary muscle contraction and EMG was assessed with a focus on the cortical silent period (CSP), compared to TMS induced CSPs. In Experiment 2, TMS motor evoked potentials (MEP) were measured before and after the same TUS protocol applied at rest to assess after-effects (i.e., “offline” changes) in corticospinal excitability. While the justification for Experiment 1 is clear, the justification for the protocol in Experiment 2 (with relatively long inter-trial intervals) is unclear, as there is no indication in the literature that such a protocol with TUS (nor TMS) would result in offline effects that outlast the stimulation. Nonetheless, Experiment 1 provides interesting null results about the (lack of) TUS effects on inhibitory interneurons at previously used parameters, which are relevant for the TUS community. However, the methods section requires several clarifications, and a number of statements seem to oversimplify conclusions from previous literature. I have the following specific comments. 1\. Line 64 – we don’t know whether these intensities are unsafe. They just haven’t been used in humans yet. 2\. Lines 61 to 69 – would be good to cite Fomenko et al. 2020 and Xia et al. 2021 which also looked at TUS effects on M1. 3\. Lines 70 to 77 – it would be helpful to discuss the effects (or lack thereof) of TUS on SICI (Legon et al. 2018 and Fomenko et al. 2020). 4\. Lines 81 to 82 – better to state explicitly here already that Experiment 2 is about the after-effects (offline effects) of TUS 5\. In Experiment 1, was the order of TUS vs TMS blocks counterbalanced? Were there any sham/ control trials between TUS or TMS trials? 6\. Lines 105-115 – It sounds like the motor hotspot search was entirely based on a couple for predefined targets and not really optimized iteratively to find the best coil position and orientation for good, stable, and well-formed MEPs. Is it possible that the optimal TME MEP hotspot was not determined and application of them more focal TUS at those locations was therefore ineffective? Please show the MEP waveform averages for all subjects in the supplement. 7\. Lines 129 to 136 – was the TMS-induced MEP measured before and after an entire block of 20 TUS trials? 8\. Lines 137 to 144 – What specific coupling medium was used (gel, gel pads, water bladder, etc.) and how was good coupling ensured in the presence of hair? 9\. Line 139 – is the focus 3 cm long, or 3cm away from the transducer face? Please provide both values, in addition to the width of the focus. 10\. Line 153 – what attenuation factor is used to estimate in the intracranial intensity? Was it the equation on line 254? Please also provide the Ispta. 11\. Lines 169 to 182 – the descriptions of cSP onset and offset determination is difficult to understand. For instance, why would the ‘first rising EMG value’ be used as onset? Please try to use a figure to explain this. And also mention whether similar methods have been used in other papers. 12\. Line 189 – why the mean of the two flanking valleys instead of the larger valley? Does this method have any advantages, and has it been previously used? 13\. Line 210 – what is the criteria for determining ‘silence’? Is it different from the threshold used to determine cSP onset and offset? 14\. Please specify the exact MR acquisition parameters for the anatomical MRIs that were used for neuronavigation and acoustic simulation. Please also provide examples of the skull masks overlayed on the subject specific MRIs. 15\. Line 279 to 282 – what about the TUS trials? 16\. Line 283 – can you please elaborate how ‘spike rate’ is measured? Is this the ‘rate of EMG peaks’ described earlier? Does this measure have any specific physiological significance? 17\. Line 396 – it is likely that some early studies recorded responses to the acoustic rather than neuromodulatory effects of TUS. However, recent papers have confirmed that neuromodulation is also present (see for eg. Mohammadjavadi et al. 2019, Brain Stimulation). I recommend adding some nuance to this statement. On the topic of sensory confounds, did any participants in your study report auditory and/ or somatosensory perception accompanying the TUS trials? Have they been asked explicitly? 18\. Lines 408 to 409 – again, I recommend adding some nuance to this statement. While TUS may preferentially stimulate inhibitory neurons at some parameter combinations, there is also evidence that it can stimulate excitatory neurons (see Yu et al. 2021, Nature Communications). 19\. Fig. S3 – it seems like this figure shows fewer peaks in the TMS compared to TUS trials. I presume the amplitude of the peaks where higher in TMS? 20\. Fig. 5 – Why are MEP AUCs shown on the y-axis as if they only had a very limited number of discrete values? Please alternatively also show (and used for statistics) the more well-established MEP peak-to-peak amplitudes. 21\. Fig. 7. – from the caption it does not become clear what the difference between (a) and (b) is supposed to be. In (a), both TUS beams seem to have “hit” the central sulcus in the hand knob, if anything more post-centrally, whereas in (b) it looks like the focal spot is very anterior and rather in the premotor cortex. None of the presented foci seem to be particularly well targeting the anterior wall of the central sulcus where M1 corticospinal output neurons are located. Simulated acoustic pressure maps for all targets and should be presented for all subjects in the Supplement. 22\. Fig. 7 – it looks like the skull image for the subj-01 is not complete. Maybe the FOV was cut-off? This has major implications for the quality of the simulations. Please discuss these limitations. 23\. Some information seem to be missing in the data availability statement: “All EMG files are available from the XXX database (url: XXX). Code to run acoustic simulations is available via the MATLAB package TUSX ([www.tusx.org](http://www.tusx.org)).” \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0267268.r002 Author response to Decision Letter 0 17 Jan 2022 (Full response attached as file) We would like to thank the reviewers and the editor for their helpful feedback on the manuscript. Your feedback was thorough, including both focused points to improve understanding of specific sections and broader points to touch on questions concerning the field. In particular, we would like to thank you all for bringing Fomenko, Chen et al., 2020 further into our attention. We feel that your feedback has allowed us to improve clarity throughout the manuscript. Additionally, it has helped us more fully cover details that tUS researchers may wish to know about our work, as well as strengthening the manuscript’s scientific discussion. For that, we are truly appreciative. 10.1371/journal.pone.0267268.r003 Decision Letter 1 Chen Robert Academic Editor 2022 Robert Chen This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 1 Mar 2022 PONE-D-21-31823R1Ultrasound stimulation of the motor cortex during tonic muscle contractionPLOS ONE Dear Dr. Heimbuch, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Both reviewers felt that the paper has improved but there are still points that needs to be addressed. The control experiment is reasonable. The experiment should be performed or at a minimum the issues discussed. Please submit your revised manuscript by Apr 15 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'.If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Robert Chen Academic Editor PLOS ONE Journal Requirements: Please review your reference list to ensure that it is complete and correct. If you have cited papers that have been retracted, please include the rationale for doing so in the manuscript text, or remove these references and replace them with relevant current references. Any changes to the reference list should be mentioned in the rebuttal letter that accompanies your revised manuscript. If you need to cite a retracted article, indicate the article’s retracted status in the References list and also include a citation and full reference for the retraction notice. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#1: All comments have been addressed Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: Yes Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: Review was uploaded as an attachment. Reviewer \#2: Tha authors have adequately addressed most of my concerns. I have only some minor comments remaining: 1\. The heading exposure vs. excitability is confusing. A suggestion is to say ‘association between exposure and excitability’ instead. 2\. Can you please provide some supporting evidence for this statement – ‘In a scenario of on ongoing voluntary motor activity, a subset of interneurons in M1 will have already been recruited, so perhaps the channels or cells that tUS would have affected are already being actively modulated by the time tUS energies into the tissue (e.g. channels have already been opened).’ For instance, is there evidence from TMS studies suggesting that inhibitory interneurons are already active during a voluntary muscle contraction? 3\. Please include the information about counter-balancing and sham in the manuscript. \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: No Reviewer \#2: No \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0267268.r004 Author response to Decision Letter 1 28 Mar 2022 We would like to thank the reviewers and the editor for their helpful feedback on the revised manuscript. Responses are uploaded in ResponsetoReviewers_Revision2.docx 10.1371/journal.pone.0267268.r005 Decision Letter 2 Chen Robert Academic Editor 2022 Robert Chen This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 6 Apr 2022 Ultrasound stimulation of the motor cortex during tonic muscle contraction PONE-D-21-31823R2 Dear Dr. Heimbuch, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Robert Chen Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: 10.1371/journal.pone.0267268.r006 Acceptance letter Chen Robert Academic Editor 2022 Robert Chen This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 11 Apr 2022 PONE-D-21-31823R2 Ultrasound stimulation of the motor cortex during tonic muscle contraction Dear Dr. Heimbuch: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Robert Chen Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Tag football is an international game that is gaining rapid popularity around the world, particularly in Australia, New Zealand and the United Kingdom. The game is best described as a modified version of rugby league where players aim to remove Velcro tags attached to each player’s shorts rather than tackle their opponent. Elite-level competition is contested at state-, national- and international-representative levels and is played over 1–3 day tournament competitions. The rules of tag football vary slightly depending on which form of the game is played, however a typical match is contested by two teams of 16 players on a 70 m by 50 m playing field over two 20 min halves. There are only 8 players allowed on the field at any one time, although teams are allowed unlimited interchanges during a match. The unlimited interchange and modified contact laws promote a fast-paced running tempo and with the increasing competition standards of tag football, a better understanding of the physical and physiological demands of the game is becoming necessary to guide training practices. The use of microtechnology devices incorporating global positioning system (GPS) and accelerometer technology have provided a useful tool to examine the match- activity profiles of team-sport athletes and guide sport-specific training practices. Several recent studies have determined the match-activity profiles of tag football players during state- and national-level competition have shown the match-activity profile to be characterised by highly intermittent running efforts during short field rotations (3.5 to 4 min) that equate to average match work-rates comparable to professional rugby league players. However, studies have yet to identify the fluctuations in work-rate or the most demanding passages of match-play where players may be required to perform successive high- intensity running efforts with minimal recovery. Previous research in similar team sports has shown that work-rate variables (e.g. m/min) can fluctuate considerably during a match and an understanding of the most demanding passages of play, although occurring infrequently, is required to adequately prepare players for competition. Several researchers have aimed to quantify the most demanding passages of play in team sports by determining the occurrence of repeated high-intensity effort (RHIE) bouts, defined as a minimum of three consecutive high-intensity running, maximal acceleration or physical collision efforts with less than 21 s between consecutive efforts. For example, an investigation by Spencer et al. in elite field hockey showed players performed 30±14 sprinting efforts during a single match equating to approximately one sprint every 2 min. However, players also performed up to four repeated sprint bouts during the match, defined as a minimum of 3 sprint efforts with less than 21 s recovery, with several bouts consisting of up to six to seven consecutive efforts. These results demonstrate the importance of determining the most intense passages of match-play in order to prescribe training activities that adequately prepare players for competition. Despite there being an increased research interest in the physical and physiological demands of tag football competition, there has been little attention given to the typical physical fitness profiles of players and the relationship between physical fitness and the match-activity profile in tag football. Physical fitness tests can provide a useful assessment of an athlete’s well-being, training progression and readiness to compete. Several researchers in similar team sports such as soccer and rugby, have examined the relationship between physical fitness tests and players’ match-activity profile to determine which tests provide a useful assessment of sport-specific fitness. For example, a study in youth soccer players demonstrated Yo-Yo intermittent recovery test level 1 (Yo-Yo IR1) to have strong associations with high-intensity running (r = 0.56; p\<0.01) and sprinting (r = 0.63; p\<0.01) distances during matches whilst maximal oxygen uptake (VO²max) determined during a laboratory treadmill test was found to have poor associations (r = 0.00; p = 0.99) with these measures. Similarly, a recent investigation in tag football found players selected into the state-representative team had greater Yo-Yo intermittent recovery test level 2 (Yo-Yo IR2) performance than non-selected players in the same team, and Yo-Yo IR2 was associated with greater very high-intensity running (VHI) distance per minute (r = 0.77; p\<0.01), VHI effort frequency (r = 0.75; p\<0.01) and peak running speed (r = 0.77; p\<0.01) during matches. These studies demonstrate the usefulness of physical fitness tests to assess sport- specific fitness and provide information pertaining to an athlete’s game readiness and training progression in team sports. Previous studies have provided important information on the match-activity profiles of representative tag football players at state- and national-level. However, further research is required to determine the most demanding passages of match-play and the physiological responses of players to contribute to our understanding of the match-activity profile in tag football. There is also limited information on the typical physical fitness profiles of tag football players and the relation between physical fitness and match performance. Previous research in tag football has shown Yo-Yo IR2 performance to contribute to high-intensity running performance during tag football match-play, although it is uncertain whether this relationship is dependent upon playing position which has been shown to influence the match-activity profile. Gaining a better understanding of the relationship between physical fitness measures and the match-activity profile may have important implications for athlete testing and training practices in tag football. The purpose of this study was to a) examine the physical fitness, match-activity profiles and physiological responses of representative tag football players in relation to playing position and b) determine the relationship between physical fitness measures and the match- activity profile. # Materials and Methods ## Study Design This study employed a prospective and observational study design. Match-activity profiles and heart rate responses were analysed using microtechnology devices (MinimaxX S4, Catapult Sports, Melbourne, Australia) combined with heart rate (HR) chest straps (Polar, Krakow, Finland). Sixteen male players from the same team were analysed for five matches played over two consecutive days of competition. Match-activity profiles, heart rate responses and perceived match- intensities were examined based on players’ playing position (inside or outside players) and partial correlation coefficients were used to determine the relationship between selected match performance variables and physical fitness tests assessed prior to competition. ## Participants Sixteen male tag football players from the same open men’s representative team participated in this study (Mean ± standard deviation (SD); age 24±4 yrs; height 1.78±0.06 m; body mass 75.9±9.4 kg). The participating team placed second out of sixteen teams during the 2014 Australian National Championships which is an annual tournament competed by regional-representative teams from Australia and New Zealand over three days of competition. It is the highest standard of tag football competition played in Australia and provides players with a pathway to represent Australia during the Triennial Tag Football World Cup. Players were informed of the study requirements prior to the tournament and all players and coaching staff gave their written informed consent to participate. All procedures were approved by the University of the Sunshine Coast’s Human Research Ethics Committee (HREC) in the spirit of the Declaration of Helsinki, HREC approval number S/13/471. ## Procedures ### Preliminary fitness testing Participants were required to complete a fitness test battery five days prior to the commencement of the Australian National Championships. The fitness test battery was performed before combined squad training following a two day unloading phase and prior to the team travelling interstate for the tournament. All players performed the tests at the same time of day on an outdoor playing surface and were instructed to maintain their usual diet prior to the commencement of training. Lower body muscular power was assessed using a vertical jump test. Vertical jump height was determined using a yardstick vertical jump device (Swift Performance Equipment, Brisbane, Australia). Players were instructed to stand with feet flat on the ground with ankles in line with the yardstick, fully extend their inside arm and hand (dominant), and mark the standing reach height. Players where then instructed to perform a rapid countermovement jump from a standing start and touch the highest point possible. Vertical jump height was calculated as the distance from the highest point reached minus the initial starting height following three maximal attempts. The vertical jump test was shown to have acceptable reliability in this study’s participant cohort (CV = 3.3%; ICC = 0.95). Straight line running speed was assessed using dual beam electronic timing gates (Fusion Sport, Sumner Park, Australia) with 10 m and 20 m splits. Players started 30 cm behind the first timing gates placed at 0 m and instructed to run as quickly as possible along the 20 m distance from an athletic start. Speed was measured to the nearest 0.01 s with the fastest value over 20 m used as the score. Straight line running speed measures had acceptable reliability over 10 m (CV = 1.9%; ICC = 0.92) and 20 m (CV = 1.5%; ICC = 0.96) in this study’s participant cohort. Yo-Yo intermittent recovery test level 2 (Yo-Yo IR2) was performed on an outdoor playing surface following the vertical jump and straight line running speed assessment. The procedures of the test have been described in detail elsewhere. All players had been familiarised with the testing procedures of the Yo-Yo IR2 during previous training sessions. The Yo-Yo IR2 was shown to have acceptable reliability within this study’s participant cohort (CV = 6.6%; ICC = 0.96). ### Match-activity profiles Micortechnology devices worn by players during matches provided GPS data sampling at 10 Hz (MinimaxX S4, Catapult Sports, Melbourne, Australia). A total of 80 match-files were collected over five matches played by the team during two days of competition. However, 9 incomplete match-files were discarded from analysis due to battery failure or player injury during match-play resulting in a total of 3–5 match-files per player being included in data analysis. The microtechnology device was worn in a small custom made vest in between the shoulder blades on the upper back of each player. Players were familiar with the device and the custom made vest which they had worn on numerous occasions prior to the tournament. The device was switched on and locked to satellites prior to the team warm-up. Recorded data was compiled and analysed in Sprint 5.1 (Catapult Sports, Melbourne, Australia). Periods of time where players were off the field, including interchanges and half-time periods, were excluded from data analysis. This study categorised GPS data into low-speed running (LSR = 0.4–8.0 km/h), moderate-speed running (MSR = 8.1–14.0 km/h), high-speed running (HSR = 14.1–18.0 km/h) and very high-speed running (VHSR = \>18.1 km/h). These speed zones were assigned as they have previously been used to describe match-activity profiles in tag football and the microtechnology device used in this study has been shown to provide accurate and reliable distance and effort measurements at velocities ranging from 0–14 km/h and 14–20 km/h during stimulated team sport running patterns. High acceleration and deceleration efforts \>2.78 m/s² and lasting a minimum of 0.4 s were also recorded. The microtechnology device used in this study has been shown to detect instantaneous changes in velocity during acceleration with a percentage bias of -3.6% to -2.1% and a typical error of 2.6% to 5.9% against a criterion measure of a laser. Occurrences of acceleration and deceleration efforts were only recorded if they lasted a minimum of 0.4 s as this method has been shown to improve the ecological validity of acceleration measures derived from GPS. The RHIE bouts performed by of players were recorded as a minimum of three high-speed running (\>14 km/h) or high acceleration efforts (\>2.78 m/s²) performed with less than 21 s recovery between consecutive efforts as similarly recommend for other team sports. Mean and peak match heart rates (HR) were expressed relative to players’ peak HR (%HRpeak), determined during the Yo-Yo IR2 (HRpeak = 197±8 bpm). As previously employed in similar team sports, HR data were categorised based on the percentage of on-field time spent in each of the following HR zones: Zone 1 (\<60% HRpeak), Zone 2 (61–70% HRpeak), Zone 3 (71–80% HRpeak), Zone 4 (81–90% HRpeak), Zone 5 (91–95% HRpeak) and Zone 6 (\>95% HRpeak). Players’ rating of perceived exertion (RPE) for matches was obtained using Borg’s CR-10 RPE scale. Players provided there scores individually, so that their scores were blinded from other team mates, approximately 10–15 min following matches. The RPE score was then multiplied by players’ on-field playing time to provide a measure of internal load (Session-RPE, AU). Players were familiarised with the Borg’s CR-10 RPE scale to assess perceptual match and training intensity prior to the commencement of the tournament. ## Statistical Analyses Data was log transformed prior to analysis to reduce non-uniformity of error and back transformed to attain descriptive statistics. Standardised differences were calculated and classified as trivial (\<0.2), small (0.2–0.6), moderate (0.6–1.2), large (1.2–2.0) and very large (\>2.0) as per previously standardised criteria. When the 90% confidence limit (CL) crossed the threshold for both substantially positive (0.2) and negative (-0.2) values the difference was reported as unclear. Pearson’s correlation coefficients were calculated to determine the relationship between physical fitness measures and mean and peak match-activity measures for the combined group (n = 16) and inside (n = 8) and outside (n = 8) playing positions. Correlations are reported as trivial (\<0.1), small (0.1–3.0), moderate (0.3–0.5), large (0.5–0.7), very large (0.7–0.9) and almost perfect (\>0.9). When the 90% CL crossed the threshold for both a substantially positive (0.1) and negative (0.2) association the correlation was determined as unclear. # Results The physical fitness scores for players based on playing position are presented in. There were moderate differences between inside and outside playing positions for vertical jump and measures of straight line running speed. There was no difference in Yo-Yo IR2 scores between playing positions. presents the activity profiles of tag football players and differences based on playing position. There were moderate differences between inside and outside players for peak running speed (ES = 0.95), mean and maximal VHSR effort distances (ES = 0.84–0.86), and maximal acceleration and deceleration frequency (ES = 0.60–0.80). Although there were only small differences in distance/min between playing positions, outside players covered greater VHSR m/min and less MSR m/min compared to inside players. Inside and outside players reported similar effort frequencies at different running speeds, except inside players performed more MSR efforts/min. The RHIE characteristics of players based on playing position are presented in. Inside and outside players reported a similar number of total RHIE bouts and mean and maximum efforts per RHIE bout. However, there were small to moderate differences for mean and maximum effort durations (ES = 0.69–1.15) and mean bout recovery (ES = 0.56). Players’ perceived match-intensities and heart rate responses are presented in. Inside and outside players reported no differences in RPE or Session-RPE for matches, however there was a moderate difference in mean HR (ES = 0.67) between playing positions. There were also differences in the distribution of time spent at various heart rates between playing positions, with inside players spending a greater percentage of time at heart rates above 90% HRpeak compared to outside players spending a greater percentage of time at heart rates between \<60–90% HRpeak. Vertical jump and straight line running speed (20 m speed) showed moderate to large correlations with VHSR m/min (r = 0.44–0.69), VHSR efforts/min (r = 0.44–0.50) and peak running speeds (r = 0.48–0.69) for the combined group. Inside players (r = 0.54–0.81) showed stronger correlations between vertical jump and straight line running speed measures and VHSR m/min, VHSR efforts/min and peak running speeds than outside players (r = 0.05–0.55). Further, vertical jump and straight line running speed measures showed moderate to very large correlations with RHIE total bouts for inside players (r = 0.46–0.85) and weaker correlations with outside players (r = 0.01–0.49). Yo-Yo IR2 was positively correlated with total distance (r = 0.59–0.73), relative distance (0.26–0.50), VHSR m/min (r = 0.38–0.61), VHSR efforts/min (r = 0.55–0.58), peak running speeds (0.40–0.56), RHIE total bouts (r = 0.48–0.52), and mean efforts per RHIE bout (r = 0.27–0.48) for inside and outside players whilst showing negative correlations with match RPE (r = -0.25 to -0.31). # Discussion This study is the first to determine the normative data on a range of physical fitness tests in representative tag football players and provides coaching and support staff with useful information for designing training programs. Firstly, this study found vertical jump, straight line running speed and Yo-Yo IR2 performance to be associated with the match activities of tag football players suggesting that these tests may provide a useful assessment of a player’s well- being, training progression and readiness to compete. Importantly though, there were a number of differences in both physical fitness and match-activity profiles between inside and outside playing positions. For greatest transference to match performance, training programs for tag football players should be individualised based on playing position to prioritise the development of certain physical fitness qualities. The RHIE characteristics identified in this study may be used by coaches to design sport-specific fitness drills that replicate the most intense passages of tag football match-play and modify certain training parameters to promote positional-specific training adaptations. The current study found a number of differences in physical fitness measures between playing positions. Outside players were found to have greater lower body muscular power and faster straight line running speed than inside players. This finding can likely be attributed to the different positional-specific demands of inside and outside players in tag football. For instance, outside players were found to have greater mean and maximal VHSR effort distances and reach higher peak running speeds than inside players and the ability to produce higher peak running speeds during matches may be more beneficial for players positioned towards the edges of the field where there is typically more space between them and their opposing player. Further, outside players likely have a greater chance to recover between repeated VHSR efforts compared to inside players, as evident through differences in high acceleration and deceleration frequencies, MSR effort frequency , and time spent above 90% HRpeak between inside and outside playing positions. Although outside players exhibited faster straight line running speed, it is interesting to note that inside players showed stronger correlations between straight line running speed and VHSR m/min, VHSR effort frequency and peak running speed than outside players. These results may be explained by the test distance of 20 m, as outside players had a maximal VHSR effort distance of 39±10 m. Therefore, the ability of outside players to produce higher running velocities between 20–40 m distances may have stronger correlations with their match-activity profile, whilst the ability to produce faster running speeds between 10–20 m is more beneficial for inside players \[, , \]. Further research should examine straight line running speed over short distances (10–20 m) as well as longer distances more typical of the maximal sprint distance (≥40 m) in tag football to provide a better understanding of the relationship between straight line running speed and match running performance for different playing positions. This study found the Yo-Yo IR2 to be a useful determinant of high-intensity running performance during tag football match-play. The combined group showed moderate to large correlations between Yo-Yo IR2 test distance and match- activity profile variables including total distance (r = 0.44–0.63), VHSR m/min (r = 0.31–0.34), VHSR effort frequency (r = 0.44–0.50) and RHIE total bouts (r = 0.50–0.55). This is in agreement with previous research in tag football that found the Yo-Yo IR2 to be a useful determinant of high-intensity running performance at state-level competition. In the current study, inside players showed stronger correlations between Yo-Yo IR2 and distance/min (r = 0.50 vs. r = 0.26), VHSR m/min (r = 0.61 vs. r = 0.31), RHIE mean efforts per bout (r = 0.48 vs. r = 0.27) and RHIE maximum efforts per bout (r = 0.47 vs. r = 0.07) than outside players, suggesting that the Yo-Yo IR2 may be more beneficial to the match-activities of inside players. Indeed, higher Yo-Yo IR2 may allow players to better maintain high-intensity running performance in-between more frequent high acceleration and deceleration efforts and MSR efforts as characteristic of the match-activity profile of inside players. There were also moderate differences between inside and outside playing positions for the amount of time spent \>90% HRpeak suggesting that inside players may require higher aerobic and anaerobic capacities than outside players due to their positional- specific demands. Therefore, although Yo-Yo IR2 appears to be related to the match-activities of both inside and outside players, the results of this study suggest that training programs should be adjusted based on playing position to prioritise the development of certain physical fitness qualities. It is recommended that training programs for inside players prioritise the development of physical fitness qualities related to the Yo-Yo IR2 whilst training programs for outside players have a greater focus on developing lower body muscular power and straight line running speed. This study is the first to determine the RHIE characteristics of tag football and can be used to guide positional-specific conditioning practices for tag football players. Generally, there were no differences in the number of efforts performed for bouts or the mean recovery between consecutive efforts based on playing position. Therefore, it is recommended that RHIE training drills for tag football players incorporate 4–8 repeated efforts with approximately 6 s recovery between efforts. There were differences in the mean and maximal effort durations between inside and outside playing positions, with outside players typically having longer mean and maximal effort durations in line with the greater VHSR effort distances performed by these players. Therefore, RHIE training drills may be modified for positional groups, with outside players covering 15–40 m effort distances and inside players covering 10–30 m effort distances to mimic their positional-specific demands. Further, to allow outside players to comply with the increased high-intensity running performed during RHIE drills due to greater effort distances, outside players should have greater recovery between bouts (75±20 s) compared to inside players (50±20 s). Although the RHIE characteristics determined in this study can be used to guide sport- specific conditioning practices in tag football, it should be recognised that not all conditioning needs to be sport-specific. Due to the fatiguing nature of this type of exercise, high volumes of training replicating the RHIE characteristics of tag football match-play may lead to adverse training adaptations such as injury, illness or overtraining. Further, previous research in similar team sports has shown measures of aerobic and anaerobic fitness to be positively related to payers’ repeated-sprint ability. It is recommended that the inclusion of training drills replicating the RHIE characteristics of tag football match-play is appropriately planned based on the amount of high- intensity running that players are performing during other training and competition activities and following the development of aerobic and anaerobic capacities. Despite tag football being a recreational sport, players competing at elite- representative levels appear to have comparable physical fitness qualities than professional and semi-professional athletes of similar team sports. For example, in the current study tag football players reported slightly faster 10 m sprint times (1.65±0.08 s) than international rugby sevens players (1.68–1.74 s) and professional rugby league players (1.66–1.72 s); and covered distances in the Yo-Yo IR2 (1023±346 m) comparable to professional Australian football players (1028±190 m) and greater than first division recreational Australian football players (880±260 m). Although caution should be taken when making comparisons between these studies due to differences in testing conditions, the results of this study provide interesting normative data from a single team and suggest representative tag football players competing at national-level exhibit high levels of physical fitness. Previous research in tag football showed players from the same team competing at state-level competition covered 892±170 m during the Yo-Yo IR2 under similar testing conditions than the current study. This study also showed that players selected into the state-representative team covered greater distances during the Yo-Yo IR2 than non-selected players suggesting that the Yo-Yo IR2 may be a useful test to assess player capabilities and select playing rosters in tag football. Similarly, the higher Yo-Yo IR2 scores shown for national-level players in the current study compared to state- level players (1023±346 m vs. 892±170 m) suggests that there may be a relation between higher Yo-Yo IR2 and playing standard attained in tag football. Whilst this study provides several important findings for coaching and support staff working with tag football players, there are a number of limitations that warrant discussion. Firstly, this study included data that was from a single team and it is possible that the results of this study may not be evident in other teams competing at national-level due to differences in team dynamics and playing strategies. Further, this study collected limited data on the physical fitness qualities of players due to restrictions applied to the scheduling and time allocation of performance testing. We reconciled these limitations with the knowledge that the participating team was of a high-standard, finishing second out of sixteen teams at the conclusion of the national championships. This study was also the first to determine the relationship between physical fitness measures and the match-activity profiles of tag football players competing at national-level and provides important information for the development of future research studies. Future studies that include larger sample sizes from different competition standards and examine the relevance of different physical fitness tests (e.g. Yo-Yo IR2 vs. 30–15 IRT) may allow for further developments in athlete training and testing practices in tag football. # Conclusions This study provides important information for coaching and support staff working with tag football players. Firstly, this study provides the first normative data for a range of physical fitness measures for male tag football players competing at national-level. Lower body muscular power, straight line running speed and Yo-Yo IR2 were found to be important physical fitness qualities related to the match activities of both inside and outside players. Therefore, these tests provide a useful assessment of players’ physical fitness and may be used to evaluate the effectiveness of training in tag football. Coaching and support staff working in tag football may use this data to identify key areas of physical fitness that a player or group of players may need to develop based on their playing position. The results of this study suggest that training programs for inside players should have a greater focus on the development of physical capacities related to the Yo-Yo IR2 whilst outside players should focus on developing lower body muscular power and straight line running speed over short distances (10–50 m). The results of this study may also be used to guide sport-specific conditioning drills for tag football players. Based on the RHIE characteristics of players’ match-activity profile it is recommended that repeated-sprint drills incorporate 4–8 repeated efforts with approximately 6 s recovery between efforts. Repeated- sprint drills may be modified for positional groups by adjusting effort distances and bout recovery. Inside players should perform 10–30 m efforts and have 50±20 s of recovery between consecutive bouts. Outside players should perform consecutive efforts over longer distances (15–50 m) whilst having longer recovery between consecutive bouts (75±20 s) to mimic their positional-specific demands. Importantly, the volume of this type of training should be adjusted during certain training phases following appropriate planning and the development of aerobic and anaerobic capacities to avoid increases in the risk of injury, illness and overtraining. The authors would like to thank the coaching staff and playing squad of the Sunshine Coast open men’s team for participating in this study. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: LH BB MM. Performed the experiments: LH. Analyzed the data: LH BB MM. Wrote the paper: LH BB MM.

# Introduction Schizophrenia is a chronic and disabling disease in which repeated relapses have a negative impact on patients’ functioning and are disruptive for patients’ education and employment. Unemployment is a large burden for patients as well as for society, and on average, the worldwide proportion of patients with schizophrenia who are not undertaking any paid employment is around 80%. Unemployment has socioeconomic consequences for the patient’s ability to live independently and participate actively in the community and thus can adversely impact quality of life. Improved quality of life and functioning is an essential long-term treatment goal in schizophrenia for patients, caregivers, clinicians, and payers that is receiving increasing scientific and societal attention. Therefore, evaluating the ability to function within a working environment (i.e. work readiness) in schizophrenia is of clinical relevance, and increasing the functional capacity to work should be considered a valuable goal in the treatment of schizophrenia, increasing patients’ social integration in the community. Systematic assessments of capacity to work and work readiness, such as the clinician-rated Readiness for Work Questionnaire (WoRQ), constitute a practical and validated approach to assess treatment-induced improvement of functioning in patients with schizophrenia. The simplicity of the Yes/No question as to whether the patient is ready to work is potentially useful in clinical practice, where more lengthy, detailed scales are unlikely to be administered routinely. The QUALIFY (QUAlity of LIfe with AbiliFY Maintena^®) study is one of the few randomized studies directly comparing two different long-acting injectable antipsychotics (LAIs), and was the first to compare the effects of two atypical LAIs on a measure of health-related quality of life and functioning as the primary outcome. The Heinrichs-Carpenter Quality-of-Life scale (QLS) is a detailed assessment, broadly defining functioning as an individual’s ability to perform normal daily activities required to meet basic needs, fulfil usual roles, and maintain their health and well-being. The QLS is one of the relatively few rating scales designed to assess aspects of functional impairment (social, occupational, and psychological) associated with schizophrenia, and it is sensitive to subtle change over time, as well as to the psychopharmacological effect of treatment. Thus, QLS is the most widely used instrument for assessing health-related quality of life in schizophrenia. A drawback of the clinician- rated QLS is that the long administration time (30–45 minutes) can be burdensome for the patient as well as the clinician. The primary analysis of the QUALIFY study, performed using a mixed model for repeated measures (MMRM), showed non-inferior and superior improvements with AOM 400 vs PP on QLS total score over 28 weeks. Furthermore, the QUALIFY study was the first study to apply the WoRQ instrument to assess differences between two treatments for schizophrenia. The effect of AOM 400 and PP on work readiness was analyzed post-hoc using logistic regression adjusted for baseline status to compare odds of work readiness after 28 weeks of treatment. We recently reported significantly greater improvement on WoRQ total scores with AOM 400 vs PP, as well as significantly more patients improving in work readiness status after AOM 400 vs PP treatment. As a randomized phase IV study, QUALIFY is a rich data source enabling cross- validation of instruments used in the evaluation of functioning in schizophrenia. This study is particularly well-suited to investigate the association between improvements on different scales of health-related quality of life and functioning after switching to LAI therapy. In this post-hoc analysis of the QUALIFY study, we investigated the association between response to LAI treatment with AOM 400 or PP on measures of patient functioning and work readiness (QLS and WoRQ). # Methods ## Study design This was a post-hoc analysis of data derived from the QUALIFY study, a randomized controlled trial comparing the atypical LAIs, AOM 400 and PP in stable adult patients, ages 18 to 60 years, with schizophrenia (defined by DSM- IV-TR). The study design and patient population of the QUALIFY study were previously described in detail; the protocol was approved by the relevant institutional review board for each country in which the trial was conducted, all patients provided written informed consent, and the study was conducted in accordance with the Declaration of Helsinki. Briefly, included patients were switching from oral to LAI antipsychotic treatment and had clinical global impression—severity (CGI-S) scores ≥3 (mildly ill) and ≤5 (markedly ill) at the screening and baseline visits. Patients with a diagnosis of psychiatric disorder or DSM-IV-TR axis I disorder other than schizophrenia or acute exacerbation of psychotic symptoms, or hospitalization for \>3 months before the screening visit were excluded. This post-hoc analysis was longitudinal and observational in nature, and combined all patients in the QUALIFY study, regardless of treatment group, to determine the effect of treatment response on patient functioning and work readiness. ## Assessments The QLS was the primary outcome of the QUALIFY study, assessed at baseline and at weeks 4, 8, 16, and 28 (end of study) by a rater blinded to the treatment. The QLS comprises 21 items in 4 domains. In addition, a number of secondary outcomes were evaluated in the QUALIFY study, including WoRQ which was rated at baseline and at week 28 (end of study). WoRQ was administered by a rater who was not blinded to the treatment (a different rater from the one assessing QLS). The CGI-S scale quantifies the clinician’s impression of the patient’s current illness severity on a scale ranging from 1 (normal, not at all ill), to 7 (among the most extremely ill patients). In the QUALIFY study, CGI-S (included as secondary endpoint) was used to assess symptom severity at all visits, namely screening, baseline and post-baseline at weeks 2, 3, 4, 8, 12, 16, 20, 24, and 28 (end of study). ## Statistical analysis All analyses were conducted in the full analysis set, which is a modified intent-to-treat set comprising treated patients with valid baseline QLS and WoRQ assessments, at least one valid post-baseline QLS assessment, and a valid week-28 WoRQ assessment. Thus, analyses were conducted in a subset of patients from the QUALIFY study. It should be noted that the final population included in the current post-hoc analysis (n = 208) comprised those who completed the original QUALIFY study (n = 183) as well as 25 additional patients who discontinued before completing the study but did complete a week-28 WoRQ assessment. Irrespective of AOM 400 or PP treatment, patients were categorized based on work readiness (Yes or No) at baseline and week 28: shift from No to Yes (n = 41), Yes to Yes (n = 49), or No at Week 28 (n = 118; comprising No to No patients \[n = 102\] and shift from Yes to No patients \[n = 16\]), see, item 8. Due to the low number of patients shifting in work readiness from Yes at baseline to No at week 28 (n = 7 for AOM 400; n = 9 for PP), these patients were grouped with those not ready to work either at baseline or at week 28 in the “No at Week 28”-group. The changes in QLS total, domain, and item scores were compared in work readiness shift groups using a MMRM with an unstructured covariance matrix including baseline score-by-visit interaction, geographic region (Europe/North America), age group (≤35/\>35 years), visit, and treatment-by-visit interaction as fixed effects. This post-hoc analysis used MMRM methodology similar to what was used in the primary analysis. The time-dependence of the associations between work readiness status and QLS total scores or CGI-S scores were analyzed with logistic regressions applied to assess the predictive effects of absolute scores of QLS total and CGI-S at individual visits on the positive outcome (“Yes”) in work readiness at week 28. The logistic regressions used baseline status of readiness to work and QLS total or CGI-S scores as covariates, and input scores (QLS total or CGI-S) were standardized using mean and standard deviation for each visit to account for the numerical differences in the score ranges. As a consequence of standardization, the size of parameter estimates for QLS total and CGI-S is comparable between scales and indicates the strength of the effect of the particular score at the given visit in predicting the work readiness outcome at week 28. The positive or negative parameter estimates indicate the direction of improvement on each scale (clinical improvements correspond to higher QLS total and lower CGI-S scores), and the parameter estimates were tested for differences from zero. For the exploratory analyses presented here, p-values were considered nominal and were not corrected for multiple comparisons. # Results Patients who were rated as ready to work at baseline and week 28 generally showed baseline scores corresponding to higher functioning (QLS \[total, and all four domains\] and WoRQ) and milder disease severity (CGI-S) as compared to patients rated not ready to work at baseline. Patients who shifted from No to Yes in work readiness showed least squares mean (LSM) change from baseline to week 28 (±SE) on QLS total scores of 14.3±2.2. This change was significantly greater than in the group of patients who were No at week 28 in work readiness (LSM change from baseline to week 28: 2.7±1.4; LSM difference: 11.6±2.6, 95%CI: \[6.5; 16.7\], p\<0.0001). Patients with Yes in work readiness both at baseline and week 28 also showed significantly greater LSM changes on QLS total scores (10.5±2.2) compared with patients who were No at week 28; LSM differences: 7.9±2.7, 95%CI: \[2.5; 13.2\], p = 0.0045. In similar analysis as for QLS total scores, patients who were ready to work at week 28 (either No to Yes or Yes to Yes) also had significantly greater improvements relative to patients who were not ready to work at week 28, across all QLS domains. QLS instrumental role domain scores were significantly improved in patients shifting from No to Yes in work readiness compared to patients who were No at week 28; LSM difference: 3.3±0.7, 95%CI: \[1.8;4.8\], p\<0.0001. In contrast, patients judged as ready to work both at baseline and at week 28 (Yes to Yes) did not show greater improvement on QLS instrumental role domain scores compared to patients who were No at week 28 in work readiness; LSM difference: 1.1±0.8, 95%CI: \[-0.4; 2.6\], p = 0.150. Across QLS items, patients rated ready to work at week 28 generally showed significant improvement vs patients not ready to work at week 28. On the specific items of QLS instrumental role (Items 9–12: occupational role, work functioning, work levels, work satisfaction), improvements of nearly 1 point on each item were found in the patients shifting from No to Yes in work readiness. In contrast, improvements on QLS items of work level and occupational role were not significantly different between patients Yes in work readiness both at baseline and week 28 versus patients who No at week 28 (Items 9 and 11). The logistic regression describes how QLS total and CGI-S scores predicted readiness to work at week 28. Significant associations between QLS total scores and work readiness at week 28 occurred at all visits. CGI-S scores at baseline did not significantly predict work readiness at week 28 (p = 0.2805), although CGI-S scores at subsequent visits were predictive (p≤0.0307). Furthermore, parameter estimates were generally higher when using QLS total scores as predictors (absolute values 0.686 to 1.033) as compared to CGI-S scores (absolute values 0.175 to 0.996). This suggests a closer association to work readiness, with faster onset between measured improvements, for QLS as compared to CGI-S. The parameter estimates for the predictive effect of QLS total and CGI-S scores generally increased during the study, indicating better predictions of work readiness when scores were measured closer in time to the outcome at the end of the study. The increase in the parameter estimates during the study was observed particularly for the CGI-S scores. # Discussion This post-hoc analysis of the QUALIFY study showed that after switching to atypical LAI therapy strong associations between improvement in functioning and work readiness were revealed, as measured with QLS and WoRQ, respectively. Robust and clinically relevant 14-point improvements from baseline to week 28 on QLS total score were observed in patients who concurrently shifted from No to Yes in work readiness. The 14-point improvement in QLS total score corresponds to more than twice the minimal clinically important difference. The association between the QLS domain of instrumental role function and work readiness was particularly noteworthy. Furthermore, QLS total scores better predicted outcome on work readiness than did the measure of symptom severity (CGI-S). This finding demonstrates that quality of life and functional improvement can lead to a tangible and meaningful outcome for the patient (i.e., the ability to work). Overall, these results suggest that improving work readiness is achievable in patients with schizophrenia, and thus can be an important goal of treatment. The QUALIFY study was the first to assess improvements in health-related qualify of life with concurrent systematic assessments of functional capacity to work and the data from this study provide a unique opportunity to investigate the association between treatment-related improvements in two distinct measures of functioning in schizophrenia. The patients who shifted from No to Yes in work readiness showed the largest changes on the QLS domain related to work—Instrumental role items on work role, level, performance, and satisfaction—and it is not surprising that the association between improvement in QLS and work readiness was strongest for QLS items related to work function. These results are also in line with a published outpatient study showing that the highest correlations between hours worked and QLS were also in the Instrumental role domain. Our results also indicate that the independent raters in the study (QLS-raters were blinded to treatment while the WoRQ-raters were not blinded to the treatment) assessed patient improvements similarly and highlight the consistency between these two scales. The large and consistent improvements in QLS total scores in patients shifting from No to Yes in work readiness suggesting a strong association between health- related qualify of life (QLS) and the functional capacity to work (WoRQ) is also supported by the QLS total scores at every visit significantly predicting work readiness at week 28 in the logistic regression. Although CGI-S scores also significantly predicted work readiness, the estimates of prediction were generally lower as compared to those obtained with QLS total scores. Thus, these associations suggest that general status in functioning (QLS) strongly predicts the possibility of attaining an important functional milestone (work readiness) in the treatment of schizophrenia. With regard to the CGI-S analysis, however, it should be noted that the study inclusion criteria restricted CGI-S scores to between 3 and 5. The limited range allowed for baseline CGI-S may have affected the parameter estimate of predictability and resulted in the absence of a significant association between CGI-S scores at baseline and work readiness at week 28. Nevertheless, the association between baseline CGI-S and work readiness at week 28 strengthened during the study, as illustrated by increases in parameter estimates for CGI-S scores at each subsequent post-baseline visit. Despite the time-dependent increases in CGI-S predictability, larger parameter estimates at all visits for QLS still suggest stronger predictive value of QLS total scores than for CGI-S scores on improvements in work readiness. The QUALIFY study used CGI-S as a surrogate measurement of clinical status, instead of symptom severity as measured by the Positive and Negative Syndrome Scale or the Brief Psychiatric Rating Scale. Previous studies have shown a close correlation between Positive and Negative Syndrome Scale, Brief Psychiatric Rating Scale and CGI-S. These post-hoc analyses suggest that improvements in health-related quality of life and functioning have a stronger association with readiness to work when compared to the association between global clinical impression and work readiness. This is in line with a previous report demonstrating that improvements in functioning are distinct from symptom improvements and improvements in functioning and quality of life extend beyond psychopathological changes. Furthermore, these recent guidelines for the treatment of schizophrenia also reinforce improvements in functioning and quality of life as main goals of treatment during the stable phase after ensuring that symptom remission or control is sustained. It should be noted that the present post-hoc analyses are exploratory in nature. Furthermore, the analyses were done with both treatment groups combined, and thus do not serve to compare AOM 400 and PP in the treatment of schizophrenia. However, we have previously reported pre-specified analyses showing superior improvements with AOM 400 (n = 136) vs PP (n = 132) on QLS total score, as well as significantly greater improvements on WoRQ total scores and work readiness with AOM 400 vs PP. In summary, the large and consistent improvement in mean QLS scores in patients shifting from No to Yes in work readiness suggests a strong association between health-related qualify of life (QLS) and functional capacity to work (WoRQ). These findings highlight that quality of life improvements can translate to tangible and meaningful outcomes for the patient, and importantly, suggest that increasing patients’ capacity to work may be a more realistic goal in the treatment of schizophrenia than previously realized. Evaluating capacity to work may further optimize functioning in schizophrenia by directing rehabilitation services to the subgroup with evaluated capacity to work. The WoRQ scale may be useful as a practical, brief, and systematic measure of work readiness that reflects broader functioning in patients with schizophrenia. # Supporting information [^1]: The authors declare the following conflicts of interest: Potkin SG has received grant support, funding, or honoraria or served as a consultant for the following companies that conducted scientific or medical research and/or marketed medications related to psychiatric and neurodegenerative disorders: Alkermes, Amgen, Baylor University, Eisai, Eli Lilly and Company, Forest, FORUM Pharmaceuticals, Lundbeck, Merck & Co., Inc., NIH, Novartis, Otsuka, Roche/Genentech, Sunovion Pharmaceuticals Inc., Takeda Pharmaceutical Company Ltd., Toyama, University of Southern California, UCSD, UCSF and Vanda Pharmaceuticals. Naber D: Advisory boards for Janssen/Cilag, Eli Lilly, Lundbeck, Otsuka and Servier; speaker honoraria from Astra Zeneca, Janssen/Cilag, Eli Lilly, Lundbeck and Otsuka. Loze JY: full-time employee of Otsuka Pharmaceutical Europe Ltd. Forray C, Sapin C, Beillat M, Nylander AG, Hertel P, Schmidt SN, Ettrup A, Eramo A, Hansen K: full-time employees of Lundbeck. Baker RA, Peters-Strickland T: full-time employees of Otsuka Pharmaceutical Development & Commercialization, Inc. Authors' employment by study sponsors does not alter our adherence to PLOS ONE policies on sharing data and materials. [^2]: **Conceptualization:** SGP JYL CF RAB CS TP-S MB AGN PH SNS A. Ettrup A. Eramo KH DN. **Data curation:** SNS. **Formal analysis:** SNS. **Funding acquisition:** TP-S AGN. **Investigation:** A. Ettrup SNS. **Methodology:** A. Ettrup SNS. **Project administration:** A. Ettrup SNS. **Resources:** SNS. **Software:** SNS. **Supervision:** TP-S AGN. **Validation:** SNS. **Visualization:** A. Ettrup. **Writing – original draft:** A. Ettrup. **Writing – review & editing:** SGP JYL CF RAB CS TP-S MB AGN PH SNS A. Ettrup A. Eramo KH DN.

# Introduction About a third of patients with localized renal cell carcinoma (RCC) treated by surgical resection will experience recurrence. However, there is currently no established adjuvant treatment for patients after complete tumor resection. A randomized trial conducted in the early 1980s comparing adjuvant radiotherapy after nephrectomy with observation showed no benefit of radiotherapy, with significantly increased post-radiation complications. Based on promising data regarding the management of metastatic RCC, several randomized trials subsequently compared adjuvant interferon-α (IFN-α), high-dose interleukin-2 (IL-2) or cytokine combinations with observation alone in patients with locally advanced, completely resected RCC. However, none of these trials showed any benefit of adjuvant treatment in terms of time to relapse or improved survival. Several phase III randomized controlled trials are currently investigating adjuvant treatment with tyrosine kinase or mammalian target of rapamycin inhibitors after nephrectomy in high-risk RCC. The first of these “new generation” studies, the ASSURE trial, randomized 1943 patients with RCC to sunitinib, sorafenib, or placebo following complete resection, reported that adjuvant treatment with sorafenib or sunitinib did not improve relapse-free or overall survival (OS) compared with placebo. Effective adjuvant treatment after nephrectomy, together with criteria for selecting suitable candidates, is therefore to be explored. A recent phase III randomized trial comparing adjuvant immunotherapy with low- dose IL-2 plus IFN-α with observation alone after nephrectomy reported that pT3a (compared with other pT stages) could be a positive predictive factor in patients treated with adjuvant immunotherapy, maintaining its prognostic role in those not receiving adjuvant treatment. In this context, the present study aimed to elucidate the optimal setting to maximize the benefit of adjuvant immunotherapy after nephrectomy in a Japanese population with RCC. # Materials and methods ## Patients and treatments This retrospective study was approved by the institutional review board and was conducted in accordance with the Declaration of Helsinki. We retrospectively reviewed 528 patients with pathologically confirmed RCC who underwent radical or partial nephrectomy at The University of Tokyo Hospital between 1981 and 2009. Distant metastasis at initial diagnosis (*n* = 49), prior nephrectomy at other institutions (*n* = 2), von Hippel-Lindau disease (*n* = 9) and insufficient clinical information (*n* = 32) were excluded from this analysis. A total of 436 patients with pT1-3N0-2M0 sporadic RCC who underwent either radical or partial nephrectomy with curative intent were finally reviewed, including 98 (22.5%) who received adjuvant IFN-α treatment after surgery (adjuvant IFN-α group) and 338 (77.5%) who did not (control group). Each treatment was assessed by physicians’ discretion. The adjuvant IFN-α regimens were as follows: Sumiferon^® (Sumitomo Dainippon Pharma, Osaka, Japan) 3–6×10⁶ IU; OIF^® (Otsuka Pharmaceutical, Tokyo, Japan) 5×10⁶ IU; or Intron^® A (Merck Sharp & Dohme, Tokyo, Japan) 3–6×10⁶ IU, injected subcutaneously two to three times per week. Pathological stage was re-evaluated according to the 7^th TNM classification of the Union for International Cancer Control (UICC) and the American Joint Committee on Cancer (AJCC) Guidelines. This TNM re-evaluation was conducted in a comprehensive manner, based on pathology reports, medical charts, radiogram interpretation reports, and so on. Histological subtype and tumor grade were assessed according to 3^rd World Health Organization Classification of Tumours and the Heidelberg classification, respectively. These are the main criteria currently used in Japan for the pathological diagnosis of RCC. All patients underwent preoperative and postoperative (every 1–6 months) evaluations, including routine blood tests, chest x-rays, and computed tomography. Bone scintigraphy was performed when indicated. Postoperative monitoring included routine chest x-rays every 3 months and/or chest and abdominal computed tomography every 6 months in the first 3 years, and yearly thereafter. ## Statistical analyses The primary endpoint was cancer-specific survival (CSS). Secondary endpoints were OS and recurrence-free survival (RFS). Survival curves were drawn using the Kaplan—Meier method. Univariate and multivariate analyses were performed using log-rank tests and Cox proportional hazards model, respectively. All statistical analyses were performed using JMP Pro version 11.0.0 (SAS Institute, Cary, NC, USA). A value of *P*\<0.05 was considered significant. Follow-up information was obtained up to December 2015. # Results ## Patient characteristics Fifty-two (11.9%) patients died from RCC with a median follow-up period of 96 months. The patient characteristics are summarized in. pT stage, pN stage, and grade were significantly higher in the adjuvant IFN-α group, but there were no significant differences in age, sex, histological subtype, and follow-up period between the two groups. The surgical procedures were open radical nephrectomy in 253 (58.0%), open partial nephrectomy in 120 (27.5%), laparoscopic radical nephrectomy in 61 (14.0%), and laparoscopic partial nephrectomy in two (0.5%) patients; there was no difference of the surgery type among the two groups. The data on adjuvant IFN-α duration were available in 51 of 98 (52%) patients: their median IFN-α duration was 10 months (range: 1–175 months). ## Treatment outcomes We conducted preliminary univariate analyses to compare CSS between the adjuvant IFN-α and control groups in each TNM setting. CSS in the control group was equal or superior to that in the adjuvant IFN-α group in earlier stages (pT1aN0, pT1bN0, pT2aN0), but the opposite trend was observed in more advanced stages (pT2bN0, pT3aNo, pT3b-cN0, pTanyN1-2). Based on these findings, we evaluated the TNM cutoffs and demonstrated that adjuvant IFN-α had maximal benefit in patients with pT2b-3cN0 (*P* = 0.0240). presents this result in another way: CSS in patients with pT2b-3cN0 was similar to that for pT1a-2aN0 in the adjuvant IFN-α group, but significantly worse in the control group. Similar trends were observed for the secondary endpoints (OS and RFS). In addition to TNM, histological subtype (clear cell vs. non-clear cell) and grade (G1-2 vs. G3) were also associated with CSS in univariate analysis. However, multivariate analysis only identified ≥pT3 and pN1-2 as independent predictors of poor CSS in the overall population. For reference, multivariate analysis in the subgroup of patients with ≥pT2 disease (*n* = 123) detected pN1-2 and omission of adjuvant treatment as independent poor prognostic factors. # Discussion The present study demonstrated that the benefit of adjuvant IFN-α after nephrectomy was detected in patients with pT2b-3cN0 RCC. CSS was significantly prolonged in this subgroup following adjuvant IFN-α treatment, compared with the control group. Adjuvant immunotherapy improved the prognosis of patients with pT2b-3cN0 tumors to a similar risk level to those with lower pathological stages. The CSS curves for patients with lower or higher pT were not significantly affected by adjuvant therapy, apart from a slight tendency towards a detrimental effect for pT1a-2aN0. Similar trends were also observed for the other endpoints of OS and RFS. Furthermore, although adjuvant IFN-α was not prognostic in the study population as a whole, it was a good independent prognostic factor in patients with ≥pT2 disease. To the best of our knowledge, six previously published randomized trials have compared cytokine-based (IFN-α and/or IL-2) adjuvant treatment with observation after nephrectomy, all of which failed to show any survival benefit of adjuvant immunotherapy. The results of previous retrospective studies assessing the efficacy of adjuvant immunotherapy were also generally disappointing. However, unplanned subgroup analysis of a randomized trial by Passalacqua et al. reported that pT3a (compared with other pT stages) could be a positive predictive factor in patients treated with adjuvant therapy, maintaining its prognostic role in the control group. The authors developed a scoring model comprising pN (N0 vs. N1-2), tumor grade (Fuhrman G1-2 vs. G3-4), pT stage (pT3a vs. others \[pT1-2 & pT3b-3c\] according to the 6^th edition of the UICC-AJCC TNM staging system), and age (≤60 vs. \>60 years), and observed better RFS and OS outcomes in the adjuvant-treatment arm in patients with higher scores (i.e., ≥2 vs. 0–1 factors among pN0, G1-2, pT3a, and age ≤60 years). The results of the current study were generally in accordance with this Passalacqua’s report, indicating maximal benefit of adjuvant immunotherapy in patients with pT2b-3c (around pT3a) but without nodal metastasis (pN0). It is necessary to point out that the TNM classification for pT3a has been modified over time, from being defined as the extension to “perinephric tissue, renal sinus, or contiguous into adrenal gland” (T3b for renal vein involvement) in 2002, to include “perinephric tissue, renal sinus, or renal vein” (the adrenal gland involvement was attributed to T4) in 2010. In the present study, the pathological stage was revised according to the 7^th TNM classification. IFN-α has established roles in the treatment of RCC in the metastatic setting and malignant melanoma in both adjuvant and metastatic settings, but its mechanism of action has not been fully elucidated. Researchers have speculated that IFN-α may exert its antitumor efficacy mainly by indirect immunomodulatory effects, involving several mechanisms. These include an increase in tumor- infiltrating cells, decrease in circulating regulatory T cells, manifestations of autoimmunity and development of autoantibodies, changes in cytokine concentrations, modulation of signal transducer and activator of transcription (STAT) 1/STAT3 balance in tumor cells and host lymphocytes, and normalization of T cell STAT1 signaling defects in peripheral blood lymphocytes. These indirect immunomodulatory effects are assumed to be enhanced under certain levels of tumor burden. Our results suggested that tumor-associated antigens might be highly presented to the systemic circulation in patients with ≥pT2b disease, which could strengthen the antitumor actions of adjuvant IFN-α (e.g., preventing further growth of micrometastases). Conversely, these indirect effects are less powerful and might thus be unable to improve the anti-tumor response in more advanced settings (pN1-2), resulting in an optimal response in patients with moderately advanced tumors (pT2b-3cN0). According to our analysis, adjuvant IFN-α was associated with poorer CSS compared with the control group in patients with earlier stage RCC, such as pT1bN0. It might be caused by the selection bias that patients with higher grade tumors were more assigned to adjuvant IFN-α treatment. However, the previous randomized trial also reported a similar trend of poorer outcomes associated with adjuvant treatment in earlier stages, suggesting that adjuvant immunotherapy may indeed have a detrimental effect in earlier stage RCC. Another randomized trial conducted in Japan also reported a similar trend, with higher RFS in the observation group compared with the interferon group in T1 or T2 subjects, but higher RFS in the interferon group over 3 years in T3 subjects, though the difference was not significant. No other study of the said six trials assessing cytokine-based adjuvant therapy after nephrectomy compared treatment outcomes in a TNM subgroup. Nevertheless, its minimal or negative impact on survival, together with its well-known adverse effects such as fatigue, headache, muscle pain, and depression, means that adjuvant IFN-α should be cautiously indicated. Attention should also be paid to the fact that these adverse effects of IFN-α are quite different from those of tyrosine kinase inhibitors including hypertension, hand-foot syndrome, rash and fatigue. This study had some limitations, including its retrospective design and potential selection bias. Given the mechanism of action of immunotherapy, the results of this study might be useful for trials of emerging immune checkpoint inhibitors. # Conclusions The benefit of adjuvant immunotherapy was most significant in RCC patients with pT2b-3cN0. Careful consideration is, however, required for interpretation of this observational study because of its selection bias and adverse effects of IFN-α. Further studies are required to validate these results and aid the establishment of optimal patient-selection criteria for trials of adjuvant immunotherapy after nephrectomy. # Supporting information We received no funding/grant support for this study. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** ST SB HF. **Data curation:** ST HF. **Formal analysis:** ST HF. **Investigation:** ST HF MO. **Methodology:** ST SB HF. **Project administration:** ST SB HF. **Resources:** TM. **Software:** ST. **Supervision:** YH. **Visualization:** ST. **Writing – original draft:** ST SB HF MB. **Writing – review & editing:** HM TN TF HK YI YH.

# 1 Introduction The formation of stable aggregates is very common in nature. For example, long- range attraction through chemotaxis can lead to aggregation of Dictyostelium cells or eukaryotic cells during development to form organs and blood vessels). But the formation of aggregates can also arise from Brownian motion and contact adhesion. Numerous examples can be cited, from inert particles such as colloids to living cells, but also in ecology where animals like mussels produce stable patterns by clustering. It has been shown that the living entities can, through this process, optimize at the same time protection against predation and access to food. In cancer, tumor cells circulating in the blood stream form aggregates that will become a metastatic tumor when settling in an organ. The merging of metastatic lumps, forming a larger aggregate, can also occur. It is now recognized that cells cultured in 2D at the bottom of a plastic Petri dish do not behave as they would do in their natural environment. For example, in vitro, an organization in 3D-clusters makes the aggregates more resistant to treatments compared to the same cells plated in 2D, in a Petri dish. Several factors can explain these different behaviors: first, the fact that the dimensionality is not the same (2D versus 3D) is important; second, cell-plastic interactions are often very strong and prevail over cell-cell interactions; finally, the plastic dish has a very high stiffness, often non realistic (for brain cells for example). Therefore, new approaches that allow cells to grow in 3D aggregates *in vitro* are being pursued. Aggregates and spheroids *in vitro*, formed on non-adhesive substrates, are considered as pseudo-tumors that can be used to study tumor development in more realistic conditions. Recently, in the context of brain tumors (gliomas), we developed a new PEG-based hydrogel that allows the formation of a tumor-like structure, which can be used to study the effect of drugs in conditions more realistic than those of a 2D Petri dish. In some of those gels, we grafted poly(L-lysine) (PLL), because of its ability to promote unspecific cell adhesion via electrostatic interactions between the polyanionic cell surfaces and the polycationic layer of adsorbed polylysine. Addition of PLL in the gels allows us to modulated cell-substrate adhesion. The stiffness of the gel can also be tuned, in order to mimic the stiffness of the natural matrix (in our case, the brain) corresponding to a specific cell line. In, we chose to study the behavior of two glioblastoma (which is the most aggressive type of gliomas) cell lines. Glioblastomas are currently non-curable and the development of an *in vitro* system that could mimic the development of these tumors could be used in order to test new drugs or radiotherapeutic strategies. Compared to other tissues, the stiffness of the brain is low (lower than 1 kPa), so we chose to design soft gels. We observed a significant difference in cell growth between PLL-containing (adhesive substrate) and PLL-free soft PEG hydrogels (non-adhesive substrate), showing the role of non-specific adhesion factors such as PLL in the migration, proliferation and aggregation in two glioblastoma cell line cultures. More precisely, we showed that on a non-adhesive substrate, the aggregates are larger and less numerous than on an adhesive substrate. The formation of aggregates has been studied theoretically with a perikinetic equation, first in the context of colloid aggregation. In, the same concepts are used to study and fit the evolution of the number of cell aggregates on a non adherent substrate. The evolution of the mean projected aggregate area is more difficult to model, especially when aggregates are composed of cells that can deform, modulate their cell-cell adhesion and reorganize in 3D. For example, it has been observed that after their formation, cell aggregates often go through a compaction phase that reduces their projected area. Other theoretical approaches consider the formation of aggregates under the point of view of phase separation: like two immiscible liquids, when mixed in a liquid medium cells move and seek a lower energy state through adhesion with other cells. The evolution of a system from a state where the concentration of particles is uniform to a final state where patterns appear is a spontaneous phase transition driven by motion of particules, the latter being either passive by diffusion (for example, in colloids), or active (as for mussels or cells) and adhesion. This phase separation corresponding to the formation of aggregates has been described by mean-field models, based on the Cahn-Hilliard. With chemotaxis, Keller-Segel equations can also be used to describe pattern formation. In development biology, discrete approaches, in particular with cellular Potts models have been used to model the formation of patterns or the segregation of two cell types in aggregates. However, to our knowledge, a model that describes the formation of aggregates from the early stages of a population of individual migrating cells to their aggregation and the late aggregate compaction, does not exist. For instance, in, the decrease of the mean aggregate area is modeled with an exponential function, but there is no direct connection with processes at the cellular level. In order to be able to describe the individual cells, an agent-based model should be chosen. One advantage of agent-based models is that one can easily implement the local rules of cell-cell interaction (\[–\], for a good review, see). In, we presented the snapshots of already formed aggregates. Here, we add new experimental results by following the whole process, from the early stage where the cell population is composed only of individual cells, to their aggregation, and later to the compaction of the aggregates. We confirm the differential migration and aggregation of cells on the substrates with different adhesivity and for two different cell lines. We combine these experimental results with a theoretical study based on two models: first, we show that a spaceless model of perikinetic aggregation can reproduce the experimental evolution of the number of aggregates. Second, we developed a minimal off-lattice agent-based model, whose rules are defined in order to reproduce the important phenomena that drive the behavior of cell assemblies: cell and aggregate motion, cell-cell adhesion, cell proliferation and aggregate compaction. We show that this model reproduces very well the experimental temporal evolution of both the number of aggregates and their area, on adhesive and non-adhesive soft gels, for the two cell lines and that it gives access to quantitative values of three parameters. # 2 Materials and methods ## 2.1 Preparation of the hydrogels and glioma cell lines In, we showed that the PEG concentration of our artificial substrate optimal for the survival and growth of the two glioma cell lines is around 3% PEG. This concentration corresponds to an elastic modulus around 300 Pa, close to the value measured for brain tissue. We use this concentration in all the following experiments. All the experimental methods can be found in. Briefly: poly(L-lysine) hydrobromide (PLL-HBr 30,000 Da, Sigma-Aldrich, Saint-Quentin Fallavier, France) was first functionalized with an acrylate residue. Hydrogels were prepared from 3% (w/v) PEG-DA 6 kDa precursor (Sigma-Aldrich), dissolved in DPBS with 0.01% (w/v) of DMPA solubilized in VP. Precursor solutions were photopolymerized under UV (UV-LED LC-L1; Hamamatsu, 2 W/cm², λ = 365 nm) for 40 s in homemade cylindrical dishes. Photopolymerized hydrogels were then incubated during 1 day in a high volume of DMEM for the hydrogel structure to be hydrated and thermodynamically stable before cell seeding. After this day of hydration and two rinsings with fresh medium, cells were seeded upon hydrogels at 2 10⁵ cells/well (or 10⁶ for the high-density experiments). To avoid cell medium acidification, the cell culture medium was replaced by fresh medium every day. The two glioma cell lines, F98 from rat model and U87-MG from human glioma, were provided by ATCC (CRL 2397 and HTB-14, respectively). Cells were maintained in Dulbecco’s Modified Eagle’s Medium (DMEM; Life Technologies, Courtabœuf, France) added with phenol red as pH indicator, supplemented with 4.5 g/L of D-glucose and pyruvate, 10% (v/v) fetal bovine serum (Life Technologies), and 1% (v/v) of penicillin and streptomycin antibiotics (Pen-Strep, Life Technologies). Cells were disposed into sterile culture flasks with anti-fungal filters to limit contamination and maintained in culture at 37°C, under a humidified atmosphere of 95% relative humidity with 5%CO₂. Cells were replicated when they attained 80%–90% cell confluence. In what follows, we call “non adhesive gels” gels composed of PEG only and “adhesive gels” gels containing PEG and PLL. ## 2.2 Microscopy and image processing Microscopy image acquisition was performed using a 10x Nikon water-immersion objective placed on an Eclipse 80i Nikon microscope (Scop Pro, Marolles-en- Hurepoix, France) which was equipped with Differential Interference Contrast (DIC) device. The sample, the stage and the objective were completely enclosed in a chamber that allows the fine control of temperature, humidity and CO₂ pressure on the living sample (Box and Cube system by Life Imaging Services, Basel, Switzerland). Cells were deposited on the gel surface and rapidly placed under the microscope. We performed time-lapse imaging (1 picture every minute) to follow the formation of the aggregates. Micrographs series were obtained using a Zyla 5.5 MPX Andor SCMOS cooled camera (Scop Pro, France) and MetaMorph acquisition software (Molecular Devices, Sunnyvale, CA). Image processing (from experiments and simulations) was performed with the Fiji software. Customized Fiji macros were developed to detect cells and aggregates in experiments and simulations. We used Python for further analysis of the data. More details are available in the. ## 2.3 The spaceless model The number of aggregates during a perikinetic aggregation process can be predicted by the model of Smoluchowski: $$\begin{array}{r} {\frac{dN_{k}}{dt}(t) = \frac{1}{2}\sum\limits_{i = 1}^{k - 1}K_{i,k - i}N_{i}(t)N_{k - i}(t) - \sum\limits_{i = 1}^{\infty}K_{ik}N_{i}(t)N_{k}(t)} \\ \end{array}$$ where *N*_*k* is the number of aggregates of size *k*, (*K*_*ij*) is the kernel aggregation rate, where the element *K*_*ij* is the aggregation rate between clusters of size *i* and clusters of size *j*. We tried two different scenarii for the aggregation process. In the first scenario, all the aggregates move and interact with other aggregates with the same constant rate *K*. In this case, the model is solvable analytically and the evolution of the total number of aggregates at time *t*, *N*(*t*), is described by the equation: $\frac{dN}{dt}(t) = - KN(t)^{2}$, and thus *N*(*t*) = *N*₀/(1 + *K N*₀ *t*). This model was used in to fit the evolution of aggregate number on a non-adhesive substrate. In the second scenario, only individual cells move and can interact with other individual cells or with aggregates (*K*_1*j* = *K*_*j*1 = *K*_1*i* = *K*_*i*1 = *K*₁ ∀*i*, *j* and *K*_*ij* = 0 if *i* \> 1 and *j* \> 1). The total number of aggregates, *N*(*t*) is given by the following equations, that we solved numerically: $$\begin{array}{r} {\frac{dN}{dt} = K_{1}N_{1}\left( \frac{1}{2}N_{1} - N \right)} \\ \end{array}$$ $$\begin{array}{r} {\frac{dN_{1}}{dt} = - K_{1}c_{1}N} \\ \end{array}$$ $$\begin{array}{r} {\frac{dN_{2}}{dt} = K_{1}\left( - \frac{1}{2}{N_{1}}^{2} - N_{1}N_{2} \right)} \\ \end{array}$$ $$\begin{array}{r} {\frac{dN_{i}}{dt} = K_{1}N_{1}\left( N_{i - 1} - N_{i} \right)\text{for}\mspace{720mu} i \geq 3.} \\ \end{array}$$ where *N* is the total number of aggregates. ## 2.4 The spatial model We define an agent-based model that involves a collection of agents evolving in a continuous 2D surface. Each agent is a cell, modeled by a disk. The disk radius is the same for all cells in a given simulation and its value stays constant during the simulation. The simulation space is lattice-free, i.e. the position of each cell is an ordered pair of real numbers corresponding to the coordinates of the center of the sphere. In experiments, the field of view is a rectangle of size 1280 per 1080 pixels (each pixel is a square of side length 0.658 *μ*m). In the simulations, we choose a square as 2D surface by simplicity. Since the experimental field of view represents a small part of the whole surface of the substrate, it is devoided of boundary effects; therefore, we choose periodic boundary conditions in the simulations (walls or other closed boundaries would induce strong artefacts). The unit of length in simulations is set so that the side of the square has the same length as the length of the region observed in experiments (842.24 *μ*m). At each iteration, all the cells are updated, one by one and in a random order in order to avoid undesirable correlations. At each iteration, each cell participates in the following processes: motion, both individual and collective, influenced by cell-cell adhesion and aggregate compaction, and proliferation. One iteration corresponds to one minute. For the sake of clarity, in the following, we call “individual cells” cells that are not part of an aggregate (individual cells have no neighbors). An aggregate is thus an assembly of at least two cells. ## 2.5 Rules of the spatial model ### 2.5.1 Cell motion At each iteration, individual cells choose a direction uniformly at random and move by a step *a*₀ in that direction. For a given simulation, this step length is constant during time and is the same for all the cells. Once cells are part of an aggregate, they continue to move, and their motion is the composition of two motions, the motion of each cell inside the aggregate and the motion of the whole aggregate. The diffusion coefficient of a spherical particle of size *r* in a viscous medium is proportional to 1/*r*; thus, for a 2D aggregate comprising *N* cells, it is proportional to $1/\sqrt{N}$. We keep the same dependency here: the motion of the whole aggregate is assumed to be random and the step of this motion is chosen so as to decrease as $1/\sqrt{N}$. Regarding the motion of each cells inside the aggregate, the length of the step is chosen as a decreasing function of the number of neighbors *n*: *a* = *a*₀/(1 + *n*²). The choice of this function is purely phenomenological. Another choice could have been that of where the probability of migration decreases with the number of neighbors, being proportional to (1 − *q*)^*n*, where 0 \< *q* \< 1 is the adhesion parameter and *n* is the number of neighbors. It turns out that the precise choice is irrelevent, since the results do not depend on the exact dependence of the step length on the number of neighbors (for reasonable choices of the corresponding expression). In the non-adhesive case, individual cells are sometimes subject to a hydrodynamic flux (due to gel local heterogeneity) that can bias the cell motion: when the flux is “on” in the simulation, the direction of the motion in the simulation is limited to a half plane (defined by the direction of the flux), both for individual cells and aggregates. ### 2.5.2 Superimposition In order to model the fact that cells are deformable and can be organized in three dimensions in aggregates, cells are allowed to partially superimpose in the model. The maximum superimposition is quantified by a parameter *α*_max that is defined as the ratio of the overlapping length (i.e. the difference between the diameter of a cell and the minimum possible distance between two cell centers) to the diameter of a cell. If *α*_max is close to 1, two cells can superimpose almost completely, whereas if *α*_max is close to 0, the cells stay well-separated. The value of this parameter is constant during a given simulation but can vary from one simulation to the other, in order to model different cell lines. This parameter is similar to the stacking index used in to indicate the formation of aggregates with some vertical stacking, forming multilayered clumps. ### 2.5.3 Cell-cell adhesion If during its motion, the position of a moving cell would break the superimposition rule, the motion is prematurely stopped and the two cells adhere to each other. There is no break-up mechanism and so cells cannot detach from their neighbors (except if its step leads it to a position where it has at least the same number of neighbors). Therefore, if the cell chooses a direction of motion that would lead to a smaller number of neighbors after performing the step, this step is canceled and the cell does not move. ### 2.5.4 Compaction We do not have precise details on what happens inside aggregates during compaction, so we chose to model the effect of cell compaction and 3D organization with a simple yet efficient empirical rule: we bias the individual cell motion in an aggregate towards the center of mass of the aggregate. Since this reorganization is much more visible for larger aggregates, we decided to modify accordingly the cell motion. When the aggregate is very small, its cells may move towards any direction. However, when the aggregate becomes more massive, the cell motion is biased towards the center of mass of the ensemble. At the limit of a very massive aggregate, the motion is possible only in a ± 90° sector around the line joining the cell with the center of mass. In practice, at each iteration and for each cell, a random angle for the cell motion is drawn in the angular sector ±90(1 + *e*^−*n*/40) (in °) with *n* being the number of cells in the aggregate. The overall effect does not depend crucially on the precise form of the function. This bias in the direction is similar to the one in where the direction choice is biased towards the direction with the higher number of cells within a distance corresponding to several cell diameters. ### 2.5.5 Proliferation We model the cell division by the following process: after the cell has moved it has a certain probability to try to proliferate. That probability per iteration is later on referred as “proliferation rate”. If the cell division process is engaged, a random position around the dividing cell is chosen. If that position is compatible with the superimposition rule (meaning that the distance between the daughter cell and all the other cells should be larger than the minimum distance allowed by the superimposition coefficient), the daughter cell is created. If not, the division process is aborted. ## 2.6 Typical workflow Simulations yield the positions of the cells, that are used to produce images representing the same observed area as in experiments. The size of individual cells in simulations is chosen equal to the mean size of cells in the experiments, so that the simulation results can be analyzed the same way as the experimental results, and we can then directly compare the dynamics of the mean area and of the numbers of aggregates in the two approaches. More details are available in the. # 3 Results ## 3.1 Experimental observations On top and on the different movies of aggregation, we observed several phases in the aggregation process ( for F98 and for U87). The first phase is the random motion of individual cells. Cells remain round, do not polarize, and do not adhere strongly to the substrate. However, they move, extend small filopodia, and perform a random motion (see and Videos). After analyzing the motion of 25 cells (F98) during one hour, we plotted the mean square distance covered versus time and we deduced that the cells have a diffusion coefficient of 1.6 ± 0.58*μ*m² min⁻¹ on the adhesive substrate, close to the value found in. The second phase is the formation of aggregates: during this random motion, individual cells encounter other cells or already formed aggregates and new aggregates begin to form. When this happens, the individual cells stick to the other cells or to the aggregates; becoming unstuck is so rare that we neglected this phenomenon. The third phase corresponds to the dynamics of formed aggregates: they move as a whole, with small aggregates exhibiting a global motion larger than bigger ones. Moreover, the inside of the aggregates is also dynamic: the cells inside move and reorganize constantly (see and Videos). During the aggregation phase, which lasts around 2 hours, a few events of proliferation are visible among individual cells. This proliferation continues within aggregates, increasing their size continuously. The last phase corresponds to the compaction of the already formed aggregates: a few hours after the formation of aggregates, they compact and reorganize into a three-dimensional shape. The projection of this shape in 2D is close to a disk. Experimentally, compaction occurs only for the F98 cell line. For the U87 cell line, this aggregate compaction does not occur and aggregates stay in a 2D configuration, see. We measured, in the experimental field of view, the mean area of the aggregates and their number as a function of time, for adhesive and non-adhesive substrates, for the F98 and the U87 cell lines. We define the normalized number of aggregates as the raw number of aggregates divided by the number of individual cells at initial time. In, the normalized number of aggregates (top) and the mean area of aggregates (bottom) are represented as a function of time, for non-adhesive (blue curves) and adhesive (brown curves) substrates, for the F98 cell line. The number of aggregates (respectively mean aggregate area) decreases (resp. increases) faster in the case of the non-adhesive substrate. ## 3.2 Comparison between experimental data and the spaceless model We compared our experimental results with the theoretical model for aggregation developed by Smoluchowski. In the case of a non-adhesive substrate, we found that the best fit was obtained with a constant aggregation kernel *K*_*ij* = *K*. From the fitting procedure, we found *K* = 2.6 10⁻¹³ m²s⁻¹, see, top (the dashed blue curve is obviously a better fit of the experimental data than the dotted blue curve). In the case of an adhesive substrate, we found that the best fit is obtained with the solution of the equations corresponding to the scenario where only individual cells can move and interact with the other aggregates. In this case, we found *K* = 6.4 10⁻¹⁴ m²s⁻¹ = 3,8 *μ*m²min⁻¹, see , top (the dotted brown curve is obviously a better fit of the experimental data than the dashed brown curve). This value is of the same order of magnitude as the cell’s diffusion coefficient on adhesive substrate calculated above. ## 3.3 Comparison between experimental data and the spatial model We model two cell lines: F98 cells (13.2 *μ*m mean diameter) and U87 cells (21.1 *μ*m mean diameter). ### 3.3.1 Qualitative comparison The rules of our model (sketched in have been defined in order to mimic what happens in the experiments: therefore, in the model, cells can move, adhere to other cells, form aggregates that can compact subsequently, and proliferate. In, it is clear that cells move inside an aggregate, and since aggregates also have a motion on their own, the motion of each cell should be a composition of the two motions, see, left. It is well known that aggregation limited by diffusion only, without any surface tension, leads to clusters with fractal shape. In our case the number of cells is not large enough to lead to fractals, but without any rule of compaction, especially in simulations with a high initial cell density, aggregates have the shape of long and branched thick filaments and do not organize themselves into more compact shapes. But in experiments at high cell densities, after about 12 hours after the beginning of the aggregation process, the aggregates compact and become more circular (in 2D) (see, middle). The effect of the compaction rule in the case of a high initial cell density is visible on and leads to aggregates that have a compact shape, close to the experiments. The qualitative comparison of the experimental and the simulation results can be performed from, which shows side by side an experiment and a typical simulation result, obtained with our model. ## 3.4 Quantitative comparison between experimental data and the results of the spatial model The experimental results consist of 9 24-hour experiments: 6 experiments for the F98 cell line (3 in the adhesive and 3 in the non-adhesive condition), and 3 experiments for the U87 cell line (one in the adhesive and two in the non- adhesive condition). We decided to fit each experimental curve (not only the mean), so the parameters were independently set for each experiment. ### 3.4.1 Choice of parameters The number of initial cells *N*₀ was deduced from the first image taken in each 24-hour experiment and was used as the initial condition of the corresponding simulation, a uniformly random distribution of *N*₀ individual cells. The superimposition parameter was chosen constant for all the experiments with the same cell line. There is a net qualitative difference between the behavior of the two cell lines: the aggregates of F98 cells compact and clearly organize in three dimensions, whereas the U87 cell aggregates keep an almost two-dimensional organization. So the superimposition parameter *α*_max should be larger for the F98 than for the U87 cells, see Figs and. We can clearly detect when the parameter is too small, because then, the mean aggregate area in the simulations is too large, even for very small aggregates at the beginning of the experiment and even if the evolution of the number of aggregates is correct, see the red stars with *α*_max = 0.2 in. It is more difficult to detect when the parameter is too large: the difference between the cyan stars with *α*_max = 0.95 and the green stars with *α*_max = 0.7 in is not obvious. Suppose that *α*_max = 1, all the cells in an aggregate could in theory superimpose and the area of any aggregate could be reduced to the area of a single cell. But since the motion step length of cells diminishes with the number of neighbors in aggregates, this process takes a lot of time, and it is not possible to see the complete superimposition of all the cells in an aggregate, during the time of experiments. We thus decided to infer the value of *α*_max from images of superimposition of two cells, see. From these images, it is clear that F98 cells allow a minimal distance between the cell centers smaller than the U87 cells, and we chose the value of *α*_max = 0.7 for the F98 cell line and *α*_max = 0.2 for the U87 cell line. Two parameters still need to be set: the step length of individual cells and the proliferation rate. We determined the step length of individual cells so that the decrease of the number of aggregates corresponds to experimental data: if cells move too slowly, this number decreases also too slowly compared to experimental data (see, cyan stars). If the step length is too large, the number of aggregates decreases too fast compared to experimental data (see, red stars). The green stars correspond to the best value of the step. The proliferation rate was chosen so that the mean area in simulations would fit the corresponding experiment at large times (the increase in the aggregate area after formation is only due to proliferation), see. The value *κ* = 10⁻⁴ min⁻¹ (cyan stars) is too small, the value *κ* = 1.4 10⁻³ min⁻¹ is too large and the value *κ* = 7 10⁻⁴ min⁻¹ (green stars) is correct. We added the bias of the flux if visible in the experiments. Without any flux, aggregates still move but their motion is very small and the distance between them is too large to allow any collision. When the flux on individual as well as on aggregates is on, that corresponds to the green stars in, collisions are possible between large aggregates and the number of aggregates decreases even at large times. We managed to reproduce the dynamics of both the mean area and the number of aggregates of each experiment and for both cell lines used. Results are shown in for the F98 cell line and in for the U87 cell line and for both adhesion conditions. # 4 Discussion We present here a combination of experimental and simulation results on the behavior of a cell population on soft hydrogels. On these soft gels, cells stay round, move and stick to each other to create aggregates, within one day (this is much shorter than the formation of glioma cell aggregates on an adherent rigid substrate such as plastic, where the time scale is one month). The shape and the size of these aggregates depend on the nature of the gels (adhesive or not adhesive), but also on the cell line: U87 aggregates are less cohesive than F98 aggregates, and for both cell lines, aggregates are smaller and more numerous on adhesive substrate (with PLL). First, we compared the experimental data with the solutions of perikinetic equations. We found that the experimental non-adhesive and adhesive cases correspond to two different scenarii: in the non-adhesive dynamics of aggregation, a constant kernel leads to a better agreement with the experimental data, whereas the adhesive case is well fitted by a kernel that is non-zero only for particles of size 1. We found the kernel value *K* = 2.6 10⁻¹³ m²s⁻¹ in the non-adhesive case, and a four-time smaller value of *K*₁ in the adhesive case *K*₁ = 6.4 10⁻¹⁴ m²s⁻¹. These values are consistent with other studies, where only the case of a constant kernel is compared to experimental data. The agreement with the theoretical spaceless model is fair, but the model describes only the evolution of the number of aggregates as a function of time, whereas in our experiments, the area was also recorded. We thus developed an agent-based model with simple rules that could reproduce as well the first stages of the experiments, when cells still move as individual cells, as the late stages where cells are in aggregates, and that could reproduce the experimental evolution of both the number and the mean area of the aggregates. To model all these stages, without describing precisely the shape of the cells, we estimated that a cellular Potts model was less adapted to our problem, compared to a classical agent-based cellular automaton. We introduced four rules: the motion rule (for individual cells, for cells inside an aggregate and for aggregates, in the presence or not of a flux), the superimposition rule, the proliferation rule and the compaction rule. The superimposition and the compaction rules may need further justifications: Since the precise 3D organization in aggregates concerns only the F98 cell line, and to follow the approach of where a stacking index is defined, we decided to keep our model in 2D and introduce an effective parameter of superimposition, that describes the strength of cell-cell adhesion and their ability to organize in 3D. This approach has the advantage of simplicity, since only one parameter can resume the differences between the behavior of the two cell lines. The compaction that arises after the formation of cellular aggregates is a collective effect of a cell population. In experiments, compaction of aggregates is due to individual cells that deform and flatten their membranes against each other, increasing cell–cell contact and minimizing intercellular spaces. This process is similar to the embryo’s compaction, see. Compaction may also be due to the possible formation of supracellular stress cables, at the scale of the whole aggregate. This made us define a compaction rule that is non-local: the cells’ motion is biased towards the center of their aggregate and this bias increases with the size of the aggregate. Actually, the limitation of the motion is not severe: the maximum bias (in very big aggregates) restricts only the cell motion to a half plane towards the center of the aggregate. With this agent-based model, in 2D and with simple rules, we were able to reproduce the behavior of two cell lines, namely the evolution of the number of aggregates and of their projected area, on two different substrates, one adhesive (with PLL) and one not (without PLL). More importantly, by fitting the number of aggregates and the mean area of aggregates as a function of time, we were able to infer quantitatively several properties of the two cells lines, on the two substrates: their speed of motion, their proliferation rate, their superimposition coefficient and the capacity of aggregates to compact. First, our model allowed us to conclude that the effect of the presence of PLL in the gel (more adhesive substrate), for both cell lines, could be modeled as a simple slowing effect on cells. On non-adhesive hydrogels, there is often a flux (probably due to inhomogeneities of the gels) which gives to the cells and aggregates a motion bias (direction in only a half plane), that was taken into account in the model. We found that in the cellular automaton, in order to model the adhesive substrate, we had to decrease the step and remove the flux (i.e. remove the restriction of the motion to a half plane). For F98 cells, the mean speed motion of F98 is 4.7 ± 0.7 pixel length min⁻¹ = 3.1 ± 0.4 *μ*m min⁻¹ on non-adhesive substrates, whereas it is equal to 1.5 ± 0.3 pixel length min⁻¹ = 1.0 ± 0.2 *μ*m min⁻¹ on adhesive ones. For U87 cells the mean speed of motion are respectively 6 pixel lengths min⁻¹ = 3.9 *μ*m min⁻¹ and 2 pixel length min⁻¹ = 1.3 *μ*m min⁻¹. The surface density of PLL molecules estimated from the volume concentration of 0.001% (w/v) PLL in the hydrogel precursor solution for a hydrogel thickness of 2 mm, is about 5 10¹¹ molecules mm⁻². About 5 10⁵ PLL molecules are found every *μ*m². The F98 and U87 MG cells have a radius between 8 and 20 *μ*m so they move on a quasi-homogeneous surface of PLL molecules. The cells make smaller steps on PLL hydrogels because they are constrained in their motion by the electrostatic interactions they form with the PLL. We also had to change the proliferation rate between the two cell lines. We found that the proliferation rate (in aggregates) is 7.10⁻⁴ min⁻¹ for the F98 and 3.10⁻⁴ min⁻¹ for the U87 cell line. Our results also reveal that the adhesive properties of the substrate do not impact the proliferation rate in aggregates: it is the same in the two conditions (adhesive and non-adhesive substrate), for each cell line. The U87 cells are characterized by a weak adhesion between cells, leading to loose aggregates, whereas the F98 cells are much more cohesive. Moreover, there is no late compaction of the aggregates and the aggregates stay in 2D instead of organizing in 3D as in the F98 case. The results for both lines could be reproduced: in order to describe the U87 cell line we had to use a superimposition parameter smaller (*α*_max = 0.2) than the one used for the F98 line (*α*_max = 0.7) and to remove the compaction rule. For two cell lines, we show here that by using our spatial model to fit the temporal evolution of the number of aggregates and their mean area, it is possible to infer the quantitative values of speed motion, proliferation and the qualitative abilities of cells to adhere to each other and to deform. It should be possible to study any cell line, providing that the gel stiffness is optimized for this cell line (indeed, carcinomas develop on a much stiffer substrate than gliomas, this was confirmed by a preliminary study of ours on the breast cancer cell line MCF7 which could not form aggregates on soft gels, dying rapidly). It has been shown for example that the cohesivity of aggregates (due to cell-cell adhesion) could be a clinically important parameter, since it seems to be inversely proportional to the *in vitro* invasive potential. One promising direction of research is that of the study of cell lines from cancers which are known to develop metastases, such as breast cancer. In that case, our experimental technique should be adapted. Another interesting direction of future studies would be to modulate the adhesion. This could not be done in the present studies using PLL since a higher concentration of this molecule becomes toxic for the cells. Using an other adhesion molecule, such as RGD instead of PLL, we expect to be able to vary adhesion and study the aggregation phenomena for a wide range of values of the latter. We expect to return to address these problems, both experimentally and through modeling, in some future work of ours. # Supporting information This work was supported through a grant from Région Ile-de-France (‘DIM Problématiques transversales aux systèmes complexes ISC-2014-PME-003”) and a grant from the GEFLUC association (association des Entreprises Françaises dans la Lutte contre le Cancer). [^1]: The authors have declared that no competing interests exist.

# Introduction The immune system of fish provides important information about conserved processes in the mammalian immune system. Studies of antibody production in fish have revealed much about the phylogeny of acquired immunity and immunoglobulins and this has led to a better understanding of the overall functionality of the immune system in all vertebrates (reviewed in). Fish are an excellent model for studying innate immunity since their innate immune components are homologous to those of mammals. Their acquired immunity differs from more advanced vertebrates in the length of time needed to initiate a specific immune response because of their poikilothermic nature. Furthermore, other than the mouse model, the rag1^−/− mutant zebrafish is the only animal model available for investigating T and B cell deficient immunological responses. Interestingly, fish are not immunologically mature when they hatch. Acquired immunity, utilizing fully functional T and B cells, does not develop until 3 to 6 weeks post-hatch, depending on the species. In previous work channel catfish (*Ictalurus punctatus*) larvae that had minimally organized lymphoid tissue produced a protective secondary response to a bacterial pathogen. This suggested that in fish, there is an adaptive component to innate immunity. Historically, immunological dogma described the innate immune response as acting naïvely to each encounter with a pathogen depending on the recognition of conserved molecular patterns and exhibiting only weak specificity. T and B cells mediate protective secondary immune responses of the acquired immune system, and immune-deficiencies develop in their absence. However, evidence of adaptive responses of cells of the innate immune system of mice to haptens and cytomegalovirus have been demonstrated. Natural Killer cells, an innate lymphocyte population, can mount antigen-specific immunological memory. Innate immune system memory may be present in, and a more critical component of, lower vertebrate immunity. We used zebrafish (*Danio rerio*) to investigate an adaptive component of innate immunity because specific mutants are available and zebrafish are recognized as infectious disease models. Numerous regions of synteny between the zebrafish and human genomes have been identified allowing immunological findings in zebrafish to be translated to higher vertebrates. Rag1**^−/−** mutant zebrafish created by a reverse genetic approach have been shown to lack VDJ recombination. After establishing a breeding colony, we further characterized the rag1^−/− mutant zebrafish to confirm lack of T cell receptor (TCR) and immunoglobulin (Ig) transcript expression. Therefore these fish do not have mature T and B cells and thus are a unique model for characterizing innate immune system memory in fish. In this study, rag1^−/− mutant zebrafish were given a primary (vaccination) exposure to a low dose of a bacterial pathogen, or a sham exposure. To determine protection, or how well the vaccination worked, a secondary high dose exposure was delivered at either four weeks (Trial 1) or eight weeks (Trial 2) after the primary. The pathogens used were members of the Enterobacteriaceae family: *Edwardsiella ictaluri, Yersinia ruckeri* and *E. tarda*. *Edwardsiella ictaluri* causes Enteric Septicemia of Catfish and a comparable disease in zebrafish. *Edwardsiella ictaluri* RE-33 is an attenuated live vaccine **(**AQUAVAC-ESC® Intervet, Inc.**)** that was used for the vaccination (primary) exposure of *E. ictaluri*. *Yersinia ruckeri* causes Yersiniosis or Enteric Red Mouth Disease (ERM) primarily in salmonids. Infection trials in our lab demonstrated susceptibility of zebrafish to *Y. ruckeri* following intramuscular (IM) injection. *Edwardsiella tarda* produces localized and systemic infections in a wide variety of vertebrates and has been shown to establish infections in zebrafish. The aim of this study was to utilize T and B cell deficient rag1^−/− mutant zebrafish to investigate adaptive protection of the innate immune system in response to repeated bacterial exposure. # Materials and Methods ## Animal Care Rag1^−/− mutant zebrafish were housed in the Mississippi State University College of Veterinary Medicine (MSU-CVM) specific pathogen free (SPF) fish hatchery. Fish were propagated according to modified standard protocols posted at: <http://www.cvm.msstate.edu/zebrafish/index.html>. The Institutional Animal Care and Use Committee at Mississippi State University approved all experimental animal protocols. ## Genotype Control All rag1^−/− mutant zebrafish used for this study were bred at the CVM-SPF fish hatchery. We established a homozygous rag1^−/− mutant zebrafish breeding colony, and all of our experimental fish are progeny from this colony. For an additional genotype control we sub-sampled zebrafish (10 fish per spawn) shortly before the experiment and the genotype, rag1^−/−, was confirmed using previously established PCR protocols. ## Preparation of Bacterial Cultures *Edwardsiella ictaluri*, *Yersinia ruckeri* and *Edwardsiella tarda* were case isolates from fish submitted to the Fish Diagnostic Lab at CVM-MSU. Culture identifications were confirmed by biochemical analysis using the bioMerieux api20E strip (BioMerieux, 69280 Marcy l’Etoile, France). Aliquots (0.5 ml) were stored in 20% glycerol at −80°C until needed for trials, at which time one aliquot was thawed and added into Brain Heart Infusion broth and incubated in a shaker incubator at 30°C overnight. Logarithmic phase cultures were obtained by dilution of the overnight culture 1∶10 and grown until the optical density was 0.4 at 540 nm which corresponds to 10⁸ colony forming units (CFU) per ml. Culture purities were assessed and bacterial concentrations determined by plating serial dilutions on 5% sheep blood plates. ## Lethal Dose (LD\>80) Determination In separate trials, rag1^−/− mutant zebrafish were injected with *E. ictaluri*, *Y. ruckeri,* or *E. tarda* (10⁶, 10⁵, 10⁴, 10³, 10², or 10¹ CFU/fish) to determine LD_\>80 dosage for the secondary exposure, referred to as the protection exposure (to determine if the vaccination exposure provided protection). Injections of rag1^−/− mutant zebrafish were performed using four replicate tanks per treatment with 15 fish per replicate. Additionally 15 control fish per strain were sham injected. Mortalities were recorded for 18 days post injection (dpi), and LD_\>80 dosages were determined from these data. ## Bacterial Injections and Experimental Observations Primary and secondary injections were carried out as described in. In all trials, adult (6–9 month old) rag1^−/− mutant zebrafish were anesthetized in 110 mg/L buffered tricaine methane sulfonate (MS222). Each fish was IM injected on the lateral line above the anal fin using an insulin syringe. All primary vaccinations were 10⁴ CFU/fish RE33, a commercial, live, attenuated *Edwardsiella ictaluri* vaccine strain (AQUAVAC-ESC® Intervet, Inc.). The secondary exposure was delivered at either 1 month (Trial 1) or 2 months (Trial 2) post-vaccination and consisted of one of the following bacteria: 10⁴ CFU/fish *Edwardsiella ictaluri*, 10⁶ CFU/fish *Yersinia ruckeri*, or 10² CFU/fish *Edwardsiella tarda*. These dosages have been established in our laboratory as the LD_\>80 dosage for each bacteria. This secondary injection is referred to as protection exposure (to determine if the vaccination exposure provided protection). All injections were delivered in a total volume of 10 µl phosphate buffered saline (PBS). Sham injected fish received 10 µl PBS. After recovery from anesthesia, fish were moved to tanks in a flow–through water system and maintained at 27°C±1°. All fish were held under the same conditions during all experiments and were observed 3 times a day for clinical signs of disease. Moribund fish were euthanatized in 340 mg/L MS222, and sampled for bacterial re-isolation. Mortalities were recorded for 10 days post-protection exposure. ## Re-isolation of Bacteria After protection exposures, deaths were recorded, and a 10 µl loop of each dead fish’s brain was plated on 5% sheep blood plates. After 24 to 48 h at 28°C, bacterial identifications were confirmed by biochemical analysis using the bioMerieux api20E strip (BioMerieux, 69280 Marcy l’Etoile, France). ## Overview of Exposure Trials Five different trials were designed to progressively determine the basis of adaptive immunity in rag1^−/− mutant zebrafish. Lethal Dose trials determined the dose of bacteria per fish required to kill 80% of a naïve population in the secondary exposure. Throughout Trials 1–4 we followed the general set-up as outlined in. For negative controls, fish were injected with sterile PBS (sham). For each trial, a tank of non-injected sentinel fish was also included. Since infected fish can shed bacteria, all tanks were on flow- through (0.5 L/min) throughout the experiments to ensure good water quality and to minimize the risk of bacterial accumulation. In a previous study we showed that immersion exposed zebrafish did not establish an infection when exposed to 10⁵ CFU/mL of tank water for 2 hours. Therefore, the likelihood of bacterial transmission through the water is very low. In Trial 1, we compared the susceptibility and adaptive protection of rag1^−/− mutant zebrafish when challenged with *E. ictaluri*. For the primary vaccination exposure, rag1^−/− mutant zebrafish were injected with 10⁴ CFU/fish of the attenuated strain of RE33 *E. ictaluri*, or PBS only (controls). For the secondary or protection exposure, rag1^−/− mutant zebrafish were injected with 10⁴ CFU of *E. ictaluri*/fish 4 weeks after the vaccination exposure. Naïve rag1^−/− mutant zebrafish were also injected with 10⁴ CFU/fish *E. ictaluri*. Negative controls included mutants injected with PBS only at the primary and secondary exposure, and mutants that were not injected at all. It is known that for a short time after the primary exposure the innate immune response is heightened, and if that heightened response is present when the secondary exposure is given, protection could be non-specific. To rule out this effect in Trial 2, we performed the same experiment as in Trial 1, except we increased the time between the vaccination and protection exposures to eight weeks. Set-up and procedures performed were the same as described in Trial 1. Trial 3 was performed to confirm that there was not a persistent low-level infection that was continually stimulating the innate immune system. In previous trials, sub-samples of vaccinated fish were cultured for bacteria and found to be negative, but a low level of infection may escape detection. Therefore, we administered oxolinic acid, an antibiotic used as a feed additive in fish. The fish received feed supplemented with oxolinic acid (7 mg/gm) daily for 7 days (fed to satiation) starting at 10 days post the primary vaccinations. In Trial 4, we determined the specificity of protection by performing heterologous (different bacteria species in primary and secondary exposures) bacteria challenges. Rag1^−/− mutant zebrafish were vaccinated with 10⁴ CFU *E. ictaluri* RE33/fish or sham injected. Four weeks later these fish were injected with 10⁴ CFU *E. ictaluri*/fish, 10⁶ CFU *Y. ruckeri*/fish or 10² CFU *E. tarda*/fish; these were dosages that resulted in greater than 80% mortality of naïve fish in Lethal Dose trials. Post-exposure procedures were the same as described in the other trials. Trial 5 was performed to determine if we could transfer protection by transferring innate immune cells from vaccinated rag1^−/− mutant zebrafish into naïve rag1^−/− mutant zebrafish before they were exposed to bacteria. In addition to blood filtering functions, fish kidney tissue is functionally equivalent to vertebrate bone marrow and is a primary lymphoid tissue. It consists of renal corpuscles and collecting tubules with an interstitial matrix of hematopoietic tissue that includes hematopoietic stem cells, macrophages, neutrophils, NCC and NK cells in rag1*^−/−* mutant zebrafish. To validate adoptive cell transfer experiments in rag1^−/− mutant zebrafish, we performed preliminary transfer experiments using carboxyfluorescein diacetate succinimidyl ester (CSFE) labeled cells. The mutant zebrafish population used in this trial was in-bred for 5 generations to minimize recognition of transfused cells by the recipient. ## Preparation and Recovery of Carboxyfluorescein Diacetate Succinimidyl Ester (CFSE) Stained Donor Cell Suspensions Kidney tissue was dissected from 10 rag1^−/− mutant zebrafish and prepared following established procedures in our lab. The cell suspension was diluted to 10⁶ cells/ml in PBS. The CFSE probe was prepared according to manufacturer’s directions (CellTrace™ CFSE Cell Proliferation Kit (C34554) Molecular Probes™). Five mM of CFSE probe per ml of cell suspension was added, and then incubated for 15 min at 30°C. The cell suspension was centrifuged at 400 g for 5 min, and the cell pellet re-suspended in tissue culture medium to obtain 10⁸ cells/ml. CFSE stained cells (10 µl) were injected intraperitoneally (IP) into recipient rag1^−/− mutant zebrafish delivering 10⁶ cells/fish. Control fish were injected with 10 µl of PBS. Fish were held in flow through tanks until sampled, and recipient kidneys were sampled as previously described. Flow cytometric analyses of adopted cells were performed on 3 replicates of adopted stained cells and 3 control replicates at 1, 24, and 48 hours, and 3, 4, 7, 10, 18 days post-injection. In Trial 5, we performed adoptive transfers of renal interstitial cells from vaccinated and naïve rag1^−/− mutant zebrafish. All adoptive transfers were carried out by infusing all the interstitial kidney leukocytes from the kidney of a single donor into a naïve recipient. Primary vaccinations were carried out as described in Trial 1 to provide donor rag1^−/− mutant zebrafish. Four weeks post-vaccination, kidneys of vaccinated rag1^−/− mutant zebrafish were removed, dissociated and cells were rinsed. Vaccinated donor leukocytes were IP injected into naïve recipient rag1^−/− mutant zebrafish (Group 1). Cells from naïve rag1^−/− mutant zebrafish were transferred into Group 2, and naïve control rag1^−/− mutant zebrafish did not receive cells (Group 3). Appropriate control groups included a transfused cell control group to rule out the effects of enhanced anti-bacterial activity resulting from non-specific cellular activation due to MHC or other transplantation antigen differences. Twenty-four hours later, fish received a secondary homologous exposure and were designated Groups 1a, 2a and 3. Control groups received PBS and were designated Groups 1b and 2b. ## Statistical Methods Cumulative mortality was calculated in trials 1, 2, 3 and 5. Relative percent survival (RPS) \[(1- mean mortality of vaccinated/mean mortality of sham vaccinated) x 100\] was calculated for Trial 4, homologous and heterologous exposures, because RPS is a more robust analysis for protection to determine specificity. Treatment groups for each trial are described above. Number of tanks per treatment and number of fish per tank are shown in. Each tank was treated as a biological replicate. Mortality data of treatments between groups were analyzed by one-way analysis of variance (ANOVA) with post hoc Tukey HSD correction for multiple comparisons with a level of significance at p≤0.05. Statistical analyses were performed using SPSS for Windows 15.0 (SPSS Inc., Chicago, IL). # Results In Trial 1, a significantly lower cumulative mortality of 28% was seen in vaccinated rag1^−/− mutant zebrafish, while naïve rag1^−/− mutant zebrafish suffered a cumulative mortality of 75%. No losses occurred in the control fish and randomly selected rag1^−/− mutant zebrafish cultured negative for *E. ictaluri*. These findings demonstrate that in vaccinated rag1^−/− mutant zebrafish, a form of adaptive protection occurred. In Trial 2, we saw similar results as in Trial 1. Cumulative mortality in naïve rag1^−/− mutant zebrafish was 52.5%. As expected, mortality of vaccinated rag1^−/− mutant zebrafish decreased significantly to 17.5%. These results demonstrate that protection in rag1^−/− mutant zebrafish is not due to a temporarily heightened post-primary response. The results also demonstrate that a component of innate immunity in rag1^−/− mutant zebrafish is capable of mediating adaptive protective immunity following vaccination. In Trial 3, randomly sampled fish cultured negative for *E. ictaluri* before evaluating protection. This trial resulted in a similar significant cumulative mortality as in Trial 1 and 2. Vaccinated rag1^−/− mutant zebrafish demonstrated significantly greater protection with a cumulative mortality of 37.5% compared to 72.5% in naïve rag1^−/− mutant zebrafish. *Edwardsiella ictaluri* isolated from moribund fish post the secondary exposure were sensitive to oxolinic acid. These findings demonstrate that protection in the rag1^−/− mutant zebrafish was not due to bacteria persisting within the fish. In Trial 4, rag1^−/− mutant zebrafish vaccinated with *E. ictaluri* demonstrated significantly reduced mortality following secondary *E. ictaluri* exposure, but were not protected against secondary *Y. ruckeri* or *E. tarda* exposures. Thus, homologous exposures provided specific adaptive protection whereas heterologous exposures did not. To validate adoptive cell transfer experiments in rag1^−/− mutant zebrafish, we performed preliminary transfer experiments using carboxyfluorescein diacetate succinimidyl ester (CSFE) labeled cells. We modified standard procedures to perform adoptive cell infusions, or transfusions, in zebrafish. We successfully isolated kidney leukocytes, CFSE stained these cells, and IP injected them into recipient fish. After re- isolating stained cells from recipient kidney tissue and performing flow cytometric analyses on them, we compared their tissue distribution to naïve adopted cells. Our findings demonstrate that transfused donor cells were stable for greater than 7 days in recipient fish. In Trail 5, group 1a received vaccinated cells and the cumulative mortality was significantly less (29.5%) when compared to rag1^−/− mutant zebrafish that received naïve cells, or no cells at all. Vaccinated kidney interstitial cells from rag1^−/− mutant zebrafish mediated specific adaptive protection in naïve recipients. Significantly higher cumulative mortalities were seen in exposed rag1^−/− mutant zebrafish that received naïve cells and exposed rag1^−/− mutant zebrafish that did not receive any cells, 78% and 58% respectively. Since these two treatments were not significantly different from each other, an increased number of naïve cells alone did not provide protection. # Discussion We have utilized T and B cell deficient rag1^−/− mutant zebrafish to investigate adaptive protection in response to bacterial infections in the absence of an acquired immune system. These mutant fish have increased numbers of neutrophils and increased expression of genes associated with innate defenses. This is the likely explanation for the ability of lymphocyte deficient zebrafish to resolve primary infection. More intriguing is our finding of significant differences in mortality upon secondary exposure, which demonstrates that rag1^−/− mutant zebrafish are able to develop and maintain protective immunity following a primary vaccination exposure. Furthermore, specificity was demonstrated when infection success was significantly reduced after previous contact with the same pathogen (homologous exposure), but not to a different pathogen (heterologous exposure). The cells that mediated specific recognition and a protective response upon secondary exposure performed these same functions when transferred into naïve rag1^−/− mutant zebrafish. Our finding of adaptive protection mediated by the innate immune system of zebrafish parallels similar findings in lymphocyte deficient mice. Furthermore, our findings are unique because they are the first demonstration of a T and B cell deficient rag1^−/− mutant vertebrate to mount an adaptive protective response to bacteria. Natural Killer cells are the most likely innate immune cell mediating protective immunity in rag1^−/− mutant zebrafish. Natural Killer cell genes that encode pathogen recognition receptors do not undergo gene rearrangement. Research on the functions of NK cells in mice have demonstrated adaptive, acquired immune responses. A hapten-based hypersensitivity study was the first to suggest NK cells had the capacity of memory in *Rag*-2 deficient Severe Combined Immunodeficient (SCID) mice. Severe Combined Immunodeficient mice possess NK cells but are devoid of T and B lymphocytes. These mice demonstrated substantial contact hypersensitivity responses to haptens that persisted for 28 days and was elicited only by haptens to which mice were previously sensitized. No contact hypersensitivity was induced in another type of mouse lacking NK cells (and T and B cells), suggesting that NK cells were mediating a true adaptive secondary response. Further, adoptive transfer studies indicated NK cells were the specific leukocyte involved. However, a specific receptor or mechanism imparting this memory was not found. Another investigation of the involvement of NK cells in epidermal responses suggested that NK cells do not mediate specific memory in a murine skin transplant model, but will mediate acute skin allograft rejection after IL-15 stimulation in the absence of any adaptive immune cells. Sun and Lanier utilized B6 mice and the mouse cytomegalovirus to investigate the role of NK cells in a viral infection. Following initial infection, NK (Ly49H receptor+) cells proliferated 100x in the spleen and 1000x in the liver. After a contraction phase, these NK cells resided in various tissues for several months. Following secondary exposure, or viral reactivation, memory NK cells were found to rapidly degranulate and produce cytokines, resulting in protection. Adoptive transfer of these NK cells also conveyed protective immunity. For the first time, immune responses of NK cells were found to undergo all four phases: expansion, contraction, memory maintenance and recall response, previously attributed only to cells of the adaptive immune system. Unlike T and B cells that express one antigen-specific type of receptor after encountering a pathogen, NK cells have been found to express an array of receptors with distinct specificity. Natural Killer cells may preserve a more general adaptive protection similar to what is observed in memory T cells, where interleukin-12 produced by dendritic cells triggers interferon-gamma production in the absence of cognate antigen. Receptors on mammalian NK cells that could provide memory function and cross- reactivity have been discussed. Human NK cells can be activated by direct contact with *Mycobacterium* via the NKp44 natural cytotoxicity receptor. This activation occurs in the absence of monocytes/macrophages and IL-12. In bony fish the NK cell receptor functional orthologs are novel immune-type receptors (NITRs). Like NK cell receptors in mammals, NITRs can function to either inhibit or activate NK cell cytotoxicity and/or cytokine release. A group of secreted NITRs have been suggested to dimerize with membrane-bound NITRs or other membrane-bound molecules involved in immune recognition, or they may bind to foreign body surfaces. The structure of an activating NITR on a cytotoxic NK- like cell line has been characterized as resembling antigen binding receptors that demonstrate specific recognition, and these receptors might undergo lineage-restricted somatic variation conveying specific protection upon re- encounter with the same pathogen. Based on our studies that have demonstrated specific secondary immune responses of T and B cell deficient rag1^−/− mutant zebrafish, and evidence that fish NK cells demonstrate the capacity of specific recognition of diverse molecules, we believe that a population of zebrafish NK cells mediate memory, and that zebrafish NK cells have a mechanism for enhanced discrimination of a bacterial target following primary exposure. In addition, other receptors have been shown to be important in recognizing intracellular pathogens. Two are of particular interest, the tripartite motif (TRIM) proteins and NOD-like receptor (NLR) molecules. Both have been shown to be present and very diverse in zebrafish. The zebrafish genome encodes 240 TRIMs. Many of them have the B30.2 domain and an important ligand binding domain. This region displays a high level of positive selection for diversification in zebrafish. The zebrafish genome also encodes 5 NLR A family members, 6 NLR B family members and several hundred NLR C family members. Like the diverse TRIMs, the NLR Cs contain the B30.2 domain with high diversity. Involvement of cytosolic receptors in the host response to *E. ictaluri* is suggested by the demonstration of up regulation of the NLR designated NOT1 in infected channel catfish. Because these receptors are cytosolic, any protection imparted would likely be phagocyte mediated. ## Conclusion We used rag1^−/− mutant zebrafish to examine immunological memory of the innate immune system to a bacterial pathogen. Memory is a term that has been only used in acquired immunity therefore we referred to innate immune system memory as adaptive protection throughout this text. Heterologous bacterial challenges demonstrated that the immune system of rag1^−/− mutant zebrafish exhibits adaptive characteristics and specificity. Adoptive cell transfers demonstrated that kidney interstitial cells mediated the specific adaptive protection. Our research demonstrates that innate immune cells are capable of mediating specific adaptive protection and that vaccination in rag1^−/− mutant zebrafish results in significantly increased survival if the fish are re-exposed to the same pathogen. Even if it is ultimately revealed that adaptive protection is more important in fish than mammals, understanding this immune response in fish will help to understand its phylogenetic development. The broader spectrum provided by memory of innate immunity would influence an individual’s ability to respond to classes of pathogens and direct the type of acquired response induced. Innate memory could be partially responsible for variation and alterations of immune function in immune-associated diseases and could be directed to help with deficiencies in acquired immune components. # Supporting Information The authors wish to acknowledge Lorelei Ford, Aparna Krishnavajhala, John Stokes and Casey Varner for technical assistance. The authors would also like to thank Drs. L Hanson, S Pruett and X Wan for critical reviews of this manuscript. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: LPH. Performed the experiments: CMH LPH. Analyzed the data: CMH LPH. Contributed reagents/materials/analysis tools: LPH. Wrote the paper: CMH LPH.

# Introduction Collisions with anthropogenic structures are a significant and well documented source of mortality for avian species worldwide. Collision risk is highest when the flight paths of birds intersect with man-made structures. Collisions of migrating birds with communication towers, buildings, power lines, and wind turbines are well documented in the literature. These studies reported mortalities after the flight hazards were unknowingly installed in bird movement corridors and where limited mitigation measures were available to decrease mortality rates. Migrating raptors are assumed at the greatest risk of collision when flight hazards, like wind turbines, are concentrated along landscape features attractive to long-distance migrants. Raptor migration corridors typically form around leading lines, narrow topographic features like ridgetops and coastlines, which produce updrafts that assist in soaring and gliding flight. These same ridgetops and coastlines also produce some of the highest wind power classes utilized for commercial wind power generation. The projected growth of the wind industry and potential increase in impacts on bat and avian species of concern resulted in the formation of federal Land-Based Wind Energy Guidelines to assist wind energy developers and wildlife agencies in assessing and mitigating adverse effects of proposed wind projects. Mitigation includes actions to avoid or minimize impacts, or compensate for impacts to wildlife. A key focus of the guidelines is on site selection because risk to wildlife is not evenly distributed across the landscape and risk can be site- and species-specific. In the United States, federal guidelines require wind developers to use the best available data on bird species abundance, distribution, and migratory behavior to forecast risk and assist in preliminary site evaluation as part of the Land- based Wind Energy Guidelines and the Eagle Conservation Plan Guidance. The guidance documents establish a standardized process for site selection, planning, and pre- and post-construction monitoring. Wind developers and operators are also encouraged to apply for an Eagle Take Permit (50 CFR 22.3) from the U.S. Fish and Wildlife Service to cover liability for potential collisions considered “take” under the Bald and Golden Eagle Protection Act (BGEPA, 16 USC 668-668c). The bald eagle (*Haliaeetus leucocephalus)* is known to be vulnerable to collision with wind turbines and electrical lines. Migrating bald eagles fly 25–600 m above ground height (AGL), which is within the rotor swept zone of utility-scale turbines. Although documented turbine collisions in bald eagles through 2016 have been low (*n* = 15), collision rates in similar eagle species suggest that the potential for collision is high for bald eagles. Collisions with turbines have been documented in golden eagles (*Aquila chrysaetos*) and white-tailed eagles (*Haliaeetus albicilla*), species similar to bald eagles in body size, flight style, and foraging techniques. One reason collision risk is so high is the exponential increase in bald eagle populations in many portions of their range, including in the Chesapeake Bay. Currently, few turbines are located in high-use areas for bald eagles, but this is rapidly changing as the both the wind industry and eagle populations expand in the Western Atlantic Flyway. Researchers have prioritized the development of tools to assess avian risk associated with wind turbines and inform the pre-construction siting process. Species risk and sensitivity maps have been developed for some species but are currently unavailable for migrating bald eagles. To address the need for risk assessment, in this study, we 1) identified areas of northeastern North America utilized by migrating bald eagles, and 2) compared them with high wind-potential areas to identify potential risk of bald eagle collision with wind turbines. Identifying areas with highest potential collision risk could facilitate site selection by wind energy developers and allow the wind energy sector to continue its expansion while minimizing risk to migrating eagles. # Material and Methods ## Study Area Our study area included northeastern North America at latitudes between 38°5N and 57°N, including eastern Canada and the New England and Mid-Atlantic bald eagle management units in the United States. The Mid-Atlantic eagle population, estimated at approximately 10,000 individuals (B. Watts unpubl.), is mostly resident with a small portion of juveniles and subadults migrating north to New England and Canada for the summer months. The New England and southern Maritime province population (including New Brunswick, Nova Scotia, and Gaspe Peninsula of Quebec) are roughly estimated at 6,200 individuals based on breeding surveys and survival rates. These populations mostly winter in the Mid-Atlantic region, including the Chesapeake Bay. In addition, a third population of unknown size migrates south from northeastern Quebec and western Labrador into New England and the Mid-Atlantic each winter. Eagles migrate along the Western Atlantic Flyway through topographic leading lines (coastlines or mountain ranges) on the Atlantic Coast, Saint Lawrence River, and Appalachian Mountains to reach summering and wintering areas. Potential for wind power generation within the study area is highest in narrow bands of ridgetops in the Appalachian Mountains, Monts Notre-Dame, and Laurentian Mountains and in broader coastal areas along the Atlantic Ocean, Bay of Fundy, and St. Lawrence River/Gulf of St. Lawrence. ## Telemetry We captured and marked bald eagles on Aberdeen Proving Ground, Maryland, in the northern Chesapeake Bay between August 2007 and May 2009. Of the 63 eagles reported in Watts *et al*., 17 migrated north from the Chesapeake Bay within the Western Atlantic Flyway. This included eagles banded as nestlings in Maryland (*n* = 2), eagles banded as nestlings in New York and captured in Maryland during their first winter (*n* = 2), and eagles whose morphometric measurements suggested were from the Chesapeake Bay breeding population (*n* = 3) or northeastern U.S. and Canada breeding populations (*n* = 10), Watts unpublished data. Five of the eagles maintained annual summer breeding territories in northern Quebec and Labrador, 52°N-56°N latitude. Eagles were fitted with 70-g solar-powered global positioning system-platform transmitter terminal (GPS-PTT) satellite transmitters (Microwave Telemetry, Inc. Columbia, MD). Transmitters were programmed to collect GPS location data (±18 m manufacturer estimated error) every hour during daylight and once at midnight. Flight altitude data were not collected. Argos satellites (CLS America, Largo, MD) processed GPS locations and data were archived online by the Satellite Tracking and Analysis Tool. Movement data were preprocessed and formatted by Movebank ([www.movebank.org](http://www.movebank.org/)) and downloaded for the analysis. Eagle capture and handling complied with Institutional Animal Care and Use Committee protocols at the College of William and Mary (IACUC20051121-3), Maryland scientific permit 42687, and USGS Bird Banding Laboratory permit 21567. ## Movement Modeling We identified migration movements for individuals as continuous directional movements north or south ≥100 km (*n* = 132 tracks) and extracted these tracks using ArcMap 10.1. Two eagles made repeated migratory flights within the same year and season and each of these movements was included as a separate track in the analysis. We used the Move package in R 3.1.2 to produce utilization distribution (UD\] surfaces for individual migration tracks using the dynamic Brownian bridge movement model (dBBMM). We set dBBMM parameters to a window size of 17, margin of 7, location error of 18 m, and raster cell size of 1 km. Window size of 17 was based on the maximum number of GPS locations received per day for an individual eagle. The margin was set in proportion to half the window size. Location error was determined by the transmitter manufacturer as ±18 m. We set the cell size to 1 km to generate the most detailed output for the geographic scale of our study area. We excluded one migration track because the number of GPS locations was less than the dBBMM window size of 17. We included approximately 24 hours of additional locations prior to the start and after the end of each migration track to ensure that the entire migration movement was included in the model output. UDs were exported as rasters and overlaid in ArcMap on a grid of 1 km cells (*n* = 1,976,935) that spanned the study area. We combined UD raster maps produced for individual migration tracks by averaging probabilities for each 1 km cell to create a population-wide UD for the study area. Because the number of locations varied among individual migration tracks, we weighted, combined, and standardized UD surfaces according to the number of locations per track. We chose to weight the UD surfaces based on number of fixes rather than weighting each track equally to relate exposure risk to the amount of time an eagle potentially interacted with a flight hazard. We avoided pseudoreplication by combining all individuals and their tracks to create a single map without comparing tracks to each other. We ordinated UD values per cell and categorized them from highest (top 20% of cells) to lowest use (100% of cells) for display purposes. ## Flight Hazards We examined the overlap of bald eagle migration movements with wind power density maps to identify locations where estimates of on-shore wind power density (w/m²) were available at 50 m AGL. Canadian wind power density maps were not available publically at a scale fine enough to be comparable so we limited the flight hazard analysis to the United States portion of the study area. Wind power classes of 3 or greater were included since these are typically used for planning utility-scale wind facilities. Existing wind turbine locations were plotted on the UD map of eagle migration using available digitized turbine data. Locations of existing turbines were validated with high- resolution aerial imagery \[Microsoft Bing Maps 2014\]. We mapped new turbines not included in U.S. government databases using aerial imagery and ESRI user data in ArcMap 10.1. We identified locations where wind turbines overlapped with areas of high use by migrating eagles to determine sites of highest potential collision risk for eagles. # Results ## Movement Modeling Seventeen bald eagles migrated during the study period (2007–2014), producing 132 migration tracks in the study area. The number of tracks per eagle ranged from 1–13 and the number of locations per track ranged from 34–1,380 ($\overline{x}$ = 252 ± 19.3 SE). Migration movements were concentrated within two main corridors, one along the Appalachian Mountains (inland corridor, 2,175 km long) and the other along the Atlantic coast (coastal corridor, 1,620 km long;). From the northern end of the Chesapeake Bay, a primary movement corridor widens from approximately 70 km to 230 km, stretching from the Atlantic Coast of New Jersey to Harrisburg in the Ridge and Valley region of eastern Pennsylvania. The single corridor then diverges into the inland and coastal corridors in Dutchess County, New York along the Hudson River Valley. A coastal corridor approximately 50 km wide branches northeast through central Connecticut and Massachusetts until it reaches the Atlantic Coast in New Hampshire. The movement corridor then widens to 90 km along the coast of Maine and into the coast of southern New Brunswick. It turns north-northwest along the Maine-New Brunswick border and ends at the Gaspe Peninsula in Quebec. The inland corridor is along the Appalachian Mountains in eastern Pennsylvania, New York, Vermont, and into southeastern Quebec. This inland corridor is approximately 145 km wide in northeastern Pennsylvania through northern New York until the corridor splits at Lake Champlain. A short branch of the inland route ends at the St. Lawrence River upstream (west) of Montreal. The inland corridor continues north from Lake Champlain narrows to approximately 70–110 km wide as it continues north into Quebec. At the St Lawrence River, the corridor parallels the northwest coast of the Gulf of St. Lawrence and continues north into Saguenay, Cote-Nord, and ends in Nord du Quebec and western Labrador. ## Flight Hazards We documented a total of 3,123 wind turbines ≥100 m AGL in the study area, with 1,405 turbines located in the United States and 1,718 in Canada. There were 1,185 turbines (38%) located in UD 20, and 971 turbines (31%) in UD 40. In the United States portion of the study area, commercially viable wind power classes overlapped with 2% of the UD category 20 (i.e., the areas of highest use by migrating eagles) and 4% of UD category 40. The coastal migration route had only 1 concentration of turbines in the UD category 20 (near Bull Hill, Hancock County, Maine) compared to 21 clusters of turbines within the inland route. # Discussion Predicting potential eagle collision fatalities is a key part of the Eagle Conservation Plan Guidance stage 1 planning process for wind facilities in the United States, yet published information on movements of eagle populations is limited. Here we provide a UD map of bald eagle migration corridors in northeastern North America for inclusion in eagle collision risk assessment. This UD map of eagle migration provides the first analysis to evaluate collision risk of migrating eagles over a broad geographic scale and is the first to incorporate eagles of mixed age class, breeding status, and breeding population. The scale and scope of this study can support future assessments of potential impacts of wind energy on migrating bald eagle populations in northeastern North America. This eagle UD map can be used in planning placement of structures that pose a collision risk to eagles. Site-specific characteristics heavily influence collision risk and predicting flight behavior of eagle migrants could identify potential site conflicts. Bald eagles have been documented colliding with distribution and transmission lines and are expected to be most at risk when a flight hazard is not shielded by vegetation, when eagles are distracted during flying (foraging, chasing, or fighting), or during migratory flight. We know of no other studies of bald eagles and wind turbine collision risk; however, a recent study documented bald eagles successfully avoiding a new stationary flight hazard erected in a known migratory corridor. In this instance eagles adapted their flight altitudes pre- and post-construction with 96% of eagles flying over a 60 m high transmission line bisecting Kittatinny Ridge, New Jersey. A similar pre- and post-construction study of wind turbines in British Columbia, Canada documented golden eagles detecting and avoiding turbines during migration with fewer flight paths in the collision risk zone after turbines were installed. Bald eagles may also exhibit similar avoidance behavior around turbines, but it has yet to be documented in the literature. Eagle migration routes described in this study were similar to routes published on juvenile eagles from Labrador and Georgia, and juvenile and subadult eagles from Florida. And thus, though our sample size is relatively small we believe our results have broad implications to eagles in the Western Atlantic Flyway. The Chesapeake Bay is a convergence area for bald eagle populations along the flyway supporting 3 distinct populations (northeast, southeast and Chesapeake Bay) throughout the year. Because the Chesapeake Bay acts as an activity hub for migrants on the flyway, we suggest the two northeast migration routes likely represent the main pathways for eagles entering and exiting the Bay region. In addition, eagles migrating through the southern Appalachians use one or more of these routes once they reach Pennsylvania or New York. We produced a map with greater detail than previous doppler satellite transmitter studies using higher accuracy of the GPS data to define eagle migration corridors. In addition, the broad range of age and breeding populations in our sample of tracked eagles created a comprehensive migration map detailed at the 1 km scale useful for project planning. Our analysis identified distinct bald eagle migration corridors with limited overlap with commercially viable wind power class areas in northeastern North America. This is encouraging because it suggests that wind energy development can still occur in the study area at sites that are most viable from a wind power perspective and are unlikely to cause significant mortality of migrating eagles. In siting new turbines, wind energy developers may wish to avoid the high-use migration corridors (UD categories 20 & 40) and focus new wind energy projects on lower-risk areas (UD categories 60–100). Our extent of inference is in UD 20 and UD 40 where we are reasonable certain eagles flew based on the ±18 m accuracy of GPS transmitter locations. Presumed collision risk was not equal between migratory routes in our study. The coastal route had fewer wind farms than the inland route, which is not surprising since the northeast coast (Delaware to Maine) has fewer areas of commercially viable onshore wind than in the mountains. In coastal areas eagles presumably migrate using thermals, which typically increase flight altitudes over 1,000 m, well out of the rotor-swept zone of turbines and above other flight hazards like communication towers and transmission lines. Eagles using the inland route are primarily using orographic lift, which limits flight to lower altitudes on slopes and ridges where updrafts can subsidize powered flight especially during the cooler fall period when thermals are unavailable. While our transmitter data did not record altitude, we assume bald eagles have similar flight altitude to golden eagles in the study area, which have almost identical body size, and flight style, and inland migration route. Based on the number of wind farms currently within the inland migration route and the potential for future construction within available wind power classes ≥3, we believe bald eagles migrating through the inland route are at the highest risk of collision. Bald eagle collisions have been documented at wind facilities, indicating towers or turbine blades are a new flight hazard for the species. It is unknown whether documented collision rates for bald eagles represent a true low collision risk for the species or are a result of poor carcass retrieval rates in heavily vegetated areas, low carcass searching effort, or both. This level of uncertainty in current collision rates in bald eagles restricts our ability to assess overall collision risk. Refinement of collision rates could be accomplished with sampling designs targeting the 1,185 turbines within the 20 UD. Future studies should increase turbine sample sizes, larger search plots around turbines to search for injured eagles, and longer search intervals during migration periods to better estimate collision and fatality risk to bald eagles during this period. In North America, wind energy is one of the fastest growing energy sources, adding more electricity generating capacity than any other power source in 2013. Canada currently has 4% (9.6 GW) of its domestic energy from wind and the United States has 2% (65.8 GW) with national capacity goals of 20% by the years 2025 and 2030, respectively. The projected growth of this industry includes continued construction of wind farms in northeastern North America. We believe the results from this study will be valuable for both planning of future turbine siting and for evaluating collision risk at existing wind facilities. The UD map produced from this study will be made available to planners on the American Wind and Wildlife Institute’s interactive Landscape Assessment Tool <http://www.wind.tnc.org/> for preparing risk assessments in the Eagle Conservation Plan Guidance stage 1 planning process. # Supporting Information The authors thank B. Kranstauber, B. Paxton and J. McClain for their assistance in the analysis. E. Lawler provided contract support for the project. We appreciate comments on early versions of the paper by T. Alison, J. Dwyer, T. Katzner, and C. Duberstein. Sources for the boundary GIS file are ESRI, TomTom, and National Atlas of the United States of America. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BDW EKM. Performed the experiments: EKM CLT. Analyzed the data: EKM CLT. Wrote the paper: EKM CLT. [^3]: Current address: EDM International, Inc, 4001 Automation Way, Fort Collins, Colorado, United States of America [^4]: Current address: Yale University, Department of Psychology, New Haven, Connecticut, United States of America

# Introduction *Vibrio parahaemolyticus* is a Gram-negative, halophilic bacterium that is found in estuarine environments worldwide. It can cause food-borne gastroenteritis most frequently associated with the consumption of raw or undercooked seafood. Pathogenic *V*. *parahaemolyticus* strains are responsible for the majority of seafood-associated infections in the United States, many Asian countries and South America. Compared to the Asian continent and the USA, *Vibrio parahaemolyticus* infections are rarely reported in Europe. This may be due to a low incidence of illnesses or may be the result of the lack of epidemiological systems for monitoring *Vibrio*-associated illness or *Vibrio* occurrence in seafood. Nevertheless several sporadic outbreaks in UK and Spain and single clinical cases in other European countries have been reported. The pathogenicity of *V*. *parahaemolyticus* is mainly correlated to the possession of genes encoding the thermostable direct hemolysin (TDH) and/or TDH-related hemolysin (TRH). Five variants of the *tdh* gene have been found, sharing an identity of 96 to 98%. In contrast, *trh* genes possess a significantly broader nucleotide sequence variation and can be subdivided in two main subgroups (*trh1* and *trh2*) which share 84% identity. Both epidemiological studies and animal-based studies indicate that at least one of the two type three secretion systems (T3SS) also plays a major role in pathogenicity of *V*. *parahaemolyticus*. T3SS1, located on chromosome 1, is present in all *V*. *parahaemolyticus* strains and involved in cytotoxicity in eukaryotic cell lines *in vitro*. However, analyses suggest that it plays a relatively minor role in intestinal disease. T3SS2, located on chromosome 2, is strongly correlated with the presence of *tdh* and/or *trh* and further subdivided with T3SS2α associated with *tdh* on a pathogenicity island (VpaI-7) and T3SS2β located close to *trh* on a different pathogenicity island. The occurence of environmental strains which lack *tdh/trh* but carry T3SS2 and transposase elements flanking the gene cluster leads to the assumption that the correlation is not 100%. Additionally, *trh* is genetically linked to an urease (*ure)* gene cluster associated with a nickel transportation system. The rare urease-positive phenotype of *V*. *parahaemolyticus* within the species *Vibrio* makes the production of urease a good diagnostic marker for potential pathogenic *trh* positive *V*. *parahaemolyticus* strains. As detection of *tdh* positive *V*. *parahaemolyticus* strains in coastal waters of Germany is extremely rare, we focused on the analysis of *trh* positive German *V*. *parahaemolyticus* strains. In general, *trh* harbouring strains are detected in a range of about 3 to 5% at coastlines of Northern Europe with tendancy to rise in coastal areas of France. Indigenous German *trh* positive strains were compared to a variety of other *trh* positive strains from different countries. In an effort to assess the genetic relationship among *trh* positive isolates we determined the complete coding region of the *trh* genes of all strains and performed multilocus sequence typing (MLST) analysis. Furthermore, all isolates were screened via multiplex PCR for potentially pathogenic targets present in the genome of pandemic strain RIMD2210633. Hemolytic activity and resistance towards human serum were investigated as a measure of virulence-associated phenotypic traits. Strains were cultivated under different growth conditions to examine possible inductive effects on the expression of hemolysins and the T3SS2β effector gene *vopC* with quantitative reverse transcription PCR. # Material and Methods ## Bacterial strains A total of thirty-one *V*. *parahaemolyticus* isolates positive for *trh* using the primers Trh-forward and Trh-reverse were obtained from different sources (clinical, environmental, and food retail) in various countries. Ten isolates were indigenous strains from the Baltic and North Sea (seven from seawater and two from mussel primary production) and one strain from a patient after consuming fish from the Baltic Sea. Four more strains from Germany were either from imported seafood (three strains) or travel-associated (Tanzania). Two *tdh* positive strains (RIMD2210633 and ATCC43996) and one *tdh/trh* negative strain (VN-0022) were used as controls. Species identification of all strains were performed by standard biochemical assays and species specific PCR targeting the *toxR* gene. ## Strain cultivation Strains were routinely grown on Luria-Bertani (LB) agar (Merck, Darmstadt, Germany) at 37°C overnight and further cultivated in LB medium (Merck, Darmstadt, Germany) containing 0.3 M NaCl at 37°C and 200 rpm shaking. For bile and urea induction experiments bacteria were grown for 2 h in LB medium containing 0.3 M NaCl with or without 0.04% of crude bile (bile bovine, Sigma- Aldrich, Steinheim, Germany) or 0.1% urea (Urea, Roth, Karlsruhe, Germany). For testing urease activity single colonies from the agar plate were transferred on urea-dextrose-agar (0.9% caseinpeptone, 0.1% K₂HPO₄, 0.5% glucose, saline solution (358 g/l), 1% bromothymol blue, 1% agar, 1% CO(NH₂)₂, pH 6.8) and incubated at 37°C overnight. ## Multilocus sequence typing (MLST) Genomic DNA was extracted using the RTP Bacteria DNA Kit from STRATEC Molecular, Berlin, Germany. MLST was performed as described by Gonzalez-Escalona *et al*. (2008) detecting partial sequences of seven housekeeping genes from both chromosomes. PCR primers targeted *recA* (recA protein), *dnaE* (DNA polymerase III, alpha subunit), *gyrB* (DNA gyrase, subunit B), *dtdS* (threonine dehydrogenase), *pntA* (transhydrogenase alpha subunit), *pyrC* (dihydroorotase) and *tnaA* (tryptophanase). PCR reactions were carried out using primers and applying conditions detailed on the *V*. *parahaemolyticus* MLST website (<http://pubmlst.org/vparahaemolyticus/info/protocol.shtml>). PCR products were purified using QIAquick PCR purification Kit (Qiagen, Hilden, Germany). Sequencing reactions were performed commercially at Eurofins MWG Operon, Ebersberg, Germany. Primers were synthesized by Metabion International AG, Planegg/Martinsried, Germany. The obtained sequences were queried against the pubMLST database to determine the allele designations and sequence type (ST) of each isolate. Phylogenetic analysis of MLST sequences was performed using Mega 6.0 software. Minimum evolution (ME) trees for concatenated sequences were constructed using the kimura-2 parameter model to estimate the genetic distances. ## PCR genotyping and *trh* sequencing PCR reactions were performed using a Mastercycler EP Gradient (Eppendorf, Hamburg, Germany) in a volume of 25 μl with 1 x PCR buffer (2 mM MgCl₂), 0.2 mM of each dNTP, 0.2 μM of each primer, and 1.5 U of Dream Taq DNA Polymerase (Fermentas, St. Leon-Rot, Germany). The PCR running conditions were: an initial denaturation of 95°C for 3 min, 30 cycles of 94°C for 30 s, 50–68°C for 30 s and 72°C for 1 min, and a final elongation step of 72°C for 5 min. To analyze the distance from *trh* to *ureC*, PCR was performed using Long PCR Enzyme Mix (Thermo Scientific, Schwerte, Germany) according to Park *et al*. with minor modifications. PCR reaction was carried out in a 50 μl volume with 1 x Long PCR buffer with 1.5 mM MgCl₂, 0.2 mM of each dNTP, 0.5 μM of each primer and 1.5 U Long PCR Enzyme Mix. PCR running conditions were as follows: an initial denaturation of 94°C for 3 min, 30 cycles of 98°C for 20 s, 53°C for 20 s and 68°C for 7 min, and a final elongation step of 68°C for 10 min. The PCR primers, annealing temperatures, target genes and amplicon sizes are shown in. Targets for genotyping were selected from website: VFDB- comparative pathogenomic organization of Vibrio (<http://www.mgc.ac.cn/cgi- bin/VFs/comp_graph.cgi?Genus=Vibrio&mode=o>). PCR based amplification of the complete *trh* gene was accomplished by using different primer combinations. Annealing temperature was 54°C for all primer combinations. For sequencing reactions additional internal *trh* targeting primers were used. Phylogenetic analysis of the coding sequence of *trh* gene was performed as described above. ## Cultivation and RNA isolation Strains were grown in 20 ml of LB medium containing 0.3 M NaCl at 37°C and 200 rpm shaking to an OD₆₀₀ 0.6. The cultures containing bacteria at a late logarithmic phase (OD₆₀₀ 0.6) were split into four 5 ml subcultures. All four subcultures were twofold diluted with LB medium (0.3 M NaCl) and exhibited a bacterial concentration of approx. 3 x 10⁸ cfu/ml. To investigate temperature effects on transcription one subculture was incubated at 37°C and one subculture at 20°C. To study the influence of bile and urea at 37°C the other two subcultures were additionally supplemented to a final concentration of either 0.04% bile or 0.1% urea. The supplements had been dissolved and filter-sterilized in LB medium at a concentration of 10%. All four subcultures were incubated for 2 h with 200 rpm shaking at either 37°C or 20°C. After incubation 200–500 μl volume of subculture was mixed with RNA Protect Bacteria Reagent (Qiagen, Hilden, Germany) and proceeded according to the manufacturer\`s protocol. Before further processing samples were kept frozen at—80°C. RNA was isolated using RNeasy Minikit (Qiagen, Hilden, Germany) according to the manufacturer\`s protocol. The removal of DNA was achieved by using RNase free DNase Set (Qiagen, Hilden, Germany) in a reaction volume of 50 μl (2.5% 10 x DNase buffer, 2% DNase, 15.5% RNase free water and 5% RNA). Reactions were incubated for 1.5 h at 37°C with moderate shaking. After incubation RNA was purified by ethanol precipitation. Briefly, 5 μl of sodium acetate (3 M, pH 5.2) and 150 μl of ice-cold 96% ethanol were added to the samples and incubated overnight at—20°C. RNA was pelleted by centrifugation at 14.000 x g for 30 min at 10°C and washed with 75% ethanol. For the complete removal of ethanol RNA pellets were dried in a speedvac and redissolved in 20 μl of RNase free water. To verify that DNA was completely removed RNA samples were tested in a real-time PCR reaction. RNA was isolated in three separate experiments for each growth condition of each strain. ## Real-time quantitative reverse-transcription PCR of virulence genes (qRT-PCR) RNA concentration was measured using a NanoDrop spectrophotometer (ND-1000, PEQLAB, Erlangen, Germany) and equally diluted to a concentration of 50 ng/μl RNA before starting synthesis of cDNA. For reverse transcription reaction High- Capacity cDNA Reverse Transcription Kit (Life Technologies, Darmstadt, Germany) was used in a volume of 20 μl with 1 x RT buffer, 1 x dNTP mix, 1 x random primers, 50 U reverse transcriptase and 250 ng RNA. Reverse transcription reaction was performed in a Mastercycler EP Gradient (Eppendorf, Hamburg, Germany) following the guidelines of the manufacturer. To perform a relative quantification of mRNA of the genes *trh1*, *trh2*, *tdh* and *vopC* we chose the comparative C_T (ΔΔC_T) method with the housekeeping gene *gyrB* as the reference gene and growth at 20°C as a control. Primers and TaqMan probes for specific detection of the target genes were synthesized by Metabion and are shown in. TaqMan probes were tested in standard curve experiments prior to main experiments. The qRT-PCRs were carried out in ABI PRISM 7500 (Applied Biosystems, Darmstadt, Germany) with 1 x TaqMan Universal Master Mix (with ROX) (Applied Biosystems), 0.25 μM primers, 0.2 μM TaqMan probe and 1 μl cDNA as template. PCR conditions were as follows: after initial denaturation at 95°C for 15 min, a cycle of 95°C for 15 s, 56°C for 60 s and 72°C for 30 s was repeated 40 times. Each sample was measured in triplicate. For data analysis we used ExpressionSuite v1.0.3 software (Life Technologies) which includes the student\`s t-test for sample group comparisons. Confidence level was chosen to be 0.95. ## Cell-free protein synthesis of TRH variants and TDH2 TRH1 (VN-0038), TRH2–2 (VN-0293), TRH2–3 (VN-0029) and TDH2 (control) were expressed in a prokaryotic cell-free system. Transcription and subsequent translation were directly initiated from PCR templates generated from chromosomal DNA of the strains. Cell-free synthesized hemolysins were mature proteins without signal peptide and protein tags. Synthesis of hemolysins was performed using the batch-formatted cell-free transcription/translation system based on *E*. *coli* lysate. Primers and PCR conditions are listed in. The detailed procedure of DNA template generation and cell-free expression of toxins is described in Bechlars *et al*. (2013). Briefly, PCR products were generated in a two-step expression PCR (E-PCR), E-PCR products were added to *E*. *coli* lysates and coupled transcription-translation reactions took place in a 90 min incubation step. Reactions were supplemented with ¹⁴C- labeled leucine for determining total yield and soluble protein via radioactive analysis in a scintillator. To determine the size and homogeneity of radioactively labeled toxins, all proteins were analyzed by *SDS-PAGE* (sodium dodecyl sulphate-polyacrylamide gel electrophoresis) and visualized using a phosphor- imager (Typhoon Trio+, GE Healthcare, Munich, Germany). The detailed analytical procedures have been described previously. All experiments were performed twice. ## *In silico* analysis of TRH variants For *in silico* analysis amino acid sequences of mature protein of all TRH variants and TDH2 were aligned using Accelrys Gene version 2.5 software (Accelrys Ltd., Cambridge, UK). Homology modelling was performed using the SWISS Model server (<http://swissmodel.expasy.org/interactive>). The crystal structure of TDH2 (PDB: 3A57) was used as a template for modelling 3D structure of TRH variants. ## Hemolytic activity of cells, culture supernatants and cell-free synthesized proteins Hemolytic activity of strains was tested with human blood (blood donation service of the German Red Cross, Berlin-Wannsee, Germany) and sheep blood (Thermo Scientific, Schwerte, Germany). Prior to hemolysis assay strains were cultivated with and without bile extract as described for the transcription analysis. Briefly, strains were grown in LB medium to an OD₆₀₀ 0.6 followed by a split and twofold dilution into two subcultures which both exhibited a bacterial concentration of approx. 3 x 10⁸ cfu/ml. One subculture contained just LB medium (0.3 M NaCl) with bacteria and the other subculture contained LB medium (0.3 M NaCl) supplemented with 0.04% bile and bacteria. Both subcultures were incubated for 2 h with 200 rpm shaking at 37°C. Hemolysis assay was performed according to Mahoney *et al*. with modifications. After 2 h incubation step 150 μl amounts of cultures were combined with 800 μl of 2% erythrocyte suspension in PBS and cells were pelleted at 7000 x g for 4 min. In parallel 1 ml of cultures were centrifuged at 5000 x g for 10 min at 4°C, culture supernatants were filtered (0.2 μm) and also 150 μl amounts were mixed with 800 μl of 2% erythrocyte suspension. Samples were incubated at 37°C for 4 h with 200 rpm shaking. After incubation cells were centrifuged again at 7000 x g for 4 min and the release of hemoglobin into the supernatant was determined by measuring the absorbance at 540 nm in 700 μl volume. By adding 4% Triton X-100 in LB medium and only LB medium to erythrocyte suspensions the maximum and spontaneous hemoglobin release was determined. All experiments were performed twice. Hemolytic activity of cell-free synthesized TRH variants and TDH2 was tested with sheep, human and rabbit blood. Prior to hemolysis assay, the toxin concentrations of soluble protein were adjusted to 5 μg/ml with 1 x PBS after centrifugation at 16.000 x g for 10 min. Aliquots of 60 μl soluble protein (approx. 300 ng) were mixed with 60 μl PBS containing 4% of erythrocytes (either sheep or rabbit or human) and incubated for one hour at 37°C as decribed by Bechlars *et al*. (2013). Cell debris was sedimented by centrifugation at 400 x g for 5 min followed by spectro-photometrical measurement (OD₅₇₀) of the released heme present in the supernatant. Extinction values were set in proportion to the maximum loss of heme in the positive control (4% Triton X). ## Serum resistance The serum resistance test was carried out as described previously. Isolates able to grow in the presence of 60–80% human serum were classified as serum resistant, while growth in the presence of 20–40% and 0–10% were designated as intermediate and sensitive, respectively. All strains were tested in triplicate. *E*. *coli* strain K-12 with the plasmid pKT107 carrying a serum resistance determinant served as a positive control, while *E*. *coli* strain K-12 without pKT plasmid served as a negative control. # Results ## Sequence analysis of complete *trh* genes Thirty-one *V*. *parahaemolyticus* strains had been identified as *trh* positive by using the primers Trh-forward and Trh-reverse which amplify an internal 251 bp sequence of *trh1/trh2* genes. Sequence analysis of the complete *trh* coding sequences (570 bp) of thirty-one *trh* positive isolates revealed a total of ten different sequence types. Phylogenetic analysis (Minimum Evolution Tree, Mega 6.0) of nucleotide sequences were performed and revealed a clustering of *trh1*, *trh2* genes and a third *trh* sequence variant. This variant has a modified codon (ATA) instead of an ATG start codon and an early stop codon due to a deletion at position 382 bp. We assume that this sequence is a pseudogene (designated *ψtrh*) which does not encode a functional protein. The ten isolates from Germany consisted of seven environmental strains from seawater, two strains from mussel primary production, and one from a patient (VN-0029) and carried alleles of the *trh2* gene. *Trh2* genes of German strains could be divided in *trh2–2* and *trh2–3* adopting the nomenclature from Ellingsen *et al*.. It should be remarked that *trh* sequences published by Ellingsen *et al*. are not complete and are missing 72 bp at the end of sequence. A new *trh2* variant (VN-0053 and VN-0084) which has not been described yet was named *trh2–6*. This new allele encoded a protein that differed in up to 30 amino acid residues from TRH1 variants, in three amino acids from TRH2–3 and in only one amino acid from TRH2–2. In total six strains of our collection contained a pseudogene *ψtrh*. Five of these strains additionally possessed a *tdh* gene while one strain (VN-0070) only had the pseudogene. ## Multilocus sequence typing Twenty-nine *trh* positive strains were successfully subtyped at seven loci using multilocus sequence typing (MLST) as described by Gonzalez-Escalona *et al*. and allele numbers and sequence types were assigned according to the PubMLST database. Sequence data were concatenated in the order of loci used to define the allelic profile to produce a single sequence of 3682 bp for each strain. Amplification of *recA* gene was difficult in some cases as previously described by Ellingsen *et al*.. For strains VN-0295 and VN-0296 which were from imported retail seafood we could not obtain a complete *recA* sequence. These two strains are therefore not included in the MLST analysis. The main focus of the MLST analysis was to establish the genetic relationships among the ten *trh* positive *V*. *parahaemolyticus* strains isolated from German sources. Results of the phylogenetic analysis based on concatenated MLST sequences are illustrated in a Minimum Evolution Tree shown in. The ten strains were differentiated into four different sequence types (STs), all of which were already present in the MLST database. All isolates carrying *trh2–2* possessed ST79, while most isolates with *trh2–3* had ST73, except for VN-0394 (ST6) and VN-10300 (ST761). also depicts information of *trh* sequences of other strains of these STs (see, *trh*^*+* indicates sequences with no detailed sequence information). STs of all remaining *trh* positive *V*. *parahaemolyticus* strains of this study are given in in. In total four new sequence types were found (strains VN-0028, VN-0053, VN-0084, and VN-4016). Detailed information of the allelic profiles are depicted in. ## Genotyping of virulence-associated traits The presence of a number of virulence-associated genes and putative accessory virulence-associated factors which are present in the genome of pandemic RIMD2210633 strain or were published for *trh* positive strains was tested via multiplex PCR. Reference strain RIMD2210633 which carries *tdh* genes was included as a control. Of the 31 *trh* positive strains nine additionally possessed *tdh* sequences (positive in diagnostic PCR using primers Tdh-forward and Tdh-reverse). Testing for the presence of T3SS1 and T3SS2 (α and β) revealed that T3SS1 was present in all strains. While T3SS2α was not detected in any of the strains (except for RIMD2210633), three components of T3SS2β (*vscC2*, *vscS2*, and *vopB2*) were present in all *trh* positive strains with the exception of strain VN-0070 which harboured *Ψtrh*. *VopC* was identified in almost all strains, except for the *Ψtrh* carrying strains. Furthermore, we studied the strains for the presence of genomic islands (VpaI-1 to VpaI-6). VpaI-1, VpaI-3, VpaI-4, VpaI-5, and VpaI-6 could not be detected in any strain, while gene *vp0643* of VpaI-2 was present in five strains (VN-0028, VN-0046, VN-0055, VN-077, and VN-4016). The distribution of type six secretion system (T6SS) genes showed a diverse pattern. While genes *vp1390* and *vp1405* were not detected in any of the tested strains, *vp1401* was found in 20 strains and *vp1409* in 19 strains. The following putative accessory virulence-associated genes could be identified in all strains: *vp0950*, *vp0952*, and *vp0962* associated with biofilm formation, capsular polysaccharide biosynthesis glycosyltranferase gene *cpsA* (*vp1403*), *hutA* and *hutR* encoding heme receptors, and *ure* (urease) and *trh-ure* cluster. The coupling of *trh* sequences to the urea biosynthesis cluster (trh-ure) was only missing in one strain (VN-0070) which harboured *Ψtrh* but lacked a *tdh* gene. The gene encoding MTase and the *acfD* gene encoding an accessory colonization factor could not be detected in any of the strains except for the control strain RIMD2210633. *MshD* gene encoding a putative MSHA pilin protein, was detected in four of five *Ψtrh* carrying strains while only two of the remaining strains had *mshD* gene. ## Quantification of virulence gene expression by qRT-PCR Expression of virulence genes *trh1*, *trh2*, and *tdh*, as well as *vopC* was measured in relation to the *gyrB* gene as a constitutively expressed housekeeping gene. The expression pattern of the genes at 20°C was set as reference and compared to the expression pattern at 37°C with and without addition of bile extract or urea. For the analyses fourteen strains were selected: Eight *trh2* strains (four *trh2–2* and four *trh2–3* strains, seven from German sources), two *trh1* strains, two strains with *trh1* and *tdh* genes, one strain with the combination *Ψtrh* and *tdh*, and one strain with *Ψtrh* only. Transcription analyses showed that strains exhibited diverse transcription patterns of the studied virulence factors and responded differently to the tested growth conditions. *Trh2* transcription was slightly increased (approximately 4 to 6 fold) at 37°C compared to 20°C in some strains (VN-0393, VN-3933, and VN-4016), but no influence of bile nor urea on *trh2* expression was observed. *Trh1* transcription was induced in the presence of bile in strains VN-0038 (30 fold) and VN-0049 (ca. eight fold), while the other two *trh1* positive strains showed no induction of transcription (VN-0028) or no transcription at all (VN-0050). Urea did not have any influence on *trh1* transcription. *Tdh* transcription of strain VN-0050 was strongly induced (80 fold) in the presence of bile, while no induction of expression was observed for the two other *tdh* positive strains (VN-0038 and VN-0055). Urea had no effect on *tdh* transcription. *Ψtrh* showed no transcription at all (VN-0070) or low transcription rate (VN-0055). Induced transcription of *vopC*, an effector of T3SS2β, in the presence of bile was observed for *tdh/trh1* positive strains (VN-0038 and VN-0050). While an induced transcription of *vopC* in the presence of urea was detected in strains harbouring *trh2–3* variant (VN-0029, VN-0030, VN-0393, and VN-3933), strains harbouring the *trh2–2* gene showed almost no difference of transcription levels at the three growth conditions at 37°C. ## Hemolysis assay and serum resistance To verify the expression results of *trh* and *tdh* genes, hemolytic activity of cultures was tested against human and sheep erythrocytes. Prior to hemolysis assay strains were cultivated with and without bile at 37°C. Additionally, cultures and supernatants were tested separately. We observed no hemolytic activity against both types of erythrocytes for all *trh2* carrying strains. The two *trh1* strains behaved differently; while strain VN-0049 showed strong hemolytic activity against sheep erythrocytes and moderate hemolytic activity on human erythrocytes in response to the induction of bile, VN-0028 exhibited no or very low hemolytic activity. *Trh1/tdh* positive strains revealed strong hemolytic activity in response to bile against both blood types. Relatively low hemolysis was caused by *ψtrh/tdh* (VN-0055) strain and *ψtrh* strain (VN-0070) showed no hemolysis at all. Generally, observed hemolytic activity was always stronger when the bacteria were incubated with erythrocytes than in samples with only culture supernatants added to the erythrocytes. All tested strains were classified as resistant towards human serum since they were able to grow in the presence of 60–80% serum (data not shown). Only pandemic strain RIMD2210633 was serum sensitive under the tested conditions. ## Cell-free synthesis of TRH variants and hemolytic activity on erythrocytes Three TRH variants and TDH2 as a control were synthesized in a prokaryotic *in vitro* transcription-translation system with lysates from *E*. *coli*. All PCR products with the expected sizes were produced with similar efficiency (see.). By incorporation of ¹⁴C-labeled leucine into toxins the rate of protein synthesis was determined. Yields of soluble protein of TRH variants ranged from 12 μg/ml (VN-0029 and VN-0038) to 56 μg/ml (VN-0293), while protein yields of TDH2 were higher (150 μg/ml). Aliquots of the translation mixtures (TMs) and supernatants (SNs) were analyzed for homogeneity and size using SDS- PAGE followed by autoradiography (see.). Results of hemolytic activity of cell-free expressed toxins on three types of erythrocytes (sheep, human, and rabbit) are summarized in. Under our test conditions the TRH1 variant (VN-0038) showed hemolytic activity against sheep erythrocytes but not against human and rabbit erythrocytes. Both of our tested TRH2 variants did not show any hemolytic acitivity against all three blood types. TDH2 revealed a stronger hemolytic activity against rabbit erythrocytes (= 50%) than human erythrocytes (= 10%) which is in agreement with Honda *et al*. (1988). # Discussion ## *Trh* sequences and MLST Sequence analysis of the complete coding regions of all *trh* positive *V*. *parahaemolyticus* strains revealed a clustering into *trh1* and *trh2* genes and a pseudogene *ψtrh*. All strains from German coastal waters possessed alleles of the *trh2* gene and belonged to four STs (ST 73, ST 6, ST 761 and ST 79). These STs were already described in the pubMLST database and included Norwegian and German strains which were all *trh* positive with one exception. Only ST 6 contained a strain isolated outside of Northern Europe (Chile) that lacked a *trh* sequence (not included). Given the time period of isolation and the geographical distribution the phylogenetic analysis indicates that the *trh2* positive strains are persistent in northern European waters and probably belong to the autochthonous microflora of this region. In the ME tree published *trh2* variants of Norway and some more German isolates with *trh* genes are displayed. Few *trh* negative strains with ST 121 and ST 761 were incorporated into the ME tree. Presence or absence of *trh* genes in strains with the same ST could indicate that these virulence factors may be acquired through horizontal gene transfer, as has been shown for *tdh* genes. Some strains analyzed in our study carried a *trh* pseudogene (*ψtrh*). The frequent occurrence of *trh* pseudogenes has recently been described by Nishibuchi. It should be noted that most of the isolates of our study were *tdh* positive and did not originate from Germany. We assume that *ψtrh* pseudogenes do not encode functional proteins. Interestingly, *V*. *parahaemolyticus* infections in Europe have been mostly associated with *tdh* positive strains so far and not *trh* except for three clinical cases in Norway and two clinical isolates of our study. Strain VN-0028 which harboured the *trh1* gene, was obtained from a patient after returning from travel to Tanzania. VN-0029 exhibited a *trh2–3* gene and was isolated from a 70-year old patient showing symptoms of diarrhea who had been eating fish from the Baltic Sea. Interestingly, one clinical isolate from Norway (strain 173) also exhibited the *trh2–3* gene and shared the same ST as VN-0029. In other parts of the world the detection of *trh* in clinical isolates is dominant with 22% (*trh* only) and 59% (*trh/tdh*) in Canada, 8.3% (*trh* only) and 24.3% (*trh/tdh*) in North America, and 8% (*trh* only) and 34% (*trh/tdh*) in Thailand. In the cited publications, there was no differentiation between *trh1* and *trh2*. ## Genotyping of virulence-associated traits PCR analysis of *trh* positive *V*. *parahaemolyticus* strains confirmed the genetic linkage of *trh* sequences to the urease gene cluster which has already been noted by others. Only strain VN-0070 with pseudogene *ψtrh* was shown to lack the urea gene cluster and urease activity. VN-0070 did not possess the T3SS2β, which was found by PCR in all other *trh* and *trh/tdh* positive strains. The strains were negative for T3SS2α as has been described already. Interestingly, five strains which were *tdh* positive and possessed the pseudogene *ψtrh*, were negative for the *vopC* gene which encodes an effector protein of T3SS2β, but possessed three other tested proteins of T3SS2β. This finding suggests that in the five strains virulence might be attenuated due to an incomplete set of effector genes of the T3SS2β. Based on the observation that the *vopC* gene could be detected only in strains with intact *trh* genes we chose this gene for expression analysis. Genotyping of further putative virulence-associated traits did not hint to a specific pattern for the *trh* positive strains. Studies showed that genomic islands VpaI-1 comprising MTase gene (Vp0394), VpaI-3, VpaI-4, and VpaI-5 were unique to post-1995 pandemic *V*. *parahaemolyticus* strains. This was confirmed by our finding that no *trh* harbouring strain was positive for these VpaIs and also VpaI-6. Gene *vp0643* of VpaI-2 could be detected in five *trh* positive strains. The presence of some VpaI-2 genes in pathogenic but also non-pathogenic *V*. *parahaemolyticus* strains has been described by Chao *et al*.. The type six secretion system (T6SS) was suggested to be involved in adhesion to host cells and helps to compete against other bacterial species in the environment therefore promoting fitness. Chao *et al*. reported that most pandemic strains have the complete set of T6SS genes. Our tested strains harboured a partial set of T6SS genes as it was described for most non-pathogenic strains. The ability to grow in biofilms is a survival strategy used by many bacteria causing infectious diseases. All strains were positive for the tested genes associated with biofilm formation. When analyzing *trh* positive strains for new defined targets (selected from: <http://www.mgc.ac.cn/cgi- bin/VFs/comp_graph.cgi?Genus=Vibrio&mode=o>) present in the genome of pandemic strain RIMD2210633 no distinct pattern was observed. In conclusion, the genotyping of all strains did not indicate significant differences between clinical strains and strains with environmental or food origin. ## Transcriptional analysis of *trh*, *tdh*, and *vopC* genes For expression analysis we compared transcription at 20°C to transcription at 37°C and tested the influence of two components present in the human intestine on transcription at 37°C. Bile is an important factor for the digestion of lipids and is excreted by the liver into the small intestine. Urea is the major nitrogenous waste product and is also produced by the liver. 20–25% of all urea produced enters the intestinal tract. In addition, urea is used in agriculture and constitutes \>50% of global nitrogenous fertilizer usage. The high input of urea into coastal waters could have an impact on *Vibrio* populations. Urease of *V*. *parahaemolyticus* is expressed and active when urea is present. As in other enteropathogens protective effects of urease were demonstrated and because of the proximate location of the *ure* gene cluster to *trh* and T3SS2β genes in the genome of *V*. *parahaemolyticus*, we hypothesized that urea might have an effect on expression of virulence genes. Our results suggest that the expression of hemolysin genes follows no general pattern. *Trh2* gene expression was slightly increased at 37°C in some strains without the influence of bile and urea. *Trh1* gene expression could be strongly induced by bile which is to our knowledge the first report of bile-inductive effects on *trh* expression in *V*. *parahaemolyticus*. However, the induction was strain-specific with several strains not responding to bile addition. For *V*. *cholera* (non-O1/non-O139) bile-inductive *trh* expression has already been described. We also observed an increase of *tdh* expression in the presence of bile in some strains which has already been described for pandemic *V*. *parahaemolyticus* strains. Urea addition did not have any significant influence on expression of any of the hemolysin genes. The expression of the T3SS2β effector gene *vopC* was also analyzed in our transcription analysis. VopC mediates host cell invasion of *V*. *parahaemolyticus* in cultured cells but VopC-mediated invasion is not a required step for pathogenicity in an animal model of infection. Nonetheless, it has been shown that VopC deamidates and therefore activates Rho family GTPases e.g. promoting the formation of actin stress fibres. GTPases play a role in regulation of numerous cellular processes, cell cycle progression and apoptosis and it has been demonstrated that GTPase-mediated signalling is often modulated by various bacterial factors in infection. So, it is possible that VopC combined with other factors takes part in the infection process of *V*. *parahaemolyticus*. Interestingly, we could observe a bile responsive induction of *vopC*, but exclusively for *tdh/trh* bearing strains suggesting a different regulation of the T3SS2β in these strains compared to the remaining strains. The collective triggering of expression of *vopC* with the adjacent located hemolysins in the presence of bile might indicate a mutual regulation of T3SS2β and hemolysins. A bile responsive induction of *tdh* and T3SS2α gene expression of *V*. *parahaemolyticus* has already been demonstrated by Gotoh *et al*.. For a subgroup of *trh2* positive strains—*trh2–3* (VN-0030, VN-0029, VN-0393, VN-3933)—an induction of *vopC* expression in the presence of urea was found. In the other *trh2* positive subgroup of strains—*trh2–2*—*vopC* transcription was slightly increased at 37°C at the three growth conditions suggesting that temperature is a critical factor for triggering *vopC* expression. The upregulation of *vopC* transcription in some strains might contribute to pathogenesis in human infection since it has been demonstrated that T3SS2β is involved in enterotoxicity in animal experiments. An inductive effect of urea on gene transcription is an interesting observation. As a result of overland transport of agriculture fertilizers, urea is also found in coastal waters. Urea in the natural environment of vibrios could have an effect on population structure, as *trh* bearing, urease positive *V*. *parahaemolyticus* strains are unique within the genus *Vibrio* enabling them to use urea as a nitrogen source and could therefore lead to a growth advantage compared to other *Vibrio* strains. An active T3SS2β—although its function in the environment is not known yet- could have an impact on potential predators like protists and increase the environmental fitness as it has been reported for T3SS2α carrying *V*. *parahaemolyticus* strains. ## Hemolytic activity on erythrocytes To verify the transcription studies we tested hemolytic activity of the selected strains against sheep and human erythrocytes with and without bile. Additionally, we examined the activity of three cell-free synthesized TRH variants against different erythrocyte types (sheep, human, and rabbit). The cell-free protein expression allows the synthesis of active hemolysins from PCR products which can be used directly for testing of hemolytic activity. All *trh2* bearing strains (cultures and supernatants) showed no hemolytic activity against human and sheep erythrocytes, although increased *trh2* transcription was observed in some strains at a temperature of 37°C (VN-0393, VN-3933, and VN-4016). The two cell-free synthesized TRH2 variants also did not show any hemolytic activity towards sheep, human or rabbit erythrocytes. This indicates that lack of hemolytic activity of *trh2* strains is probably due to the missing functionality of the *trh2* gene product and not caused by lack of expression. This is in agreement with the suggestion by Kishishita *et al*. that low or no hemolytic activity is a result of protein functionality and/or binding activity. The cell-free synthesized TRH1 variant of strain VN-0038 was highly hemolytic against sheep erythrocytes, while under our test conditions no hemolysis of human and rabbit erythrocytes was observed. Similarly, purified TRH1 was shown to be 40 to 50 fold more active against sheep erythrocytes than against human and rabbit erythrocytes. When testing culture and supernatants of this strain, activity against sheep and human erythrocytes was found. This might probably be due to the fact that the strain harboured *tdh* and *trh* genes. *In silico* analysis of the protein sequences provides one possible interpretation for non-hemolyic activity of the two TRH2 variants since they share very hydrophobic amino acids (aa) at position 156 and 165. TRH forms a tetramer in solution. Performing homology modelling on the basis of TDH structure these two positions (156 and 165) are predicted to be located in the central pore of the tetramer structure which is thought to function as an ion channel. Furthermore, analysis of the tetrameric structure of TDH2 by Yanagihara *et al*. suggested an interaction between the conserved amino acid glutamic acid at position 138 (E138, also conserved in all TRH variants) with the amino acid glutamine at position 165 (Q165), which contributes to the stabilization of the tetrameric structure. The amino acid changes in the two TRH2 variants (as in all TRH2 proteins of this study) could therefore lead to a dysfunction of the hemolysins since it has been demonstrated that maintenance of tetrameric structure with a central pore is indispensable for biological activity of TDH and TRH. Noticeably, most TRH1 variants shared the same amino acids at positions 156 and 165 with VN-0038, while the TRH1 variant of strain VN-0028 had the same aa variation as the TRH2 variants and was not hemolytic. While we could not detect hemolytic activity of the TRH2 hemolysins, it cannot be excluded that other biological activities may be present. It has been demonstrated that TRH1 variants show various biological activities including fluid accumulation in rabbit ileal loops, increase of rabbit skin vascular permeability, and cardiotoxicity on cultured myocardial cells. When combining the results of the transcription studies with the results obtained from exposing erythrocytes to cultures or supernatants of strains carrying *trh1* or *trh1*/*tdh* or *ψtrh/tdh*, no clear picture arises. Bacterial cells used in hemolysis assays probably express various virulence factors (e.g. T3SS) which could lead to lysis of erythrocytes. However, when only supernatants of cultures were used, the hemolysin activity could be subject to proteolytic activity of extracellular and cell-associated proteases and vary during the growth phase. # Conclusion *Trh* is considered as a virulence factor since there is a strong association with gastroenteritis. The investigation of all thirty-one *trh* carrying strains for the presence of virulence factors by PCR did not reveal any significant differences between isolates from clinical, environmental, and food sources. Strains originating from Germany only possessed alleles of the *trh2* gene. Although clinical strains carrying *trh2* genes have been found and one German strain (VN-0029) was isolated from a patient that had diarrhea due to consumption of fish from the Baltic Sea, we could not detect hemolytic properties of TRH2 proteins. This may question the role of TRH2 as a virulence factor. However, it cannot be ruled out that other biological activities associated with virulence described for TRH1, like e.g. fluid accumulation in rabbit ileal loops, increase of rabbit skin vascular permeability, and cardiotoxicity on cultured myocardial cells, may still be functional in TRH2 proteins. We will address these questions in the future using different human intestinal cells as targets for TRH2 proteins. Nevertheless, it seems likely that other virulence factors may contribute to infection and illness caused by *V*. *parahemolyticus* strains. This is supported by the reports about *tdh/trh* negative *V*. *parahaemolyticus* strains which caused acute gastroenteritis. It is an interesting finding that all intact *trh* genes were linked to the presence of *vopC*, an effector protein of the T3SS2β, while in strains harbouring a pseudogene (*ψtrh*) *vopC* was not detected. We conclude that a further differentiation of *trh* variants and including the detection of T3SS2β components like *vopC* could be an important step to improve the characterization of potential pathogenic *V*. *parahaemolyticus* and could lead to a refinement of the risk assessment in food analyses and clinical diagnostics. ## Nucleotide sequence accesion number The *trh* sequences of the following *V*. *parahaemolyticus* strains were deposited at the European Nucleic Archive (<http://www.ebi.ac.uk/ena/data/view/accesion> number): VN-0028 (Accession#: LM993801), VN-0029 (Accession#: LM993802), VN-0038 (Accession#: LM993803), VN-0049 (Accession#: LM993804), VN-0057 (Accession#: LM993808), VN-0084 (Accession#: LM993807), VN-0295 (Accession#: LM993805) and VN-2897 (Accession#: LM993806). # Supporting Information The curation work for the assignation of new alleles or STs reported here in, was performed by Dr. Gonzalez-Escalona and supported by the FDA Foods Program Intramural Funds. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SB RD ES. Performed the experiments: SB CJ SD DAW. Analyzed the data: SB RD ES SK DAW. Contributed reagents/materials/analysis tools: SB CJ SD DAW SK. Wrote the paper: SB ES RD SK.

# Introduction Inhibitory motor control — i.e. the ability to suppress unwanted behavioral responses — provides crucial flexibility in goal-directed behavior, allowing individuals to quickly adjust to a changing environment and to overcome pre- potent responses when they are inadequate or inappropriate (see for a review). Interest in this topic has dramatically increased over the past several years in accord with the central role of this function in normal human behavior and development, as well as in a range of neurological and psychiatric conditions, such as attention-deficit hyperactivity disorder and substance abuse. One of the most prominent experimental paradigms designed to investigate response-inhibition capabilities is the Stop-signal task. In this task, a choice-reaction Go-stimulus is rapidly followed, on a minority of trials, by a Stop-stimulus requiring participants to withhold the response to the Go- stimulus. Variants of this and related tasks have been used extensively with a variety of methodological approaches to investigate brain processes underlying response inhibition. Converging evidence from these studies has led to the view that a mostly right-hemisphere network of brain areas plays a critical role in response inhibition (but see). This network includes the inferior frontal gyrus (IFG; especially the frontal operculum extending into the insula) and the pre- supplementary motor area (pre-SMA), which in turn interacts with the basal ganglia and the thalamus (for reviews see,). Although it is very likely that the loop between the frontal cortex and the basal ganglia/thalamus described above is a core structure subserving response inhibition, it is increasingly recognized that other mechanisms play an important role leading up to response inhibition and in determining whether it will be successful or not. Specifically, it has been reported that selective attention to the task-relevant stimuli can play an important role in determining trial outcome in the Stop-signal task. Numerous studies have reported transient modulations of sensory processing of the relevant stimuli in the time-range of the sensory evoked N1 ERP component, which precedes the implementation of response inhibition, and that these modulations are predictive of the outcome of the process. Due to the timing and the posterior topography, such effects can be compellingly attributed to differences in sensory processing, including due to attentional modulations of that processing. Unfortunately, such conclusions are much more difficult to draw for activity at later time-ranges and in other brain areas, so that a separation of perceptual/attentional processes from those that are directly related to response inhibition has proven difficult. A case in point relates to the right IFG, which has received a lot of experimental support as a key structure in response inhibition. This area is reliably activated in human fMRI studies investigating response inhibition (for a recent comprehensive review, see), while lesion and electrophysiological studies in humans have provided corroborating evidence. Recent studies, however, have challenged the view that the right IFG is directly involved in response inhibition (but see). For example, strong right IFG activations have also been reported in response to other rare stimuli besides task-relevant Stop-stimuli (–, see also –), consistent with its reported participation in the ventral attention system that has been implicated in bottom-up attentional processes triggered relatively automatically by salient environmental events. An important distinction in this context that has not yet been established (neither for the right IFG nor for other involved brain areas) is the degree of automaticity with which Stop-trial stimulation elicits neural activity. Specifically, in the existing attention literature it is appreciated that the presentation of rare, and/or physically salient, stimuli (note that Stop-trials meet both criteria) tend to automatically capture attention and activate at least parts of the ventral attention system. Such processes can even occur if these stimuli are entirely task-irrelevant (for a recent discussion, see). In the context of the Stop-signal task, however, the degree to which neural activity is related to such salience-triggered processes is not clear, versus how much such activation may depend on the general task context, in which the behavioral relevance of stimuli other than Go-stimuli needs to be determined. Establishing such a distinction is important in order to gain insights into which areas and processes are under active top-down control during a Stop-signal task, even if their function is related to control processes that are not directly related to response inhibition. Moreover, because these other functions could also be derailed in psychopathology, thereby potentially mimicking deficits directly in motor control, disentangling and understanding these processes better is an important goal. In the present report, we have carried out additional sets of analyses of the data from a recent study that included, as a control condition, task-irrelevant Stop-trials from separate task blocks. In our previous report, this control condition was used specifically to subtract out activity related to the sensory processing of Stop-stimuli. Here, we expanded our analyses of these data in order to gauge, on a brain-wide level, the degree to which activity in different brain areas is related to the sensory and attention-attracting features of Stop- stimuli. Additionally, we performed an ROI analyses to investigate the relative degree of activation by Go-trials, task-irrelevant Stop-trials, and task- relevant Stop-trials in key brain areas. These analyses provide activity profiles indicative of separable neural operations that have important implications towards a better understanding of the specific systems-level neural circuits that lead to and implement response inhibition. # Methods ## Participants and Ethics Statement Eighteen participants took part in this study, two of which had to be excluded due to technical problems, and another one due to particularly poor behavioral performance. The 15 remaining participants (nine female) had a mean age of 22.9 years, all with correct or corrected-to-normal visual acuity, and none reporting a history of psychiatric or neurological disorders. All participants gave written informed consent and the study was approved by the Duke University Health System Institutional Review Board. Participants were compensated \$20 per hour. ## Task The present experiment entailed two variants of the typical Stop-signal task that differed only in the instructions given to the participants. During Stop- relevant blocks, participants were instructed to try to withhold their response when a Stop-stimulus followed a Go-stimulus, whereas in Stop-irrelevant blocks the visual stimulation was identical, but participants were instructed to ignore the Stop-stimuli and thus to respond to all Go-stimuli irrespective of whether they were followed by a Stop-stimulus. Each task was performed once per experimental run (approximately 2.5 minutes each), separated by a 16-sec break (i.e., task break, with continuing MR data acquisition). Odd runs began with the Stop-relevant task, followed by the Stop-irrelevant task, with even runs in the opposite order. Ten runs were collected for each participant, yielding a total of 943 trials across all conditions per participant. The time between trial onsets was varied pseudo-randomly between 2 and 8 seconds (gamma distribution; average 3.2 sec) to allow for the separation of different conditions in an event-related fMRI analysis. ## Stop-relevant blocks Stop-relevant blocks used a standard Stop-signal task (using German traffic- light signs, see), entailing a random sequence of frequent Go-trials and less- frequent Stop-trials. On Go-trials (80% of all trials), only a Go-stimulus was presented, requiring a rapid choice response. On Stop-trials (20% of trials), the Go-stimulus was followed shortly after by the presentation of a Stop- stimulus, indicating that the response to the Go-stimulus was to be canceled. On Go-trials, a green symbol was presented for 800 ms, and participants had to decide whether it was oriented to the left or right (mapped to the right index and middle finger). Stop-trials started identically, but after a variable stimulus onset asynchrony (SOA) the Go-stimulus was replaced by a red Stop- stimulus until the end of the total stimulus duration of 800 ms. The SOA between the Go- and the Stop-stimulus is an important determinant for whether participants are able to withhold the response to the Go-stimulus (successful Stop-trials, **SST**) or not (unsuccessful Stop-trials, **UST**; see). Note that for discussion of the SST and UST trials below, the respective block is usually not specified because these conditions are exclusive to the Stop- relevant blocks. A common approach for controlling performance is to titrate the Go-Stop SOA using an adaptive staircase procedure to yield approximately equivalent numbers of SST and UST for each participant. We implemented such a procedure here, increasing the SOA by 17 ms (one refresh screen) after SSTs and decreasing it by the same amount after USTs (starting SOA: 200 ms). This procedure allowed us to calculate the Stop-signal response time (SSRT), which is viewed as reflecting the mean amount of time that is required to implement the inhibition of a motor response and is derived by subtracting the mean Go-Stop SOA from the average Go- trial response time. ## Stop-irrelevant blocks During Stop-irrelevant blocks, visual stimulation was identical to the Stop- relevant ones, but participants were instructed to respond to all Go-stimuli irrespective of the occurrence of Stop-stimuli. To equate the sensory stimulation as much as possible between the two block types, we also varied the Go-Stop SOA during Stop-irrelevant blocks. Specifically, the SOA value resulting from the staircase procedure of the preceding Stop-relevant block was used as the initial value, which was then varied in a random one-up/one-down fashion after each Stop-trial, staying within +/- three 17-ms steps of the initial value. Stop-relevant blocks used the end value of the preceding Stop-relevant- block staircase as their starting value. ## Data acquisition and basic analysis MR data was acquired on a 3-Tesla GE Signa MRI system. Functional images were acquired with a reverse spiral imaging sequence (TR = 2000 ms, TE = 25 ms; flip angle  = 75°; 32 slices with 3×3×3 mm resolution; AC-PC orientation providing coverage approximately from the top of the brain down to the pons). The first five functional images were excluded from the analysis, to allow the scanner to reach steady-state magnetization. For anatomical reference, a high-resolution structural T1 (3D Fast Spoiled Gradient Recalled (FSPGR); 1×1×1 mm resolution) was obtained. The fMRI data were analyzed using SPM5 (<http://www.fil.ion.ucl.ac.uk/spm/>). All functional images were corrected for acquisition time delay, spatially realigned, and normalized by applying the normalization parameters used to warp the high-resolution T1 image to the SPM template. Images were resliced to a voxel size of 2×2×2 mm and smoothed with an isotropic 8-mm full-width half-maximum Gaussian kernel. For each participant, a statistical model was computed by applying a canonical hemodynamic response function (HRF) combined with time and dispersion derivatives for each of the conditions, including a 128-sec high-pass filter. All conditions were modeled separately, restricting the analyses on the trials with correct responses (or with a successfully withheld response in the case of successful Stop-trials). Additional regressors were included to model trials with incorrect responses, misses, and break onsets, as well as for modeling the six realignment parameters measuring the participants’ movements during the experiment. For visualization purposes, activation maps were rendered on the SPM single-subject template. ## Data analysis The parameter estimates resulting from each condition/contrast and participant (first-level analysis) were entered into a second-level, random-effects group analysis using one-sample t-tests. In order to test for areas that were more active for Stop-irrelevant Stop-trials than Stop-irrelevant Go-trials on a brain-wide level, a voxel-wise analysis was performed. The respective group- level results were thresholded at T\>3 (uncorrected) and a minimum cluster size of k = 10 contiguous voxels. Additionally, cluster-level correction for multiple comparisons was performed. Clusters surviving this correction (p\<0.05) are highlighted in the Results tables, and strong inferences are limited to these areas. Despite the danger of false positives, we also report those activations that did not survive this correction. Such two-stage procedure was employed to meet our inferential goals to simultaneously not *underestimate* activity differences in areas that are typically associated with the Stop-signal task (i.e., to not make strong claims about the absence of activity differences based on a highly conservative threshold), while also highlighting which activations are quite certainly not false positives. Additionally, a region of interest (ROI) analysis was performed to compare activity elicited by the different conditions in the key regions involved in this task. In order to define ROIs that would allow for a comparison between Stop-trials from the Stop-relevant and the Stop-irrelevant task blocks, a t-contrast was employed that tested the average of these Stop-trial responses across the blocks against the average of all Go-trial responses across the blocks. Due to the very robust and widespread activations identified by this contrast, the group-level effects were thresholded comparatively conservatively (p\<0.01; FDR-corrected corrected on the voxel level with an extent threshold k = 50 contiguous voxels; note that the resulting clusters also survived cluster-level multiple-comparison correction). Defining ROIs on the basis of this contrast enabled quantitative comparisons between the Stop-relevant and Stop-irrelevant Stop-trials in these regions, because the ROI selection was not biased in favor of either of those conditions. Ten 4-mm-radius spherical ROIs were selected. Marsbar (<http://marsbar.sourceforge.net/>) was used to extract percent-signal-change values from these ROIs. Statistical assessment of the ROI data and the behavioral data was accomplished using repeated-measures analyses of variance (rANOVAs), with non-sphericity correction of the degrees of freedom (Greenhouse-Geisser algorithm) where necessary, or using paired t-tests reporting two-tailed p-values if not indicated otherwise. For the ROI analysis, our inferential goals also had an influence on the choice of statistical significance criterion. Specifically, p-values are reported without correction for multiple comparisons. Although this risks some false-positive results, applying strict correction would have artificially biased our results towards the conclusion that there are no differences between Stop-trials from the two task-blocks (as well as between SST and UST). Nevertheless, we note that some differences could represent false positives and need to be interpreted with a degree of caution. # Results ## Behavioral Results Participants performed very accurately during both the Stop-relevant and Stop- irrelevant task blocks. No significant differences in accuracy were observed for the three trial types that always required a response (i.e., Stop-relevant Go- trials \[97.6%\], Stop-irrelevant Go-trials \[96.6%\], and Stop-irrelevant Stop- trials \[97.1%\]; F(1.5,21.4) = 2.2, p = 0.15). Response times were slower on Stop-relevant Go-trials (520 ms) relative to unsuccessful Stop-trials (446 ms) and relative to Stop-irrelevant Go- and Stop-trials (436 and 439 ms; overall F-Test: F(1.8,25.3) = 32.3, p\<0.001), but were similar between the latter three conditions (F(1.3,17.61) = 0.9, p = 0.37; see). During Stop-relevant Stop- trials, participants managed to withhold their behavioral response on approximately half of the trials (52.7%), indicating the success of our staircase SOA-adjustment procedure. The average SSRT across subjects was 230 ms. ## fMRI Results ### Activity related to Stop-irrelevant Stop-trials In order to identify brain areas that respond differentially to Stop-stimuli, as compared to Go-stimuli, even if those stimuli are entirely task-irrelevant, we performed a voxel-wise comparison between Stop-trials and the Go-trials in the Stop-irrelevant blocks (T\>3; k = 10; additionally, cluster-level correction for multiple comparison was employed, and clusters surviving this procedure are highlighted below and). Differences were not only found in lateral occipital areas, as can be expected based on the differences in sensory stimulation between these trial types, but also bilaterally in widespread clusters in the inferior parietal lobules (IPL;). Importantly, these main posterior clusters all survived multiple-comparison correction on the cluster level. Turning to the frontal cortex, two clusters were identified, one in the right IFJ, and another one in the right pre-SMA. Importantly, these clusters were too weak/small to survive the multiple-comparison correction employed. Although these could reflect false-positive results, the locations of these clusters are well in line with typical activations in the Stop-signal task in these areas (see also ROI analysis below), thus giving some more credibility to both effects. Due to the failure to reach cluster-level-corrected significance, however, these activations need to be interpreted cautiously. Taken together, activity that is purely triggered by the perceptual and attention-attracting aspects of Stop- trials are not limited to ventral sensory areas, but are also present in inferior parietal areas, with possible contributions from the right IFJ and pre- SMA. ### ROI selection and predicted activity profiles While the above analyses provide a formal brain-wide test for which areas are activated during Stop-trials (as compared to Go-trials) even when these stimuli are task-irrelevant, the relationship to activity triggered by task-relevant Stop-trials is hard to evaluate without direct reference to these other trial types. Importantly, activity in some areas might not be triggered in an all-or- none fashion by Stop-stimuli. Rather it is possible that some areas may be activated in a graded fashion, wherein a certain amount of activity is triggered even by task-irrelevant Stop-stimuli, which gets more pronounced if those stimuli are in fact task-relevant. A good example process for which such a pattern might be present is attentional capture, although other processes might also be engaged in a graded fashion. Specifically, attentional capture has an automatic component that does not depend on task-relevance. However, attentional capture effects can get enhanced if the capturing stimulus is furthermore relevant to the task. Accordingly, in order to provide a more detailed analysis of the contributions of different key areas to the processing of task-relevant and task-irrelevant Stop-trials, we performed additional analyses within ROIs that were delineated based on both kinds of Stop-trials. Another advantage of such an ROI analysis is that voxel-wise comparisons are necessarily quite conservative, whereas ROI-analyses can focus on the most relevant areas derived from orthogonal contrasts, thus ameliorating the multiple-testing problem. In order to allow for an unbiased comparison between the Stop-trials from the different task blocks, ROIs were selected on the basis of a contrast comparing all Stop-trials from the two task blocks (i.e., Stop-relevant and Stop- irrelevant) against all Go-trials from those tasks. The present ROI analysis is related to an ROI analysis of some of these data applied in our earlier paper. As compared to this earlier report, however, we used the Stop-irrelevant Stop- trials here as an active part of the ROI-defining contrast instead of using it as a baseline condition. Moreover, our earlier report used a conjunction of SST and UST for the ROI definition (rather than their average), which could have introduced some bias towards finding similar activity estimates for the two conditions in the subsequent analysis. Such bias would be avoided with the present analysis. Due to the very robust and widespread activations that were identified by this contrast, we opted for a comparatively conservative voxel-level threshold (FDR corrected p\<0.01; k = 50). This contrast yielded eight activation clusters (note that all clusters furthermore survived cluster-level correction for multiple comparisons; note also that the present set of areas is very similar to other studies that have compared Stop-trials with Go-trials, which presumably indicates that activity levels in Stop-relevant Stop-trials were sufficient to identify typical stopping-related areas even when averaged with Stop-irrelevant Stop-trials that may have failed to elicit substantial activity in some of these areas.). The respective maxima of seven of these were highly distinctive and were thus directly used for further analysis (see, and). Five of these were found in frontal cortex, including right-hemispheric lateral frontal areas IFG (protruding into the anterior insula) and the inferior frontal junction (IFJ). Additionally, this contrast revealed activity in a middle frontal gyrus (MFG) area, along with right pre-SMA and the left anterior insula. In the left hemisphere two additional substantial clusters were found, namely in the lateral middle occipital gyrus (MOG) and in the inferior parietal lobule (IPL). A final eighth cluster was identified in the posterior part of the right hemisphere, but seemed to be a grouping of three subclusters. Accordingly, the three main local maxima in this cluster were each analyzed separately (right MOG, right IPL, and a cluster in the superior temporal gyrus close to the right temporo-parietal junction (TPJ/STG), yielding 10 locations total. (Note that we will use the combined abbreviation TPJ/STG here because the present local maximum is a bit ventral to the typical TPJ location. However, a slightly more dorsal local maximum displayed a very similar activity pattern. Moreover, there is some heterogeneity between studies reporting activity in TPJ, so that the present activation seems quite likely to relate to the functions that tend to be ascribed to the right TPJ.) Percent signal change values were determined for spherical ROIs around these ten maxima (see ; also see, for other functional contrasts of data from this experiment). Among these ROIs, we predicted finding three distinctive activity profiles for the different conditions: (1) Sensory-driven activity that would be present for Go-trials and further enhanced for Stop-trials (due to the extra sensory stimulation), but not differing significantly between Stop-relevant and Stop- irrelevant Stop-trials (dark blue bars in). (Note that the ROI selection favored Stop-trials, as it is based on a direct comparison of Stop-trials with Go-trials. Therefore, statistical tests between Stop- and Go-trials within the ROIs were avoided in our analyses here. Similarly, tests comparing activity estimates for Go-trials against zero were also not performed, and the respective results are only displayed to serve as an approximate reference and for qualitative comparisons between the activity profiles in different areas. To highlight this fact and to further set them apart from Stop-trials, bargraphs referring to Go-trials are represented without a fill color in and.) (2) Activity associable with automatic attentional capture by the rare Stop-stimuli, showing little or no activity on Go-trials but strong responses on Stop-trials irrespective of their task relevance (i.e., not differing significantly for stop-irrelevant versus stop-relevant Stop-trials; light blue bars in). (3) Activity profiles dominated by responses to the Stop-relevant Stop-stimuli, with significantly stronger responses to task-relevant than task-irrelevant Stop- trials and little or no response to Go-trials (red bars in). This activity pattern would be indicative of top-down control processes in response to the Stop-stimuli, including both directed attention toward them due to their relevance and response inhibition following their detection. ### Posterior brain regions Analyses of the five posterior ROIs revealed three different activity profiles that were largely symmetrical for the bilaterally activated areas. Occipital ROIs (left and right MOG) revealed a pattern in line with simple visual processing, in that activity estimates were similarly prominent for both types of Go-trials (i.e., in both the Stop-relevant and Stop-irrelevant task blocks), but were substantially larger for Stop-trials of either task block, reflecting the presentation of an additional salient stimulus on all Stop-trials. In contrast, the bilateral IPL regions produced very little activity for either kind of Go-trial, but showed strong responses for all Stop-trials. These IPL activations, however, did not differ significantly between the Stop-relevant and Stop-irrelevant Stop-trials (p\>0.1 in both hemispheres). Finally, the TPJ/STG ROI also displayed little activity related to Go-trials from either trial block type. More importantly, however, this area yielded a statistically significant difference between the average of Stop-relevant Stop-trials vs. Stop-irrelevant Stop-trials (t(14) = 2.2; p = 0.04), distinguishing this area from all other posterior regions. ### Frontal brain regions Of the five frontal clusters identified, all areas except the right IFJ displayed qualitatively the same pattern of activity. More specifically, these areas did not respond strongly to Go-trials in either task block, nor to the Stop-trials from the Stop-irrelevant blocks, but responded strongly to Stop- relevant Stop-trials. In all these areas the Stop-relevant Stop-trials yielded significantly stronger activations than the Stop-irrelevant Stop-trials (all p\<0.05; see for further significant differences). Comparisons among only the Stop-relevant Stop-trials indicate that the right IFG displayed a trend for stronger activity for successful than for unsuccessful Stop-trials (t(14) = 2; p = 0.07). Interestingly, the pre-SMA showed a significant effect in the opposite direction (t(14) = 2.4; p = 0.03). The right IFJ displayed a different general activity profile, with stronger activity for successful Stop-trials than for unsuccessful Stop-trials (t(14) = 2.9; p = 0.01) or for Stop-irrelevant Stop-trials (t(14) = 2.4; p = 0.03), but the average of all Stop-relevant Stop- trials did not differ significantly from Stop-irrelevant Stop-trials in this area (p\>0.3). # Discussion The present fMRI study aimed at delineating the neural processes that are involved in the context of response inhibition during the Stop-signal task and to distinguish different neural underpinnings of the various cognitive processes engaged during such tasks. In an attempt to identify areas that respond to the rare and salient sensory stimulation of Stop-trials in an automatic fashion, we found that occipital and inferior parietal areas respond more strongly to Stop- trials than to Go-trials, even if the Stop-stimuli are completely task- irrelevant. Interestingly, this analysis also identified clusters in the right IFJ and the right pre-SMA that responded in a similar fashion, albeit only on a comparatively lenient uncorrected significance level. An additional ROI analysis that focused on the comparison of neural responses to Stop-trials from task blocks in which the Stop-stimuli were versus were not task-relevant identified three major activity profiles for different cortical areas. These profiles indicate a hierarchy in which pure sensory processing is mostly restricted to occipital areas, whereas some degree of automatic attentional capture by rare Stop-stimuli regardless of their task-relevance occurs in the inferior parietal lobules. In contrast, in the third profile, activity in a wide range of frontal areas and the right TPJ/STG was prominent only for Stop-trials that were task- relevant. These findings provide an important step towards establishing a framework in which sensory processes, bottom-up attentional processes, and top- down control functions can be attributed to specific portions of the wider cortical network that is typically associated with response inhibition during the Stop-signal task. ## Visual stimulation, attention, and response inhibition At least three recent publications have highlighted the difficulty of distinguishing processes directly involved in response inhibition from those related to the attentive processing of the relevant stimuli, focusing on the role of right IFG (–, see also –). The majority of these studies concluded that the right IFG, which has typically been considered a crucial node in response inhibition, may only be indirectly related to this function, and that this area may actually be more generally involved in the attentive processing of the Stop- stimuli. Using functional connectivity patterns derived from Granger causality analyses of fMRI data, Duann and colleagues observed that the right IFG influenced activity in the motor system only indirectly via the pre-SMA (see also). They concluded that the connectivity pattern of the right IFG suggested an attentional role based on its close functional relationship with temporal and parietal brain structures (see also). Two subsequent studies have made similar arguments by investigating modified Stop-signal tasks that used high-level control stimuli that did not require response inhibition. In addition to an attentional account of the function of the IFG, the ubiquity of its activation across different tasks might also relate to recent accounts that assign more global control functions to this area (for recent reviews, see). Moreover, it has recently been suggested in the context of a modified Go- NoGo task that the requirements for response control irrespective of response inhibition can also strongly activate the right IFG. Specifically, these authors report strong activations of the right IFG (exceeding the activation level of NoGo trials) for a control condition where subjects had to press an additional button in response to a third class of stimuli that were equally infrequent as NoGo trials. Note, however, that there is still controversy about whether the right IFG is really not directly related to response inhibition. Either way, the example of the right IFG thus highlights the importance of further distinguishing related functions on a brain-wide level, both for understanding the basic underlying cognitive functions and because psychopathological derailment of other functions could potentially mimic deficits in motor control. For example, a number of studies have reported that fluctuations in attentional engagement can strongly influence the outcome of Stop-trials in this task (–; see also), thereby mimicking fluctuations in response-inhibition processes. Importantly, none of the studies mentioned above included a condition in which Stop-stimuli were entirely task-irrelevant, thus leaving open the question whether the right IFG and other areas would respond to Stop-stimuli even when they are entirely task-irrelevant. Such activation could arise by means of bottom-up attentional capture by the Stop-stimuli simply due to their rarity and physical salience. The present study identified activation by task-irrelevant Stop-stimuli bilaterally in occipital areas and in IPL, as well as in the right IFJ and pre-SMA (albeit on a more lenient, uncorrected significance threshold). Other frontal areas such as the IFG, however, did *not* respond in this condition, thus indicating that their task involvement depends on some level of task-relevance of the Stop-stimuli. This relevance, in turn, does not have to be based on the necessity to withhold a motor response, but at a minimum on the requirement that such a stimulus needs to be discriminated from the other stimuli in the sequence to determine its task-relevance. Thus, the area that appeared to be most consistently activated by the mere rarity and salience of Stop-stimuli, irrespective of their task relevance, was the bilateral IPL, thus arguing against notions that have ascribed a direct involvement of this area in response inhibition (e.g.,). Activity in the IPL was highly similar in response to task-relevant and task-irrelevant Stop- stimuli, even when directly comparing activity estimates in the ROI analysis (i.e., avoiding conservative voxel-wise tests), thus also arguing against a graded involvement in the sense of a weak involvement in task-irrelevant Stop- trials that is enhanced for task-relevant ones. Based on this activity pattern, we would suggest that the IPL’s role would be described in terms of automatic attentional capture by rare salient events. Given this, it is slightly surprising that there is in fact no *behavioral* effect of attentional capture for task-irrelevant Stop-trials (i.e., no RT decrement as compared to the corresponding Go-trials). Nevertheless, the IPL appears to be mostly responsive to Stop-trials, thus arguing against a simple sensory role. However, it is not entirely atypical to find neural indications of attentional capture in the absence of a significant behavioral effect. Moreover, independent of its precise function, it appears that IPL is fulfilling a role during the Stop- signal task that is quite exclusive to Stop-trials, yet not directly related to response inhibition. For areas that respond to Stop-trials only when they are task-relevant, however, response inhibition processes cannot be easily distinguished from those related to enhanced attentive processing of the Stop-stimuli, or those related to increased response control demands that are not inhibitory. This paradigmatical problem is difficult to overcome if it is the case that task-relevant Stop- stimuli require more attention or other top-down control mechanisms than control stimuli that require a different response. Consequently, areas that are more active during task-relevant Stop-trials could generally subserve several different functions. The present data suggest that for most of the frontal activations, as well as for those in the TPJ/STG, Stop-stimuli do not elicit a robust neural response if they are completely task-irrelevant. However, while this defines a “lower limit” of basic processes that do not elicit activity in typical response-inhibition areas, the present data cannot further distinguish between different high-level operations such as top-down attentional engagement, response demand complexity, and response inhibition. ## The role of right IFJ and MFG While the right IFG and pre-SMA have commonly been discussed in the context of response inhibition, other frontal areas such as the right IFJ, MFG, and left anterior insula have also been frequently implicated in such tasks. Considerably less is known about the functional role of these other areas, however. For example, the right IFJ has been argued to play a role in detecting infrequent NoGo-trials in a Go-NoGo task, including because it has been shown to also respond to an additional type of Go-trials during a Go-NoGo task when they are presented as infrequently as the NoGo-trials. This finding dovetails with our observation that the right IFJ also responds to Stop-trials that are entirely task-irrelevant (see also). The present data furthermore revealed that the right IFJ was more active during successful than during unsuccessful Stop-trials (with the latter triggering similar levels of activity as the task-irrelevant Stop-trials), echoing the results from another recent report. Interestingly, it has also been reported that the right IFJ is more active for Go-trials that *might* turn into a Stop-trial than for Go-trials that could not do so (“conditional” Stop-signal task, see), suggesting that such activations may represent a process related to the preparation to inhibit a response , rather than being part of the inhibition-generating processes itself (see also). More generally, the right IFJ has been implicated in maintaining, updating, and/or activating task sets. Our finding that the right IFJ activity is enhanced during successful versus unsuccessful Stop-trials suggests that it may play a role either specifically in preparing response inhibition, or in representing and enforcing the task rules more generally to influence the outcome of Stop-trials. The current results also revealed a large cluster of activity in the right MFG during task-relevant Stop-trials, in line with results from earlier response- inhibition paradigms,. In general, the role of this area during response inhibition has not been well characterized, but it has been suggested to be involved in task-related, top-down control processes, potentially related to working memory demands. Given that the present MFG activation did not differentiate between successful and unsuccessful Stop-trials, we cannot attribute a more specific role to it, beyond the fact that it only responds to Stop-trials when they are task-relevant. ## The cortical control of response inhibition Given recent reports of right IFG activation by control stimuli that do not require response inhibition, it is possible that the role of this brain area in response inhibition may be relatively indirect, which would be counter to most earlier notions that identified it as being critical for this function. Accordingly, it would not be clear which cortical area, if not the right IFG, actually initiates the cancellation of a motor response. Although one possible candidate is the right pre-SMA, it is noteworthy that in our study, as well as in another recent report, the pre-SMA was *more* active during unsuccessful than during successful Stop-trials. This raises the question whether inhibition- related activity is really stronger in successful than in unsuccessful Stop- trials, as is usually assumed. Alternatively, it is conceivable that other control-modulating factors, such as fluctuations in the perceptual/attentional processing of the task-relevant stimuli, are critical in determining the outcome of the process, while activity fluctuations related to actual response inhibition may not be the critical determining factor as to whether a Stop-trial is responded to successfully or not. Another possibility would be that the fMRI signal, due to its low temporal resolution, additionally includes processes that occur in the pre-SMA before or after the response has been (or has not been) generated, which could also differ between the successful and unsuccessful Stop- trials without reflecting response inhibition processes. The fact that the pre- SMA might perform additional operations during this task that are not related to response inhibition is further supported by our finding that there appears to be a weak pre-SMA response even for task-irrelevant Stop-trials. Although it is important to note that the respective cluster was only weakly activated and did not survive multiple-comparison correction, the fact that the peak coordinate was identical to the one that resulted from the analysis that also included task-relevant Stop-trials would appear to give some more credibility to this activation. Based on the observation that the pre-SMA generally tracks response speed in this context, even for Go-trials, it is possible that this activity is related to the small amount of slowing that occurs during task-irrelevant Stop- trials (as compared to the respective Go-trials). Nonetheless, other functions are conceivable, including for example a role in attentional shifting. Regardless, the current findings underscore that further research is needed to more clearly disentangle the functional components of processes engaged during the inhibition of a motor output in response to the detection of a unique stimulus type in a stream of other stimuli. [^1]: Conceived and designed the experiments: CNB RMK MGW. Performed the experiments: CNB LCC. Analyzed the data: CNB LGA RMK LCC. Wrote the paper: CNB LGA RMK LCC MGW. [^2]: The authors have declared that no competing interests exist.

# Introduction Pain is the most commonly reported symptom in the general population and medical settings. Persistent or chronic pain is a major universal health problem, prompting the WHO to endorse a global campaign against pain. Pain has been associated with lower degree of health-related quality of life and may lead to psychosocial distress, insomnia, and depressive symptoms. ## Cannabinoids Cannabinoids have lately emerged as a potential alternative to other analgesics, e.g., opioids for the treatment of pain. Cannabinoids are most commonly consumed via smoked, inhaled vapour, or oral routes of administration. Sublingual administration is used for some medical cannabis preparations (e.g., nabiximol). The endocannabinoid system consists of two types of cannabinoid receptors in the human body, type I and type II. Cannabinoid receptor type I are most abundant in the central nervous system, especially in areas stimulating nociception and short-term memory, and in the basal ganglia. Cannabinoid receptor type II is mostly found in the periphery, often in conjunction with immune cells, but may appear in the central nervous system predominantly in association with microcytes during inflammation. These receptors is thought to suppress the pain stimulus through different mechanisms. Neurochemical, behavioral, and electrophysiological studies all demonstrated the modulation of inflammatory nociception through cannabinoid receptor type II activation. ## Pain Acute pain usually has a well-defined onset and most often a readily identifiable etiology (i.e., surgery, etc.). Acute pain is expected to run its course in a relatively short time frame and management of acute pain typically focuses on providing symptomatic relief until pain is reduced to an acceptable level. Chronic pain is typically defined as pain lasting for more than three months. Chronic pain may also have a well-defined onset related to tissue injury (e.g., surgery) and be mediated through an intact nervous system. It may, however, also be caused by nerve damage and dynamic changes in the nervous system, and be characterized by an ill-defined onset and a prolonged, fluctuating course. Pain may also be classified based on whether it is cancer-related or non-cancer- related. Cancer-related pain is caused by the cancer itself (primary tumor and metastases) or its treatment (e.g., radiation therapy). Previous reviews have reported on serious adverse events (e.g., agitation, impaired memory, abuse, dissociation, acute psychosis, and death) and non- serious adverse events (e.g., sedation, dizziness, dry mouth, increased appetite, somnolence, confusion and nausea) \[–, –\] in users of cannabinoids for pain. According to Canada’s Drug and Health Technology Agency (CADTH) there is some suggestion of benefit with cannabis-based medicines specifically for neuropathic pain. However, such benefits needs to be weighed against harms. Before healthcare systems ought to endorse the applicability of cannabinoids for pain globally, the potential short- and long-term benefits and harms of cannabinoids must be investigated. We conducted this systematic review with meta-analysis and Trial Sequential Analysis (TSA) and based on available randomised trials to evaluate the effectiveness and safety of cannabinoids in participants with pain. # Methods The objective of our systematic review was to assess the benefits and harms of cannabinoids versus placebo or no intervention for participants with any type of pain (acute and chronic pain, cancer pain, or any other types of pain). Our methodology is described in detail in our protocol published prior to conducting the literature search. In short, we carried out this systematic review following the recommendations of the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines. We included all randomised clinical trials comparing cannabinoids versus placebo or no intervention for participants with any type of pain. Two authors (JB, SKK) independently searched for trials identified prior to January 2022 see ‘**Supplement 1 in S1 File**’ for a detailed list of databases and ‘**Supplement 2 in** ’ for the search strategy. We included randomised clinical trials regardless of trial design, setting, publication status, year, language, and reporting of outcomes. Four authors (JB, SKK, JBF, and MM) working in pairs independently extracted data and assessed the risks of bias in included trials. We contacted trial authors by email if data were unclear or missing. Disagreements were resolved through discussion or by consulting a third author (JCJ). ## Outcomes ### Primary outcomes - All-cause mortality - Pain assessment on visual analogue scale (VAS) or numerical rating scale (NRS) - Proportion of participants with one or more serious adverse events. Serious adverse events were defined as any untoward medical occurrence that resulted in death; was life threatening; was persistent or led to significant disability, hospitalisation, or prolonged hospitalisation. As we expected the trialists’ reporting of serious adverse events to be heterogeneous and not strictly according to the International Conference on Harmonisation—Good Clinical Practice (ICH-GCP) recommendations, we included the event as a serious adverse if the trialists either: (1) used the term ’serious adverse event’ but did not refer to ICH-GCP, or (2) reported the proportion of participants with an event we considered fulfils the ICH-GCP definition. If several of such events were reported then we choose the highest proportion reported in each trial - Quality of life measured on any valid (published validation) continuous scale ### Secondary outcomes - Cannabinoid dependence (as defined by trialists) - Psychosis (as defined by trialists) - Proportion of participants with one or more adverse event not considered to be serious - Quality of sleep measured on any valid (published validation) continuous scale ### Exploratory outcomes - Each type of serious adverse event separately - Each type of adverse event not considered serious separately - Twenty-four hours morphine consumption (as defined by trialists) - Physical function (as defined by trialists) - Depressive symptoms (e.g., Hamilton Depression Rating Scale) Adverse events were included in our analysis regardless of whether it was defined as an outcome by the trialists. For all outcomes, we used the trial results reported at maximal follow-up. ### Patient and public involvement We conducted email correspondence with several patient associations in Denmark to select the most patient-relevant outcomes. The associations were the Danish Diabetes Association, the Danish Rheumatism Association, the Danish Multiple Sclerosis Society, and the Danish Cancer Society. ### Sub-group analyses We pre-defined subgroup analyses for our primary outcomes assessing risk of bias, risk of vested interests, type of pain, type of chronic pain, and type of cannabinoids used. ### Assessment of risk of bias We assessed risk of bias according to the Cochrane Handbook for Systematic Reviews of Interventions 5.1. We evaluated the risk of bias in the domains ‘random sequence generation’, ‘allocation concealment’, ‘blinding of participants and treatment providers’, ‘blinding of outcome assessment, ‘incomplete outcome data’, and ‘selective outcome reporting’. We used these domains to classify the included trials as being at overall low risk of bias or at overall high risk of bias as described in our protocol. The domains ‘blinding of outcome assessment’, ‘incomplete outcome data’, and ‘selective outcome reporting’ were further assessed separately for each outcome result. ### Assessment of effect We calculated risk ratios (RRs) with 95% and 98% confidence intervals (CIs) for our continuous and dichotomous outcomes (see "**Assessment of statistical and clinical significance")**. For our dichotomous outcomes we used the conventional direction with RR \> 1.0 representing higher risk in the experimental intervention group. We calculated mean differences (MDs) and standardised mean differences (SMDs) for our continuous outcomes. For pain assessment, we used the numerical rating scale to measure the mean difference between groups. Visual analog scale (VAS) (0 to 100) was converted into the numerical rating scale (NRS) (0 to 10) by dividing with 10. ### Assessment of heterogeneity We investigated forest plots to visually assess any sign of heterogeneity. We secondly assessed the presence of statistical heterogeneity by Chi² test (threshold P \< 0.10) and measured the quantities of heterogeneity by I² statistic and tau (τ)² statistic. We also investigated heterogeneity through subgroup analyses. Ultimately, we decided whether the assessment of heterogeneity showed that meta-analysis should be avoided. ### Assessment of reporting biases We used funnel plots to assess reporting bias, although it should be noted that funnel plots assess small-study effects. Funnel plots were performed if 10 or more trials were included. For dichotomous outcomes, we assessed asymmetry with the Harbord test if τ² was less than 0.1 and with the Rücker test if τ² was more than 0.1. For continuous outcomes, we used the regression asymmetry test. ### Assessment of statistical and clinical significance We performed all meta-analyses using Review Manager 5.4.1 and STATA 16.1. We assessed our intervention effects with both random-effects meta-analyses (DerSimonian method) and fixed-effect meta-analyses (DeMets method). We primarily reported the more conservative point estimate of the two and the less conservative result as a sensitivity analysis. To control random errors when analysing our primary outcomes, we adjusted the threshold for statistical significance using the procedure suggested by Jakobsen and colleagues. We used four primary outcomes and therefore considered a P-value of 0.02 as the threshold for statistical significance. When analysing secondary and exploratory outcomes, we considered a P-value of 0.05 as the threshold for statistical significance as these outcomes were considered hypothesis-generating only. We used Trial Sequential Analysis (TSA) to control for the risks of random errors. Traditional meta-analysis runs the risk of random errors due to sparse data and repetitive testing of accumulating data when updating reviews. We wished to control the risks of type I errors and type II errors. By conducting TSA on the outcomes, we could calculate the required information size, i.e., the number of participants needed in a meta-analysis to detect or reject our anticipated intervention effects. For dichotomous outcomes, we estimated the required information size based on the observed proportion of participants with an outcome in the control group, a relative risk reduction of 20% in the experimental group, an alpha of 2.0%, a beta of 10%, and the diversity suggested by the trials in the meta-analysis. For the outcome pain assessment on VAS or NRS, we used an analgesic effect equivalent to 10 mm on VAS or 1 point on NRS. For the outcome 24-hour morphine consumption we used an effect equivalent to at least 5 mg morphine. For the remaining continuous outcomes, we used the observed standard deviation (SD), a mean difference of the observed SD/2, an alpha of 2.0% for our primary and secondary outcomes, a beta of 10%, and the diversity suggested by the trials in the meta-analysis. The above-mentioned intervention effects used in the TSA were also predefined as the minimal important differences (MIDs) of the review. We used a ’best-worst case’ and a ’worst-best case’ sensitivity analysis to assess the impact of missing data (incomplete outcome data bias). We used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) to assess the certainty of the evidence. # Results Our literature search identified 5766 records. After removing duplicates, 4302 records remained. We excluded 4196 records based on title or abstract. We excluded another 41 records based on the full texts. We included a total of 65 clinical trials randomising 7017 participants. All 65 trials compared cannabinoids versus placebo. The mean age of included participants was 50.4 years ranging from a mean of 20.7 years to 63.7 years (see for more details). The participants were included in the trials based on the different pain diagnoses. Forty-four trials randomised 4306 participants with chronic pain (34 trials in neuropathic pain, and 10 trials in chronic nociceptive pain); 9 trials randomised 1681 participants with cancer pain; 10 trials randomised 965 participants with acute pain; and 2 trials randomised 65 participants with fibromyalgia-related pain. Length of maximum follow-up of the included trials varied between 7 hours and 16 weeks with a mean length of follow-up of 7.3 weeks. Only 35 of the 65 trials provided data that could be included in our meta- analysis. The primary reason that trial data could not be included in the meta- analyses was that the trials were designed as cross-over trials and that they did not provide data at the end of the first phase. These trials are described qualitatively in the paragraph ‘**Trials not contributing with data in our meta- analyses’.** Fifty-nine of the 65 included trials were at high risk of bias, see ‘**Supplement 4 in** ’ and ‘**S4 File’** for more information on bias assessment. ## Primary outcomes ### All-cause mortality Seven trials randomising 2073 participants reported on all-cause mortality. Meta-analysis showed no evidence of a difference between cannabinoids versus placebo (RR 1.20; 98% CI 0.85 to 1.67; P = 0.22; low certainty of evidence; **S1 Fig in**). Visual inspection of the forest plot, I²-statistic (I² = 7%), and tau² statistic (τ² = 0.01; τ = 0.1) indicated low heterogeneity. TSA showed that there was not enough information to confirm or reject that cannabinoids versus placebo increased the risk of all-cause mortality by 20% or more (**S2 Fig in**). We assessed this outcome result at high risk of bias, see ‘**Supplement 4 in** ’ and **S3 and S4 Figs in.** Test for interaction showed no evidence of a difference when comparing trials at high risk of bias to trials at low risk of bias (P = 0.87) (**S5 Fig);** trials at risk of vested interests to trials at no risk of vested interests (P = 0.87) (**S6 Fig in**); trials assessing different types of pain, i.e., cancer pain and chronic pain (P = 0.43) (**S7 Fig in**); trials comparing the effects of different types of cannabinoids (P = 0.78) (**S8 Fig)**. The remaining planned subgroup analyses were not possible to conduct due to lack of relevant data. ### Pain Twenty-six trials randomising 4110 participants assessed pain using either VAS (8 trials) or NRS (18 trials). We converted all pain measures to NRS as described in the ‘**Methods**’ section. The visual inspection of the forest plot and test for subgroup difference (P = 0.02), showed that the effects of cannabinoids seemed to differ between trials randomising participants with acute pain, cancer pain, and chronic pain (**S9 Fig in**). It was therefore not justifiable to meta-analyse the results of trials including the different types of pain. Hence, we chose to report results separately for each group of trials (acute pain; cancer pain; and chronic pain). ### Acute pain Four trials randomising 530 participants suffered from acute pain. Meta-analysis showed no evidence of a difference between cannabinoids versus placebo (mean difference NRS 0.52; 98% CI -0.40 to 1.43; P = 0.19; very low certainty of evidence; **S10 Fig).** Visual inspection of the forest plot, I²-statistic (I² = 90%), and tau² statistic (τ² = 0.67, τ = 0.82) indicated high heterogeneity that could not be resolved. TSA showed that we had not enough information to confirm or reject that cannabinoids versus placebo reduced acute pain (**S11 Fig in**). We assessed this outcome result at high risk of bias, see ‘**Supplement 4 in** ’. Test of interaction showed evidence of a difference when comparing trials at high risk of bias to trials at low risk of bias (P = 0.035) (**S12 Fig in**). When trials at low risk of bias and trials at high risk of bias were analysed separately we found no evidence of a difference between cannabinoids versus placebo (**S12 Fig in**). Test of interaction showed no evidence of a difference when comparing trials at risk of vested interests to trials at no risk of vested interests (P = 0.25) (**S13 Fig)**. All remaining planned subgroup analyses were not possible to conduct due to lack of relevant data. A post-hoc subgroup analysis comparing trials with long-term follow-up to trials with short-term follow-up showed no evidence of a difference (P = 0.10) (**S14 Fig in**). ### Cancer pain Six trials randomising 1550 participants suffered from cancer pain. Meta- analysis showed no evidence of a difference between cannabinoids versus placebo (mean difference NRS -0.13; 98% CI -0.33 to 0.06; P = 0.1; low certainty of evidence; **S15 Fig).** Visual inspection of the forest plot, I²-statistic (I² = 2%), and tau² statistic (τ² = 0.00) indicated low heterogeneity. TSA showed that there was enough information to reject that cannabinoids versus placebo reduced cancer pain (**S16 Fig in**). We assessed this outcome result as high risk of bias, see ‘**Supplement 4 in** ’. Test of interaction showed no evidence of a difference when comparing trials assessing the effects of different types of cannabinoids (P = 0.71) (**S17 Fig in**). All remaining planned subgroup analyses were not possible to conduct due to lack of relevant data. A post-hoc subgroup analysis comparing trials with long-term follow-up to trials with short-term follow-up showed no evidence of a difference (P = 0.06) (**S18 Fig in**). ### Chronic pain Sixteen trials randomising 2030 participants suffered from chronic pain. Meta- analysis showed that cannabinoids reduced chronic pain, but the effect size was below the predefined MID and the MID was not included in the 98% confidence interval (mean difference NRS -0.43; 98% CI -0.72 to -0.15; P = 0.0004; low certainty of evidence;, **S19 Fig).** Visual inspection of the forest plot, I²-statistic (I² = 64%), and tau² statistic (τ² = 0.11) indicated moderate heterogeneity that could not be resolved. TSA showed that there was enough information to confirm that cannabinoids versus placebo reduced chronic pain with a statistically significant effect (**S20 Fig in**). We assessed this outcome result as high risk of bias, see ‘**Supplement 4 in** ’. Test for interaction showed no evidence of a difference when comparing trials at high risk of bias to trials at low risk of bias (P = 0.31) (**S21 Fig in**); and trials at risk of vested interests to trials at no risk of vested interests (P = 0.57) (**S22 Fig in**). Test for interaction showed evidence of a difference when comparing trials assessing different types of chronic pain, i.e., neuropathic pain (including central, peripheral, and mixed), fibromyalgia, and visceral nociceptive pain (P = 0.01) (**S23 Fig in**); and trials comparing the effects of different types of cannabinoids (P \< 0.001) (**S24 Fig in**). The funnel plot showed no clear signs of small-study effects (**S25 Fig in).** A post-hoc subgroup analysis comparing trials with long-term follow-up to trials with short-term follow-up showed no evidence of a difference (P = 0.59) (**S26 Fig in**). ### Serious adverse events Eighteen trials randomising 3980 participants reported serious adverse events. Meta-analysis showed no evidence of a difference between cannabinoids versus placebo (RR 1.18; 98% CI 0.95 to 1.45; P = 0.07; low certainty of evidence;, **S27 Fig in**). Visual inspection of the forest plot, I²-statistic (I² = 0%), and tau² statistic (τ² = 0.00) indicated low heterogeneity. TSA showed that there was not enough information to confirm or reject that cannabinoids versus placebo increased the risk of serious adverse events by 20% or more (**S28 Fig in**). We assessed this outcome result at high risk of bias, see ‘**Supplement 4 in** ’ and **S29 and S30 Figs in**. Test for interaction showed no evidence of a difference when comparing trials at risk of vested interests to trials at no risk of vested interests (P = 0.97) (**S31 Fig in**); trials assessing different types of pain, i.e. acute pain, cancer pain and chronic pain (P = 0.96) (**S32 Fig in**); trials assessing different types of chronic pain, i.e., neuropathic pain (including central and peripheral) and nociceptive pain (P = 0.16) (**S33 Fig in**); and trials comparing the effects of different types of cannabinoids (P = 0.79) (**S34 Fig in**). The remaining planned subgroup analyses were not possible to conduct due to lack of relevant data. The funnel plot showed no clear signs of small-study effects (**S35 Fig in**). When analysing each specific serious adverse event, cannabinoids did not seem to increase or decrease the risk of any single serious adverse events compared with placebo, see ‘’. ### Quality of life Only four trials randomising 548 participants assessed quality of life using either EuroQol 5-D (3 trials) or EORTC-QLQC30 (1 trial). Meta-analysis showed no evidence of a difference between cannabinoids versus placebo (mean difference -1.38; 98% CI -11.8 to 9.04; P = 0.75; very low certainty of evidence; **S36 Fig in**). Visual inspection of the forest plot, I²-statistic (I² = 86%), and tau² statistic (τ² = 61.98, τ = 7.87) indicated high heterogeneity which could not be resolved. TSA showed that there was not enough information to confirm or reject that cannabinoids versus placebo improved quality of life (**S37 Fig in**). We assessed this outcome result as high risk of bias, see ‘**Supplement 4 in** ’. Test for interaction showed no evidence of a difference when comparing trials assessing different types of pain, i.e., cancer pain and chronic pain (P = 0.33) (**S38 Fig)** and trials comparing the effects of different types of cannabinoids (P = 0.56) (**S39 Fig in**). All remaining planned subgroup analyses were not possible to conduct due to lack of relevant data. Meta-analysing the standard mean difference showed no evidence of a difference between cannabinoids versus placebo (SMD -0.19; 95% CI -0.70, 0.32; P = 0.46). ## Secondary outcomes ### Dependence and psychosis None of the included randomised clinical trials reported results on the outcomes ‘dependence’ or ‘psychosis’ that could be analysed through a meta-analysis. ### Non-serious adverse events Twenty-nine trials randomising 5536 participants reported non-serious adverse events. Meta-analysis showed evidence of a harmful effect of cannabinoids (RR 1.20; 95% CI 1.15 to 1.25; P \< 0.001; very low certainty of evidence; **, S40 Fig in**). The number needed to harm (NNH) was seven. Visual inspection of the forest plot, I²-statistic (I² = 27%), and tau² statistic (τ² = 0.00) indicated low heterogeneity. TSA showed that there was enough information to confirm that cannabinoids versus placebo increased the risk of non-serious adverse events by 20% or more (**S41 Fig in** ). We assessed this outcome result at high risk of bias, see ‘**Supplement 4 in** ’ and **S42 and S43 Figs in**. The funnel plot showed signs of small-study effects (**S44 Fig in**). When analysing each specific non-serious adverse event, cannabinoids seemed to increase the risk of five non-serious adverse events versus placebo, see ‘’. These were ‘dizziness’, ‘fatigue’, ‘vertigo’, ‘nervous system disorders’, and ‘gastrointestinal disorder’. Cannabinoids did not seem to decrease the risk of any non-serious adverse events. ### Quality of sleep Seventeen trials randomising 3291 participants assessed quality of sleep using numerical rating scale (NRS). Meta-analysis showed that cannabinoids versus placebo improved quality of sleep, but the effect size was below the predefined MID and the MID was excluded in the confidence interval (mean difference NRS -0.42; 95% CI -0.65 to -0.20; P = 0.0003; low certainty of evidence; **, S45 Fig in**). Visual inspection of the forest plot, I²-statistic (I² = 74%), and tau² statistic (τ² = 0.15) indicated moderate heterogeneity that could not be resolved. TSA showed that there was enough information to confirm that cannabinoids versus placebo improved quality of sleep with a statistically significant effect (**S46 Fig in**). We assessed this outcome result at high risk of bias, see ‘**Supplement 4 in** ’. The funnel plot showed no clear signs of small-study effects (**S47 Fig in**). ## Exploratory outcomes ### Twenty-four hours morphine consumption Six trials randomising 1546 participants assessed 24-hours morphine consumption. Meta-analysis showed no evidence of a difference between cannabinoids versus placebo (mean difference -0.01; 95% CI -1.60 to 1.59; P = 0.99; low certainty of evidence; **S48 Fig in**). Visual inspection of the forest plot and I²-statistic (I² = 21%) and tau² statistic (τ² = 29.78, τ = 5.46) indicated low heterogeneity. TSA showed that there was enough information to reject that cannabinoids versus placebo reduced 24-hours morphine consumption by five mg morphine equivalents or more (**S49 Fig in**). We assessed this outcome result at high risk of bias, see ‘**Supplement 4 in** ’. ### Physical function (measured by activities of daily living) Three trials randomising 320 participants assessed ability to perform activities of daily living (ADL) using Barthel Index. Meta-analysis showed no evidence of a difference between cannabinoids versus placebo (mean difference 0.21; 95% CI -0.10 to 0.51; P = 0.18; very low certainty of evidence; **S50 Fig in**). Visual inspection of the forest plot and I²-statistic (I² = 0%) and tau² statistic (τ² = 0.00) indicated low heterogeneity. TSA could not be performed due to inadequate information size. We assessed this outcome result at high risk of bias, see ‘**Supplement 4 in** ’. ### Depressive symptoms Five trials randomising 651 participants assessed depressive symptoms using either HADS-D (3 trials), MADRS (1 trial) or BDI-II (1 trial). Meta-analysis showed no evidence of a difference between cannabinoids versus placebo (mean difference -0.03; 95% CI -0.12 to 0.06; P = 0.49; low certainty of evidence; **S51 Fig in**). Visual inspection of the forest plot and I²-statistic (I² = 0%) and tau² statistic (τ² = 0.00) indicated low heterogeneity. TSA showed that there was enough information to reject that cannabinoids versus placebo improved depressive symptoms (**S52 Fig in**). We assessed this outcome result at high risk of bias, see ‘**Supplement 4 in** ’. ### Trials not contributing with data in our meta-analyses Nineteen trials randomising 603 participants concluded in favor of an analgesic effect of cannabinoids compared with placebo \[, ,, –\]. Fifteen trials assessed chronic pain including neuropathic pain (13/15) and visceral nociceptive pain (2/15), two trials assessed acute pain, and two trials assessed cancer pain. Six trials assessed the use of plant-derived THC extract, ten trials assessed the use of synthetic THC, and three trials assessed plant-derived THC/CBD extract. Cannabinoids were generally assessed as safe. Only one trial found a beneficial effect of cannabinoids on quality of life compared with placebo. Mortality was not assessed by any of the trials. Eleven trials randomising 391 participants concluded against an analgesic effect of cannabinoids compared with placebo. Eight trials assessed chronic including neuropathic pain (4/8), visceral nociceptive pain (3/8), and fibromyalgia (1/8), two trials assessed acute pain, and one trial assessed cancer pain. Two trials assessed the use of plant-derived THC extract, four trials assessed the use of synthetic THC, one trial assessed the use of CBD and four trials assessed plant-derived THC/CBD extract. Cannabinoids were generally assessed as safe, however, cannabinoids were not assessed in regards to mortality or quality of life. # Discussion The objective of this review was to assess the beneficial and harmful effects of cannabinoids in any dose, formulation, and duration versus placebo or no intervention for participants with any type of pain. Meta-analysis and TSA showed no evidence of a difference between cannabinoids versus placebo on all-cause mortality, serious adverse events, and quality of life. Meta-analyses showed that cannabinoids compared with placebo reduced chronic pain (in particular central and peripheral neuropathic pain) and improved quality of sleep, but both effect sizes were below our predefined minimal important differences and the minimal important differences were excluded by both confidence intervals. Meta-analyses and TSA showed that cannabinoids increased the risks of non-serious adverse events, which corresponds to a number needed to harm of seven. None of the included trials reported on cannabinoid dependence or psychosis. According to the European Medicines Agency, P-values are of limited value as relative or absolute differences in terms of adverse effects. A non-significant difference between treatments will not allow for a conclusion on the absence of a difference in safety. In cases of adjustment for multiplicity, the European Medicines Agency state that this can be counterproductive for considerations of safety. Hence, even though meta-analysis and TSA did not show a statistically significant difference when assessing serious adverse events, our results show indications of a harmful effect which should be considered before prescribing cannabinoids for pain. Our findings are in contrast to the majority of previous reviews which have indicated an adequate analgesic effect of cannabinoids and supported the use of cannabinoids for the treatment of chronic pain \[–, , \]. Our findings suggest that cannabinoids reduced chronic pain when compared with placebo, however, whether this effect is clinically important seems questionable. Our findings are in agreement with position statement of the International Association for the Study of Pain (IASP). IASP have identified important research gaps and due to the lack of high- quality clinical evidence IASP does not currently endorse general use of cannabis and cannabinoids for pain relief. Our findings regarding cancer pain and acute pain are in line with previous reviews on this topic. ## Strengths Our present review has several strengths. Our methodology was predefined and was described in detail in our published protocol. To control the risk of random errors, we used TSA and adjusted our thresholds for statistical significance. We included more trials than any other previous review, which has increased our power and precision and strengthened our analyses. We assessed the risk of bias of all included trials to assess the risks of systematic errors, and we used our eight-step procedure to assess if the thresholds for statistical and clinical significance were crossed. Furthermore, we predefined MIDs for all outcomes to assess the clinical implications for patients of our results. Pain level thresholds for acute pain and for chronic pain were predefined based on Olsen et al.. The MID of one point on NRS is considered lenient in comparison to previous reviews, and our predefined lenient quantification decreases the risk of erroneous rejections of cannabinoids’ beneficial effects on pain. We were also in contact with several relevant patient associations in Denmark at the protocol stage to select the most relevant patient-important outcomes. ## Limitations Our review also has several limitations. All trials except five were at high risk of bias. Therefore, there is a high risk that our results overestimate the beneficial effects and underestimate the harmful effects of cannabinoids. We decided to combine the VAS scores and the NRS scores by converting the results to NRS scores. Even though the two scoring systems correspond very well, some information may be lost in the conversion. Furthermore, a limitation of this systematic review is that we only compared the cannabinoid intervention with placebo. Hence, our results do not assess the effects of cannabinoids compared with other types of analgesics (e.g., opioids). When conducting a subgroup analysis, one is at risk of study-level confounding because the subgroup analyses are based on aggregate data on a trial level and we did not obtain any individual patient data which may decrease the statistical power of our subgroup analyses. This is a common potential limitation in a meta- analysis of aggregate data. ## Minimal important difference Pain and quality of sleep measurements are subjective measures, why imprecision could be present when assessing these outcomes. We, therefore, need to be careful before dismissing such outcome results as clinically unimportant. Nevertheless, any outcome result should be related to a predefined minimal important difference (MID) to ensure the scientific validity of trial results and to avoid focus on statistically significant results without importance to patients. If a large number of trial participants are randomised, small and clinically irrelevant intervention effects may lead to statistically significant results and rejection of the null hypothesis. Jaeschke et al. defined the MID as ‘the smallest difference in score in the domain of interest which patients perceive as beneficial’. Olsen et al. have conducted two systematic reviews on this matter in order to gather the evidence and present an estimate of the MID for pain. Olsen et al. conducted a systematic review on the MID in patients with acute pain and concluded that the median of the studies’ results was 17 mm on VAS (IQR 14 mm to 23 mm). Another systematic review conducted by Olsen et al. was on the MID in patients with chronic pain and the results showed a median of 23 mm on VAS (IQR 12 mm to 39 mm) when using the within-patient anchor-based method, while the median in studies using the sensitivity-based and specificity- based method was 20 mm on VAS (IQR 15 mm to 30 mm). Our MID for quality of sleep was not based on previous studies, because such studies have not been conducted. We, therefore, based our estimation of MID on Cohen’s D higher than 0.5. The 0.5 SMD threshold was originally proposed by Cohen (as a minimum for a ‘moderate’ effect) and has been used as a MID in several studies across medical specialties. Nevertheless, it has to be considered that the MID estimation for quality of sleep compared to the MID for pain is more unclear because studies assessing the MID are lacking. Several countries have recently either expanded or introduced the medical use of cannabinoids. Of the many different approaches introduced globally only a few have been presented by the European Monitoring Centre for Drugs and Drug Addiction (EMCDDA) in their report. Canada was one of the first countries to establish a national programme for the medical use of cannabinoids in 1999 and have since evolved to an expanded access programme. New legislation in 2014 licensed more cannabis producers, allowed doctors greater latitude in prescribing, removed federal oversight of prescribing, and permitted patients to receive cannabis directly from licensed producers. Similarly, numerous European countries have allowed the usage of cannabinoids for medical purposes. The different national regulatory frameworks are complicated, but the most common initial approach implemented is to use some form of special access scheme. Examples of countries that have established some form of exceptional use programme or access programme to allow access to cannabinoid preparations for the treatment of pain are Croatia, Denmark, Finland, Norway, Poland, Sweden, Czechia, Germany, Italy, and The Netherlands. According to our present results, the medical use of cannabinoids for pain ought to be reconsidered. The NASEM report from 2017 on the health effects of marijuana reviewed the evidence on the use of cannabinoids as a treatment for chronic pain. The primary source of information for this summary was based on the works of Whiting et al. suggesting that cannabinoids demonstrate a modest effect on pain. Our present review represents the most comprehensive systematic assessment of the effects of cannabinoids on pain. This systematic review will furthermore work as recommendation for where focus needs to be in future randomised clinical trials. # Conclusions Cannabinoids reduced chronic pain and improved quality of sleep, but the effect sizes are of questionable importance. Cannabinoids had no effects on acute pain or cancer pain and increased the risk of non-serious adverse events. The harmful effects of cannabinoids for pain seem to outweigh the potential benefits. The expanded medical use of cannabinoids for pain is at this point questionable. # Supporting information 10.1371/journal.pone.0267420.r001 Decision Letter 0 Siddiqi Tariq Jamal Academic Editor 2023 Tariq Jamal Siddiqi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 16 Jun 2022 PONE-D-22-10156Cannabinoids versus placebo for pain: a systematic review with meta-analysis and Trial Sequential AnalysisPLOS ONE Dear Dr. Barakji, Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process. Please submit your revised manuscript by Jul 31 2022 11:59PM. If you will need more time than this to complete your revisions, please reply to this message or contact the journal office at <plosone@plos.org>. When you're ready to submit your revision, log on to <https://www.editorialmanager.com/pone/> and select the 'Submissions Needing Revision' folder to locate your manuscript file. Please include the following items when submitting your revised manuscript:A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). You should upload this letter as a separate file labeled 'Response to Reviewers'.A marked-up copy of your manuscript that highlights changes made to the original version. You should upload this as a separate file labeled 'Revised Manuscript with Track Changes'.An unmarked version of your revised paper without tracked changes. You should upload this as a separate file labeled 'Manuscript'. If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter. Guidelines for resubmitting your figure files are available below the reviewer comments at the end of this letter. If applicable, we recommend that you deposit your laboratory protocols in protocols.io to enhance the reproducibility of your results. Protocols.io assigns your protocol its own identifier (DOI) so that it can be cited independently in the future. For instructions see: <https://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory- protocols>. Additionally, PLOS ONE offers an option for publishing peer-reviewed Lab Protocol articles, which describe protocols hosted on protocols.io. Read more information on sharing protocols at <https://plos.org/protocols?utm_medium=editorial- email&utm_source=authorletters&utm_campaign=protocols>. We look forward to receiving your revised manuscript. Kind regards, Tariq Jamal Siddiqi Academic Editor PLOS ONE Journal Requirements: When submitting your revision, we need you to address these additional requirements. 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at  <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and  <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf>. 2\. Please ensure that you refer to Figures 2-5 in your text as, if accepted, production will need this reference to link the reader to the figure. 3\. Please include your tables as part of your main manuscript and remove the individual files. Please note that supplementary tables (should remain/ be uploaded) as separate "supporting information" files. 4\. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: [http://journals.plos.org/plosone/s/supporting- information. ](http://journals.plos.org/plosone/s/supporting-information. ) 5.  We noticed you have some minor occurrence of overlapping text with the following previous publication(s), which needs to be addressed: \- <https://bmjopen.bmj.com/content/9/10/e031574> In your revision ensure you cite all your sources (including your own works), and quote or rephrase any duplicated text outside the methods section. Further consideration is dependent on these concerns being addressed. \[Note: HTML markup is below. Please do not edit.\] Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 2\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 3\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#1: Yes \*\*\*\*\*\*\*\*\*\* 4\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#1: No \*\*\*\*\*\*\*\*\*\* 5\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#1: 1. Don’t use short forms like TSA directly in the abstract. 2\. Introduction: Please improve upon the Introduction. There should be use of a variety of sentence structures, if possible. Importantly, the gaps in current literature and consequently the rationale for this review remain unclear and should be elaborated upon further in the Introduction. Secondly, please discuss both acute and chronic pain while briefly outlining the comparison of cannabinoids with other relevant treatment therapies for pain. Thirdly, briefly outline the please discuss current guideline recommendations, if any. You may find the following paper useful: <https://www.cadth.ca/sites/default/files/pdf/h tis/2019/RC1153%20Cannabis%20Chronic%20Pain%20Final.pdf> 3\. Methods: Please cite “visual analogue scale (VAS) or numerical rating scale (NRS)”. Additionally, please elaborate upon the use of 95% and 98% confidence intervals for dichotomous outcomes – “We calculated risk ratios (RRs) with 95% and 98% confidence intervals (CIs) for our dichotomous outcomes”. 4\. Results: The detailed analysis of the data at hand is commendable. Please state an insignificant p value as there being no difference between the cannabinoid and placebo groups, for instance for “Meta-analysis indicated that cannabinoids increased the risk of serious adverse events (RR 1.18; 98% CI 0.95 to 1.45; P = 0.07; low certainty of evidence; Supplementary figure 27).” 5\. Use the short form for “Trial Sequential Analysis” as TSA after mentioning the full-form once initially. Also, state the full form of GRADE. 6\. Discussion: Please discuss, compare, and contrast findings of other studies with your findings instead of merely stating “Our findings are in contrast to the majority of previous reviews which have indicated an adequate analgesic effect of cannabinoids and supported the use of cannabinoids for the treatment of chronic pain. 20-23 111 112 Our findings suggest that cannabinoids reduced chronic pain when compared with placebo, however, whether this effect is clinically important seems questionable. Our findings regarding cancer pain and acute pain are in line with previous reviews on this topic.113- 115”. Also, improve upon the structure and quality of the Discussion section in general. 7\. References: Avoid the use of references more than 10 years old. \*\*\*\*\*\*\*\*\*\* 6\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#1: **Yes: **Warda Ahmed \*\*\*\*\*\*\*\*\*\* \[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files.\] While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, <https://pacev2.apexcovantage.com/>. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email PLOS at <figures@plos.org>. Please note that Supporting Information files do not need this step. 10.1371/journal.pone.0267420.r002 Author response to Decision Letter 0 14 Jul 2022 1\. Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at <https://journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main _body.pdf> and <https://journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_titl e_authors_affiliations.pdf>. This has been addressed. We have changed the reference style. The headings have been changed according to the PLOS ONE requirements. Figure and table caption have been inserted directly after they have been cited in the text. 2\. Please ensure that you refer to Figures 2-5 in your text as, if accepted, production will need this reference to link the reader to the figure. Figures 2-5 are now referred to in the text. 3\. Please include your tables as part of your main manuscript and remove the individual files. Please note that supplementary tables (should remain/ be uploaded) as separate "supporting information" files. We have now included the “Table 1: Table of included studies” in our manuscript. 4\. Please include captions for your Supporting Information files at the end of your manuscript, and update any in-text citations to match accordingly. Please see our Supporting Information guidelines for more information: <http://journals.plos.org/plosone/s/supporting-information>. We have now included a paragraph of our supporting information files at the end of the manuscript. This includes captions of all the supplemental files. 5\. We noticed you have some minor occurrence of overlapping text with the following previous publication(s), which needs to be addressed: \- <https://bmjopen.bmj.com/content/9/10/e031574> We have now rephrased and changes the sentence structure of the introduction section in order to avoid overlapping text with the above mentioned publication. In your revision ensure you cite all your sources (including your own works), and quote or rephrase any duplicated text outside the methods section. Further consideration is dependent on these concerns being addressed. We have carefully checked that all sources are cited and that there is no duplicated text outside the methods section. Reviewer response: 1.Don’t use short forms like TSA directly in the abstract. We have changed this accordingly 2\. Introduction: Please improve upon the Introduction. There should be use of a variety of sentence structures, if possible. Importantly, the gaps in current literature and consequently the rationale for this review remain unclear and should be elaborated upon further in the Introduction. Secondly, please discuss both acute and chronic pain while briefly outlining the comparison of cannabinoids with other relevant treatment therapies for pain. Thirdly, briefly outline the please discuss current guideline recommendations, if any. You may find the following paper useful: <https://www.cadth.ca/sites/default/files/pdf/h tis/2019/RC1153%20Cannabis%20Chronic%20Pain%20Final.pdf> The introduction has now been revised and separated into different sections to provide a better overview. We have tried to describe the difference between acute pain, chronic pain and cancer pain more in depth while trying to included current guideline recommendations. We thank the reviewer for this suggestion. We have also tried to clarify the rational for this systematic review further in the end of the introduction section. 3\. Methods: Please cite “visual analogue scale (VAS) or numerical rating scale (NRS)”. Additionally, please elaborate upon the use of 95% and 98% confidence intervals for dichotomous outcomes – “We calculated risk ratios (RRs) with 95% and 98% confidence intervals (CIs) for our dichotomous outcomes”. We have now cited the visual analogue scale (VAS) or numerical rating scale (NRS) in the methods section. We have also elaborated upon the use of two different confidence intervals. This rationale has also been described in the section “Assessment of statistical and clinical significance" previously, therefore we have also referred to this section. 4\. Results: The detailed analysis of the data at hand is commendable. Please state an insignificant p value as there being no difference between the cannabinoid and placebo groups, for instance for “Meta-analysis indicated that cannabinoids increased the risk of serious adverse events (RR 1.18; 98% CI 0.95 to 1.45; P = 0.07; low certainty of evidence; Supplementary figure 27).” We have changed this to “Meta-analysis showed no evidence of a difference between cannabinoids versus placebo (RR 1.18; 98% CI 0.95 to 1.45; P = 0.07; low certainty of evidence; Fig 3, Supplementary figure 27)”. 5\. Use the short form for “Trial Sequential Analysis” as TSA after mentioning the full-form once initially. Also, state the full form of GRADE. We have changed this accordingly. 6\. Discussion: Please discuss, compare, and contrast findings of other studies with your findings instead of merely stating “Our findings are in contrast to the majority of previous reviews which have indicated an adequate analgesic effect of cannabinoids and supported the use of cannabinoids for the treatment of chronic pain. 20-23 111 112 Our findings suggest that cannabinoids reduced chronic pain when compared with placebo, however, whether this effect is clinically important seems questionable. Our findings regarding cancer pain and acute pain are in line with previous reviews on this topic.113- 115”. Also, improve upon the structure and quality of the Discussion section in general. We have included the position paper of the IASP and compared our results to the findings of the IASP. Further, we have divided the discussion section into several parts to improve the structure. Therefore the following sections appear in the discussion section: “Strengths”, “Limitations” and “Minimal important difference”. 7\. References: Avoid the use of references more than 10 years old. We thank the peer reviewer for this advice. This has been taken into consideration. 10.1371/journal.pone.0267420.r003 Decision Letter 1 Siddiqi Tariq Jamal Academic Editor 2023 Tariq Jamal Siddiqi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 17 Aug 2022 Cannabinoids versus placebo for pain: a systematic review with meta-analysis and Trial Sequential Analysis PONE-D-22-10156R1 Dear Dr. Barakji, We’re pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it meets all outstanding technical requirements. Within one week, you’ll receive an e-mail detailing the required amendments. When these have been addressed, you’ll receive a formal acceptance letter and your manuscript will be scheduled for publication. An invoice for payment will follow shortly after the formal acceptance. To ensure an efficient process, please log into Editorial Manager at <http://www.editorialmanager.com/pone/>, click the 'Update My Information' link at the top of the page, and double check that your user information is up-to- date. If you have any billing related questions, please contact our Author Billing department directly at <authorbilling@plos.org>. If your institution or institutions have a press office, please notify them about your upcoming paper to help maximize its impact. If they’ll be preparing press materials, please inform our press team as soon as possible -- no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact <onepress@plos.org>. Kind regards, Tariq Jamal Siddiqi Academic Editor PLOS ONE Additional Editor Comments (optional): Reviewers' comments: Reviewer's Responses to Questions **Comments to the Author** 1\. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation. Reviewer \#2: All comments have been addressed \*\*\*\*\*\*\*\*\*\* 2\. Is the manuscript technically sound, and do the data support the conclusions? The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 3\. Has the statistical analysis been performed appropriately and rigorously? Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 4\. Have the authors made all data underlying the findings in their manuscript fully available? The [PLOS Data policy](http://www.plosone.org/static/policies.action#sharing) requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 5\. Is the manuscript presented in an intelligible fashion and written in standard English? PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here. Reviewer \#2: Yes \*\*\*\*\*\*\*\*\*\* 6\. Review Comments to the Author Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters) Reviewer \#2: (No Response) \*\*\*\*\*\*\*\*\*\* 7\. PLOS authors have the option to publish the peer review history of their article ([what does this mean?](https://journals.plos.org/plosone/s/editorial- and-peer-review-process#loc-peer-review-history)). If published, this will include your full peer review and any attached files. If you choose “no”, your identity will remain anonymous but your review may still be made public. **Do you want your identity to be public for this peer review?** For information about this choice, including consent withdrawal, please see our [Privacy Policy](https://www.plos.org/privacy-policy). Reviewer \#2: No \*\*\*\*\*\*\*\*\*\* 10.1371/journal.pone.0267420.r004 Acceptance letter Siddiqi Tariq Jamal Academic Editor 2023 Tariq Jamal Siddiqi This is an open access article distributed under the terms of the Creative Commons Attribution License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited. 19 Aug 2022 PONE-D-22-10156R1 Cannabinoids versus placebo for pain: a systematic review with meta-analysis and Trial Sequential Analysis Dear Dr. Barakji: I'm pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department. If your institution or institutions have a press office, please let them know about your upcoming paper now to help maximize its impact. If they'll be preparing press materials, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact <onepress@plos.org>. If we can help with anything else, please email us at <plosone@plos.org>. Thank you for submitting your work to PLOS ONE and supporting open access. Kind regards, PLOS ONE Editorial Office Staff on behalf of Dr. Tariq Jamal Siddiqi Academic Editor PLOS ONE [^1]: The authors have declared that no competing interests exist.

# Introduction Humans accomplish a wide variety of motor and cognitive tasks in everyday life. Furthermore, we continue acquiring new complex motor skills to overcome various challenges. Therefore, it is important to study the underlying changes in cerebral hemodynamics during motor-cognitive adaptation. In recent years, various functional brain imaging techniques that enable noninvasive visualization of brain activity have advanced greatly. Cerebral dynamics during motor-cognitive task adaptations have been investigated by employing various neuroimaging techniques, such as positron emission tomography (PET) or functional magnetic resonance imaging (fMRI), while the subject is in a supine position. However, in daily life, motor-cognitive adaptation tasks are usually performed in upright positions such as sitting or standing. To address the issue of limited ecological validity and enable task performance in a more natural setting, we used near-infrared spectroscopy (NIRS) as a less-limiting imaging method that enables measurement of cortical activation during activities of daily life. Although NIRS is characterized by marked limitations in temporal and spatial resolution, its safety, non-restrictiveness, and portability enable broader and more flexible use than other brain imaging methods. A number of studies have examined cerebral hemodynamics using NIRS during a wide variety of motor activities such as a walking or running, cycling, apple peeling, and finger tapping. Additionally, researchers have examined cerebral hemodynamics during cognitive tasks such as trail making, the rock-paper-scissors game, maze navigation, and sequential finger touching. To our knowledge, NIRS studies that examine motor skill learning with ongoing adaptation processes are rare. Two NIRS studies have reported changes of hemodynamics over time related to motor skill learning using eye-hand coordination in pursuit rotor or target reaching tasks. These NIRS studies evaluated specific sensorimotor tasks. However, these studies did not address changes in hemodynamic activity during motor learning of complex skills with concurrent cognitive processing such as working memory and executive function. Gentili et al. reported changes in cerebral hemodynamics as measured by NIRS during performance of a motor-cognitive adaptation task and demonstrated activation only in the prefrontal cortex (PFC). However, the NIRS probe in their study only covered the forehead. Therefore, no studies are available that examine the hemodynamic responses of other cortical regions during performance of motor-cognitive adaptation tasks using a NIRS system that also includes the posterior half-head. Information technology has substantially affected modern society, and researchers concerned with neurorehabilitation should examine interventions relevant to the lifestyle of modern people. In recent years, the number of touch-screen terminal users has increased remarkably and smartphones have spread all over the world. eMarketer, a U.S. research company, speculated that the number of smartphone users will total 1.75 billion in 2014 and further increase to 2.5 billion in 2017. The operation of a smartphone requires flick inputs to create an email sentence in Japanese, and the entry of the required characters is difficult and will need to be considered in the field of neurorehabilitation in the future. However, there are no studies measuring cerebral hemodynamics related to motor skill learning that address the underlying complex cognitive processing in tasks such as character entry into a touch-screen terminal. Therefore, in this study, we examined cerebral hemodynamics using NIRS associated with acquisition patterns of motor-cognitive skills in healthy subjects using character entry into a touch-screen terminal. Based on previous studies, we hypothesized that the behavioral improvements resulting from adaptation would be accompanied by distinct patterns of activation over the various cortical regions. # Methods ## Subjects Twenty healthy subjects (9 men and 11 women; mean age, 27.5±5.5 years) participated in this study. All subjects were self-reported as right-handed. Exclusionary criteria included any medical illness affecting central nervous system function, psychiatric or neurological disorders, history of head trauma, or current substance abuse. None of the subjects had previous experience with character entry using a touch-screen terminal. Written informed consent was obtained from each subject. The study was approved by the local ethics committee of Nagasaki University Graduate School of Biomedical and Health Sciences. All of the experimental procedures were conducted in accordance with the Declaration of Helsinki. ## Tasks and procedures The subjects sat on a chair 80 cm away from a PC monitor (Epson LD1957S, Japan, 19-inch, resolution: 1024 × 768 pixels). All subjects performed the task with their right thumb for 15 s alternating with 25 s of rest for 30 repetitions (cycles 1 to 30). Gains in motor skills were evaluated according to the number of characters entered into the touch-screen terminal (Apple iPod Touch 4, Japan, 3.5-inch, resolution: 960×640 pixels). All subjects were asked to enter the characters of a randomly formed Japanese syllabary presented on a PC monitor, starting with the character from the upper left, into the touch-screen terminal as quickly as possible. The character string in the monitor was constructed in five lines and nine rows, resulting in 45 characters. The characters of the randomly formed Japanese syllabary were changed every cycle. All subjects were asked to fixate on a single point at the center of the screen during rest and to stay relaxed. In English, the letter combinations “P Q R S” and “W X Y Z” involve a three-way operation in the upper, right, and left direction, whereas the input otherwise involves operation in two ways. In Japanese, the combinations “ya yu yo” and “wa wo n” are two-way operations in the right and left directions, whereas the input otherwise operates in four ways in the upper, lower, right, and left directions. Therefore, operation of these devices using Japanese input is more complicated than that using English input. The task of the present study was not simply a matter of entering as many characters as possible; the only characters that were meant to be entered were those displayed. There was thus a trade-off between the number of characters and the number of incorrect entries. Thus, the task performance of each subject was calculated by subtracting the number of errors from the number of correct answers in each cycle. NIRS measurements were performed using a continuous wave system (ETG-4000, Hitachi Medical Corp., Tokyo, Japan) equipped with 4 × 4 optode probe sets (8 incident light and 8 detector fibers), resulting in a total of 24 channels with an inter-optode distance of 3.0 cm. The continuous-wave NIRS system utilizes two different wavelengths (\~625 and 830 nm). Relative changes in the absorption of near-infrared light were sampled at 10 Hz, and these measures were converted into related concentration changes for oxygenated hemoglobin (oxy-Hb) and deoxygenated hemoglobin (deoxy-Hb), based on the modified Beer-Lambert approach. The moving average method (window: 5 s) was used to exclude short-term motion artifacts in the analyzed data. The obtained data were analyzed in the integral mode, which calculates average waveform. Pre-task baseline was defined as the 5-s period immediately prior to task onset. In this study, we used changes in oxy-Hb concentration as an indicator of changes in regional cerebral blood volume, as an earlier NIRS signal study using a perfused rat brain model proposed that oxy-Hb, rather than deoxy-Hb, is the more sensitive parameter for the study of activation. The NIRS channels were placed according to the international 10–20 system. Regarding the positions of the optodes, we followed previous NIRS studies of motor-related areas. The optodes were positioned using a custom-made cap that covered the right and left PFC, presupplementary motor area (preSMA), supplementary motor area (SMA), dorsal premotor cortex (PMC), and sensorimotor cortex (SMC). The areas and optodes were as follows: left SMC, channels 18 and 22; right SMC, channels 21 and 24; motor area, channels 19, 20, and 23; SMA, channels 9, 12, 13, and 16; preSMA, channels 2, 5, and 6; left PMC, channels 8, 11, and 15; right PMC, channels, 10, 14, and 17; left PFC, channels 1 and 4; and right PFC, channels 3 and 7. The Cz position in the international 10–20 system was used as a marker for ensuring replicable placement of the optodes. # Data analysis Changes in performance across task cycles were analyzed by calculating Spearman’s rank correlation coefficients; serial changes (1–30 cycles) in the level of oxy-Hb associated with cycle repetition in the various regions were also evaluated using Spearman’s rank correlation coefficients. These correlation coefficients tested for associations between the number of task cycles and changes in oxy-Hb level for each region. # Results shows the average performance (i.e., number of correct entries minus incorrect entries into the touch-screen terminal) during the touch-screen task over 30 cycles. There was a highly significant positive correlation between task cycle and performance (ρ = 0.924, p \< 0.001). shows the mean changes in oxy-Hb concentration for three cortical regions over the 30 cycles of the touch-screen task. A significant positive correlation between task cycle and oxy-Hb concentration for the channels covering the area of the left SMC (ρ = 0.387, p \< 0.05) was observed. In contrast, significant negative correlations between task cycle and oxy-Hb concentration for the channels covering the SMA (ρ = -0.513, p \< 0.01) and right PFC (ρ = -0.364, p \< 0.05) were obtained. There were no significant correlations between task cycle and oxy-Hb changes for the channels covering the other areas (left PFC, preSMA, bilateral PMC, and right SMC). # Discussion Flick input is one of the entry methods for Japanese characters in a touch- screen terminal. However, in general, flick input operation is difficult. The user of the touch-screen terminal has to learn to input a letter quickly to effectively convey information. Because the subjects of this study were inexperienced in touch-screen terminal operation, this task was equivalent to motor learning combined with complex cognitive processing. In recent years, activities involving complex cognitive processing during operation of touch- screen terminals have become abundant in everyday life. However, the cerebral blood flow dynamics during motor learning with complex cognitive processing in humans are not well understood. Therefore, we used NIRS to examine the cerebral blood flow dynamics of various cortical regions during motor learning of flick input operation of a touch-screen terminal. ## Role of SMC in motor learning The left SMC was activated with an increasing number of task cycles, whereas the right SMC did not show such an activation pattern. The left SMC is equivalent to a region including primary motor and primary sensory areas primarily controlling operation of the hand. PET and NIRS studies have reported that the cerebral blood flow volume of the contralateral primary motor area of the operating hand increased with the frequency of finger tapping. In addition, the primary motor and primary sensory areas contralateral to the operating hand became activated during motor learning of the finger. In the present study, the number of character entries significantly increased with the task cycle. During the course of one experiment, motor learning of the finger occurred in all subjects, as confirmed by the improvement of the flick input operation. Therefore, we speculate that increases in the momentum of the finger reflect motor learning. ## Role of SMA in motor learning The SMA extracts motor programs that depend on memory information and the initiation of spontaneous movements. Thus, the SMA plays an important role in situations where a compound movement is controlled. The examination of SMA activity during motor learning can be accomplished by various neuroimaging techniques. However, there is little consensus on the activity of the SMA during motor learning. In a previous NIRS study, SMA activation gradually increased with motor learning. That study used a pursuit rotor task requiring simple motor learning and therefore differs from the motor learning task of the current study that involved complex cognitive processing. Although the SMA showed greatly increased activity immediately after experiment initiation, its activity was gradually attenuated with increasing task cycles. We suggest that these changes in cerebral blood flow were caused by motor inhibition. All subjects of this study owned a feature phone and inputted characters using ten keys that operated in a toggle manner. During the experiment, all subjects were required to input letters using a method that differed from the operation they had mastered previously. We speculate that the two divergent character input methods competed, leading to motor inhibition that occurred in all subjects during this task. Preliminary research has suggested that the SMA is activated by motor inhibition. We further suggest that the SMA is activated initially by motor inhibition; as learning progresses, SMA activity is attenuated. Although our findings demonstrate activation of SMA by motor inhibition, we could not find similar reports that examined changes of cerebral blood flow showing motor inhibition with cognitive learning over time. Therefore, this study may be the first to report cerebral blood flow dynamics in SMA with cognitive learning involving motor inhibition. However, the task required complex cognitive processes including visuomotor adaptation and working memory. Therefore, the current task was not specifically designed to test motor inhibition. In future experiments, we thus need to examine the dynamics of cerebral blood flow over the SMA during more specific tasks requiring motor inhibition. This may be the first study to report the activity of the SMA during a motor learning task with complex cognitive processing using character entry into a touch-screen terminal over time. Therefore, further research is needed to examine the role of SMA during motor learning using other complex tasks (e.g., character input on a personal computer, trail making, serial reaction time, or other visuomotor adaptation and motor sequence learning tasks that involve complex cognitive processing). ## Role of the PFC in motor learning The PFC is an important neural substrate for visual working memory. As the task of this study consisted of entering a letter presented on a PC screen into a hand-held touch-screen terminal, working memory was required to successfully accomplish this task. Memory encoding has been associated with lateral PFC activation across a variety of experimental paradigms in functional neuroimaging. In the current study, the right PFC activation was gradually attenuated with increasing task cycle. However, activation of the left PFC continued without attenuation as the task cycles progressed. Activity of PFC regions has been reported to gradually decrease with learning. Therefore, acquisition of the cognitive skill associated with working memory in our subjects may have attenuated the activity of the right PFC. As all subjects in our study were right-handed, we assume that their left hemisphere was language- dominant. Thus, the left PFC was likely involved in linguistic information manipulation and continued to be recruited during all task cycles, thereby maintaining the activity of the left PFC. Previous studies have reported PFC asymmetry for memory encoding of verbal and nonverbal code. The right PFC is activated during a nonverbal task, whereas the left PFC is activated during a verbal task, supporting the validity of our hypothesis. ## Role of other regions The PMC is involved in choice and control of movements that depend on sensory information. During the 1–30 cycles, the PMC continued to be activated bilaterally. As the task of this study consisted of a visuomotor problem, we speculate that the activity of PMC was maintained because the subjects depended on visual information throughout all task cycles. The preSMA controls the aspect of the task that uses the rich entry from the premotor area, including order decision of the complex movements during preparation and movement choice; this also includes motor learning of changes in movement style and timing of the movement initiation. Furthermore, the preSMA is activated by cognitive control in the absence of movement control. The activity of the preSMA was previously reported to be attenuated with motor learning. However, in this study, the activity of preSMA continued during all task cycles. The subjects performed a movement control problem with complex cognitive information processing by translating the letter presented on a monitor to the movement direction of their finger. We suggest that this step represents early stages of cognitive-motor learning. In the current task, sustained attentional engagement from the subjects was required during cognitive processing, and we speculate that this led to the continued activity of preSMA without attenuation. ## Limitations of this study First, as our measurements included only the initial learning period of 30 cycles, it was not possible to determine how cerebral blood flow dynamics might have changed after completion of the 30-cycle learning stage. Therefore, further studies are needed to investigate cerebral blood flow dynamics after task training. Second, we used the number of characters entered minus the number of incorrect entries as the performance index in this study, and investigated its relationship with Oxy-Hb concentration changes in each region in this study. However, it is also possible that the Oxy-Hb concentration changes for each region would show a different course if the speed of entering the characters and reaction time (i.e. the time to enter a character appearing on a monitor upon confirmation) were investigated separately as a performance index. Further studies should be performed to address this issue. Third, activation in deeper structures such as the basal ganglia, which are closely linked with the frontal cortices, cannot be detected because of the technical limitations of NIRS. We thus could not address functional connectivity between brain areas, but it would be interesting to investigate functional connectivity between the right PFC and SMA during complex motor learning tasks. In future research, it will be necessary to consider other neuroimaging techniques that measure cerebral blood flow dynamics in humans during motor learning of flick input operation of a touch-screen terminal. # Conclusions We examined cerebral blood flow dynamics during motor learning of flick input into a touch-screen terminal using NIRS. The number of character inputs significantly increased with repetition of task cycles. These results show that motor learning occurred in all subjects during the course of one experiment. In the left SMC, SMA, and right PFC, there was a significant change of cerebral blood flow dynamics as the task cycles progressed, indicating motor learning over time. Changes in activity over the left SMC, SMA, and right PFC likely reflect distinct aspects of acquisition of the motor task such as increase in finger momentum, motor inhibition, and visual working memory, respectively. # Supporting Information The authors would like to thank all the participants in this study. Takehito Yonezawa, Kengo Fujiwara, and Akira Nakashima contributed significantly with their flexible support of the participants during data acquisition. We thank Hitachi Medical Corp., Japan for their skilled technical support whenever needed. Moreover, we would particularly like to thank Takumi Inakazu for his continuous support during the data acquisition and processing phase. We would like to thank Editage ([www.editage.jp](http://www.editage.jp/)) for English language editing. NIRS near-infrared spectroscopy PFC prefrontal cortex PFC left dorsolateral prefrontal cortex preSMA presupplementary motor area SMA supplementary motor area PMC dorsal premotor cortex SMC sensorimotor cortex [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AS TT KO TH. Performed the experiments: AS NI TM TH EK. Analyzed the data: AS NI KT. Wrote the paper: AS TH TT.

# Introduction The story of language acquisition is, in large part, the story of how children move beyond simply repeating words and phrases that they have learned from their caregivers (e.g., *Bye+bye*; *Drink*!) and acquire the ability to produce utterances that they have never encountered in exactly that form; a point perhaps most famously made in Chomsky’s review of Skinner’s *Verbal Behavior*. At the heart of this ability lies what many researchers have referred to as a paradox (,,,) In order to achieve adult-like productivity with language, children must set up generalizations that allow them to use verbs in unattested constructions. For example, on the basis of hearing pairs of utterances like *The ball rolled* and *The man rolled the ball*, the child might set up a rule that allows any verb attested in the *intransitive inchoative* construction to be generalized into the *transitive causative* construction, and vice versa. Thus, upon encountering an utterance like *The vase broke*, the child could use her generalization to produce *Daddy broke the vase*, even if she has never encountered *break* in the transitive causative construction. The paradox arises because children must learn that, of the verbs that are as yet unattested in the target construction, many can be generalized into that construction if the communicative need arises (e.g., *The car moved → Mummy moved the car*), while others cannot (e.g., *The baby giggled* → *\*Daddy giggled the baby*). The problem of how children learn to avoid these overgeneralizations of verb argument structure is sometimes referred to as the problem of the *retreat from overgeneralization*. Indeed, findings from several naturalistic and experimental studies suggest that at least some children pass through a stage in which they produce such errors before subsequently retreating from them (eg,,, ) Some examples of these errors are given in. However, it is important to note that the paradox of partial productivity applies whether or not a particular child happens to produce errors from which to “retreat”. Thus, while children certainly appear to make some use of corrective feedback from adults (eg,), this is unlikely to be a complete solution. The present study investigates two solutions to this problem It is important to also acknowledge the existence of a third, and potentially complementary, proposal: Pinker’s semantic verb class hypothesis, which holds that learners form classes of verbs that are semantically (in)consistent with particular constructions For example, verbs of semi-voluntary emotional expression may appear in the intransitive and periphrastic causative constructions (eg, *The baby laughed/giggled; Daddy made the baby laugh/giggle*), but not the transitive causative (eg, *\*Daddy laughed/giggled the baby*) The present study does not investigate this hypothesis, and we will say no more about it here, other than to note that an effect of verb semantics—though not necessarily of discrete semantic classes per se—has been observed in a number of previous studies (eg, ,,,,) The two proposals that are investigated in the present study are both forms of statistical learning The first is the *preemption* hypothesis: “ If a potential innovative expression would be precisely synonymous with a well-established expression, the innovation is normally pre-empted by the well established term, and is therefore considered ungrammatical (: 798) ” Preemption was first proposed to account for the retreat from word-level errors involving derivational morphology. For example, children’s novel coinages such as *\*spyer*, *\*driller* and *\*unappear* are gradually preempted by the adult forms *spy*, *drill* and *disappear* (perhaps even semi-explicitly, as the child “notices” a mismatch between the adult form and her own). Because preemption is a probabilistic process, the prediction that follows from this proposal is that the greater the frequency of the competitor form, the less likely children will be to produce the error, and the more unacceptable they will rate this error in a judgment task. Indeed, recent studies of overgeneralizations of derivational morphology have provided support for this claim (see, for errors of verbal *un-*prefixation—e.g., *\*unclose*—and for errors of adjectival *a-*prefixation—e.g., *\*The asleep boy*). While *preemption* works well for word-level morphological overgeneralizations, attempts to apply the proposal to utterance-level syntactic overgeneralizations (eg,)—the focus of the present study—have met with mixed success. Again, the prediction is of a negative correlation between the production probability/rated acceptability of a particular error (e.g., *\*Daddy laughed the baby*; a transitive causative overgeneralization) and the frequency of this verb in the single most nearly synonymous construction (e.g., *Daddy made the baby laugh*; the periphrastic causative construction). Similarly, overgeneralizations into the intransitive construction (e.g., *The book lost*) are held to be probabilistically preempted by passive uses (e.g., *The book was lost*); See for more examples. Although novel verb laboratory studies of the intransitive and transitive constructions have provided support for this prediction (,) there is reason to doubt that the relevant preempting constructions are sufficiently frequent in the input to which children are exposed. Indeed, on the basis of the counts obtained for the present study, periphrastic causative and passive uses are extremely rare in child directed speech (0.4% and 2.2%). Neither does preemption seem likely to constitute a plausible mechanism by which children could avoid errors with the passive construction (e.g., *\*An hour was lasted by the film; A kilogram was weighed by the box; Marge was looked like/resembled by Lisa*). These verbs are not particularly frequent in the potentially-preempting active construction (e.g., *The film lasted an hour*) and indeed, are presumably less frequent in this construction than many verbs that passivize easily (e.g., *Homer was pushed by Marge*). Potentially more straightforward are the three- argument constructions: datives and locatives, for which—in each case—the two alternate syntactic forms seem to express almost identical meanings (see), and hence are good candidates for preemption. Indeed, a grammaticality judgment study of errors involving the DO-dative construction (e.g., *\*I said her no*) observed a clear effect of preemption (e.g., *I said no to her*), even after controlling for another statistical predictor; entrenchment (discussed below) On the other hand, a similar study with errors involving the locative constructions (e.g., *\*The boy poured the cup with water; \*The boy filled water into the cup*) found an effect of preempting alternatives (e.g., *The boy poured water into the cup; The boy filled the cup with water*), but one that disappeared after controlling for entrenchment Thus the status of preemption as a construction-general solution to the paradox of partial productivity is unclear. While it seems to work well for morphological overgeneralizations and those involving the dative constructions, this is not necessarily the case for other argument structure constructions. In the present study we investigate whether preemption operates across a range of argument structure constructions, using statistical techniques that allow us to generalize beyond the particular constructions tested: mixed effects models (,) A more recent statistical learning proposal—and one that perhaps does not share the intuitive appeal of preemption—is *entrenchment*. The general idea goes back to Langacker who posits a continuous scale of entrenchment in cognitive organization. Every use of a structure has a positive impact on its degree of entrenchment. Units are variably entrenched depending on the frequency of their occurrence. When applied to the domain of the retreat from overgeneralization (eg,), the idea is that overgeneralization errors with a particular verb (e.g., *\*Daddy laughed the baby*) are probabilistically blocked by *any* use of the relevant verb (e.g., *The baby laughed*), and not solely—as for preemption—by uses in a nearly-synonymous construction (e.g., *Daddy made the baby laugh*). In other words, the entrenchment of a verb *in any number of constructions* probabilistically blocks its generalization into constructions in which it has not been attested. In intuitive terms, one can imagine the learning mechanism making a kind of inference from absence (e.g., “Given how frequently I’ve encountered laugh, then if this verb *could* appear in the transitive causative construction, I would surely have heard it by now”). However, entrenchment need not necessarily be framed in terms of deductive reasoning. For example, connectionist networks can show entrenchment type behaviour (eg,) simply because increasing the strength of the connection between an input node representing *laugh* and output nodes representing other constructions (e.g., the intransitive, the periphrastic causative, the single-word imperative) necessarily reduces the strength of the connection between *laugh* and the output node representing the transitive causative construction. Entrenchment enjoys an important advantage over preemption: Because erroneous uses are probabilistically blocked by *any* use of the relevant verb, regardless of construction, it does not rely on learners encountering particular verbs in very low frequency constructions (e.g., the periphrastic causative and passive). Indeed, for all verb argument structure overgeneralizations, preempting evidence is always, by definition, a subset of entrenching evidence (and often a very small one). In support of the entrenchment hypothesis, many studies have observed the predicted negative correlation between the acceptability or production probability of errors and overall verb frequency (,,,,) In general, these studies have revealed an effect of entrenchment even after controlling for preemption, though a production study of *un-*prefixation observed the opposite pattern, perhaps because—as noted above—preemption is particularly powerful in the case of word-level overgeneralizations of derivational morphology (e.g., *open* preempting \**unclose*). In summary, previous findings and theoretical considerations suggest that entrenchment may hold more promise than preemption as a construction general solution to the problem of the retreat from overgeneralization error. At present, however, this conclusion remains tentative for two reasons. The first is that, since measures of entrenchment and preemption (e.g., overall verb frequency and frequency in a particular construction) are invariably highly correlated, the two mechanisms are difficult to distinguish empirically. Indeed, many of the studies cited above did not even attempt to do so. Those that did used regression techniques to partial out the effect of each individual predictor on the dependent variable. However, this solution produces unreliable results when the correlation between the two predictors is very high (e.g., *r* = 0.7 in, 2011; *r* = 0.9 in). In the present study, we address this problem by running separate statistical analyses for each predictor. The second is that, with a single exception, each of these studies has focused on a single construction pair (or “alternation”). This is an important shortcoming, since we already know that—for example—preemption works well for *at least some* types of errors (e.g., morphological overgeneralizations); the issue at stake is the ability of these two statistical learning mechanisms to provide a general solution to the retreat from error across a range of different error types. More generally, as Herb Clark (co-originator of the preemption account) pointed out in a famous paper, one cannot simply assume that one’s “findings generalize beyond the specific sample of language materials…chosen” (: 335). Although Clark focuses mainly on single words, he explicitly notes that the need to demonstrate generalizability beyond the specific items used in the study holds for “words, sentences and other language materials” (p.335), presumably including abstract syntactic constructions. This clearly cannot be done in a study that includes only a single construction pair In the present study we address this problem by using mixed effects models with crossed random effects for participants and items, where “items” includes both verbs and syntactic constructions. This strategy allows us to infer that any observed effect of preemption or entrenchment generalizes beyond the particular verbs and constructions included in the present study; clearly a prerequisite for any satisfactory solution to the partial-productivity paradox. In summary, the present study tested the preemption and entrenchment hypotheses across a range of verb argument structure constructions. The former predicts a negative correlation between the rated acceptability of a particular overgeneralization error and corpus frequency of the relevant verb *in the single most nearly synonymous construction* (see for examples). The latter predicts a negative correlation between the rated acceptability of a particular overgeneralization error and the overall frequency of the relevant verb. # Method ## Ethics statement The study was approved by the University of Liverpool Ethics Committee. Informed written consent was obtained from adult participants and from parents of participating children (who also gave verbal consent). ## Participants Participants were 72 children aged 5;2–6;8 (*M* = 5;10), 72 children aged 9;2–10;6 (*M* = 9;11) and 72 adults aged 18;1–22;2 (*M* = 19;1), all reported as showing typical language development. Participants were recruited from schools and a university in the North West of England. The study was approved by the University of Liverpool ethics committee, and informed consent was obtained from all participants. Oral consent was obtained from children, written consent from parents, and from adult participants. ## Verbs and Sentences Since the design of the study is rather complex, it is probably best understood by consulting the relevant table. The description below is intended to outline the logic of the design set out in the table, rather than to constitute a complete free-standing explanation of the design in its own right. The study used nine different sentence-level verb argument structure constructions grouped into four alternations: Dative (PO-Dative, DO-dative), Causative (Intransitive inchoative, Transitive causative, Periphrastic causative), Locative (Figure locative, Ground locative) and Passive (Active, Passive) Morphological construction and overgeneralizations (eg, *un-*VERB; *\*unclose*) were not included because the relationship between entrenchment and preemption is different for these error types, in that the relevant preempting form generally has a different lexical root (eg, *open* for \**unclose*) Thus while our conclusions generalize across different types of overgeneralization *of verb argument structure*, they do not generalize across *all* different types of overgeneralization error Indeed, there is some evidence that preemption may be more important for overgeneralization errors at the morphological level (,) For each construction (e.g., PO-dative) we selected (a) four verbs that are grammatical in that construction but not the other construction in the alternation (e.g., *carry*, *haul*, *scream*, *shriek*), (b) four verbs that show the opposite profile (e.g., *cost*, *fine*, *refuse*, *deny*) and (c) four verbs that may appear in both constructions (e.g., *give*, *hand*, *show*, *teach*). As these examples indicate, each set of four verbs comprised two higher-frequency verbs and two lower-frequency near synonyms (e.g., *carry+haul*, *scream+shriek*). We then created a sentence for each construction+verb combination. The sentence for each alternation pair and each high/low frequency synonym pair used the same noun phrases (e.g., *Bart carried/hauled the box to Lisa; \*Bart carried/hauled Lisa the box*). For each of these sentence quadruples, we created, using Anime Studio Pro 5.5, a single cartoon animation for which all four sentences would constitute an appropriate description (e.g., Bart lifting a heavy-looking box, carrying it to Lisa and placing it at her feet). The main purpose of the animations was to maintain children’s interest, but they also served to illustrate the intended meaning of the accompanying sentence, and to demonstrate that its veracity was not in doubt (only its grammaticality). In fact, the design was not quite as balanced as this description implies, due to (a) the inclusion of periphrastic causatives in the causative alternation (b) the non-existence of passive-only verbs (c) the unavailability of close synonyms for non-passivizable verbs and (d) the fact that the active construction in the passive alternation is the same construction as the transitive construction in the causative alternation (although as a non-causative transitive, it is arguably not *exactly* the same, depending on whether or not one posits multiple transitive constructions; see Ambridge &, Lieven, in press, for discussion). Nevertheless, as shows, it was still possible to devise sentence stimuli that are consistent with the overall design. ## Rating scale and Procedure The dependent variable was the acceptability rating for each sentence on a five- point “smiley face” scale The expressions on the faces ranged from sad (leftmost) to neutral (middle) to happy (rightmost). The two leftmost faces were red, the two rightmost faces green and the middle face split with the left-hand half red and the right-hand half green. The scale can be downloaded from <http://journalsplosorg/plosone/article/figure/ image?size=large,id=info:doi/101371/journalpone0110009g002> Children indicated their judgments by selecting a red counter (for ungrammatical) or a green counter (for grammatical) and placing it on the relevant face to provide a graded judgment, with responses noted down by the experimenter. Children were told that the red and green counters could be placed on only the red and green faces respectively, except that either counter could be placed on the middle face. Adults marked their ratings directly on the face scale. The procedure was the same as that used in previous judgment studies of verb argument structure overgeneralization errors (for a more detailed description, see,) In brief, children first complete a training session in which they are told that a talking dog (a toy with an internal loudspeaker connected to a laptop computer) is learning to speak English but “because he’s only a dog, sometimes gets it wrong and says things a bit silly”. The child’s task is to help the dog by telling him whether he “said it right, or a bit silly”. The training procedure consists of seven warm-up sentences; the first two completed by the experimenter, the remainder by the child: *The cat drank the milk* (intended rating 5/5), \**The dog the ball played with* (1/5), *The frog caught the fly* (5/5), *\*His teeth man the brushed* (1/5), *\*The woman said the man a funny story* (2/5), \**The girl telephoned her friend the news* (3/5) and *The man whispered his friend the joke* (4/5). Note that the final three warm-up sentences are examples of PO→DO dative overgeneralization errors. Although it would have been ideal to avoid using any of the same types of overgeneralization error as in the study proper, this was unavoidable, given the importance of providing children with practice at rating verb argument structure overgeneralization errors. Nevertheless, the warm-up sentences did not use any of the same verbs as test sentences. After completing the warm-up, children moved on to the main part of the study, which they completed in two sessions on different (usually consecutive) days. Because the total number of trials (*N* = 100) was felt to be too many for young children, each child completed only half of the total number (i.e., 50): One high-low frequency sentence pair for each cell of the design, selected at random on a child-by-child basis (i.e., for any given row in, any given child completed either the two sentences in the column “Sentence Pair A” or the two sentences in the column “Sentence Pair B”, but never both). Children completed the trials in pseudo-random order with the constraint that neither (a) the same verb (or its high/low frequency equivalent) nor (b) the same construction could occur on consecutive trials. ## Predictor variables As outlined in the introduction, the *preemption* measure was operationalized as the log frequency of each verb in the single mostly nearly synonymous construction. Entrenchment was operationalized as the log frequency of all uses of that verb (excluding uses as a noun). Frequency counts were taken from SUBTLEX-UK, a 200 million word corpus of subtitles from programmes shown on British television, which has been shown empirically (e.g., via lexical decision tasks) to be more representative of the language heard by speakers of British English than either (a) the British National Corpus (its only serious rival in terms of size) or (b) the equivalent US subtitle corpus. In order to generate these measures, we obtained counts of each verb in each of our target constructions. (In fact, since for each target construction, only one other construction was designated the preempting construction, such a level of detail was not necessary for the analysis. The aim in obtaining such detailed counts was to create a publically available resource for use in future studies). Since the SUBTLEX-UK corpus is tagged, but not parsed, these counts had to be obtained largely by hand. First we used a program (custom written by the final author) to (a) count the number of uses of each verb and (b) to extract a random sample of 100 sentences of each (or, for verbs with fewer than 100 occurrences, the full set). Two raters then classified each sentence as (a) an instance of one of the constructions shown in (PO-dative, DO-dative, Intransitive, Transitive, Periphrastic Causative, Figure-Locative, Ground-Locative, Passive), (b) an “Other” verb use or (c) a non-verb uses (in which case the sentence was replaced with another, and the counts pro-rated accordingly). Missing arguments were allowed, provided that they could be inferred on the basis of the ongoing discourse, and construction classifications were not mutually exclusive. Thus, for example, an utterance such as “John gave a card” would be classified as both a Transitive and a PO-dative. This decision was taken partly on theoretical grounds (i.e., children presumably can and do recover missing arguments from discourse) and partly on practical grounds: insisting that all arguments be explicitly realized would generally have resulted in counts of close to zero for all three-argument constructions (PO/DO-dative, Figure/Ground-locative). Two coders (Amy Bidgood and Katherine Twomey) each classified 50% of the dataset, and reliability-checked 10% of the data coded by the other. Inter-rater reliability was 87% (Cohen’s Kappa = 0.79, *z* = 16.4, *p*\<0.001). Disagreements were resolved by discussion. Raw verb-in-construction counts were pro-rated, on the basis of the overall number of verb uses, in order to yield a final estimate of the frequency of each verb in each construction in the corpus. All raw data are available in. # Results As the predictions of the entrenchment and preemption hypotheses relate only to ungrammatical sentences, grammatical sentences were excluded from all analyses (though they play an important role as fillers and encourage use of the full scale). The data were analysed using linear mixed-effects models () in *R*, with random intercepts for (a) Participant and (b) Verb (*N* = 32; non-alternating verbs only) nested within Sentence Type (i.e., construction: PO-Dative, DO- Dative, Intransitive, Transitive, Periphrastic causative, Figure locative, Ground locative, Passive) In accordance with the recommendations of a recent methods paper, all models included by-participant random slopes, always correlated with the intercept, and by-Sentence Type/Verb random slopes, correlated with the intercept, except for a few cases where this yielded convergence failure. Random slopes for Age Group and its interactions were also excluded for this reason. Depending on the analysis, the fixed effect was either the Preemption or the Entrenchment predictor, with some models also including Age Group (5–6, 9–10, Adult) and the relevant interactions. For example, for the first analysis, the model (in R syntax) was $$\begin{array}{l} {Model1 = lmer(Rating\tilde{}\ AgeGroup*Preemption\ + \ \left( 1 + Preemption \middle| Participant \right)\ +} \\ {\left( 1 + Preemption \middle| SentenceType/Verb \right),\ data = UngrammaticalSentences)} \\ \end{array}$$ Note that because the nesting structure is rather unusual—transitive-only verbs were rated in both intransitive and periphrastic causative sentences, whereas all other verb types were rated in one sentence type only—it was necessary to specify this structure directly in the syntax. *P* values were obtained via the *t* distribution (from the *lmer* function of *lme4*), but we also confirmed that *p* values obtained using a backwards-elimination model-comparison procedure (performed automatically using the *step* function from the *lmerTest* package, eliminating fixed effects only) yielded an identical pattern of results. Indeed, in most cases the *p* values were identical to at least two decimal places (and hence we do not report them separately). It is also important to note that the present analysis tests the entrenchment and preemption hypotheses across different sentence constructions (i.e., treating construction as a random effect), but does not look for entrenchment and preemption effects across verbs within any given construction. Given that, for each particular construction, only four verbs—and hence four sentences—are ungrammatical, such an analysis would be seriously underpowered, and almost guaranteed to yield Type II errors (i.e., to fail to detect any effect present). Such an analysis strategy is not at all unusual. For example, consider a hypothetical drug trial in which 32 human participants are split across 8 treatment centers, with four participants per center. (analogous to the present situation of 32 verbs nested across 8 constructions). Mixed effects modeling (with treatment center as a random effect) could tell us that the drug is effective, and that this effect generalizes across the 8 treatment centers, but could not tell us whether or not the treatment given in any one center alone was effective. In the same way, the present study can tell us whether entrenchment and preemption effects are observed, and generalize across constructions, but not whether they hold for any particular construction individually. ## Preemption The first analysis was conducted on the combined data for all participants, and hence, in addition to the Preemption predictor, included as fixed effects Age Group and its interactions (with Adult as the reference category). This analysis revealed a main effect of age, such that both 5–6 and 9–10 year olds rated the overgeneralized ungrammatical sentences as more acceptable than did adults. However, the preemption predictor was not associated with any main effects or interactions (*t* \<1 in all cases). The null effect for the preemption predictor (collapsing across all age groups) is plotted in ; it is clear that the line is almost flat. Despite the lack of a significant interaction of Age Group by Preemption, it seemed important to verify that no individual age group showed any suggestion of a preemption effect, by running a separate model for each. These models revealed no effect of preemption for any group (*t*\<1 in all cases). Thus, in summary, the present study failed to find any evidence for preemption either for all participants combined, or for any age group individually. ## Entrenchment An equivalent set of analyses for the entrenchment predictor revealed an interaction, such that 5–6 year olds, but not 9–10 year olds, showed a significantly smaller entrenchment effect than did adults. Indeed, the follow up models revealed that a significant entrenchment effect in the predicted (negative) direction was displayed by the 9–10 year olds (*B* = -0.13, *SE* = 0.06, *t*\[28.79\] = -2.28, *p* = 0.03) and adults (*B* = -0.16, *SE* = 0.04, *t*\[20.31\] = -3.65, *p* = 0.002), but not the 5–6 year olds (*t*\<1). Presumably the null finding for the 5–6 year olds is the cause of the narrow failure of the main effect of entrenchment to reach significance (*p* = 0.10) in the all-participants analysis. Plots of the entrenchment predictor for each age group separately (Figs –) show that as overall verb frequency increases, so the rated acceptability of errors decreases (significantly so for the two older groups). Thus, in summary, the present study found an effect for entrenchment for 9–10 year olds and adults, but not 5–6 year olds. # Discussion The present study investigated the central question of how children retreat from—or, in many cases, avoid altogether—errors of verb argument structure overgeneralization (e.g., *\*Daddy giggled the baby*). The study was motivated by previous empirical findings and theoretical considerations which raise doubts regarding the feasibility of preemption as a general solution to the problem of the retreat from argument structure overgeneralization, and suggest that entrenchment may constitute a more promising approach. To investigate this possibility, participants aged 5–6, 9–10 and 18–22 rated the acceptability of overgeneralization errors with a variety of different constructions. No support was found for the prediction of the preemption hypothesis that the greater the frequency of the verb in the single most nearly synonymous construction (for this example, the periphrastic causative; e.g., *Daddy made the baby giggle*), the lower the acceptability of the error. Support was found, however, for the prediction of the entrenchment hypothesis that the greater the frequency of the verb in all constructions (e.g., *The baby laughed*), the lower the acceptability of the error, at least for 9–10 year olds and adults. Although previous studies have investigated the preemption and entrenchment hypotheses, the present study was unique in using statistical models with crossed random effects for participants and items (both verb and construction) in order to investigate the generalizability of these effects. The conclusion, then, is that entrenchment appears to be a robust effect that generalizes across different types of verb argument structure overgeneralization error. (see also) We did not, however, find any evidence that this is the case for preemption. Before considering this null effect in more detail, it is important to reemphasize that the present study investigated only sentence-level overgeneralizations of verb argument structure. It remains possible, even likely, that preemption is the major retreat mechanism for other types of overgeneralization such as morphological overgeneralizations at the lexical level (e.g., *whisk* and *open* preempt *\*whisker* and *\*unclose*). Indeed, as we saw in the introduction, the preemption account was initially proposed with exactly these types of errors in mind It is the subsequent extension of this account to overgeneralizations of verb argument structure that the findings of the present study call into question. Returning to the present findings, although it is always difficult to draw conclusions on the basis of a null effect, the lack of a preemption effect does not seem to be straightforwardly attributable to flaws in our experimental design. For example, the study does not seem to be underpowered with regard to the number of participants (72 at each age). Indeed, given that the regression line is almost flat, it does not seem likely that adding even a large number of participants would change the outcome. Another potential objection is that we failed to select a set of verbs with a sufficient spread along the dimension defined by the preemption predictor (i.e., that there is too little variance in this measure to enable it to predict variance in participants’ judgments). Certainly it is true that, by definition, the spread is smaller for the preemption than the entrenchment predictor. Nevertheless, inspection of reveals that the preemption predictor shows a relatively good spread; and, again, the finding that the predictor did not even approach significance suggests that a preemption effect could not easily be obtained simply by adding more items. A related objection is that the absolute frequency of the verbs in the relevant preempting constructions was too low for a preemption effect to be observed. However, this is exactly the point: While preemption may work in experimental novel verb studies in which participants are trained on a very large number of exemplars, the present study suggests that, for many familiar verbs, even adults may never encounter sufficient preempting evidence: occurrence in very low frequency constructions such as the passive and periphrastic causative. (Of course, learners hear enough exemplars of these constructions to eventually acquire them, but this does not necessarily mean that they hear each and every relevant verb used in one of these constructions). In summary, although it is wise to avoid drawing firm conclusions on the basis of a single experimental result, particularly when it is a null effect, the present study at least raises doubt regarding the feasibility of preemption as a general solution to the retreat from sentence-level overgeneralizations of verb argument structure. This raises the question of how to explain the finding that preemption *does* seem to work for certain verb argument structure constructions, particularly the locative constructions (though it is important to remember that the design of the present study did not allow for the investigation of pre-emption or entrenchment effects for any particular construction individually). What, in fact, does it mean to have a learning mechanism that works for some constructions, but not others? Why would children use preemption in only a particular subset of cases to which it would seem to apply? The answer, we suggest, is that it is a mistake to posit a sharp distinction between preemption and entrenchment, and perhaps even to posit preemption and entrenchment as *mechanisms* rather than *effects* at all. Consider an account under which several different constructions (e.g., active transitive, passive transitive, intransitive, periphrastic causative) compete for the right to express the speaker’s message (eg, DADDY CAUSE \[BABY LAUGH\]; *\*Daddy laughed the baby; \*The baby was laughed by Daddy; The baby laughed; Daddy made the baby laugh*), perhaps even in real time as the sentence is produced word-by-word Assume that the activation of each competitor construction is determined, at least in part, by the frequency with which the verb that the speaker intends to use has occurred in each. An “entrenchment” effect would fall naturally out of this competition process, with no need for any kind of semi-explicit inference from absence. Now, one *could* draw a circle around a particular set of verb uses (for this example, periphrastic causative uses), and label them as “preemption”, but this would seem to add little to the explanation: Both “entrenchment” and “preemption” are just labels for particular *effects* that are outcomes of the construction competition process, rather than *mechanisms*. Why then—according to previous studies—are preemption effects observed for some constructions but not others? Under the account that we have outlined above, preemption is simply a special subtype of entrenchment: A preemption effect (as opposed to solely entrenchment) will be observed when the “preempting” construction is (a) particularly frequent relative to the error construction and (b) particularly closely synonymous with the error (see for evidence that a connectionist model that implements these factors can yield entrenchment and preemption effects in this way). Presumably these conditions are met for overgeneralizations of PO-dative-only verbs into the DO-dative construction (e.g., *I said no to her* preempts *I said her no*), and indeed for morphological overgeneralization errors (e.g., *ran*, *mice* and *open* preempt \**runned*; \*mouses and *\*unclose*). Presumably, they are not met, however, for a sufficient number of the constructions used in the present study for a significant construction-general preemption effect to be observed. Essentially the same argument can be made coming from the opposite direction; by positing that entrenchment is a special extension of preemption Suppose that the notion of preemption is broadened such that errors with a particular verb (eg, *\*Daddy laughed the baby*) are probabilistically blocked not only by uses of that verb in the *single* most nearly synonymous construction (eg, *Daddy made the baby laugh*) but by *every* construction that meets some minimum threshold for near synonymy (eg, *The baby laughed*) Under this proposal, preemption and entrenchment are again collapsed into a single construction-competition process If the account that we have outlined here is along the right lines, then the aim of future research should be not so much to disentangle “preemption” and “entrenchment”, but rather to investigate the factors that determine the outcome of this putative construction-competition process. These factors might include the (a) frequency of the relevant verb and the relevant construction (independently and in co-occurrence), (b) the extent to which the construction is relevant to the speaker’s intended message (and includes a slot for every argument that the speaker intends to express) and (c)—a factor that we have not discussed here—the fit between the verb and the verb slot of the relevant construction in terms of semantics, pragmatics, phonology and any other properties exemplified by this slot (see,,,,) In conclusion, whether or not the account set out above is along the right lines, the present study has provided some preliminary evidence against the longstanding claim that preemption is the key mechanism in the retreat from verb argument structure overgeneralization error. Instead, our findings suggest the need for a learning mechanism that is sensitive to overall verb frequency, regardless of construction (whether or not this is framed as “entrenchment”). Although a number of previous studies have investigated both preemption and entrenchment, the novel and particularly important contribution of the present study is its demonstration that only the latter (or whatever takes its place) appears to be robust across a range of different argument structure overgeneralizations. We hope, therefore, that this study will inspire other researchers not only to conduct further experimental investigations into the factors that are important in the retreat from overgeneralization, but also to accept Clark’s challenge of demonstrating that the effects observed—and hence the mechanisms proposed—generalize across different types of errors, and hence hold the promise of a general solution to the paradox of partial productivity. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BA AB JMP CFR. Performed the experiments: AB KT. Analyzed the data: BA AB KT. Contributed reagents/materials/analysis tools: DF. Wrote the paper: BA AB KT JMP CFR DF.

# Introduction Cigarette smoke, including active smoking and passive smoking, has been implicated in many diseases, disability and death. Cigarette smoke consists of more than 7300 chemical constituents, many of which are potent carcinogens and tumor promoters. A number of specific infections have been associated closely with cigarette smoke, including community-acquired pneumonia, tuberculosis, Helicobacter pylori infections, inflammatory bowel disease, invasive fungal infections, periodontitis, and oral candidiasis. Even though cigarette smoke directly mediates upregulation of bacterial virulence, the pro-infective effects of cigarette smoke are believed to result primarily from interference with host defense. Innate immunity constitutes the first line of defense against microbe infection. As two main classes of innate immune receptors, the Toll-like receptors (TLR) and NOD-like receptors (NLR) serve as pattern recognition receptors that recognize conserved structures of pathogens, toxic compounds, or cellular damage known as “danger signals.” Depending on the adapter Receptor-interacting protein 2 (RIP2), NOD induces NF-κB activation and nuclear translocation. NF-κB activation promotes the production of proinflammatory cytokines, chemokines, and antimicrobial peptides. The human defensins, one group of small cationic antimicrobial peptides include the α-defensins of intestinal and neutrophil origin, and the β-defensins of skin, oral mucosa and other epithelia. The human β defensins (hBDs) play important roles in innate immune and adaptive immune, such as antimicrobial activity, antitumor effect, chemoattractive effect and immunomodulation. hBD1, 2, and 3 represent the main group of human defensins expressed and secreted by oral mucosal epithelial cells and have been most investigated. So far the best characterized proteins of NLR members are nucleotide-binding oligomerization domain-containing protein 1 (NOD1) and NOD2. As one of intracellular pattern recognition receptors (PRRs), NOD1 plays a pivotal role in pathogen microbe clearance and tissue homeostasis of oral cavity, gastrointestinal, and respiratory tract. Sugawara et al. indicated that NOD1 and NOD2 in oral epithelial cells were functional receptors that induced antibacterial responses. Cigarette smoke directly activates epithelial cells and induces chemokine and inflammatory mediator release. Nevertheless, epithelia-mediated innate immune responses to infectious pathogens are compromised by cigarette smoke. Although many studies have established that cigarette smoke exposure affects the expression of TLRs, study data about the effects of cigarette smoke exposure on NLRs remain scarce,. Aldhous et al. have determined that Cigarette Smoke Extract (CSE) delays NOD2 expression and affects NOD2/RIP2 interactions in intestinal epithelial cells. However, it remains unknown whether NOD1 expression is affected by cigarette smoke exposure. Caspases are cysteinyl aspartate-specific proteases that play a pivotal role not only in the induction of apoptotic cell death but also in the inflammatory responses against microbial infection. Caspases are divided into three functional groups: apoptosis induction (Caspase-2, -3, -6, -7, -8, -9, and -10), inflammatory responses (Caspase-1, -4, -5, -11, and -12) and differentiation (Caspase-14). Caspase-1 is activated in the inflammasome, an intracellular protein complex that is formed by the recognition of intracellular ligands or cellular stresses by sensor molecules such as NOD-like receptors. Caspase-1 activation can induce the production of mature IL-1β/IL-18 and trigger pyroptosis. Under certain conditions, Caspase-11 is required for the activation of the caspase-1 inflammasome, referred to as the noncanonical inflammasome. In addition, Caspase-8 also contributes to the production of inflammatory cytokines. Specially, only Caspase-12 can dampen the responses to bacterial infection and inhibit IL-1β, IL-18, and IFN-γ production. It had been confirmed that Caspase-12-deficiency not only enhanced bacterial clearance and sepsis resistance, but also augmented the production of antimicrobial peptides, cytokines, and chemokines to some pathogens. Previous studies have determined that cigarette smoke exposure or some components in cigarette smoke could up- regulate the expression of Caspase-12, while Caspase-12 could negatively modulate NOD1 signaling. Based on these established evidences, we hypothesized that cigarette smoke may also have direct effect on NOD1 signaling and the production of antimicrobial peptides of human oral mucosal epithelial cells by up-regulating the expression of Caspase-12. The first goal of this study was thus to investigate whether CSE affected the expression of crucial molecules in NOD1 signaling pathway, including NOD1, RIP2, NF-κB and hBD1, 2, 3 in human oral mucosa epithelial cells. Our second focus was to verify the potentially inhibitory effect of Caspase-12 on NOD1 signaling and hBD1, 2, 3 in these cells following CSE exposure. # Materials and Methods ## Reagents Keratinocyte Serum-Free Medium (K-SFM) for Culture of Human Keratinocytes (Keratinocyte-SFM) was purchased from GIBCO (Invitrogen, Carlsbad, CA, USA), phosphatase inhibitor cocktail purchased from Roche (Mannheim, Germany), protease inhibitor cocktail was purchased from Fermentas UAB (Vilnius, Lithuania) and protein assay reagent and an enhanced chemiluminescent (ECL) kit were purchased from Pierce (Rockford, IL, USA). The following antibodies were used: rabbit anti-NOD1 antibody, rabbit anti-Caspase-12 antibody, mouse anti- RIP2 antibody, rabbit anti-p-NF-κB (p-p65) antibody, rabbit anti-NF-κB (p65) antibody, mouse anti-hBD1 antibody, and rabbit anti- hBD2 antibody were purchased from Abcam (Cambridge, UK); rabbit anti-hBD3 antibody was purchased from Novus (Littleton, CO, USA); rabbit anti-GAPDH antibody was purchased from Cell Signaling (Danvers, MA, USA). Dylight Fluor 488-labeled goat-anti-mouse secondary antibody and Alexa Fluor 555-labeled goat-anti-rabbit secondary antibody were purchased from Jackson ImmunoResearch (West Grove, PA, USA). Caspase-12 small interfering RNA (siRNA) and scrambled siRNA were purchased from GenePharma (Shanghai, China). Lipofectamine 2000 and Opti-MEM I medium was purchased from Invitrogen (Carlsbad, CA, USA). RevertAid First Strand cDNA synthesis kit was purchased from Fermentas UAB (Vilnius, Lithuania), and SYBR Green PCR Master Mix was purchased from Roche (Mannheim, Germany). The hBD1, 2, 3 ELISA Kits were purchased from Jingtian (Shanghai, China). ## Preparation of aqueous cigarette smoke extracts (CSE) Aqueous CSE was prepared as previously described. Kentucky 3R4F research- reference filtered cigarettes (The Tobacco Research Institute, University of Kentucky, Lexington, KY), one of which contains 0.73 mg of nicotine and 9.4 mg of tar, were used for CSE preparation. A cigarette was smoked continuously by a peristaltic pump. Four cigarettes were bubbled through 40 ml of cell growth medium, and this solution, regarded as 100% strength CSE. The generated CSE solution was filtered (0.22 mm) to remove bacteria and large particles which was adjusted to a pH of 7.45 and used within 15 min after preparation. The content of nicotine in CSE was analyzed in the institutional laboratory using liquid chromatography-tandem mass spectrometry as previously described. The 100% CSE contained 239±45 µg/ml of nicotine in three separate samples. Working dilutions of CSE (in the range of 0.5% to 8%) were made with culture medium expressed as a percentage (v/v %). ## Cell culture Immortalized human oral mucosal epithelial (Leuk-1) cell line was a generous gift from Professor Li Mao at Department of Oncology and Diagnostic Sciences, University of Maryland Dental School, Baltimore, MD. The Leuk1 cell line was established from a dysplastic leukoplakia lesion adjacent to a squamous cell carcinoma of the tongue. It exhibits an immortalized but nontumorigenic phenotype. The cell line was expanded and passaged in keratinocyte serum free medium. This medium was supplemented with BPE (25 µg/ml), epidermal growth factor (0.2 ng/ml), CaCl₂ (0.4 mM). The passaged cells were cultured in 37°C humidified air incubators with 5% CO₂ which were stimulated with CSE of different concentration (0.5%, 1%, 2%, 4%, and 8%) in certain experiments. ## Western immunoblot analysis Western blotting was performed as described. Leuk-1 cells were washed twice with PBS and harvested by trypsinization. Cells were lysed in ice-cold lysis buffer containing 1% Nonidet P-40, 0.5% deoxycholate, 0.1% SDS, protease inhibitor cocktail, and phosphatase inhibitor cocktail. The lysates were incubated on ice for 30 min and centrifuged at 14,000 g, for 10 min, at 4°C, to remove cell debris. Total cellular protein was collected and protein concentration was measured. Next, 10 to 50 µg of total cell protein was separated by 10% SDS-PAGE gels and transferred from gels to polyvinylidene difluoride membranes (Millipore, Bedford, MA) by wet electroblotting. Membranes were blocked for 1 h at room temperature with 5% bovine serum albumin (BSA) in PBS-0.1% Tween 20 (PBST). After washing with PBST for three times, membranes were incubated with the primary antibodies (NOD1, Caspase-12, RIP2, p-NF-κB, and GAPDH antibodies were diluted 1:1,000 with PBST containing 5% BSA) overnight at 4°C. The next day, membranes were washed with PBST followed by 1 h incubation at room temperature with horseradish peroxidase-conjugated secondary antibodies (5,000-fold diluted with PBST containing 5% BSA). Following washing with PBST, immunostained protein bands were detected by using an enhanced chemiluminescence (ECL) assay kit and were visualized on FluorChem FC2 system (Cell Biosciences, Santa Clara, CA). Densitometric analyses of bands were performed using Image J software and the data of the target protein were normalized to those of the corresponding GAPDH (<http://rsb.info.nih.gov/ij/>). Data were normalized to GAPDH and expressed as the percentage or fold change compared with the corresponding control, which was set to 1 or 100. ## RNA extraction and real time quantitative reverse transcription PCR (qRT-PCR) qRT-PCR assay was performed as described previously. Total RNA was extracted from Leuk-1 cells by using TRIzol reagent (Invitrogen) according to the manufacturer's instructions, and 2μg of RNA was used to synthesize first-strand cDNA synthesis in 20 μl of reaction volume using RevertAid First Strand cDNA Synthesis Kit (Roche) according to the manufacturer's protocol. The primers used for the PCR amplifications are listed as follows: hBD1 forward TCA TTA CAA TTG CGT CAG CAG, reverse TTG CAG CAC TTG GCC TTC ; hBD2 forward TCC TCT TCT CGT TCC TCT TCA, reverse AGG GCA AAA GAC TGG ATG AC ; hBD3 forward CCA TTA TCT TCT GTT TGC TTT GCT C, reverse CCG CCT CTG ACT CTG CAA TAA TA ; Caspase-12 forward AAT GGA ATC TGT GGG ACC AA, reverse GAA CCA AAC AAT CCC AGC AC ; GAPDH forward TCA AGA AGG TGG TGA AGC AG, reverse CCC TGT TGC TGT AGC CAA AT. Real-time PCR analyses ware performed using an ABI 7300 Real Time PCR System (Applied Biosystem, Foster City, CA), and PCR amplifications were performed using the SYBR Green PCR Master Mix (Roche) according to the manufacturer's instructions. Amplification conditions were as follows: 50°C for 2 min, 95°C for 10 min, 40 cycles of 95°C for 15 s, 58°C for 30 s, and 72°C for 30 s, followed by melting curve analysis, by which the specificity of primers was confirmed. The experiment was repeated three times. The data are expressed as relative mRNA levels and were normalized to GAPDH. Fold changes in expression of each gene were calculated by a comparative threshold cycle (Ct) method using the formula 2^−(ΔΔCt). ## Immunofluorescence, confocal microscopy and densitometry image analysis Immunostaining was performed as described previously. Briefly, Leuk-1 cells were collected and pipetted onto coverslips, which had been put in six-well culture plate in advance. After overnight incubation, Leuk-1 cells were washed with PBS and fixed in 4% paraformaldehyde for 15 min at room temperature. After being washed in PBS, the cells were permeabilized in 0.5% (v/v) Triton X-100 in PBS, washed, and blocked with 5% BSA in PBS-0.1% Tween 20 for 1 h at 37°C. Next, the cells were exposed overnight at 4°C to primary antibodies. Primary antibodies against the following proteins were used: NOD1 (1:100), Caspase-12 (1:200), RIP2 (1:50), phospho- NF-êB p65 (1:100), NF-êB p65 (1:100), hBD1 (1:100), hBD2 (1:100), and hBD3 (1:100). The next day, coverslips were washed with PBS and then incubated with Dylight 488 (green) or Alexa Fluor 555 (red) - labeled goat- anti-mouse or goat-anti-rabbit secondary antibody for 1 h at room temperature. To stain the nuclei, 4′, 6-diamidino-2-phenylindole (DAPI, Sigma) was added for 5 min, and slides were examined by a confocal laser scanning microscope (FluoView FV10i, Olympus, Japan). Densitometry image analysis was performed as previously reported with some modifications. Five randomly selected discontinuous fields per slice were evaluated. The densitometry analysis of immunofluorescence results was performed by one blinded investigator using the Image J software. Briefly, the software was used to achieve the “gray image” and measure the optical density of the selected pixels within the region of interest (ROI). The calibration procedure was finished before image analysis. The mean value of the optical densities of all selected pixels was Mean Optical Density (MOD), which represented the corresponding fluorescence intensity of immunofluorescence staining. ## Enzyme-linked immunosorbent assay (ELISA) To detect the amount of hBD1, 2, 3 produced by Leuk-1 cells, ELISA kits were used to measure the levels of hBD1, 2, 3 in cell culture supernatant. hBD1, 2, 3 standard was used to construct standard curves. The experiments were performed according to the manufacturer's recommendations. Absorbances were read at 450 nm and 570 nm using a microplate reader, the absorbance at 570 nm was subtracted from the absorbance at 450 nm. ## Preparation of siRNA and cell transfection Caspase-12 silencing was obtained through transfecting FAM fluorescence-labeled siRNA, which was used to determine the transfection efficiency. Cell transfection was carried out by using siRNA as previously described. For Caspase-12 siRNA transfection, cells grew in 60 mm-diameter dishes and transfection was performed when cells were 50%∼70% confluent. For each dish, 100 pmol of Caspase-12 siRNA and 5 µl of Lipofectamine 2000 were diluted in 250 µl of Opti-MEM I reduced serum medium. Caspase-12 siRNA and Lipofectamine 2000 dilutions were then mixed and incubated at room temperature for 20 min before being added to each well containing cells and K-SFM (final concentration of Caspase-12 siRNA was 40 nM). The cells were incubated at 37°C in a CO₂ incubator for 24 h, transfection medium was replaced by complete K-SFM, and the incubation continued for an additional 24 h before the addition of CSE. Controls were treated with scrambled siRNA and Lipofectamine 2000. The sequences of siRNAs are as follows : Caspase-12 siRNA-1 sense 5′- GCA GUU AUA CAC GAG AUC ATT -3′ and antisense 5′- UGA UCU CGU GUA UAA CUG CTT -3′, Caspase-12 siRNA-2 sense 5′- GGC UCU UGC AAG GUA ACA UTT -3′ and antisense 5′- AUG UUA CCU UGC AAG AGC CTT -3′, scrambled siRNA sense 5′- UUC UCC GAA CGU GUC ACG UTT -3′ and antisense 5′- ACG UGA CAC GUU CGG AGA ATT -3′. ## Antimicrobial assay of culture supernatants of Caspase-12-silenced cells and control cells following CSE treatment The antimicrobial assay of culture supernatants was performed as described previously. *Candida albicans* strain ATCC 10231 was purchased from China Center of Industrial Culture Collection (CICC). *C. albicans* was incubated in yeast extract-peptone-dextrose (YPD) liquid media at 37°C overnight. Following various concentration of CSE treatment for 24 h, each of the culture supernatants of Caspase-12-silenced cells or control cells was then harvested, centrifuged and filtered. One hundred microliters each of the culture supernatants was added to 100 ml each of the *C. albicans* suspensions and incubated under 5% CO₂ at 37°C for 1 h. After serial dilutions (until 10⁴-fold), each *C. albicans* suspension was then applied to YPD agar plates, incubated at 37°C for 24 h and subjected to colony counting. The antimicrobial activity of each culture supernatant was measured three times. ## Statistical analyses Statistical analyses were performed using SPSS 15.0 (Chicago, IL). Values are expressed as the mean ± SE. Differences between the groups were analyzed via an unpaired *t*-test, and ANOVA was used to compare the differences between the different concentrations within the same treatment group. Two-tailed probability values of \<0.05 were considered statistically significant. Error bars on images represent SE. # Results ## CSE altered NOD1, Caspase-12, RIP2, and p-NF-κB expression in Leuk-1 cells The first goal of this study was to investigate whether CSE affected the expression of crucial molecules of NOD1 signaling pathway in human oral mucosa epithelial cells. Leuk-1 cells were treated with various concentrations of CSE for 24 h. As shown by Western blotting results, the protein levels of NOD1 decreased with increasing concentration of CSE, while the protein levels of RIP2 elevated with increasing concentration of CSE. Relatively low concentrations of CSE treatment resulted in elevated p-NF-κB levels when comparing with the control group. The protein expression of p-NF-κB reached the highest level in Leuk-1 cells following 1% CSE treatment. However, p-NF-κB levels significantly decreased when exposed to higher concentrations of CSE. Caspase-12 expression was significantly increased in Leuk-1 cells following 1%∼8% CSE treatment, respectively. Consistent with Western blotting results, immunofluorescence assays revealed that CSE down-regulated NOD1 protein levels and up-regulated RIP2 protein levels both in a dose-dependent manner. Relatively low concentrations of CSE elevated p-NF-κB levels, while relatively high concentrations of CSE reduced p-NF-κB levels. The p-NF-κB protein level reached the peak level following 1% CSE treatment. CSE activated Caspase-12 expression in Leuk-1 cells. As shown by confocal microscopy result, marked nuclear translocation of NF-κB p65 subunit was observed in Leuk-1 cells after 24 h following 1% CSE exposure. ## CSE regulated the expression and release of hBD1, 2, 3 by Leuk-1 cells To clarify the effects of CSE on hBDs expression in human oral epithelial cells, qRT-PCR and immunofluorescence were performed to detect hBDs expression at mRNA and protein levels respectively. As shown by qRT-PCR results, relatively low concentrations of CSE treatment resulted in elevated hBD1 levels when comparing with control group. Interestingly, hBD1 mRNA level was up to the highest following the treatment of 1% CSE. However, hBD1 mRNA level significantly decreased when exposed to relatively higher concentrations of CSE. The real time qPCR results revealed that hBD2 mRNA levels up-regulated following CSE exposure. As shown by qRT-PCR results, 0.5% and 4% CSE treatment significantly down- regulated hBD3 mRNA level. To further study the effects of CSE on hBDs releases from human oral epithelial cells, ELISA assays were performed to detect hBDs levels in the supernatant of Leuk-1 cells culture following CSE exposure. As showed, 0.5% CSE treatment up- regulated hBD1 release, while CSE treatment of higher concentrations down- regulated hBD1 releases. On the contrary, CSE increased hBD2 releases in a dose- dependent manner. The hBD3 releases down-regulated following CSE treatment of 0.5%, 2%, 4% and 8% concentrations. Consistent with qRT-PCR results, immunofluorescence assays indicated that hBD1 expression reached the highest level following the treatment of 1% CSE and down- regulated following relatively higher concentrations of CSE treatment. The Immunofluorescence assays revealed that hBD2 mRNA and protein levels up- regulated following CSE exposure. However, immunofluorescence staining results revealed CSE treatment completely down-regulated hBD3 protein level. ## The transfection of Caspase-12 siRNA caused Caspase-12 silencing at mRNA and protein levels To further examine the relationship between NOD1 signaling and Caspase-12, Caspase-12 was silenced by RNA interference. We selected two siRNA sequences from the region of Caspase-12 for evaluation. Non-silencing control siRNAs were synthesized using scrambled sequences as a negative control (NC) to assay the interference efficiency of Caspase-12 siRNA. These FAM labelled siRNAs were transfected into Leuk-1 cells. After 24 h, confocal microscopy showed that the transfection efficiency was ∼95%. FAM-labelled siRNAs indicated subcellular localization of cytoplasm, especially perinuclear area. As shown in, in the Western blot assay, compared with the NC group, the normalized protein level of Caspase-12 reduced remarkably in Caspase-12 siRNA-1 and -2 groups. Especially Caspase-12 siRNA-2 had greater interference efficiency than caspase-12 siRNA-1. As shown in, Caspase-12 levels in cells transfected with Caspase-12 siRNA-2 (40 nM) were ∼3% of that of NC group. Real-time PCR showed that the transfection of Caspase-12 siRNA-2 caused clear Caspase-12 silencing at mRNA level. Therefore Caspase-12 siRNA-2 was chosen in the following experiments. ## Caspase-12 silencing increased NOD1 and p-NF-κB expression and down-regulated RIP2 expression RIP2 is a downstream component of Caspase-12 and NOD1 signaling is regulated by Caspase-12 in murine intestinal epithelial cells. To determine whether NOD1 signaling was regulated by Caspase-12 in human oral epithelial cells, Leuk-1 cells were transfected with Caspase-12 siRNA-2 (40 nM). Immunoblotting demonstrated that NOD1 and p-NF-κB levels significantly increased in Caspase-12-silenced cells compared with that in NC group. Other than NOD1 and p-NF-κB, Caspase-12 silencing remarkably reduced RIP2 level. ## Caspase-12 silencing did not significantly alter the expression of hBD1, 2 and 3 at mRNA and protein levels Since Caspase-12 silencing increased NOD1 and p-NF-κB levels in Leuk-1 cells, we investigated whether Caspase-12 impacted hBD1, 2 and 3 at mRNA and protein levels. Firstly, we examined the relative alterations of hBD1, 2 and 3 at mRNA levels due to Caspase-12 silencing. As shown in, and, no statistically significant change was observed in hBD1, 2 and 3 mRNA levels due to Caspase-12 silencing. We then analyzed whether Caspase-12 impacted hBD1, 2 and 3 at protein levels by immunofluorescence staining. Consistent with qRT-PCR results, Caspase-12 silencing also did not lead to significant change of hBD1, 2 and 3 protein expression compared with that in control cells. ## Caspase-12 silencing partially attenuated inhibitory effects of CSE on NOD1 and p-NF-κB expression Caspase-12 has been reported to negatively regulate NOD1 signaling in enterocytes. To confirm whether Caspase-12 inhibits NOD1 signaling following CSE exposure, we examined the impact of Caspase-12 silencing on NOD1 signaling pathway in Leuk-1 cells following CSE exposure. As immunoblotting analysis shown, NOD1 protein levels significantly increased in Caspase-12-silenced cells compared with that in controls following 4% CSE treatment. Caspase-12 silencing markedly increased RIP2 levels following 2% and 8% CSE treatment. Like NOD1, p-NF-κB level also greatly increased in Caspase-12-silenced cells compared with that in controls following 4% CSE treatment. These results confirmed that Caspase-12 silencing partially attenuated the inhibitory effect of CSE on NOD1 signaling pathway to a certain extent. Coincidentally, immunofluorescence results indicated that NOD1 protein levels significantly increased in Caspase-12-silenced cells compared with that in controls following 4% CSE treatment. Caspase-12 silencing significantly increased RIP2 levels following relatively high concentrations of CSE treatment. Immunofluorescence results indicated that p-NF-κB protein level significantly increased in Caspase-12-silenced cells compared with that in controls following 4% CSE treatment ## Caspase-12 silencing abrogated the suppression of hBD1, 3 expression by CSE and augmented the induction of hBD2 expression by CSE In the next stage, we examined the impact of Caspase-12 silencing on CSE- stimulated hBD1, 2, 3 mRNA expressions in Leuk-1 cells. As shown in , hBD1 mRNA levels increased and reached the highest level following 1% CSE treatment in control cells and then decreased following higher concentrations of CSE treatment. Caspase-12-silenced cells expressed 1,000-fold hBD1 mRNAs than the controls following 1% CSE treatment. Strikingly, the hBD1 mRNA levels following 8% CSE treatment reached the peak level in Caspase-12-silenced cells, which expressed more than 8,000-fold hBD1 mRNAs than the controls. As shown in, hBD2 mRNA levels was up-regulated by CSE and reached the highest level following 8% CSE treatment in control cells. The hBD2 mRNA levels increased and reached the highest level following 1% CSE treatment in Caspase-12-silenced cells, which expressed more than 1,000-fold hBD2 mRNAs than the controls. The hBD2 mRNA levels following 8% CSE treatment increased in Caspase-12-silenced cells, which expressed nearly 10-fold hBD2 mRNAs than the controls. As shown in, hBD3 mRNA levels decreased following 0.5% CSE treatment and increased following 4% CSE treatment and then decreased following 8% CSE treatment in control cells. The hBD3 mRNA levels following 1% CSE treatment reached the peak level in Caspase-12-silenced cells, which expressed about 100-fold hBD3 mRNAs than the controls. Caspase-12-silenced cells following 8% CSE treatment also reached the peak level and expressed about 300-fold hBD3 mRNAs than the controls. Since Caspase-12 silencing significantly increased hBD1, 2, 3 mRNA levels in Leuk-1 cells following CSE exposure, we investigated whether Caspase-12 silencing impacted hBD1, 2, 3 protein levels following CSE exposure by immunofluorescence staining. Consistent with qRT-PCR results, the image analysis of immunofluorescence results revealed that hBD1 protein expression increased and reached the highest level following 1% CSE treatment in control cells and then decreased following relatively higher concentrations of CSE treatment. To the contrary, Caspase-12-silenced cells expressed significantly higher levels of hBD1 than the controls following 1%, 2%, 4%, and 8% CSE treatment. As shown in, CSE up-regulated hBD2 protein levels in control cells, while hBD2 protein levels following 2%, 4%, and 8% CSE treatment were remarkably higher in Caspase-12-silenced cells than the control cells. As shown in, hBD3 protein levels in control cells decreased clearly following CSE exposure, while Caspase-12-silenced cells expressed slightly higher hBD3 levels than the controls following CSE exposure. However, the difference was not statistically significant. ## Caspase-12 silencing enhanced the antimicrobial activity of culture supernatants of CSE-exposed Leuk-1 cells As shown in, the culture supernatant of CSE-exposed control cells could not inhibit *C. albicans* colonies formation. The colonies number of *C. albicans* clearly increased in control groups following the treatment with the culture supernatant of control cells exposed to 2%, 4% and 8% CSE. Interestingly, the culture supernatant of CSE-exposed Caspase-12-silenced cells exhibited significantly higher antimicrobial activity to *C. albicans* than that of control cells. The colonies number of *C. albicans* following the treatment with the culture supernatant of Caspase-12-silenced cells exposed to 2%, 4% and 8% CSE was remarkably fewer than that in the corresponding NC group. These results indicated that Caspase-12 silencing enhanced the inhibitory effect of culture supernatants of CSE-exposed Leuk-1 cells on *C. albicans*. # Discussion Many studies have demonstrated that cigarette smoke alters the expression of PRRs, especially for TLRs. However study data about the effects of cigarette smoke on NLRs remain very little. A recent study indicated that CSE delays NOD2 expression and affects NOD2/RIP2 interactions in intestinal epithelial cells. As fundamental members of NLR family, NOD1 and NOD2 have very similar structures. Our data indicated for the first time that CSE could inhibit NOD1 signal pathway in oral mucosal epithelial cells. There is a close relationship between cigarette smoke and many diseases. As is well known, smoking is one of the most important risk factors for periodontitis, second only to plaque. A recent study indicated that NOD1 is critical for commensal-induced periodontitis. Apart from periodontitis, cigarette smoke itself, or in combination with other factors, is a well recognized risk for oral candidiasis, oral leukoplakia and oral cancer. In oral cavity, oral mucosal epithelium is the first tissue that encounters cigarette smoke. Very little study data existed about the effects of cigarette smoke on innate immune of oral mucosal epithelial cells. Given the vital role of NOD1 in innate immune and tissue homeostasis, the inhibitory effect of CSE on NOD1 expression could result in reduced antibacterial peptide production and oral diseases occurrence. NOD1 may be a potential therapeutic target for some diseases in future. In the study, we found that CSE increased RIP2 expression in Leuk-1 cells. As an adapter, RIP2 acts a crucial role in NOD1-induced NF-κB activation. Moreover RIP2 also mediates cell apoptosis and autophagy according to previous studies. Recently, Wang et al. found that RIP1 expression remarkably increased in cigarette smoke-exposed mouse lung and significantly induced by CSE in human bronchial epithelial cells. It is well known that receptor-interacting protein family consists of several members RIP1-4, which play a crucial role in cell survival signaling. Based on existing evidences, increased expression of RIP2 may result from CSE-induced cell damage. Our early results indicated that low concentrations of CSE increased NF-κB expression in murine macrophages, while CSE of higher concentrations inhibited NF-κB activation. Consistent with the study data, our present findings further confirmed CSE regulated-NF-κB activation or suppression is dependent on CSE concentrations. One of possible explanations is that the exposure to relatively low concentrations of CSE may lead to the cellular stress response to toxic compound stimulation. Organisms have developed an elaborate system of defensive molecules and survival signaling pathways to counteract various toxic and environmental stresses. If the adaptive response is unable to counteract adverse exposure, cells will be eliminated by death processes such as apoptosis. The hBD is one of antimicrobial peptides that are expressed by the epithelia throughout the body including oral cavity. The expression of hBD1, 2, and 3 has been most investigated. These peptides are produced by oral epithelial cells and may control many commensal and pathogenic bacteria in oral cavity. hBD1 is constitutively expressed in epithelial cells and may be up-regulated by bacterial products. hBD2 is inductively expressed in epithelial cells and strongly up-regulated *in vitro* by commensal, pathogenic bacteria as well as proinflammatory cytokines. hBD3 is expressed in normal epithelium and is up- regulated by bacteria, IFN-γ and growth factors. Study data on effects of cigarette smoke on hBD1 expression are little. Wolgin et al. found that the expression of hBD1 and hBD2 mRNA significantly reduced in gingival samples of smokers compared to that in non-smokers. An early study indicated that mouse β defensin (mBD) 1 expression decreased in cigarette smoke- exposed mice, while the expression of mBD2 and mBD3 was greatly elevated in the lungs of cigarette smoke-exposed mice compared with air-exposed mice. Some studies indicated that CSE or whole cigarette smoke exposure modulates hBD2 and hBD3 mRNA by human gingival epithelial cells *in vitro*. The current result indicated that CSE could modulate hBD1, 2, and 3 expression levels in Leuk-1 cells. hBD1 expression in Leuk-1 cells was activated by relatively low concentrations of CSE and suppressed by relatively high concentrations of CSE. Otherwise, our results indicated CSE significantly increased hBD2 expression and inhibited hBD3 levels in Leuk-1 cells. In the present results, the difference between hBDs mRNA expression levels in Leuk-1 cells and hBDs protein levels in the supernatant was observed. On the one hand, the difference could be explained by the regulation of antimicrobial peptide expression at transcriptional, post- transcriptional and post-translational levels. On the other hand, this difference may originate from the modulation of hBDs at secretory level by epithelial cells themselves, which released distinct amount of hBDs to control defensive responses to varying extents. Many study results have confirmed that cigarette smoke or some components in cigarette smoke can activate Caspase-12 expression. In accordance to previous studies, our results suggested that CSE treatment could increase the expression of Caspase-12 in oral mucosal epithelial cells. Caspase12 is a crucial molecule associated with endoplasmic reticulum (ER) stress-induced apoptosis and inflammasome activation. LeBlanc et al. found that Caspase-12 modulates negatively NOD1 signaling in mouse colonic epithelial cells. Mechanistically, Caspase-12 binds to RIP2 and displaces Traf6 from NOD1 signaling complex. As a result, its ubiquitin ligase activity is inhibited and NF-kB activation is blunted. Supporting their findings, our results indicated that Caspase-12 silencing down-regulated RIP2 expression in Leuk-1 cells. Intriguingly, our results further showed that Caspase-12 silencing up-regulated the expression of NOD1 and NF-kB. These findings suggested clearly that Caspase-12 is a negative regulator of NOD1 signaling. Further studies are needed to investigate the complicated interacting mechanism between Caspase-12 and crucial molecules in NOD1 signal pathway. The present study in Leuk-1 cells showed that Caspase-12 silencing partially attenuated inhibitory effects of CSE on NOD1 and p-NF-kB protein expression. After 4% CSE treatment, both NOD1 and p-NF-κB levels significantly increased in Caspase-12-silenced cells compared with that in controls. Therefore, these results indicated that Caspase-12 could be involved in the inhibitory effect of CSE on NOD1 signaling in oral mucosal epithelial cells. In the present study we found for the first time that Caspase-12 silencing abrogated the suppression of hBD1, 3 expression levels by CSE and augmented the induction of hBD2 expression by CSE. Moreover our results indicated that Caspase-12 silencing enhanced the inhibitory effect of culture supernatants of CSE-exposed Leuk-1 cells on *C. albicans*, a common conditioned pathogen in oral cavity. LeBlanc et al. confirmed that Caspase-12-deficient enterocytes after infection hyper-produced antimicrobial peptides, specifically mBD1, a functional homolog of hBD1. LeBlanc and colleagues' results provide a clue that Caspase-12 regulates antimicrobial peptide production following infection stimulation. Coincidentally, our results suggested that Caspase-12 regulates antimicrobial peptide production following CSE stimulation. Antimicrobial peptide production may up-regulated in the absence of Caspase-12. Mechanistically, CSE activated intracellular Caspase-12, which negatively regulated NOD1 signaling by suppressing NOD1 and NF-κB expression and inducing RIP2 expression. As a part of downstream molecules of the signaling, hBDs production was subsequently inhibited and the defense response of human oral mucosal epithelial cells to pathogens was dampened. Saleh et al. confirmed that Caspase-12-deficient mice enhanced bacterial clearance and sepsis resistance. Caspase-12 is detrimental to *in vivo* handling of systemic bacterial infections and predisposes to sepsis, thereby making it a potentially important target for future therapeutic strategies. All together, CSE could suppress NOD1 signaling and modulate the downstream hBDs production in Leuk-1 cells. Caspase-12 silencing partially attenuated the inhibitory effects of CSE on NOD1 signaling and abrogated the suppression of CSE on hBDs expression. Caspase-12 silencing could enhance the antimicrobial activity of CSE-exposed cells. Caspase-12 may be a potential therapeutic target for some infectious and inflammatory diseases in future. # Acknowledgments We thank Professor Li Mao (School of dentistry, University of Maryland, USA) for his kind provision of oral mucosal epithelial (Leuk-1) cell line. We also thank Professor Wantao Chen (Department of Oral and Maxillofacial Surgery, Ninth People's Hospital, School of Stomatology, Shanghai Jiao Tong University School of Medicine, China) for his kind help to our experiment. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: WW XDH WZ XW. Performed the experiments: XW YQ QZ PY. Analyzed the data: XW YQ PY ND QZ. Contributed reagents/materials/analysis tools: XFH YZ JL LH WZ. Wrote the paper: XW XDH WW.

# Introduction Despite improvements in stroke mortality in Western societies, stroke remains the second leading cause of death worldwide. An embolus in the brain provides a well-recognized etiology for ischemic stroke together with decreased perfusion and thrombosis. It has been suggested that cardioembolism may constitute a major mechanism for not only cardiogenic but also cryptogenic stroke. When derived from the heart, an embolus may be induced by either chronic atrial fibrillation (AF) or paroxysmal atrial fibrillation (PAF) via predisposing clot formation and by reshaping of the left ventricle (LV), left atrium (LA), and left atrial appendage (LAA). Over 90% of all cardioemboli are formed in the LAA. The LAA has been shown to be frequently enlarged in patients with acute cryptogenic stroke or a transient ischemia attack (TIA). Both AF and PAF are known to contribute to LAA enlargement. However, the factors associated with LAA volume increase are not fully understood. For the risk stratification of stroke recurrence and accurate anticoagulation, or other treatment targeting, we need a more comprehensive understanding of stroke pathogenesis. In the current study we measured LAA volumes with cardiac CT (cCT) and adjusted the volumes for patient’s body surface area (BSA). Our aim was to identify the clinical factors and/or imaging findings that are associated with LAA volume increase. # Methods The EMBODETECT study and all participants were approved by the Kuopio University Area Research Ethics Board. Prior to participation, written informed consent was obtained from the patient or the patient’s legally authorized representative if the patient was unable to give consent due to impaired capacity caused by stroke or TIA. ## Study Design and Patients Between March 2005 and November 2009, our neurologists recruited 162 patients who had been admitted to Kuopio University Hospital with suspected cardioembolic stroke/TIA. Patients whose symptoms were not explained by a hemodynamically significant (\>50%) carotid/vertebral artery stenosis, or by AF diagnosed previously (or during hospitalization), were considered for the study. Thirteen patients were excluded due to technical errors or because they refused to participate having previously given informed consent. Altogether 149 patients (47 females; mean age 61 years; range 32–84 years) with suspected cardioembolic stroke/TIA underwent CT-angiography performed with ECG-synchronized mid-diastole cCT (16- or 64-slice, 120 kV, 190 mAs), enabling cardiac volumetric analyses. Transthoracic echocardiography (TTE) and 24-hour Holter ambulatory ECG were performed. The proportions of visceral (VAT), intrathoracic (IAT) and pericardial (PAT) adipose tissue were measured. ## Determinants for LAA Volume Study patients were dichotomized into subgroups according to their general characteristics and many other variables. These included the presence of defined (hypertension, dyslipidemia, carotid stenosis, and PAF) and potential (smoking, diabetes, prior ischemic heart disease, and mitral valve insufficiency) causal risk factors for ischemic stroke, cardiac echoparameters, cCT based volume measurements, and adipose tissue measurements. For dichotomization of mitral valve insufficiency, grade II mitral regurgitation was used as a threshold. Dimensions in echoparameters and ejection fraction were dichotomized by the guidelines of the American Society of Echocardiography (ASE). To derive upper thresholds for cCT based measurements, 243 consecutive patients who underwent coronary CT-angiography (coronary CTA) to exclude coronary stenosis were evaluated as candidates for healthy controls. Subjects with coronary stenosis of \>50%, AF, hypertension, renal insufficiency, prior stroke or malignancies were excluded, which left 40 patients available to create age- and gender-matched pairs with our stroke/TIA patients. The upper thresholds for normal variances in LAA and LA volume were set to two standard deviations above the mean of the control population. Upper thresholds for IAT and PAT were based on the same control population. The upper thresholds for VAT (\>122.8 cm² in females; \>188.4 cm² in males) were based on results from an ethnically comparable North American population (n = 1160). ## Volumetric Analysis of the LAA and LA Quantitative image analysis was performed on an IDS5 diagnostic workstation (version 10.2P4; Sectra Imtec, Linköping, Sweden) by an independent observer (MT), guided by an experienced cardioradiologist (PS). The volumetric analyses of LAA were performed three-dimensionally using the cCT stack. Planimetration of LAA covered 10.4±2.0 consecutive slices in the transversal plane. A two-chamber view and localizer tool were used to differentiate the LAA orifice from the LA. The LAA borders were traced manually on each transverse slice while the LA borders were traced from the consecutive slices on two-chamber plane using the mitral valve annulus as the landmark differentiating the LA from the LV. LAA volume was calculated with Simpson’s method by multiplying each manually traced LAA and LA area by the section thickness (3 mm) and summing up the volumes of the separate sections. Volumes of the LAA are presented with adjustment for BSA, which was calculated using Mosteller’s formula. In our previous study intra- class correlations (ICC) for the LAA volume measurements resulted in almost perfect reproducibility between different observers (ICC = 0.96) and the same observer (ICC = 0.95). ## Measurements of Pericardial, Intrathoracic and Visceral Adipose Tissue The PAT area was calculated by drawing a line through the parietal layer of the pericardial sac. The IAT area was calculated via the parietal layer of the pleural cavity from sternum to the anterior surface of the ascending aorta. The IAT included all PAT. Both measurements were performed from a single axial slice (slice thickness 7.5 mm) from the opening level of the common left coronary artery. The VAT area (slice thickness 10 mm) was calculated by drawing a line within the muscle wall that delineated the abdominal cavity at the level of the fourth lumbar vertebra. The adipose tissue surfaces were computed with an attenuation range from −30 to −190 Hounsfield Units. ## Statistical Analyses Continuous variables with normal distribution are presented as mean±SD, and categorical variables are presented as absolute values and percentages. Student’s *T*-test for individual samples was used to compare LAA volumes between dichotomized groups. Inclusion criteria for multifactoral linear regression analyses were set at *P*\<0.2, statistical significance at *P*\<0.05 and high statistical significance at *P*\<0.01. In multivariate analysis, tolerance\>0.2 was used to indicate non-multicolinearity of variables. By using the stepwise method, variables were entered if F\<0.05 and removed if F\>0.1. The change in R² and adjusted R² value were used to assess the contribution of each variable to the LAA volume. Data were analyzed using SPSS for Windows (version 19, 1989–2010 SPSS Inc., Chicago, USA). # Results shows the clinical characteristics and uncategorized values of echocardiography and CT measurements of the patients. Only BSA-adjusted values were used in regression analyses. Threshold values of LA volume, PAT and IAT area were based on the control population. In comparison between 40 age- and gender-matched stroke/TIA patients and control subjects, no significant difference was found in hyperlipidemia, diabetes, EF, LV dysfunction (no identified wall dyskinesia) or smoking. Body mass index (BMI) was 28.7±4.8 kg/m² in stroke/TIA patients and 25.3±4.1 kg/m² (*P* = 0.002) in control subjects. BSA was 2.0±0.2 m² in stroke/TIA patients and 1.8±0.2 m² (*P* = 0.033) in control subjects. All 18 threshold values are shown in. ## Correlates of LAA Volume in Linear Regression Analysis The possible correlates of LAA volume adjusted for BSA (mL/m²) are shown in. In univariate analyses, age, previously diagnosed hypertension, atrial fibrillation seen by 24-hour Holter ambulatory ECG, diabetes, mitral valve insufficiency, antero-posterior diameter of the LA, EF, LVSD and LVDD, LA volume and LV PAT were associated with LAA volume (*P*\<0.2) and were included in our multivariate linear regression analyses. No significant multi-colinearity (tolerance 0.473–0.892) was observed between the variables. To investigate independent predictors of LAA volume, stepwise linear regression analysis was performed. The best model accounted for 33% of the variability in LAA volume, whilst AF accounted for 19% (*P*\<0.001); enlarged LA volume for 7% (*P* = 0.011); enlarged LVSD for 4% (*P* = 0.007) and decreased PAT for 3% (*P* = 0.043) of the variability. When adjusted for the number of predictors, these variables accounted for 19%, 6%, 3% and 2% of the variability in LAA volume, respectively. The whole model accounted for 30% of the LAA volume variability when adjusted for the number of predictors. # Discussion An enlarged LAA constitutes an established risk factor for cardioembolic stroke. In a recent study, over half of all cryptogenic stroke/TIA patients had an enlarged LAA, indicating that cardioembolism may play a role in the mechanism of stroke in these patients. Chronic AF and PAF are generally considered causative factors for secondary LAA enlargement. To the best of our knowledge, no previous study has explained correlates of LAA volume increase in patients with stroke/TIA. We found that previously recognized risk factors for ischemic stroke/TIA combined with routine echoparameters and more novel CT parameters explain together only 30% of LAA volume increase. These factors proved to be AF, enlarged LA volume, enlarged LV systolic diameter and decreased pericardial fat. The prevalence of these risk factors could naturally explain LAA dilatation and furthermore the formation of thrombus in the LAA, which contribute to the pathogenic mechanism of cardioembolic stroke/TIA. In particular, PAF may easily remain unrecognized in routine clinical evaluation. Therefore, in patients with stroke/TIA, assessment of LAA size with cCT might help to identify patients who should be more carefully monitored for possible AF. Atrial fibrillation explained 19% of the LAA volume variation in the present study. It has been suggested that PAF plays a vital role in LAA volume increase even if not diagnosed with current methods. In our study, AF diagnosis was based on 24-hour Holter ambulatory ECG. It is likely that even more PAFs would have been recognized by using lengthened ECG monitoring such as 7-day Holter ambulatory recording or event recording. The volume increment of the LAA due to AF has been verified by MRI studies. Stroke/TIA patients with PAF in our study (9.37±3.40 mL/m² when adjusted for BSA corresponding to 18.74±6.80 mL when non-adjusted) proved to have even larger LAA volumes compared to patients with PAF (13.0±6.1 mL) or chronic AF (14.3±6.2 mL; 15.04±07.1 mL; 17.3±6.7 mL) in those previously published MRI studies,. Second to AF, LA dilatation had an impact on LAA size, explaining 6% of the volume increase. This proportion remains surprisingly small despite the anatomical connection of the structures, indicating that the function and factors influencing their size differ. Indeed, the LAA is not only a component of the LA but it constitutes a separate chamber of the heart with embryological, anatomical, and functional features distinct from those of the LA. On the other hand, LAA enlargement has been shown to be associated with LA filling pressure and hypertension. Being more compliant, the LAA is likely to increase in size prior to LA volume increase. The risk of stroke has been suggested to double for every 10 mm increment in LA diameter in males. Further investigations have confirmed the correlation between stroke/TIA risk and BSA adjusted LA volume enlargement. Our results also show that increased LV systolic diameter on TTE, indicating LV dysfunction, explains 3% of LAA volume increase. In patients with dilated cardiomyopathy, LV dimensions correlate significantly with LAA size. Importantly, in those patients, systolic dysfunction was associated with a high prevalence of LV (13%) and LAA (96%) thrombus in patients with sinus rhythm. These results parallel other studies investigating the relationship between LV systolic function and LAA thrombus formation. We hypothesize that LV dilation not only influences LA dilation and enhances LAA dilation but reduces LAA flow velocity. Elevated filling pressures in the left ventricle contribute to deterioration of the LAA flow, which in turn increases LAA volume. Interestingly, a low amount of pericardial fat was associated with increased LAA volume. The observed impact was minor and our finding is controversial in relation to previously published studies. There appears to be a graded relationship with a higher pericardial fat burden in chronic AF patients compared to PAF patients that is independent of BMI. This may partly explain why a decreased PAT area was found in patients with higher LAA volumes. Patients with chronic AF were excluded from our study while patients with lower volumes of PAT may suffer from PAF. More prospective studies for the evaluation of causalities between LAA volume, arrhythmias and adipose tissue are highly needed. It is to note, that PAT had no statistical significance (*P* = 0.133) in univariate analysis. No conclusions about the role of increased PAT and risk of recurrent stroke can be made from this study. Our study has some limitations. Firstly, the number of stroke/TIA patients included in the study was relatively small. However, patients were randomly selected without sampling bias. Secondly, 24-hour ECG monitoring is not able to reveal short-term PAFs with long intervals. Therefore, it is likely that PAF plays a more prominent role in cardioembolic stroke/TIA pathogenesis than currently known. Thirdly, the definition of hypertension and dyslipidemia were based on previously established diagnoses, and non-diagnosed cases were not perceived. A diagnosis of hypertension cannot be based on blood pressure measurements in the acute phase of stroke/TIA and comprehensive blood lipoprotein assays were not monitored. Stroke/TIA patients had higher BMI and larger BSA than control subjects. Adjusting the LAA volume with BSA can only decrease the impact of LAA volume increase in stroke/TIA patients compared to controls. This may lead to a negative bias in our hypothesis but does not overstate the result. It has also previously been shown that obesity has no significant impact on these results. While BSA is linked to patients’ weight and hence, to the amount of adipose tissue, adjusting LAA volume with BSA may interfere in the association of PAT area and LAA volume. To conclude, the extent of LAA enlargement is poorly linked with causative stroke/TIA risk factors and other imaging measurements, implying a relatively independent pathogenic mechanism for cardioembolic stroke/TIA. While PAF recorded with 24-hour Holter ambulatory ECG explains only 19% of LAA increase, it would be ideal to target more extensive ECG monitoring for stroke/TIA patients with large LAA volumes measured by cCT. If a connection between an enlarged LAA and an increased risk for cardioembolic stroke recurrence can be verified in future studies, cCT imaging may help to identify patients for more comprehensive ECG monitoring. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: MT PS PJ RV MH AM PM AMK. Performed the experiments: MT PS PJ RV MH AM PM AMK. Analyzed the data: MT PS PJ RV MH AM. Contributed reagents/materials/analysis tools: MT PS PJ RV MH AM PM AMK. Wrote the paper: MT PS PJ RV MH AM PM AMK.

# Introduction Skeletal muscle is the largest non-adipose tissue component of the tissue-system level of human body composition and is central to the study of nutritional, physiologic and metabolic processes. Magnetic resonance imaging (MRI) is currently applied as the golden standard for evaluating skeletal muscle due to its high accuracy and lack of radiation to the subjects. However, the high cost of MRI precludes its routine use in research and clinical practice. Presently, dual energy X-ray absorptiometry (DXA) is considered an alternative approach to estimate skeletal muscle *in vivo,* with substantially lower costs and less radiation exposure. So far, several studies have used DXA to predict skeletal muscle mass. Quantification of skeletal muscle mass significantly contributes to further understanding the characteristics of skeletal muscle, instructing the measurement and evaluation of physique and improving national physical fitness. China is the most populous country in the world, accounting for nearly 20% of the world’s population. However, there is no reported DXA-based skeletal muscle prediction model applicable to Chinese adults. The prediction model from Kim *et al.* by DXA was developed and validated through subjects of multiple ethnicities, including Caucasians, African- Americans, Hispanics and Asians. However, the interpretation and application of this prediction model in Chinese adults are questionable. First, the Asian subjects in Kim’s study were a mixture of multiple Asian populations. In addition to China, subjects were also recruited from India, Korea and Japan and other Asian countries. Previous studies have revealed that there were differences in body composition among people from different Asian countries. Second, there may be differences in body composition between the Asian population living in the U.S. and those in their originating countries,. Third, the Asian subjects in Kim’s study only accounted for 12.8% and 7.5% of the development and validation models, respectively. The established model was validated only in 7 Asians, and the quantitative difference between the proportions of Asian subjects in the development and validation groups are too great to conclude that ethnicity will not affect the amount of total body skeletal muscle predicted by DXA. The objective of the present study is to establish a DXA-based skeletal muscle prediction model in Chinese adults by using MRI as the reference standard. # Methods ## Protocol From the perspective of body composition, appendicular tissues primarily consist of skeletal muscle, bone, fat, skin and connective tissue. Appendicular lean soft tissue (ALST), measured by DXA, is a fat- and bone mineral-free tissue which consists of skeletal muscle, skin, connective tissue, and tendons. Because a large proportion (∼75%) of total body skeletal muscle exists in appendicular tissues and a large proportion of ALST is SM, ALST is considered as a measure of appendicular SM (ASM). A small amount of adipose tissue that locates between muscle groups and beneath the muscle fascia is defined as intermuscular adipose tissue (IMAT). The association among appendicular tissue, ALST, ASM, SM and IMAT is illustrated in. In the current study, total body skeletal muscle mass (SM) is defined as the skeletal muscle with IMAT. The association among IMAT, IMAT-free SM and SM is represented as: ## Subjects A total of 68 subjects aged from 20 to 80 years were voluntarily recruited through leaflets and posters provided by the Obesity and Body Composition Research Center (OBCRC), Zhejiang University School of Public Health between December 2009 and July 2010. The study was approved by the Ethics Committee of the Second Affiliated Hospital, Zhejiang University. Written consent was obtained from all participating subjects prior to testing. None of the individuals were on medication or participated in vigorous physical activity. One subject was excluded because of BMI \<18.5 kg/m². Another subject was excluded because one highly influential sample with cooks distance was close to 1.0 and absolute value of DFFITS was close to 2. Therefore, 66 subjects (52 men and 14 women) were finally included in the analysis. ## Body Composition Analysis Subjects underwent three evaluations within one day: anthropometry and DXA were performed at OBCRC, and MRI was performed at the Second Affiliated Hospital of Zhejiang University, China. ### Anthropometry Weight and height were measured while barefoot and wearing light clothing. Body weight was measured to the nearest 0.1 kg (Detecto, USA). Height was measured with a hypsometer to the nearest 0.1 cm. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared. ### Dual-energy X-ray absorptiometry DXA (GE Lunar Prodigy, WI, USA with software version 11.40.004) was implemented to measure lean soft tissue, fat and bone mineral in both whole body and specific regions of interest. By using specific anatomic landmarks, legs, arms and trunk were isolated on the anterior view planogram using the DXA system’s automated software. The DXA software then provided compositional estimates of the legs, arms, trunk, head and whole-body. ALST was calculated as the sum of lean soft tissue in the right and left legs and arms. DXA was calibrated daily against a phantom with the manufacturer’s precision standards of ≤0.8%. ### Magnetic resonance imaging Whole-body MRI imaging was performed by a 3.0 Tesla MRI scanner (Signa, GE Healthcare, USA). Subjects were required to lie in the magnet bore in a supine position with arms extended overhead. A T1 weighted, spin-echo, axial plane sequence was performed with a 1500 millisecond repetition time and a 17 millisecond echo time. Transverse images (10 mm image thickness) were obtained every 50 mm from hand to foot. The intervertebral space between the fourth and fifth lumbar vertebrae (L4–L5) was used as the point of origin. Three series of seven images were obtained for the lower body and three series of seven images were obtained for the upper body. The protocol involved the acquisition of 36–45 images, depending on the height of the subject. The SliceOmatic 4.3 software (TomoVision Inc, Montreal, Canada) was applied to segment tissue compartments. IMAT-free SM and IMAT were segmented automatically first by the combination of functions offered by the software including threshold, region growing, mathematic morphology and snakes. We then adjusted manually and calculated their cross-section areas. Segmentation of IMAT and skeletal muscle on MRI image is indicated in **.** Volumes of IMAT-free SM and IMAT by MRI were calculated as, where is the thickness (10 mm) of each image, (40 mm) is the distance between consecutive images and is each image’s cross sectional area. Volumes were converted to mass by multiplying the density of 1.04 kg/l for IMAT-free SM and 0.92 kg/l for IMAT. Our study shows that the technical error for repeated measurement of the same scan by the same observer of IMAT-free SM and IMAT volumes are 1.0% and 0.9%, respectively. ## Statistical Analysis Data was grouped by gender and summarized in terms of mean ± standard deviations (SD). The differences in the attributes of men and women were tested by Student’s *t* test. Pearson’s correlation was performed to explore the associations between DXA-measured ALST and MRI-measured SM. A simple linear regression model was first applied to the data as:, where is the intercept and is the slope of the equation. Then other independent variables such as age and gender were considered as the adjustment to the simple linear regression model. The adjusted *R²* and standard error of estimate (SEE) were used to evaluate model-fitting performance. To evaluate the models’ performance via out of sample prediction, we applied the method of ‘Leave One Out Cross Validation’ for regression models, which is a rotation analysis of prediction performance based on N–1 training samples and 1 validation sample where N is the total sample size. Specifically, the original data set was partitioned into a training set of N–1 samples and a validation set of the remaining one. A regression model was fitted to the training set. Based on the fitted model, a prediction was made for the validation sample and the predicted residual was calculated. This procedure was replicated N times for all possible partitions and then the prediction performance was summarized as Coefficients of Variation (CV) statistics. In the present study, the root mean square error of the predicted residuals was used to compare the out of sample prediction performance for potential models. In addition, final model was also checked by both F-test and permutation test for analysis of variance (ANOVA). Pearson correlation was also explored between predicted SM values and the corresponding actual values measured by MRI. The difference and the agreement between total body SM measured by MRI and predicted from the equation were tested by Student’s *t* test and the method of Bland-Altman plot. All statistical analyses were performed by Stata software (version 11.0 for Windows; Stata Corporation, College Station, TX) and statistical significance level was set at *P*\<0.05 (two-tailed). # Results ## Characteristics and Body Composition The characteristics and body composition of 66 subjects (52 men and 14 women) are presented in. There were no significant differences in age and BMI between men and women. Men tended to have a greater body mass, height, ALST, IMAT, IMAT- free SM and SM than women. In contrast, women had a greater percentage of body fat than men. ## Evaluation of the SM Prediction Model by Kim et al. in Chinese Adults By substituting the values of ALST, age and gender into the SM prediction model by Kim *et al.*, SM predicted according to Kim’s equation was significantly less than that measured by MRI which was underestimated by 2.3% and 3.4% in men and women, respectively. There was a significant bias between the difference of SM measured by MRI and calculated by the prediction model of Kim *et al.* (*r* = 0.43, *P*\<0.001) and two thirds (44/66, 67%) of the subjects were underestimated from Kim’s prediction equation**.** ## Prediction Model of IMAT-free SM ### Model development The ratio of IMAT-free SM to ALST was 1.15±0.06 in men and 1.13±0.05 in women with the corresponding CVs of 5.2% and 4.4%, respectively. There was a significant correlation between IMAT-free SM and ALST (*r = *0.97, *P*\<0.001). Developed models for predicting IMAT-free SM from multiple linear regression analyses are presented in **.** ALST (in kg) was the strongest predictor of IMAT-free SM (in kg), explaining 94.3% of the between-subject variance with a SEE of 1.11 kg. No significant interaction effect was observed between gender and ALST (F = −0.84, *P* = 0.41). The adjustment of gender failed to contribute to the developed model (*P* = 0.57). The inclusion of age together with ALST in the multiple linear regression model contributed 0.3% to the variance of IMAT-free SM. To further examine the effect of age on IMAT-free SM, we modeled the age effect (adjustment) as a piece-wise linear function which was remaining constant for the middle-aged and decreasing for the elderly people. The cutoff point of the piece-wise linear effect of age was estimated by minimizing the SEE of the following model. The cut-off point turned out to be 45 years old for the database. The inclusion of the age effect together with ALST in the multiple regression models captured an additional 0.6% of the variance in measured IMAT-free SM mass. With the age increase, subjects older than 45 years contributed a negative effect in IMAT-free SM (Coefficient = −0.04, *P* = 0.0053). In addition, the ratio of IMAT-free SM to ALST was 1.18±0.06 in men and 1.17±0.05 in women with the corresponding CVs of 5.1% and 4.3%, respectively. IMAT-free SM/ALST significantly decreased with age in men (IMAT–free SM = -0.001×age +1.22, *P* = 0.02) but not in women (IMAT–free SM = −.0.001×age +1.22, *P = *0.77) as indicated in. ### Model-validation Prediction performance was compared the default model 1 and age-adjusted model 2 via ‘delete-one (leave one out) cross validation’ on the root mean squared error of predicted residuals. The root mean squared errors for these two models based on the whole sample are 1.10 and 1.03 kg respectively. The root mean squared errors for the “leave one out cross validation” are 1.13 kg for the default model and 1.08 kg for the age-adjusted model. The age-adjusted model is also favorable due to out of sample prediction performance. The cut-off point of the piece-wise linear age effect was also verified by delete-one cross validation. Apart from the “leave-one-out” validation technique, the final prediction model with age and ALST adjusted, was also checked by both F-test and permutation test for analysis of variance (ANOVA). The testing results showed consistent evidence that both ALST and age effect are significant. IMAT-free SM predicted by both two models (model 1∶23.3±4.5 kg, model 2∶23.3±4.5 kg) were not significantly different from IMAT-free SM measured by MRI (23.4±4.7 kg) (both *P*\>0.05) IMAT-free SM predicted from two models were highly correlated with IMAT-free SM measured by MRI (both *r*\>0.97, *P*\<0.001). The inclusion of age effect to ALST in the prediction model contributed a slight increase in the correlation between measured and predicted IMAT-free SM with *r* values from 0.972 to 0.975. There were no significant between-methods biases and the agreement of these two methods between measured and predicted IMAT-free SM by Bland-Altman analysis is presented in **.** # Discussion The present study developed and validated DXA-SM prediction models applicable for Chinese adults. The prediction models in Chinese adults had a small but significant difference compared with prediction models for other ethnicities. The quantification of SM according to previous study underestimated SM in Chinese adults. ## Differences in IMAT, IMAT-free SM and SM between Chinese and Other Race Populations The IMAT compartment, although comparatively small, varies in magnitude with age, sex and race. In the Chinese population, the quantities of IMAT were 0.6±0.2 kg in men and 0.5±0.1 kg in women which was comparatively smaller than the quantity of IMAT in Caucasians and African-Americans. Evidence indicated that Asians have lower IMAT compared to African-Americans even after adjusting for the difference in total adiposity. There was a significant difference between SM with and without IMAT in men and women (both *P*\<0.001). Although there’s no obvious reduction of SEEs by removal of IMAT compared with the studies of Kim *et al.*, when expressed relative to SM, IMAT accounted for 0.9% ∼ 5.1% of SM in the present study, and the exclusion of IMAT could eliminate the overestimation of the actual quantity of muscle. In the current study, we established models with SM and IMAT-free SM as the dependent variable with ALST and age adjusted. Kim *et al.* also extended their study by removal of IMAT in the SM compartment in diverse ethnic groups. Prediction models from the current study and Kim *et al.* are presented in. By removal of IMAT, Kim *et al.* observed lower SEEs. In contrast, our two models with the dependent variable as SM and IMAT-free SM have similar SEEs (1.09–1.15 kg vs. 1.05–1.11 kg), yet both have a high adjusted *R²* values in the range of 0.94–0.95. ## Why Kim et al.’s Equation Underestimates SM in Chinese Adults We observed that, although the prediction model on SM proposed by Kim *et al.* is widely accepted, there was an underestimate of SM by 2.5% in Chinese subjects when ALST was substituted into Kim’s prediction equation. The quantities of SM predicted using Kim’s equation were smaller than SM measured by MRI both in men and women (both *P*\<0.05). This phenomenon even persisted in the group of subjects \<45 and ≥45 years, respectively (**,**). The observation above impelled us to explore a possible explanation. Eveleth and Tanner reported that there were considerable differences in leg lengths via the sitting height ratio among ethnic groups, with blacks having longer legs than Caucasians and Asians having shorter legs than Caucasian. Recent data from the NHANHS III survey confirmed that compared to Caucasians, blacks have longer legs and this is also the case for a national sample of black and white youths aged 12 to 17 years old. In addition, Deurenberg *et al*. reported that Asians (especially Chinese) had relatively shorter legs when compared to Caucasians. Since ASM accounts for 75% of the total-body SM in the body, subjects with longer legs are expected to have more SM with the height into consideration. Previous studies have already elucidated that prepubertal Asians had less ASM than their African-American and Caucasian counterparts. Although data is limited on the difference of ASM in adults among different ethnicities, Chinese adults would have less ASM compared with African-Americans and Caucasians with height taken into consideration. As a measure of ASM, ALST is a strong predictor in estimating SM by DXA. Thus comparatively, the distribution of SM in Chinese would turn out to be smaller in appendicular tissues than in Caucasians and blacks, because Chinese have the shortest legs, on average. This hypothesis might be one plausible explanation for the underestimation of Kim’s equation in Chinese adults. It should be noted that although the general notion indicated that Asians have the shortest legs when compared to blacks and Caucasians, there were observed differences of leg length among Chinese and people from other Asian countries. Lin *et al.* reported that mainland Chinese had moderate limbs while Taiwanese had longer legs, Japanese had shorter limbs and Korean had longer upper limbs. Given the existence of the discrepancy of leg length in different ethnic groups and in people from different Asian countries, the development of a specific SM- prediction model for Chinese in the present study is of vital significance and accuracy. There was a high correlation between SM and ALST, which is consistent with previous studies. In the current study, ALST alone explained 94% of the observed between-individual variation in IMAT-free SM with a low SEE (i.e. 1.11 kg), indicating high estimation accuracy in the prediction model. ## Effect of Age on SM Our results did find that there was a curvilinear relationship between age and IMAT-free SM mass, with a change in the slope of the regression line occurring at 45 years old. The age of 45 years was thus identified as the cut-off point, after which an obvious negative association between age and SM was observed. This finding is consistent with the study of Janssen *et al.,* who reported that there were accelerating rates of SM loss in both genders in adults of multiethnic groups (67% Caucasians, 17% African-Americans, 8% Asians, and 7% Hispanics) aged 18–88 years, averaging at 45 years of age. Evidence also indicated that the isometric, and isokinetic, strength did not change substantially until approximately 45 years of age. Thus the age of 45 years is a critical cut-off point of total body skeletal muscle mass across lifespan. Silva *et al.* had inconsistent findings, reporting a cut-off point of SM decline of 27 years of age, after which the SM begins to decline in a large and diverse sample. Gallagher *et al*. also confirmed the decline of total body potassium, which predominantly accumulates in SM, starting from the age of 31 in men and 30 years of age in women, both African-American and Caucasian. One explanation for the discrepancy of the age cut-off point when SM begins to decline between our study and the other studies by Silva *et al.* and Gallagher *et al.* is that there might be a minor decrease of SM in the third decade of adults in the present study, yet the rate of SM loss may be so slight that the decline in the quantity of SM is not noticeable. However, previous studies reported that the muscle fiber cross sectional area (i.e., contractile muscle) and body cell mass, did not change substantially until approximately at the age of 45 years. The discrepancy of observed age at which SM begins to decline may be due to ethnic differences in body composition. ## Effect of Gender on SM The observation that men had greater SM than women agrees with the finding of previous studies. Because no gender difference in the relationship between IMAT- free SM and ALST was detected, we combined men and women together in the prediction model, although gender didn’t contribute to the final developed model. One possible explanation for the insignificant effect of gender in predicting SM may be attributed to the relative small sample size of female subjects, which might offset the additional prediction accuracy by this model. However, even in a previous study with moderate sample size (145 men and 176 women), gender was found to be a borderline contributor (*P* = 0.052) in the SM prediction model by Kim *et al*.. On the other hand, the insignificance of gender in the prediction model might be just due to the ethnic difference *per se*. Nevertheless, the protocol to develop the prediction model may be useful when to develop similar total-body SM prediction model in a large scale of Chinese adults. ## SM/ALST Ratio The values of IMAT-free SM/ALST in our study, 1.15 in men and 1.13 in women, were consistent with the ratios of the previous study. In addition, we observed that IMAT-free/ALST decreased significantly with age in men, which is consistent with the finding of Kim *et al.*. However, we failed to observe the association between IMAT-free SM/ALST with age in women. The discrepancy of the association between IMAT-free SM and age in men and women might be due to the small sample size of women. Large scale studies are needed to explore and confirm whether IMAT-free SM/ALST decreased with age in both genders. On the other hand, because ALST is composed of muscle, skin, connective tissue, and the lean portion of adipose tissue, muscle decreases but other components such as connective tissue and the lean portion of adipose tissue increase during the aging process. ## Study Implications The major contribution of this study was to establish DXA-SM prediction equations applicable for healthy Chinese adults. The DXA-SM prediction equations were developed by using MRI as the reference standard and this observation suggests that DXA can provide reliable and accurate way to estimate SM in large scale studies in Chinese adults. In view of the strong influence of SM on bone mineral density, the observation that SM progressively decreased with age after the age of 45 years, highlights the importance of strength training, especially high-intensity strength training that can prevent osteoporosis by simultaneously promoting muscle mass, muscle strength and bone mineral density. Sarcopenia, defined as a decrease in SM in aging men and women is of great interest. The quantification of SM is help to establish the diagnostic index of sarcopenia in the Chinese population and further explore the prevalence of sarcopenia and its health consequences. ## Study Limitations A limitation of the present study is the relatively small size of female subjects, which may introduce a gender bias in predicting the IMAT-free SM by ALST, and it may also offset the accuracy of the prediction based on this model. Larger scale studies should be conducted in the future to explore whether there are any effects due to gender in predicting IMAT-free SM by ALST in the Chinese population. The prediction equations developed in this study may not be appropriate for athletes and subjects \<18 years. There is also a need to develop a corresponding prediction model for children and adolescents. In addition, we cannot rule out the possibility that the accuracy of this model will be compromised in subjects with disease states and conditions associated with extreme muscle atrophy, such as spinal cord injury. ## Conclusions In conclusion, SM predicted according to a previous model developed in multi- ethnic groups was underestimated by 2.3% and 3.4% for Chinese men and women, respectively. A new prediction equation by DXA has been established to estimate IMAT-free SM in Chinese adults. This observation suggests that DXA is a reliable and accurate approach to study the quantity of total body SM and its association with age and gender on a large scale in Chinese populations. We are grateful to those subjects who participated in this study. [^1]: The authors have declared that no competing interests exist. [^2]: MRI image segementation: XYZ. Conceived and designed the experiments: XYZ ZMW WH SKZ. Performed the experiments: XYZ JMH WH. Analyzed the data: XYZ ZMW JYZ. Wrote the paper: XYZ ZMW JYZ WH SKZ.

# Introduction Hepatocellular carcinoma (HCC) is the most common cause of liver cancer and the fourth most frequent cause of death in the world. The number of primary liver cancer cases, of which HCC accounts for 75–85%, is expected to increase globally by 2030; however, of the 30 modeled countries, only Japan is predicted to decline in liver cancer incidence. In contrast, the number of patients with HCC who test negative for both hepatitis B surface antigen and hepatitis C virus antibody is increasing in Japan, which is problematic as these patients are not adequately screened and, therefore, the disease is not diagnosed until it has reached late stages. Furthermore, many patients who have received curative therapies such as surgical resection or local ablation subsequently develop recurrent disease which is often advanced. Presently, the first-line treatment strategy for patients with advanced HCC is the administration of sorafenib according to several guidelines from Japan, Europe, and the United States; however, hepatic arterial infusion chemotherapy (HAIC) is also widely used throughout Asia, especially in Japan. Indeed, Japan is the only country that recommends HAIC as standard therapy in the treatment algorithm. A Japanese nation-wide survey, which highlighted the efficacy of HAIC for treatment of advanced HCC, demonstrated that patients treated with HAIC using a low-dose of cisplatin (CDDP) and 5-fluorouracil (5-FU, FP) exhibited a significantly longer median survival time (MST) than patients who did not receive active treatment (14.0 versus 5.2 months; *p* \< 0.0001). To date, no randomized controlled trials have established a guidline for clinicians to select either sorafenib or HAIC for the management of advanced HCC. Considering the poor prognosis associated with advanced HCC, the identification of patients likely to benefit from either sorafenib or HAIC is important. A number of prognostic factors have been identified for various malignancies, including the patient’s skeletal muscle and visceral fat composition. In fact, reports of patients treated with sorafenib for HCC suggested that skeletal muscle depletion was an independent prognostic factor. We recently determined that the lack of skeletal muscle depletion with a high visceral fat area (H-VFA) was a favorable prognostic predictor for the survival of sorafenib-treated advanced HCC patients. However, there are no studies of body composition in HAIC treated HCC patients. In addition, there is inconclusive evidence for whether therapeutic conversion (conversion from HAIC to sorafenib, or sorafenib to HAIC) provides a considerable survival benefit. In this study, we retrospectively analyzed the impact of body composition and therapeutic conversion on the clinical outcome of patients with advanced HCC treated with HAIC or sorafenib. Also, we investigated factors that would identify patients likely to respond to treatment with HAIC or sorafenib. # Materials and methods ## Study design and patient selection This study complied with the ethical principles of the Declaration of Helsinki. The Institutional Review Board of Yamaguchi University Hospital approved the research protocol (H28-026). This was a retrospective patient record study conducted at Yamaguchi University Hospital, and the informed consent was waived. All records were obtained as anonymized data. This study began after sorafenib was approved for use in Japan. Records of patients diagnosed with advanced stage HCC between April 2009 and December 2016 were reviewed for study inclusion by a clinician. Patient records were selected if either HAIC or sorafenib was administered as a first line treatment for advanced HCC, which was untreatable by loco-regional therapies, including hepatectomy, radiofrequency ablation, or transcatheter arterial chemoembolization (TACE). According to Japanese guidelines, in general, HAIC and sorafenib were recommended for cases with macrovascular invasion (MVI) and extra-hepatic spread (EHS), respectively. The diagnosis of HCC was based on imaging results and elevated serum levels of alpha-fetoprotein (AFP), des-gamma-carboxy prothrombin (DCP), or both. Patient records without computed tomography (CT) within 1 month of starting HAIC or sorafenib were excluded; if no CT image slices were available to assess body composition, magnetic resonance imaging (MRI) was used. Out of 133 reviewed records, 55 and 78 patients who were treated with HAIC and sorafenib, respectively, were included in the analysis. The follow-up period ended on December 31, 2017. Data on age, sex, Child-Pugh classification, tumor number and maximum size, MVI, EHS, radiological response, muscle depletion, and visceral fat area (VFA) were collected for each patient. Information on therapeutic conversion from HAIC to sorafenib, or from sorafenib to HAIC was also collected. ### HAIC and sorafenib therapy For HAIC administration, a 5-French catheter (Anthron P-U Catheter; Toray Medical Co. Ltd., Tokyo, Japan; or Piolax, G-spiral catheter; Piolax, Kanagawa, Japan) was inserted in the gastroduodenal artery or proper hepatic artery and connected to the port. All patients received HAIC with a low-dose of FP and isovorin. In brief, patients received a daily low-dose of FP, including CDDP (10 mg/body), followed by 5-FU (250 mg/body), and isovorin (6.25 mg/body) once daily on days 1–5. One course of HAIC consisted of 20 doses. The initial dose of sorafenib was 800 mg. However, when adverse effects during the treatment course occurred, the dose was reduced. Furthermore, the initial dose was started from 400 mg depending on the liver function. The initial evaluation was performed 8 to 12 weeks later. ## Assessment of body composition We evaluated body composition using CT images obtained within 1 month prior to introducing treatment with HAIC or sorafenib and 3 months after each treatment. The areas of skeletal muscle and visceral fat were measured using an AZE 3D work station (AZE-3DWS, AZE Virtual Place Raijin, AZE Ltd., Tokyo, Japan). Specifically, the skeletal muscle area and VFA were measured using a CT scan at the third lumbar vertebra level (L3), and at the umbilical level, respectively. Skeletal muscle and visceral fat were quantified using Hounsfield unit (HU) thresholds of –29 to 150, and –190 to –30, respectively. The skeletal muscle mass was normalized to height (m²) and was expressed as the skeletal muscle index (SMI, cm²/m²). Using these values for SMI, muscle depletion was defined as \< 42 cm²/m² in men and \< 38 cm²/m² in women, in accordance with the Japan Society of Hepatology (JSH) criteria. The cut-off value of VFA, following the criteria for obesity established by the Japan Society for the Study of Obesity (JASSO), was set to 100 cm², and H-VFA was defined as ≥ 100 cm² in men and women. ## Evaluation of treatment response The evaluation of treatment response was performed with dynamic CT or MRI and classified according to the modified response evaluation criteria in solid tumors. The response was evaluated after one course of HAIC, and every 3 months in the group receiving sorafenib. A subset of patients who were not evaluated for treatment response according to imaging, was designated as no evaluation (NE). The positive response and positive disease control were defined as the sum of complete response (CR) and partial response (PR), and the sum of CR, PR, and stable disease (SD), respectively. ## Statistical analysis Continuous variables were presented as the means ± standard deviation or median (interquartile range \[IQR\]) and compared using a paired or unpaired *t*-test. Categorical variables were evaluated using the chi-squared test or Fisher’s exact test. The overall survival (OS) was estimated by the Kaplan-Meier method and compared using the log-rank test. The clinical prognostic factors that were assessed in the survival analysis, included age (\< 70 or ≥ 70 years), sex (male or female), Child-Pugh classification (A or B), tumor number (\< 10 or ≥ 10 in the HAIC group, \< 7 or ≥ 7 in the sorafenib group), maximum tumor size (\< 70 mm or ≥ 70 mm in the HAIC group, \< 40 mm or ≥ 40 mm in the sorafenib group), MVI (absence/presence), EHS (absence/presence), muscle depletion (absence/presence), and VFA (H-VFA or low VFA \[L-VFA\]), Radiological evaluation response was positive or negative in the HAIC and sorafenib groups, and disease control was denominated as positive/negative in the latter group as well. Also, the therapeutic conversion was defined as yes or no for both HAIC and sorafenib groups. The cut-off values of number and size of tumors are presented as medians. A Cox regression model was used to analyze the factors associated with OS, and the results are presented as hazard ratio (HR) with 95% confidence intervals (CIs). A *p*-value \< 0.05 was considered statistically significant. All analyses were performed in the JMP software package v13.0 (SAS Institute, Cary, NC, USA). # Results ## Patient characteristics The clinical patient characteristics are summarized in. Fifty-five individuals were treated with HAIC and mean patient age was 66.7 ± 11.4 years. Liver function was Child-Pugh class A in 36 patients and class B in 19 patients. The median number and maximum size of tumors were \> 10 and 71.0 mm, respectively. Forty-six patients had MVI and eight showed EHS. Twenty-six patients converted from HAIC to sorafenib as a post-treatment. Muscle depletion was observed in 24 patients (43.6%) and H-VFA was detected in 34 patients (61.8%). In contrast, 78 individuals were treated with sorafenib and had a mean age of 72.2 ± 8.5 years. Liver function was Child-Pugh class A in 60 patients and class B in 18. The median number and maximum size of tumors were 6.5 and 40.0 mm, respectively, while 15 patients had MVI and 36 showed EHS. Six patients converted from sorafenib to HAIC as a post-treatment. Muscle depletion was observed in 32 patients (41.0%), and H-VFA was detected in 52 patients (66.7%). There were significant differences in age, tumor number, maximum tumor size, MVI, and EHS between the two groups because of selection bias. The HAIC group had significantly higher number of tumors (median, 10.0; *p* = 0.015), larger maximum tumor size (median, 71.0 mm; *p* \< 0.001), and higher proportion of patients with MVI (83.6%; *p* \< 0.001) than the sorafenib group. Whereas, the sorafenib group had significantly older (mean, 72.2 years; *p* = 0.002), and higher proportion of patients with EHS (46.2%; *p* \< 0.001). ## Response to HAIC and sorafenib The responses to HAIC and sorafenib are summarized in. In the HAIC group, 2 (3.6%), 14 (25.5%), 22 (40.0%), and 16 (29.1%) patients exhibited CR, PR, SD, and progressive disease (PD), while NE was found in one (1.8%) patient. Thus, the response and disease control rates were 29.6% and 70.4%, respectively. On the other hand, in the sorafenib group, 0 (0.0%), 4 (5.1%), 41 (52.5%), 25 (32.1%), and 8 (10.3%) patients exhibited CR, PR, SD, PD, and NE, respectively. Thus, the response and disease control rates were 5.7% and 64.3%, respectively. ## Patient survival predictor In the HAIC and sorafenib groups, the median survival time (MST) was 12.5 and 12.1 months, respectively. The prognostic factors of HAIC group are shown in, and five factors were significant predictors of survival in the univariate analysis: age \< 70 (HR, 0.365; *p* = 0.003), male sex (HR, 2.481; *p* = 0.014), Child-Pugh A (HR, 0.461; *p* = 0.024), a positive response (CR and PR, HR, 0.369; *p* = 0.004), and conversion to sorafenib (HR, 0.409; *p* = 0.005). In contrast, body composition measurement did not significantly affect OS. The MST of patients with muscle depletion was 15.8 months compared to 10.3 months for patients with no muscle depletion (*p* = 0.121), while the MST of patients with H-VFA was 10.3 months compared to 13.2 months for patients with L-VFA (*p* = 0.371). Furthermore, multivariate analysis identified positive response to HAIC (HR, 0.438; 95% CI at 0.200–0.892, *p* = 0.022), and conversion to sorafenib (HR, 0.374; 95% CI at 0.183–0.767, *p* = 0.008) as favorable prognostic factors. In the sorafenib group, univariate analysis indicated that prognostic factors were Child-Pugh A (HR, 0.487; *p* = 0.018), tumor number \< 7 (HR, 0.494; *p* = 0.007), absence of muscle depletion (HR, 0.506; *p* = 0.013), and H-VFA (HR, 0.485; *p* = 0.009). Furthermore, multivariate analysis identified tumor number \< 7 (HR, 0.475; 95% CI at 0.273–0.822, *p* = 0.008), absence of EHS (HR, 0.511; 95% CI at 0.295–0.877, *p* = 0.015), absence of muscle depletion (HR, 0.555; 95% CI at 0.317–0.983, *p* = 0.044), and H-VFA (HR, 0.483; 95% CI at 0.275–0.863, *p* = 0.015), as shown in. Patients with no muscle depletion showed significantly longer survival than those with muscle depletion (MST 13.4 vs. 11.0 months, *p* = 0.010;). In addition, patients with H-VFA also showed significantly longer survival than those with L-VFA (MST 15.0 vs. 8.4 months, *p* = 0.004;). ## Changes in body composition after HAIC or sorafenib treatment We analyzed the changes on SMI and VFA 3 months after the induction of each treatment. SMI decreased by 2.7% and 7.2% in HAIC and sorafenib groups, respectively (*p* = 0.095). However, comparison of VFA between both groups did not show significant changes at this time point. # Discussion JSH recommends administration of HAIC as a first-line treatment for unresectable advanced HCC with portal invasion; however, American or European professional organizations do not recommend it because of lack of evidence of survival benefits. Instead, the latter ones support systemic chemotherapy, including administration of sorafenib, a multi-targeted tyrosine kinase inhibitor (TKI) as a first treatment. Therefore, it is necessary to identify patient characteristics that correlate with response to HAIC. One characteristic of interest is sarcopenia, which was initially described as an age-related disorder; however, secondary sarcopenia has recently been linked to factors other than age and prognosis of various diseases, including cancer. Indeed, skeletal muscle depletion is significantly associated with the prognosis of HCC. In contrast, the predictive effect of VFA on the survival of patients with HCC is controversial. Generally, obesity promotes carcinogenesis and cardiovascular disease, and is associated with poor prognosis in several cancers. However, because these findings have been linked to strong observational bias, many investigators disagree with obesity being a predictive factor in the elderly or in patients with certain cancers. We previously performed an integration analysis between skeletal muscle and visceral fat, which indicated that the body composition characteristic of no muscle depletion with H-VFA was a favorable survival predictor in patients treated with sorafenib. In fact, both sorafenib and HAIC have been often applied as a sequential chemotherapy regimen in Japan. However, there are no studies of body composition in patients treated with HAIC. Therefore, during the same period, we assessed prognostic factors related to HAIC and sorafenib therapy, including body composition by adding radiological disease control, response, and therapeutic conversion. In the present study, body composition significantly affected the prognosis of sorafenib-treated patients with HCC, as a difference with HAIC-treated patients with HCC (Figs and, Tables). Therefore, body composition may be a useful biomarker to identify patients with HCC who are likely to benefit from either HAIC or sorafenib. Although the reasons why body composition has a different effect between the two treatments remain unclear, sorafenib therapy reduced skeletal muscle mass compared to HAIC (median, –7.2 vs. –2.7%; *p* = 0.095). This result indicated that sufficient skeletal muscle mass was required for patients to complete the sorafenib therapy, compared to HAIC therapy. As patients with advanced HCC suffer from severe energy malnutrition regardless of having Child-Pugh A, a severe progression of muscle loss may contribute to a negative prognosis in sorafenib-treated patients with muscle depletion. Furthermore, such patients may require a certain amount of visceral fat as an energy substrate because of insufficient skeletal muscle mass. Consequently, H-VFA may be associated with a positive prognosis in this group. Therefore, we attribute the difference in changes of skeletal muscle mass between both treatments to a difference in pharmacological effects. Specifically, sorafenib suppressed tumor growth over a long duration of administration, whereas HAIC showed tumoricidal activity during a comparatively short-term administration. The success of sorafenib, regorafenib, lenvatinib, cabozantinib, ramucirumab, and immune checkpoint inhibitors may change the treatment paradigm in patients with intermediate stage and advanced HCC. For intermediate HCC \[the so-called Barcelona clinic liver cancer (BCLC) stage B\], TACE is the preferred first-line treatment according to guidelines for HCC. However, because patients with intermediate stage HCC become refractory to repeated TACE procedures and liver function declines, conversion to sorafenib is initiated to improve the prognosis. Maintaining hepatic function is important for improving the prognosis of patients with intermediate HCC because TKIs are generally used in patients with Child-Pugh A HCC. In the future, determination of TACE refractory status would increase the therapeutic options, such as TKIs, immune checkpoint inhibitors, and several combination therapies. However, HAIC treatment did not significantly deteriorate liver function compared to sorafenib, but significantly improved the Child-Pugh score of HAIC responders with Child-Pugh B HCC. In addition, patients with Child-Pugh B HCC were candidates for HAIC. Therefore, HAIC may be considered an important alternative treatment because there was a difference in survival between patients under HAIC and sorafenib treatments based on body composition. Furthermore, multivariate analysis identified positive response to HAIC and conversion to sorafenib as independent prognostic factors. The considerable survival benefit of HAIC has been reported in HAIC responders. Interestingly, our study showed that conversion to sorafenib in patients receiving HAIC was a stronger prognostic factor than positive response because of the lower HR. There are few reports on conversion to sorafenib in HAIC non-responders. Miyaki, et al. reported that conversion to sorafenib in HAIC non-responders was a significant prognostic factor in a retrospective cohort study, but this study was limited by the different observation periods between conversion from the HAIC to the sorafenib group and the HAIC alone group. The same research group also conducted a prospective study, but it included only a single arm and considered a small sample size in a narrow observation period. Our study also demonstrated the significant survival benefit of conversion to sorafenib in patients with advanced HCC regardless of the response to HAIC. The MST was significantly longer in patients with conversion to sorafenib than in those without conversion to sorafenib (16.5 vs. 7.1 months, *p* = 0.004;), although, there were no significant differences in patient characteristics except for Child-Pugh class. This observation suggested that conversion to sorafenib rescued a subset of HAIC non-responders from poor prognosis; therefore, it is important to identify patients who are likely to benefit from this strategy early in the HAIC treatment process. In contrast, therapeutic conversion to HAIC was not identified as a prognostic factor in patients with HCC treated with sorafenib. However, it is impossible to draw a conclusion from the present study because a small population of patients with therapeutic conversion was used; therefore, further investigations are required. Finally, we combined the findings of this study with our previously developed scoring system for HAIC treatment to propose a guide as a treatment strategy for advanced HCC. We have reported the ACTH (Assessment for Continuous Treatment with HAIC) score as the scoring system which evaluated the response to HAIC at the mid-cycle of HAIC. The ACTH score (range, 0–3) was calculated based on three parameters: Child-Pugh class (A = 0, B = 1), AFP response (yes = 0, no = 1), and DCP response (yes = 0, no = 1). A positive response of tumor marker was defined as a reduction of ≥ 20% for the baseline to the mid-cycle of HAIC (2 weeks after HAIC induction). Thus, based on the ACTH score, patients were divided into two groups: ≤ 1 vs. ≥ 2 points; MST, 15.1 vs. 8.7 months; *p* = 0.003, which was further validated for therapeutic assessment. We considered to continue HAIC treatment for patients with a score ≤ 1 and to initiate other treatments for patients with a score ≥ 2. Our draft proposal for a treatment strategy for advanced HCC is as follows: (1) Body composition of patients with Child-Pugh A HCC should be evaluated. Patients having no muscle depletion with H-VFA should select sorafenib and regorafenib as the first and second-line treatments, respectively. If patients have other body compositions, HAIC should be selected as the first-line treatment. Furthermore, for patients with Child-Pugh B HCC, the first-line treatment should be HAIC. (2) After mid-course treatment of HAIC (i.e., 2 weeks), ACTH score can be evaluated, which is likely to increase the number of candidates for sorafenib treatment, and might allow the use of sorafenib earlier in the treatment of HCC non-responders. When used in combination, the evaluation of body composition and the ACTH score may vastly improve patient survival. Furthermore, this is the first report of HCC treatment algorithm using an assessment of body composition. As the body composition quantitatively assessed the treatment performance, it may become an effective alternative to determine performance status, which is included in the BCLC system for HCC. It has been reported that analytic morphomics, which uses semi- automated high-throughput CT image to measure body composition, is a useful approach to predict survival in patients with HCC. This novel technique may provide further prognostic information. The present study has some limitations. As this was a retrospective, and single- center study with a small population, further investigations, including a validation study with larger sample size, are necessary. In conclusion, body composition such as skeletal muscle and visceral fat was not identified as a prognostic factor for advanced HCC patients treated with HAIC. However, body composition may be a useful biomarker for the identification of patients with advanced HCC likely to experience survival benefits from sorafenib treatment. Additionally, our data suggest that the conversion to sorafenib in patients receiving HAIC may improve survival, regardless of response to HAIC treatment. The inclusion of a strategy that efficiently incorporates both HAIC and sorafenib in patients with advanced HCC has the potential to significantly improve patient survival. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction The incidence of CM has steadily increased over the last 30 years in most fair- skinned populations even if the great majority of the increase has been linked to the increase in diagnoses of thin lesions with excellent prognosis. One of the hypotheses for the discrepancy between incidence and mortality trends is that melanomas are detected at earlier stages in women than in men. Another hypothesis is that part of melanoma epidemic is made of non life-threatening melanomas that could be promoted by sun exposure. Recent sun exposure could be able to trigger skin cancers with little malignant potential. Long noted is the relationship between sun exposure and non-melanoma skin cancer (NMSC) and superficial spreading melanoma (SSM), which are often not aggressive. This last hypothesis arose from the results of a study evaluating the association of sun exposure indicators with melanoma mortality. This American study of 528 melanoma cases showed that markers of sun exposure were inversely associated with death from melanoma.In this study, we investigated if different indicators of UV exposure, collected before and after CM diagnosis, are associated with Breslow thickness and recurrence in Italy. We wanted to investigate if there could be an induction period for the most aggressive melanoma. # Material and Methods A hospital-based study of melanoma cases was initiated in 2007 at the European Institute of Oncology in Milan, Italy, which is a secondary and tertiary centre for CM diagnosis and treatment. A self-administered questionnaire was given systematically to all patients during clinics (response rate 99%) and data was collected on socio-demographic variables, season of diagnosis and CM site. Skin type was assessed with the Fitzpatrick classification. A total of 742 CM patients with histologically confirmed diagnosis of primary CM without relapse were identified via two databases: the Institutional Melanoma Database and the Tumour Registry of the European Institute of Oncology (IEO), Milan, after informed consent was obtained. Use of the data included in The IEO Tumour Registry was approved by the IEO Institutional Review Board (March 2013). Main clinical information and information on disease history of the patients are obtained from the Tumor Registry. The European Institute of Oncology is a tertiary referral centre and assessment of diagnosis, cancer characteristics and recurrence was made by highly qualified dermatologists, pathologists, surgeons and oncologists. As shown in, there were two cohorts of patients which did not overlap: cases interviewed at initial diagnosis (Group 1) and those during follow-up (Group 2), with the same questionnaire. In the latter group, median time from CM diagnosis to questionnaire was 2.6 years (1-6 years inter-quartile range). Sun exposure data for Group 1 was collected by asking CM patients if they had holidays in the 2 years preceding their melanoma diagnosis and if yes, how many weeks of holidays per year for each year. For Group 2 at follow up, they were asked the same questions for a period of 2 years before the questionnaire. We asked information on the last 2 years because this is easy to remember and usually these types of habits before diagnosis do not change. After diagnosis patients will likely change behaviour and we made the same questions for the Group 2. Data on sun exposure during peak hours (11:00AM-1:00PM) in the previous two years, sunbed exposure (current use and use before age 30) and residence of at least one year in tropical countries before age of 14 was also collected. We used ‘sunny holidays’ as indicator of high sun exposure because Italians sunbath mainly in holidays and it is a measure less prone to recall bias. In fact questions commonly used to measure sun exposure, especially those reflecting past exposures, embody attempts to recall events that are subject to both systematic recall bias and random misclassification. Recent sunny holidays will be less subject to misclassification. Furthermore recall bias characterize case-controls designs and in our study design comparing melanoma cases this information should be more reliable. Sunny holidays will be referred to holidays in the manuscript. Patients with previous CM primary, missing histology, in situ, acral melanoma, mucosal melanoma, vulvar or ano-rectal melanoma and patients with distant metastases at diagnosis were excluded (n=51, 7%), leaving 691 patients eligible for the study: 289 were interviewed at diagnosis (42%, Group 1) and 402 during follow-up (58%, Group2). For the analyses on CM recurrence, 496 patients with pT1-T3N0M0 staging were included: 225 interviewed at diagnosis (group 1) and 271 after diagnosis (group 2). For group 2 we included patients with a diagnosis of CM at the most 7 years before the questionnaire, to avoid inclusion of very long survivors. No residents in other countries were found. Questions on sun exposure refer to two years before diagnosis for group 1 and for group 2, for 2 years before the questionnaire during follow-up. In order to avoid time related bias, the competing risk analysis in group 2 started from the time of interview (free of disease). A sensitivity analysis was carried out including also long survivors. ## Statistical methods Associations between categorical variables at baseline and median Breslow thickness was evaluated by non-parametric median two sample tests. Associations between categorical variables and holidays were evaluated by the Chi-square test, Fisher exact test and the Chi-square test for trend, as appropriate. In multivariate analyses, Breslow thickness was either analysed as categorical variable or as continuous variable. Logistic models were used to evaluate the probability to have a thick CM (Breslow \>1mm) and an ANCOVA model was introduced, transforming Breslow thickness for normal distribution. All possible confounding factors were evaluated in multivariate models. For cumulative incidence, only first events of interest were considered. For the cumulative incidence of CM-related events, CM-related recurrences from a primary CM and deaths were counted as events, while second CM, NMSC and non-CM primary cancers were considered as competing events, since they may be associated to a genetic predisposition or other causes. Follow-up of patients free of events was censored at the date of last visit. Cumulative incidences were compared across groups by means of the Gray test. Multivariable Cox proportional hazards models were applied to evaluate holidays as independent prognostic factor and assumptions on proportional hazards were verified. Adjusted hazard ratios (HR) with 95% confidence intervals (CIs) were reported. All analyses were carried out with SAS software version 9.2 (SAS Institute, Cary, NC) and R software, version 2.12.2 (http://www.r-project.org). All reported p values were two-sided. # Results Six hundred ninety-one histologically confirmed primary CM, diagnosed between 1984 and 2012 were eligible for the study. Median age at diagnosis was 47 years (inter-quartile range: 37–60), 375 (54%) had a thick CM (Breslow\>1mm), 378 (55%) were women, 519 (75%) had a high level of education (high school at least) and 449 (65%) lived in Northern Italy. Seventy-three percent (n=501) had a superficial spreading melanoma (SSM) and 15% (n=101) nodular melanoma (NM). All potential factors associated with Breslow thickness are shown in. As expected men, older cases (above 60 years) and those with low levels of education had thicker CM compared to women, younger and more educated patients, respectively. CMs diagnosed by a dermatologist had a lower Breslow thickness compared to those detected by general practitioners or other specialists (median Breslow 1.0 vs.1.5 mm; p=0.002). The female frequency of thick CM was significantly lower for high education than in low education (40% vs 71%; p\<0.0001), whereas male thickness in men did not change by education. In multivariable logistic models assessing the probability of thick (\> 1mm) vs. thin (≤1mm) CM, after adjusting for age, gender, grade of clinician, BMI, family history and personal history of NMSC, the interaction between thickness and education in women remained significant (p=0.03): a 77% reduction in risk of thick CM for highly educated women (OR=0.23; 95%CI: 0.12 to 0.45) compared to low educated women, whereas this was not significant in men (OR=0.67; 95%CI: 0.33 to 1.36). Frequencies of socio-demographic, clinical features and UV exposure indicators categorised by holidays, for the 289 patients evaluated at diagnosis, are presented in. Patients with higher level of education went more often on holidays than those with low educational levels (83% vs 65%; p\<0.0001). Ulcerated CM and CM diagnosis during the summer were more common in those without holidays. Slightly younger subjects were observed in those having holidays but this did not reach significance (p=0.07). Breslow categories (pT) were associated with holidays: the proportion of very thick CM (\>4 mm Breslow thickness) was significantly lower among patients having holidays (8% for holidays vs 20% for no holidays; p for trend=0.002;). Very thick melanoma were also negatively associated with number of weeks of holidays (categorized as ‘no sunny holidays’, ‘1-2 weeks of sunny holidays per year’, and ‘\>2 weeks per year’) in a dose response manner (p for trend=0.001; ). Median Breslow thickness values are presented in for several categorical variables possibly associated with UV exposure. After adjusting for several confounding factors (age, gender, education, grade of clinician at visit, history of NMSC and season at diagnosis) holidays before diagnosis remained significantly associated with lower Breslow thickness (p=0.003). Exposure during peak hours, history of NMSC, sunbed use and CM body site, were not significantly associated with Breslow thickness in multivariate analyses, neither were skin type nor season of CM diagnosis. Frequencies of sunny holidays were significantly lower in the group evaluated during follow-up (72% vs. 62% for Group1 and Group 2 respectively; P=0.01) compared to the group at diagnosis as well as sun exposure during peak hours (44% vs. 27%; P\<0.0001), whereas sunbed exposure did not change (11% vs. 10%; p=0.28). Holidays were also significantly associated with Breslow thickness in a dose- response manner (p=0.007;). We found a significant interaction between the effect of holidays and gender (p=0.004;): women had significantly lower Breslow thickness if they had a history of holidays (p=0.0004), whereas for men this protective effect of sun exposure was not significant (p=0.88). For CM recurrence, median follow-up was 44 months (range 1-72) for group 1 and 40 months (range 2-75) for group 2. Melanoma staging was very similar according to holidays, as well as when comparing group 1 and 2. Overall, six percent of patients had a melanoma recurrence and 5% a second primary cancer. Holidays before diagnosis was not associated with risk of recurrence (p=0.64 Gray test; HR=4.19; 95%CI: 0.53 to 33.36 with p=0.18;). However, a significant trend was found when looking at holidays during follow-up. The 5-year cumulative incidence of CM recurrences was 8% for those having holidays after diagnosis compared to 17% for those without (p=0.07 Gray test;). After adjustment for gender, Breslow thickness, ulceration and grade of clinician, the beneficial impact of holidays on relapse after diagnosis was statistically significant: HR of 0.30 (95%CI: 0.10 to 0.87; p=0.03). The results were not affected by further adjustment for age, education and lag time from diagnosis to questionnaire. Median time from diagnosis to questionnaire in group 2 was 28.8 for those having holidays after diagnosis compared to 29.1 months for those not having holidays respectively. Including long survivors the difference in recurrence between exposed and unexposed is even more significant (HR=0.24, 95%CI: 0.07-0.75 and P=0.01). When restricting the analyses for those with no more than 5 years from diagnosis to questionnaire within group 2, the hazard ratio still indicated a 60% reduction in risk of recurrence, even if it lost statistical significance: HR=0.38 (95%CI: 0.12 to 1.23; p=0.12). We also found a dose-response relationship between risk of melanoma recurrence and number of weeks of holidays. The hazard ratio for up to 2 weeks or more of holidays were: HR=0.74 (95%CI: 0.16 to 3.45) and HR=0.28 (95%CI: 0.08 to 0.98), respectively, compared to patients not going on holidays. # Discussion A history of holidays in the sun for 2 years before CM diagnosis was significantly associated with lower Breslow thickness with a significant dose response in Italy. Sunny holidays after CM diagnosis were also associated with lower rates of CM recurrence. No association was found with sunbed exposure and sun exposure during peak-hours and we observed a lower frequencies of sunny holidays after melanoma diagnosis. Our results may possibly be explained by Vitamin D serum levels. However, the observed protective effect of sun exposure in relation to CM thickness reached statistical significance only among women. Vitamin D serum levels have been reported to be lower in women compared to men even in a sunny country like Italy.Thinner CM and lower recurrence rates have previously been linked to higher vitamin D serum levels.Other observational and case-control studies found a beneficial effect of vitamin D serum levels on melanoma survival.It has also been hypothesised that Vitamin D might have a beneficial influence for total mortality and incidence of different types of cancers, so this does not apply to melanoma only.The effect of sun exposure was independent from known melanoma prognostic factors as well as skin awareness or screening indicators. High educational level, that can be considered a proxy for socio-economic status, was found to be significantly associated with thin CM but only among women. A limitation of this study is the lack of information on types of occupation (e.g. indoor versus outdoor work) and socio-economic status, since education can only in part overcome this problem. In fact the reason why educated females have thinner melanomas could be that they have a different pattern of sun exposure compared to men. Male gender has already been reported to be an independent risk factor for thick CM as well as living alone. Part of the observed increase in CM incidence over the last 30 years consists in a large part of very thin CM, which could potentially be promoted by sun exposure. However this is confounded by increased awareness and screening over the last 30 years. There is a need to understand what triggers aggressive CM. Indeed, nodular melanomas are already known to be less related to sun exposure compared to superficial spreading types.It is important to notice that sunny holydays are not necessarily associated with sunbathing and and sunburns, well know risk factors for melanoma. In fact sun exposure during hot hours and residence in tropical countries in youth, that are proxy for sunbathing and sunburns more than sunny holidays, were not found to be associated with CM prognosis. Our data confirmed a greater frequency of SSM in patients going on regular holidays in the sun. However, adjusting for histology, the effect of UV exposure on CM thickness and recurrence remained. A potential bias for the effect of sun exposure on thickness could be screening bias, We took into account of this looking at grade of clinician and season of diagnosis. In fact subjects who had regular holidays in the sun before diagnosis were more prone to seek dermatological screening as in view of their sun seeking behaviour or because having naked skin in summer made them more aware of potentially malignant lesions. However frequencies of visits to dermatologists were similar in exposed and unexposed and all the analyses were adjusted for grade of clinician. Furthermore season of diagnosis, was not significantly associated with Breslow thickness and CM diagnosed during summer months were more frequent among patients with no holidays, which does not suggest a screening bias. We know that skin awareness and going on holidays are linked with education, but sunny holidays were associated with thickness and melanoma recurrence adjusting for educational status and skin awareness indicators. Another selection bias could be that patients who enjoy outdoor sun exposure are more likely to be in good health, compared to patients who are older and/or with comorbidities who may have a higher chance of relapse. However the two exposure groups have similar initial stage of disease. In order to avoid time related biases, such as immortal bias, we started the analysis from the time of the questionnaire. It is still possible that other unknown protective factors in sun seeking individuals explained the improved survival. Despite the fact that there are sources of bias that we were not able to fully address, we believe that our study shows several intriguing results that should be further investigated and are in agreement with previous epidemiological publications un sun exposure indicators and melanoma survival It is also reassuring that results on recurrence are in agreement with results at baseline: sunny holidays are inversely associated with worse prognostic factors (thickness and ulceration) and recurrence. Furthermore dose-response relationships are confirmed. Moreover, if major confounders had biased results, we should have obtained similar results with other indicators of UV exposure, such as sun exposure during peak hours and sunbed use, but it was not the case: only sunny holidays were found to be inversely associated with CM prognosis. Solar radiation is a well-established skin carcinogen so the interpretation of the protective effects of sun exposure on melanoma needs to be made with caution. These results need to be confirmed by larger studies with measurements of Vitamin D serum levels in summer and winter as well as data on Vitamin D receptor genotypes. In the meantime, there is a need to be aware of the risk of Vitamin D deficiency in melanoma patients and we recommend measuring Vitamin D levels during follow up and offer supplementation if necessary. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: SG EV AT. Performed the experiments: GT GS FB EC PFF EP FC BB AC MB. Analyzed the data: SG EB PM. Contributed reagents/materials/analysis tools: MB. Wrote the manuscript: SG EV VB EB. Interpreting the data: SG EB PM CM AJ VB.

# Introduction Marine and coastal areas are affected by human activities at sea, on land and along the coast, that cause chronic and diffuse pollution. Fisheries, shipping, agriculture, industry and waste water treatment cause contamination of seawater, sediments and biota from the coastal areas by trace elements such as metals and radionuclides, and many organic pollutants such as pesticides and Polycyclic aromatic hydrocarbons (PAHs). In this context, the European Union's Marine Strategy Framework Directive (MSFD Dir. 2008/56/EC) was adopted in June 2008. It aims to achieve “Good Environmental Status” of marine environment by 2020 and to improve the conservation status of marine biodiversity. It is therefore crucial to find reliable methods to assess and monitor the chemical contamination of coastal areas and its effects on marine ecosystems. *In situ* experiments are paramount to achieving this goal, as they complement laboratory experiments by integrating the full complexity of natural systems and the simultaneous presence of all of contaminants in the environment. The chemical assessment of different contaminants (whether in the environment or in the tissues of marine organisms) is a fundamental first step in understanding the possible impact of pollution on marine species. However, this approach should be combined with biological approaches aimed at measuring the response of organisms to environmental insults. Indeed, biological or biochemical indicators provide information on the bio-availability of contaminants and the integration of the effects of multiple exposures over time. They offer a very effective and sensitive early warning system of longer-term negative effects. For example, biomarker responses reflect defense mechanisms that might be affected by environmental contaminants. Marine bivalve mollusks are widely accepted as useful sentinels of chemical contamination in seawater because they are sedentary, filter great volumes of seawater and accumulate pollutants in high concentrations. An increasing number of biomarkers for the detection of early biological response to cellular stress response signals generated by pollution are developed in bivalves. The variegated scallop, *Mimachlamys varia*, appears as a potential sentinel species because it responds to specific biological criteria: sedentariness, ease of collection (it can be found on the intertidal zone), large distribution range (from the North Sea to the south east of the Atlantic Ocean and in Mediterranean sea) and potential of contaminants incorporation. Moreover, its relevant bioaccumulation rate of contaminants makes it a very promising species for biomonitoring studies, with the digestive gland showing higher metal concentrations than other tissues such as the adductor muscle. In that perspective, a recent *in situ* study on *M*. *varia* evaluated the responsiveness of several biomarkers that are commonly used for biomonitoring. In the present paper, we evaluate the effects of chronic chemical contamination in a marine bivalve by combining multiple approaches that provide information on individual and population health. Samples of *M*. *varia* were collected on the Atlantic coast of France from three sites exposed to inorganic contamination and compared to a reference site, which was less contaminated. The concentrations of 14 trace elements (Ag, As, Cd, Co, Cr, Cu, Fe, Mn, Ni, Pb, Se, Sn, V and Zn) were measured in the digestive glands of *M*. *varia*, since many of these metals are considered as tracers of anthropogenic inputs in coastal waters. Furthermore 45 organic contaminants were assayed in the same organ or the whole soft part of *M*. *varia*. Among them, organochlorine pesticides, such as dichlorodiphenyltrichloroethane (DDT) and its metabolites, aldrin, dieldrin and lindane are considered as POPs (Persistent Organic Pollutants), toxic, bioaccumulative and mutagenic. PAHs, including pyrene, fluoranthene and naphthalene, considered as priority pollutants and polychlorobiphenyl (PCB 153, 180) widely used in transformers or as lubricants, were also measured. Analysis of other seawater micropollutants such as radionuclides that can alter oxidative stress and bioaccumulate in marine bivalves was beyond the scope of this study. In parallel, six different biomarkers were used to monitor the modulation of oxidative stress, immune system alteration, mitochondrial respiration and general metabolism of *M*. *varia*. We relied on several biomarkers as they respond to complex mixtures of pollutants, which are expected for ecotoxicology studies conducted *in situ*. Variations in both contaminant levels and biomarker responses were analyzed as a function of the season and the location to assess the potential of *M*. *varia* as a biomonitoring species. One important consideration to keep in mind in ecological studies is ensuring that the same taxon is being investigated throughout the experiment and when comparisons are made with previous studies. A species validation step is crucial when studying natural populations *in situ*, and particularly in the marine environment where difficulties in taxonomic assignment are exacerbated. In this study, we relied on genetic tools to verify the species identity of collected scallops using a DNA barcoding approach. We also used these genetic data to assess potential long-term (*i*.*e*., intergenerational) effects of a chronic chemical contamination on *M*. *varia* by estimating genetic diversity at each sample site. Our integrative, multi-time scale approach is illustrated in. To the best of our knowledge, there is no previous information about the bioaccumulation of organic pollutants in *M*. *varia* and no previous study assessing the effects of chemical pollutants on the genetic diversity of its populations. Indeed, previous ecotoxicological studies on *M*. *varia* in this region have focused on trace metal concentration bioaccumulation processes and correlations with biomarkers. Our goal was to obtain a more comprehensive evaluation of contaminants impacting these organisms, notably by including analyses of organic pollutants, which may be particularly relevant in our sampling areas given the influence of large effluents that may carry pesticides. Thus, this study provides a novel, integrative ecotoxicological assessment by investigating the links between two types of pollutants (trace metals and organic contaminants), biomarkers, and population genetic diversity. # Materials and Methods Collection of the marine bivalve *M*. *varia* was authorized by the French Ministère de l'Ecologie, du Développement Durable et de l'Energie (Direction interrégionale de la mer Sud-Atlantique) at the four sampling sites: Loix-en-Ré, Les Palles, Port-Neuf and Les Minimes (France). These sampling sites were not private or protected. Our field study did not involve an endangered or protected species. ## 1. *In situ* sampling and preparation of biochemical material Sampling was conducted simultaneously on four sampling sites in March and September 2014 before and at the end of *M*. *varia*’s reproductive season, respectively. Sampling occurred at low tide in the infralittoral zone at four sites along the French Atlantic coast (Charente Maritime;): Loix-en-Ré (reference site), Port-Neuf (near the water treatment plant of La Rochelle), Les Minimes (near sailing harbour) and Les Palles (near the mouth of the Charente estuary). Ten to thirty scallops (adult height \>35 mm) were collected at each sampling site. summarizes how sampled individuals were either used for a single methodology or for multiple analyses. The preparation of biochemical material is described in. ## 2. Species validation We used a genetic approach to validate the species identity of collected scallops. For each of the four sites, twenty scallops collected in March and ten scallops collected in September were used for genetic analyses. Genomic DNA was extracted from approximately 15 mg of muscle tissue using the NucleoSpin^® Tissue kit (Macherey-Nagel EURL, Hoerdt, France) following the manufacturer’s protocol. Species identification was performed using a portion of the mitochondrial gene coding for cytochrome *c* oxidase I (*cox1*). This gene is commonly used in DNA barcoding studies as it is conserved across taxonomic groups and generally allows distinguishing closely related species. The methodology for PCR reactions, sequencing, and species identification methodology is given in. ## 3. Trace elements assessment Methodology for trace elements analysis was based on previous work by. Analyses of 14 trace elements (Ag, As, Cd, Co, Cr, Cu, Fe, Mn, Ni, Pb, Se, Sn, V, Zn) were realized with a Varian Vista-Pro ICP-OES and a Thermofisher Scientific XSeries 2 ICP-MS. Details on the methodology can be found in. ## 4. Organic contaminants assessment For 5 to 10 scallops per site, and for each of the two sampling periods, 21 to 23 PAHs, 14 PCBs and 8 pesticide residues were analyzed in digestive glands by stir bar sorptive extraction-thermal desorption-gas chromatography-tandem mass spectrometry (SBSE-GC-MS/MS) using a method adapted from. Details on the methodology can be found in. ## 5. Biomarkers assessment Protein assays and the methodology used for each biomarker are described in. ### 5.1. Malondialdehyde (MDA) assay Oxidative stress was evaluated through lipid peroxidation by quantifying malondialdehyde (MDA) that is a chemical metabolite of cell lipid degradation. MDA concentration was estimated in the final fraction using a commercially available MDA assay kit (Oxis International). ### 5.2. Superoxyde Dismutase (SOD) assay Superoxide dismutase (SOD) activity was measured given its important role in antioxidant response. SOD activity was assessed in the final fraction using the method developed by. ### 5.3. Laccase assay Immune system alteration was evaluated by assessing a phenoloxidase (PO) activity (laccase-type) since this enzymatic activity has been shown to be contaminant-sensitive and to play an important role in the immune defense mechanism in invertebrates. ### 5.4. Glutathion-S transferase (GST) assay Glutathion S-transferase (GST) activity was measured given its important role in contaminant excretions. Total activity of GST (GST are a group of enzymes) was determined according to the method of Sigma-kit (CS0410-1KT). ### 5.5. Citrate Synthase (CS) assay Citrate synthase activity has been extensively used as a metabolic marker in assessing oxidative, mitochondrial abundance and respiratory capacity. CS activity was determined spectrophotometrically according to the method of. ### 5.6. Evaluation of phosphatase activities As phosphatases are involved in basic biochemical processes in marine invertebrates, we screened their activity in extracts of digestive glands. Phosphatases are involved in many lytic processes associated with the work of lysosomes and underlying anabolism. ## 6. Statistical analyses Results are expressed as means ± one standard error unless otherwise specified. Statistical analyses were carried out using R v. 3.1.2 and 3.2.2. For all variables, normality was first tested on residuals using Kolmogorov-Smirnov tests and homogeneity of variances was assessed using Bartlett tests. For inorganic and organic contaminants and for phosphatase activity, values from three presumably more contaminated sites were compared to values obtained for the reference site (Loix-en-Ré) using One-Sample t-tests. For biomarkers, we conducted analyses of variance (One-way and Two-way ANOVAs) to highlight significant differences among sampling areas, sampling seasons, and interactions between these factors. If a significant difference was detected (p \< 0.05), pairwise comparisons among conditions were conducted using a HSD Tukey post-hoc test. We explored the multivariate response of biomarkers for our four sampling sites using Canonical Correspondance Analysis (CCA), in which ordination is constrained by environmental data, represented here by trace metal concentration in tissues. Stepwise model selection based on an AIC-like statistic was performed to determine which combination of trace metals optimized the fit of the CCA ordination. Individuals with missing data (n = 11) were removed from the analysis. CCA and its statistical significance (estimated using an ANOVA-like procedure based on 999 permutations) as well as model selection were performed in R using package vegan (functions cca, step and anova.cca;). ## 7. Genetic diversity assessment To estimate genetic diversity, *cox1* sequences were aligned by eye in Se-Al v. 2.0a11. Two shorter *cox1* sequences (from Les Minimes), which were used for species identification, were not included in the final alignment to estimate genetic diversity. The number of haplotypes (*H*), haplotype diversity (*Hd*), and nucleotide diversity (π) were calculated using DnaSP v. 5.10, for each sampling site separately and excluding sequence sites with missing data. Data from the two sampling seasons were pooled. A median-joining network was constructed to visualize relationships among haplotypes using PopART (available at: [http://popart.otago.ac.nz](http://popart.otago.ac.nz/)) with default settings. A one-sample t-test comparing nucleotide diversity in the reference site with nucleotide diversity observed in the three other sites was conducted in R v. 3.1.1.. Interpreting patterns of genetic diversity requires information on population structure across the geographic range that is sampled. Therefore, we tested whether individuals from the four sampling sites belong to the same panmictic population by performing an Analysis of Molecular Variance (AMOVA) in Arlequin v. 3.5.1.2. An estimator of the fixation index (*F*_ST) was calculated based on haplotype frequency differences only and statistical significance was assessed using 10,000 permutations with the null hypothesis that *F*_ST is not different from zero (*i*.*e*., there is no genetic structure). # Results and Discussion ## 1. Species validation All individuals but one (from Loix-en-Ré) successfully amplified, and good quality sequences spanning 461 to 538 bp of the gene were obtained for 118 individuals (Loix-en-Ré: 29; Minimes: 29; Les Palles: 30; Port Neuf: 30). Sequences generated for this study were deposited in Genbank (accession numbers: KU680872-83). Prior to this study, only four *cox1* sequences of variegated scallops were available in public databases. All 118 sequences obtained in this study were genetically closest (\~98% sequence identity) to the variegated scallop (*M*. *varia*) sequences already in Genbank (accession numbers EU523665-66), validating species identifications made in the field. The importance of this taxonomic verification step has been acknowledged by several ecotoxicological studies, for organisms collected in the field (*e*.*g*.,), and for toxicity studies involving test species commonly cultured in laboratories (*e*.*g*.,). Taxonomic uncertainty can be due to cryptic species (*i*.*e*., distinct evolutionary lineages lacking obvious characters distinguishing them morphologically), which, if not recognized, can confound results as they may respond differently to contaminants (*e*.*g*.,). In the present study, the low number of pairwise nucleotide differences among sequences was not suggestive of cryptic lineages. ## 2. Trace elements concentrations Trace element concentrations are presented in for both sampling dates. First, we investigated differences in concentrations between sampling periods. Differences across seasons were apparent for six heavy metals: Ag, Fe, As, Cd, Cr and Zn. Indeed, scallops of Loix-en-Ré and Les Minimes showed a higher bioaccumulation in September than in March (19% and 43% respectively) for Ag. Similar results were observed for Fe for the same sites, while the opposite pattern was detected for As, Cd, Cr and Zn bioaccumulation at all sites. For instance, a decrease in As concentration (30% for Loix-en-Ré and 53% for Les Palles) and a decrease in Cd concentration (62% for Les Minimes) were recorded. Second, we compared trace element concentrations among sites, for each sampling season (March and September). In March, Ag concentrations were significantly lower in Loix-en-Ré (reference site) than at two other sites: Les Minimes (sailing harbour) and Les Palles (near the mouth of the Charente Estuary). Concentration of seven metals (Ag, Cd, Co, Cr, Cu, and Zn) was significantly higher in scallops from Les Minimes, than in scallops from Loix-en-Ré. Higher concentrations for the three impacted sites compared with the reference site was observed for four metals: Cd, Co, Cr and Zn. Overall, metal concentrations we measured in March in Loix-en-Ré were similar to concentrations reported by at the same sampling site. In September, significant differences were less pronounced. Indeed, only Ag, Co and Cu concentrations were significantly different for two impacted sites compared with Loix-en-Ré. Concentrations of traces elements Cd, Cr, Fe and Mn showed higher values either for Les Minimes or for Port-Neuf than in the reference site. Many studies conducted on the bioaccumulation capacity of pectinids have been done in controlled condition of contamination, while our study was carried out using scallops collected *in situ*. Our results showed high bioaccumulation of Cd in the digestive gland of *M*. *varia* (25 to 81 μg/g of dry weight for March), which is consistent with previous findings on pectinids. Previous studies indicated high levels of Cd in digestive glands in contaminated areas, as well as in reference sites. Even though significant differences between the reference site and contaminated sites were not detected for all trace metals, the overall pattern that was observed did show that in Winter, most of these metals (11 out of the 14 analyzed) were present in lower concentrations at the reference site compared to one or several of the other sites. For trace elements showing higher levels in reference sites (e.g., for As), results could be explained by differences in dietary availability (e.g.,). In general, pectinids are filter feeders using different kinds of nutrients (phytoplankton, organic material and other suspended matter) with variable composition of nitrogen and carbon. Thus, differences in diet among sampling areas could also explain the different metal contents. Additionally, previous studies have highlighted an influence of biotic and abiotic factors on metal concentrations in pectinids especially with seasons: Fall and Winter are marked by high concentrations. These variations can be due to temperature, nutritional intake and sexual cycle. Bioaccumulation capacities of inorganic contaminants in bivalve mollusk organisms depend on tissue characteristics in target organs: kidney, digestive gland. Our study focused on the digestive gland, however it is important to underline an important bioaccumulation in the kidney of pectinids, mostly for essential metals. Furthermore, it is difficult to infer accumulation pathways of metals from measures in one organ. Given the role of the digestive gland in digestion, the trophic pathway may be the main pathway of accumulation for this organ. However, seawater could also play an important role in global bioaccumulation of heavy metals. An organotropism study would shed some lights on these questions but interpretations should be made with caution. ## 3. Organic contaminants A selection of 11 organic contaminants (pesticides, hydrocarbons and PCB) measured in the digestive glands of *M*. *varia* is presented in (March and September 2014): two pesticides (aldrin and dieldrin), two PCBs (PCB 153 and 180) and seven PAHs (pyrene, fluoranthene, naphthalene, acenaphthylene, acenaphthene, fluorene and anthracene). Additional compounds (six pesticides, 12 PCBs and 15 to 16 PAHs) were measured. The comparison of the two sampling periods shows that lower organic contaminants rates were found in September 2014 than in March 2014, in digestive gland of scallops. Indeed, it was possible to note in Loix-en-Ré (reference site) a decrease of 50%, 62% and 43% of the bioaccumulation in digestive gland of dieldrin, pyrene and fluoranthene compounds respectively. These results are in accordance with previous studies performed on the Atlantic coast. Indeed significant differences of PAH level accumulated in bivalves were found between Summer and Winter, with maximal levels in Winter, both in oysters and mussels collected in the Bay of Biscay and for juvenile oysters transplanted for a period of three months in different sites on the Marennes-Oléron Bay and the Gironde estuary. Concerning pesticides, dieldrin showed a significantly higher concentration in digestive gland of scallops from Les Minimes, near sailing harbor, compared to Loix-en-Ré, both in March and September, and also in Port-Neuf in September. DDE is a degradation product of DDT, whose sell is prohibited since 1972 in France. Nevertheless, both compounds are ubiquitous, as they are very persistent in the environment. 4.-4.DDE was found at the same level in Loix-en-Ré and Les Palles , although inputs from agriculture are more important in Les Palles situated near the estuary of the Charente River. This is in accordance with levels measured in oysters by the French monitoring network ROCCH (Réseau d’Observation de la Contamination Chimique du littoral), who showed equal levels for DDT and its metabolites in five different sampling areas on the Marennes-Oléron Bay and the Gironde Estuary. These organochlorine pesticides were transported along the entire coast and can now be found in equal quantities in open areas, making it impossible to find a reference site for pesticides. It has to be noted that 4.-4.DDE was found in higher quantity in scallops from Les Minimes, which constitutes a semi-open area, in which pollutants may concentrate. Concerning PCBs, PCB 180 presented a significantly higher rate in Les Minimes in March, which was not observed in September. PCB 153 showed significantly higher concentrations in digestive glands of scallops from Les Minimes compared to Loix-en-Ré in March, and in scallops from the three polluted sites (Les Minimes, Port-Neuf and Les Palles) in September. Although the use of PCB was restricted, leading to a decrease of PCB levels in recent years, values of PCB 153 found in 2014 in the digestive glands of scallops in Les Palles (17.55 and 52.91 μg/kg of dry weight) were higher than those measured by ROCCH at the same site in oysters from 2003 to 2007 (9 and 12 μg/kg of dry weight) and than values found for mussels in Arcachon Bay in 1997 (8.75 μg/kg of dry weight). This is all the more significant since mussels showed the highest bioaccumulation factor for PCB compared to razor shell and carpet shell. Our results would indicate that scallops present a high bioaccumulation potential, compared with other bivalves. This was already pointed out for metal pollutants. Regarding PAHs, Les Palles was the sampling site with the most contaminated scallops, with significantly higher concentrations in pyrene, fluoranthene, acenaphtene and anthracene compared to Loix-en-Ré, both in March (p \< 0.01) and in September (p \< 0.001). Pyrene, fluoranthene and anthracene were also measured in significantly higher concentration (p \< 0.05) in scallops from Les Minimes in both sampling periods. These two contaminated sites (Les Palles, opposite to Oléron Island and Les Minimes, near the harbour of La Rochelle) are impacted by tourism practices during September and shipping throughout the year. Finally, naphtalene was found in significantly higher concentration in scallops from Port-Neuf compared to Loix-en Ré. Pyrene and fluoranthene have the highest rates among the accumulated PAHs accumulated in digestive glands of scallops. Both of these molecules contain four aromatic rings, which classifies them as heavy PAHs. Other heavy PAHs (benzoanthracene, benzo(b)fluoranthene, benzo(k)fluoranthene, benzo(a)pyrene, benzo(e)pyrene, perylene) were found in high levels in a least one sampling site. These data are in agreement with previous results showing that other mollusks collected in the Atlantic coast, like oysters and mussels have a strong ability to accumulate more hydrophobic high molecular weight PAHs. Fluoranthene levels measured in Les Palles in the present study (135.83 ± 30 μg/kg of dry weight in September and 337.65 ± 183 μg/kg of dry weight in March) are much higher than those obtained by ROCCH between 2003 and 2007 in the same sampling site for adult oysters collected *in situ* (21 ± 0 μg/kg of dry weight in Summer and 4±2 μg/kg of dry weight in Winter). This could be due to the fact that ROCCH uses whole soft part of oyster, whereas digestive glands are used in the present study and PAHs have been shown to best store up in hepatopancreas in invertebrates. Furthermore, *M*. *varia* was found to bioaccumulate more PAHs than many other bivalves found in Charente-Maritime, including oysters (*Crassostrea gigas*) and mussels (*Mytilus edulis*). Some of the PAHs we measured in September in the digestive glands in *M*. *varia* are also much higher than those measured in the digestive gland of *M*.*edulis* at the same period of the year, in the commercial port of Brest (Brittany, France), which has a severe record of chemical contamination with concentrations of PAHs (for example, Naphtalene 2.2 ± 0.3 μg/kg of dry weight in Brest and 33.28 ± 3.57 μg/kg of dry weight in Les Minimes; Fluoranthene 14 ± 2 μg/kg of dry weight in Brest and 75.11 ± 11.29 μg/kg of dry weight in Les Minimes). Organic contaminants measurements in the digestive gland of *M*. *varia* have provided information about anthropogenic sewage in four separate sites. In the present work, the data we obtained have shown high levels of dieldrin, hydrocarbons and PCBs (153 and 180) in Port-Neuf with important differences in comparison with the reference site. This might be explained by the presence of the sewage treatment plant of La Rochelle. Recent research indicates that the major sources of PCBs came from domestic effluent and rainwater. Taken together, these different rates confirmed that Port-Neuf remained highly contaminated, which is coherent with previous results about heavy metals. The results shown in indicate that the mean total of PCB concentrations in scallops in Les Minimes during March and September were significantly higher than in the reference site (Loix-en-Ré). Moreover, measurements of pyrene and other congeners in digestive glands showed environmental forcing and harbour activity. These results might be due to the proximity of the dredging spoils of La Rochelle harbour. Concerning Les Palles near the mouth of the Charente Estuary, hydrocarbons measurement results (p \< 0.01) are unexpected and surprisingly high in comparison with pesticides low rates. Pyrene and Fluoranthene concentrations could potentially originate from other sectors around the Charente Estuary (Oléron Island, la Rochelle) with tidal currents, and in particular from Gironde Estuary. Moreover, previous research indicates that pesticides, PCBs and PAHs concentrations in sediments are higher than in the surrounding water, and this might have accounted for the observation in digestive glands of *M*. *varia*, which is a benthic species. ## 4. Biomarkers The two-way ANOVAs indicated that the overall level of response differed between the two sampling seasons for all biomarkers (p \< 0.006) except citrate synthase (F_1,69 = 0.129, p = 0.7). Additionally, there was a significant interaction between the date and site of sampling for three biomarkers: laccase (F_3,68 = 15, p \< 0.001), malondialdehyde (F_3,70 = 2.8, p = 0.045) and superoxyde dismutase (F_3,66 = 13, p \< 0.001), which signifies that differences among sites were not the same for the two seasons. Seasonal variation in biomarker levels may be due to the reproductive status of variegated scallops at time of sampling, as well as abiotic factors. Indeed, September sampling was conducted at the end of these organisms’ reproductive period, when perhaps less energy can be allocated to defense mechanisms. In some datasets (e.g., Laccase dataset, for which variance at Les Minimes was higher than at other sites), residuals deviated from normality and variances were non homogeneous. Results from one-way ANOVAs were therefore checked using Kruskall-Wallis tests, which provided the same results as the ANOVAs. Except for malondialdehyde, all biomarker levels differed among sites in March (p \< 0.03), while three biomarkers (Laccase, malondialdehyde and superoxyde dismutase) showed significant differences among sites in September (p \< 0.004). Results for each biomarker is further presented and discussed below. ### 4.1. Laccase assay In March, Laccase activity was 10.6 ± 1.7 U/mg of protein for organisms sampled in Les Minimes. When compared to this value, the three other sites presented significantly lower levels of activity for this enzyme. Conversely, Port-Neuf showed a significantly elevated activity compared to Les Minimes in September. The values obtained are in agreement with previous works in other bivalve species *Crassostrea gigas* and in scallop *M*. *varia*. Laccase proves to be a sensitive biomarker and phenoloxydase seems to play an important role in bivalve immunity when they are activated by ROS. Moreover Laccase has metal coenzymes that can be affected by heavy metals according to the same activation as SOD in short-time (part 3.4.2). ### 4.2. Superoxyde Dismutase (SOD) assay The mean value of SOD activity for March was 34.67 ± 0.89 U/mg in Loix-en-Ré. For March, scallops from Port-Neuf showed a significantly higher level of SOD activity compared with the three other sites. For the September sampling, a mean value of 55.39 ± 1.80 U/mg of protein (an increase of 37% in comparison with March) was measured in Loix-en-Ré. The activity of the enzyme significantly differed between Les Palles and the three other sites, the latter showing higher activity levels. Overall, mean SOD activity was significantly higher in September compared to March. The increase in SOD activity observed in Port-Neuf (near a water treatment plant) in March may be linked to the important role of this enzyme in inhibiting ROS production and thus, decreasing lipid peroxydation effects in cells. This finding is in accordance with different studies conducted in pectinids that showed a modulation of this enzyme activity due to inorganic contaminants and organic contaminants. This higher activity may constitute an antioxidant response (probably induced by ROS) in *M*. *varia*. ### 4.3. Malondialdehyde (MDA) assay For the first date (March), MDA content in Loix-en-Ré scallops showed an average value of 119.50 ± 13.27 μM. No significant difference in MDA content was observed among the four sites for this season. For the second sampling, lipid peroxidation showed a significant difference between Port-Neuf near water treatment of La Rochelle city and the three other sites. Indeed, MDA content in Port-Neuf scallops showed an average value of 147.0 ± 10.6 μM, representing a 63% increase compared to scallops from Loix-en- Ré (reference site). This result highlighted an increased lipid peroxidation in the scallops from one of the contaminated sites. This is consistent with the results of who also showed this phenomenon studying *Chlamys farreri* from less contaminated and contaminated sites. These results suggest that MDA in Port- Neuf, assessed in the digestive gland of *M*. *varia*, is a biomarker responding to accumulation of peroxydation product, membrane destabilization which have been related with decrease of capacities of oxidative defenses. ### 4.4. Glutathion-S transferase (GST) assay The mean GST activity for first sampling in the digestive gland of *M*. *varia* was 2546.02 ± 139.45 U/mg of protein in Loix-en-Ré scallops. Compared to this value, scallops from Les Palles near the Charente estuary showed a significantly lower level of GST activity. Furthermore, there is a significant positive difference in this enzyme’s activity between Port-Neuf and two other sites (Les Palles and Les Minimes). In September, GST activities did not show significant differences among sites. Overall, GST showed a significantly lower activity at all sites in September in comparison with March. It is important to notice that this enzyme activity showed low intra-group variability. The seasonal difference we observed for this enzyme may result from the organism’s sexual reproduction cycle and abiotic factors. A clear modulation of glutathione metabolism was mesured in mollusks collected at Loix-en-Ré, Port-Neuf, Les Minimes and Les Palles. Glutathion S-transferase (GST) plays an important role within the cellular oxidant system and in particular with the regulation pathway for detoxication of xenobiotics. A reduction in GST activity in September in digestive glands of scallops could be due to depletion levels of glutathione with an inhibition of glutathione reductase causing an imbalance of Warbour-Dickens-Horecker pathway (oxidative equilibrium). The depletion of this enzyme activity renders *M*. *varia* more susceptible to formed peroxides as also confirmed by the significant increase of malondialdehyde in Port-Neuf (see part 3.4.3). ### 4.5. Citrate synthase (CS) assay CS is one of the key regulatory enzymes in the energy-generating metabolic pathway that catalyzes the condensation of oxaloacetate and acetyl coenzyme A to form citrate in the citric acid cycle. Variable changes in CS activity indicated the possible alterations in oxidative capacity. The CS activity for both sampling seasons in the muscles of *M*. *varia* is presented in. In Loix-en-Ré, mean value was 0.024 ± 0.003 U/mg of protein in March. The activity of the enzyme did significantly decrease for the Port-Neuf site near waste-water compared to the reference site of Loix-en-Ré. However, scallops from all contaminated sites, that is, Les Minimes, Port-Neuf and Les Palles showed 12.5%, 20.0%, and 17% significantly lower specific activities when compared to Loix-en- Ré, respectively. For the second sampling in September, there was no significant difference among sites. High variation in glycogen levels during storage and gametogenic development suggests that carbohydrates are the main respiratory substrate. Outside this metabolically active period, bivalves are probably less sensitive to stress. Responses to environmental stress (temperature, salinity, pollutants…) in marine organisms induce changes in energy metabolism and oxidative stress. In these reviews, Tomanek suggests that metabolic pathways are activated to respond to increased oxydative stress or to switch metabolic fuels. The ability to control metabolic processes in response to changes is indispensable. This adaptability is necessary for conserving the stability of the intracellular environment (homeostasis), and essential for maintaining an efficient functional state. Metabolic control is accomplished by altering the activity of at least one pacemaker enzyme (or rate-determining step) of the pathway. In our study, the decrease in CS activity at impacted sites suggests a limitation of the Krebs cycle to the benefit of another pathway. CS stands as a pace-making enzyme in the first step of this metabolic pathway. ### 4.6. Phosphatase activity Screening of phosphatase activity in the digestive gland of *M*. *varia* was conducted for the two seasons and for the four sampling areas. In the digestive system of these animals, the specific activity of phosphatase was 5–13 times higher in Les Minimes (0.324 μmol/mL/min/μg) than in the other sampling sites in March (one sample t-test: t₂ = -23,8, p = 0.002). In September, specific activity of phosphatase was higher than those measured in March (\> 0.28 μmol/mL/min/μg for all sites) but we found no difference among sampling sites. Similarly to laccase activity, a high value of specific activity was recorded in March and in Les Minimes corresponding to a pre-reproductive period for animals. Phosphatases are known to be implicated in regulation of metabolic pathways such as gluconeogenesis and glycogenesis, and are sensitive to heavy metals. Therefore, we suggest that *in situ* contamination of Les Minimes by heavy metals like Cd may modify the capacity of scallops to produce glucose and glycogen, the primary carbohydrate stored in the liver and muscle cells of animals. Overall, variations in both phosphatase and citrate synthase activities suggested metabolic changes in *M*. *varia*. ## 5. Relationship among biomarker responses and contaminants CCA were run on each sampling season separately prior to attempting analyses pooling the two sampling dates. For the month of September, constrained analyses had a poor fit and were therefore not pursued, and consequently an analysis pooling both sampling seasons was not attempted. For the month of March, a model including Cd, As, Pb and Cr was selected. The ANOVA-like function of the R package vegan suggested that this constrained analysis was statistically significant (anova.cca function; 999 permutations, p \< 0.001). This model, explaining 44% of the variance in the dataset (axis 1: 40%; axis 2: 4%), suggested that Cd was the most structuring factor of the environmental matrix (anova.cca permutation test: p \< 0.001), and most particularly the variation in the Laccase and MDA biomarkers. Indeed, both Laccase and MDA levels were positively correlated with Cd concentrations in the tissues of *M*. *varia* (linear models: R²: 18%, F_1,36 = 9.0, p = 0.005 for LAC and R²: 12%, F_1,36 = 5.9, p = 0.021 for MDA), although variance in the MDA dataset was higher than in the Laccase dataset. Loix-en-Ré and Les Minimes displayed the lowest and the highest biomarkers and Cd concentrations, respectively. Enzymatic activities of Laccase and MDA are involved in immune response and lipidic peroxidation, respectively. Thus, lipidic peroxidation, which acts on the plasma membranes by destabilizing fatty acid poly-insaturated, could be notably due to accumulation of Cd. Our results are in accordance with, which reported the induction of Laccase activity in harbour areas. An increase in Laccase activity reflects the synthesis of melanins from phenolic substrates, which plays a decisive role of cellular response to metal and in particular to Cd in *M*. *varia*. Contrary to inorganic pollutants, it was impossible to explore the multivariate response of biomarkers using CCA with ordination constrained by organic pollutants concentrations in digestive glands. Indeed, whole digestive glands were used for organic contaminant measurements to maximize the sensitivity of the methodology, and biomarkers were assayed in other individuals. Therefore we can only highlight trends of simultaneous modulation of biomarkers and organic contaminants. Our study of organic contaminants in digestive glands of *M*. *varia* showed, for the first time, a high bioaccumulation capacity of this marine organism for this type of contaminants. In parallel, significant changes were observed in activities of enzymes involved in the modulation of oxidative stress (GST, SOD), immune system alteration (Laccase), mitochondrial respiration (CS) and general metabolism (phosphatase) in scallops collected from impacted sites compared with less exposed scallops (data from March). This is in agreement with the existence of some chemical pressure in Les Minimes and Port-Neuf harbors and in Les Palles. This confirms previous results obtained in Marennes-Oleron Bay and Gironde Estuary, in juvenile oysters of *C*. *gigas* transplanted in contaminated areas. Indeed, higher levels for SOD, MDA, and Laccase were measured in winter in correlation with high levels of organic contaminants. Regarding enzymes involved in the modulation of oxidative stress (GST, SOD), high GST and SOD activities were found in March in Les Minimes and Port-Neuf, where high levels of PAHs were found in the digestive glands. This suggests an enhancement of xenobiotic biotransformation processes and of antioxidant protection. This is in accordance with two studies conducted by who showed an increase of GST activity in the digestive glands of pectinids contaminated with PAHs. Laccase activity was high in Les Minimes in March, in comparison with the reference site (Loix-en-Ré). In addition to the potential response of Laccase activity to higher Cd concentrations mentioned above, this result could be linked to a modulation of enzymatic activity by organic contaminants (hydrocarbons and PCBs). A decline of citrate synthase (CS) activity and an increase in phosphatase activity in a contaminated site (Port-Neuf) compared with a reference site (Loix-en-Ré) was also found in March. These enzyme activity levels could be linked with high PAHs rates in the fishing harbor and other contaminants coming from the water treatment plant. Our results showed potential links between modulations of biomarkers (SOD, MDA, GST, Laccase, CS) and chronic chemical contamination *in situ*. Although we identified trends between biomarkers and different classes of contaminants, it is important to note other parameters might have some effect on this biological response. Several abiotic factors such as oxygen level, temperature, pH and environment dilution could modulate enzymatic response in bivalves. ## 6. Genetic diversity assessment The final alignment used for genetic diversity assessments was 477 bp including seven sites with missing data that were excluded from analyses. The alignment included 12 variable sites that were all synonymous substitutions. A total of 12 haplotypes were found within the dataset, including one central haplotype that was the most common at all sites. The greatest genetic diversity was observed in Loix-en-Ré (the reference site for inorganic contaminants), particularly in terms of nucleotide diversity. Conversely, the lowest genetic diversity was measured in Les Minimes. The difference in nucleotide diversity observed between the reference site (Loix-en-Ré) and the other three sites was statistically significant (one sample t-test: t₂ = -5.3, p = 0.03). Measuring the consequences of chemical pollution on the genetic diversity of natural populations has been recognized as an important aspect in impact assessments. However, only a limited number of ecotoxicological studies include a genetic component (but see for example:). Chemical contaminants can affect genetic diversity by increasing the mutation rate, by causing a population size reduction (bottleneck), or through directional selection favoring some genotypes. To properly interpret patterns of genetic diversity, we investigated genetic differentiation among sampling sites. The haplotype network did not reveal any genetic structure. This was confirmed by results from the AMOVA: no significant restriction of gene flow was observed among the four sites (*F*_ST = -0.007 p = 0.7). Thus, all sampled individuals belong to the same population and differences in genetic diversity among sites are not due to genetic drift or differences in effective population size. Given the short distances observed among haplotypes, we can also rule out that the presence of cryptic species could be driving genetic diversity differences. On the contrary, lower genetic diversity in more contaminated sites may result from the effect of selective pressure eliminating less resistant genotypes (a process that is plausible for mitochondrial coding genes) coupled with high larval mortality due to chemical pollution (which has been reported in marine bivalves:). It is important to note that a direct causal relationship between contamination and genetic diversity cannot be established here, but the significant relationship we observed is consistent with findings from previous studies (*e*.*g*.,). Moreover, an implication of the same processes (high mortality and selection) has been suggested in other invertebrate models (*e*.*g*.,). The consequences of our findings in terms of evolutionary potential and persistence of this population is unclear, particularly in a marine bivalve with presumably large effective population sizes. Nonetheless, a decrease in genetic diversity linked to anthropogenic impacts is generally seen as a negative population response, all the more if these impacts are expected to intensify in the near future. Evaluating the effects of chemical pollution on gene expression levels in this model would be an interesting next step, particularly using a transcriptomic approach. The promise of ‘-omic’ methodologies for the study of immune response in marine bivalves has been recently highlighted by. # Conclusions Assessing the effects of chemical contamination on natural populations collected *in situ* is complex and requires the integration of i) multiple indicators of individual and population health and stress response, ii) different types of contaminants that may be present in the environment. Our integrative study applied to natural populations of variegated scallops underlines how different locations on the infralittoral zone can display significantly distinct contamination profiles despite being geographically close. By combining analysis of organic and inorganic pollutants, we showed that locations that are considered reference (i.e., less contaminated) sites for certain contaminants (here, trace elements), may be impacted by other types of contaminants. This finding has important implications for biomonitoring studies as the organisms’ responses will reflect the effects of the mixture of all contaminants present, referred to as cocktail effects. Furthermore, including multiple sampling time points is crucial, as pointed out by the differences in contamination profiles and levels of biomarker responses we measured for March and September. These differences are likely due to the reproductive status of the organisms at time of sampling, which influences the energy that can be allocated to defense mechanisms during gametogenesis. Additionally, environmental conditions such as temperature, pH, oxygen levels and food availability may play a role in the seasonal differences we observed. Our measurements of biomarkers and phosphatase activities revealed different physiological responses among sites, including the modulation of citrate synthase, a biomarker that was evaluated in *M*. *varia* for the first time. In terms of stress response, multivariate analyses of inorganic contaminants and biomarkers revealed a significant link between the activity of the laccase-type enzymes, and Cd contamination. Additionally, MDA content was also correlated Cd concentrations. Finally, we observed a significant decrease in genetic diversity at sites impacted by heavy metal pollution, suggesting long-term (intergenerational) effects of this chronic contamination at the population level. Accumulation rates of contaminants in benthic organisms depends on various factors, such as exposure time, sediment nature, uptake and desorption by organisms and behaviour of bivalves during exposure. In contrast with the murine model, the fate of chemical organic contaminants in scallops has been poorly studied at this day. Given its bioaccumulation rates of both inorganic and organic contaminants, the variegated scallop, *M*. *varia*, appears as a good model for biomonitoring studies in the marine environment. Since *M*. *varia* is not farmed, it provides a better reflection of the natural environment than cultivated bivalves, and patterns of genetic diversity are more easily interpreted. Furthermore, monitoring contaminant levels in variegated scallops collected *in situ* is important for health and economic reasons as it is harvested for human consumption. Owing to the important bioaccumulation potential of the species, it will be interesting to elaborate a protocol for measuring both organic contaminants and biomarker responses on the same individuals, as it is already the case for inorganic contaminants. Our study provides novel baseline information on levels of organic contaminants and population genetic diversity for this bivalve on the way to becoming a sentinel species. # Supporting Information The authors want to thank people involved in sampling: Antoine Bonnet, Paco Bustamante, Armelle Combaud, Thierry Guyot, Nicolas Lachaussée, Julie Lucas, Nathalie Imbert, Philippe Pineau; Le Cedre of Brest for performing the biochemical assays, and Benoît Simon-Bouhet for his scientific advice and help in statistical analyses. We would like to acknowledge Pierre-Louis Stenger for conducting preliminary genetic data analyses, Rachel Durand, Marion Morgen- Thaler, Thomas Garcia and Claire Toucheteau for performing some biomarker assays. DNA samples were prepared at the Molecular Core Facility at the University of La Rochelle. Metal concentrations were measured at the Trace Element Facility at the University of La Rochelle. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AV MG LM VB IFA VH EP SLF HTG. Performed the experiments: MB AV MG LM VB CC IFA VH CL EP SLF HTG. Analyzed the data: MB AV LM EP. Contributed reagents/materials/analysis tools: MG LM VB CC IFA SLF HTG. Wrote the paper: MB AV MG LM VB CC IFA CL EP HTG. Performed field collections and/or scallop dissections: MB AV MG LM VB CC IFA VH CL EP SLF HTG. [^3]: ‡ These authors also contributed equally to this work

# Introduction Gorgonian corals (subclass Octocorallia) are abundant and important members of coral reef communities throughout the Caribbean. Unlike the dramatic decline of scleractinian corals in the Caribbean, gorgonian coral abundance is steady or even increasing. For example, in the Florida Keys, gorgonian octocoral abundance increased significantly since 1999. And, in the US Virgin Islands, the abundance of two of three gorgonian species studied has increased since 1992. Gorgonian corals are also abundant in the Yucatan coast of México, where gorgonian species richness can exceed scleractinian coral species richness. Despite being prominent members of Caribbean reef communities, studies on gorgonian coral physiology are sparse, predominantly focusing on the gorgonian coral hosts without addressing their symbiosis with the unicellular dinoflagellates in the genus *Symbiodinium*. Studies tracked digested material, described sclerite (microscopic skeletal elements) formation, or measured growth, and feeding rates. In addition, gorgonian secondary metabolites have been extensively studied due to their medical and economic importance. Two early studies measured the photosynthetic rates of *Symbiodinium* in a few Caribbean gorgonian species, while key photosynthetic characteristics, including light absorption efficiencies, have not been measured. Furthermore, the handful of studies that investigated the physiology of Caribbean gorgonian corals and their Symbiodinium – did not identify the *Symbiodinium* present. *Symbiodinium* are currently divided into nine phylogenetic clades, A-I , although clade E may represent a single species. Within each clade, *Symbiodinium* are often distinguished using sequences of the internal transcribed spacer regions of ribosomal DNA (*Symbiodinium* types sensu). Almost all Caribbean gorgonian species associate with *Symbiodinium* clade B types and many harbor type B1. Within type B1, multiple lineages have been identified. Hosting different *Symbiodinium* types can correlate with ecological, and physiological differences between cnidarian hosts. Often, however, physiological differences are assessed when the cnidarians face stressful environmental conditions. Collecting baseline physiological data on coral-algal symbioses, and not just data when the symbioses are stressed, are critical to evaluating the effects of stressors on symbioses. The objective of this study was to characterize the photosynthesis of Symbiodinium, in hospite and in isolation, in four common Caribbean gorgonian species: *Pterogorgia anceps*, Eunicea tourneforti, *Pseudoplexaura porosa*, and *Pseudoplexaura wagenaari*. Studying the physiology of the symbiosis between gorgonian corals and *Symbiodinium* may shed light on why gorgonian corals dominate Caribbean reefs while scleractinian coral abundance is declining. # Materials and Methods ## Sample collection and acclimation In June 2010, gorgonian branches were collected at 3 m depth from a patch reef near Puerto Morelos, Quintana Roo, México (20° 52' N, 86° 51' W, permit number DGOPA.11519,121109.3949, Secretaria de Medio Ambiente y Recursos Naturales, México). One branch was removed from each sampled colony of the four gorgonian species: Pterogorgia anceps (n = 9), Eunicea tourneforti (n = 9), Pseudoplexaura porosa (n = 9), and Pseudoplexaura wagenaari (n = 6). In order to maintain their natural orientation, the branches were attached to vertical PVC stands and were held in outdoor aquaria with flowing seawater for 11 days for acclimation. The temperature in the aquaria was determined using Hobo pendant data loggers (Onset Computer Corporation, MA, USA), and it was maintained at 29.5±0.5°C, similar to the ambient temperature on the reef, by using an aquarium chiller (0.5hp Delta Star, Aqua Logic, USA) and heaters (1000 W and 1800 W EasyPlug heater, Process Technologies, USA). Light levels in the aquaria were maintained at levels similar to those at the collection site (∼900 µmol quanta m⁻² s⁻¹ at solar noon) by shading the aquaria with window screening. ## Photochemical efficiency of photosystem II in Symbiodinium Throughout the acclimation period, the photochemical efficiency of photosystem II of the Symbiodinium in each branch was measured using a diving pulse amplitude modulated (PAM) fluorometer (WALZ, Effeltrich, Germany) at solar noon (Δ*F/F_m'*; effective yield) and after sunset (*F_v/F_m*; maximum yield). Photochemical efficiency was measured approximately 2 cm below the branch tip, on the side of the branch, at a constant distance from the surface of the branch. Using the effective and maximum yield values obtained for each species, we calculated the maximum excitation pressure over photosystem II, *Q_m,* whereby *Q_m* = 1-((Δ*F/F_m'*)/(*F_v/F_m*)). ## Oxygen flux of *Symbiodinium* in gorgonian branches After 11 days of acclimation, the oxygen fluxes of the branches were measured using Clark-type oxygen electrodes in twin 0.5 l acrylic chambers. The electrodes were connected to a 782 Oxygen Meter (Strathkelvin Instruments Ltd., North Lanarkshire, Scotland). The chambers were filled with 0.45 µm filtered seawater with 4 mM sodium bicarbonate. Water was circulated inside each chamber using a small pump. Electrodes were calibrated by bubbling oxygen and nitrogen gas to define 100% and 0% oxygen concentrations, respectively. A water jacket, connected to a water recirculator, maintained the temperature of 29.0–29.5°C inside the chambers. To measure the oxygen flux of a gorgonian branch, a branch was sealed in a chamber. Following a 20 min acclimation, respiration was measured for 10 min in darkness. Then, the branch was illuminated from one direction with three 6 W LED light bulbs, while white acrylic on the sides and rear of the chamber reflected the light within the chamber. The branch was exposed to 11 irradiance levels (0–2200 µmol quanta m⁻² s⁻¹) by progressively removing sheets of window screening. The irradiance levels in each chamber were measured using a 4π quantum sensor (WALZ, Effeltrich, Germany). A gorgonian branch was exposed to each irradiance level for up to 15 min, or until a linear change in oxygen concentration was recorded in the chamber. Following the last light level, respiration was measured again in darkness. Oxygen flux measurements required 2.5 h per sample and were collected at approximately 09:00 or 12:00 local time. Due to the lengthy data collection time, it took 10 days to measure the oxygen fluxes of all the gorgonian branches. To account for changes in photosynthetic rate at different times of day, the photosynthesis of Symbiodinium in each gorgonian species was measured at alternate times on consecutive days. We then used the oxygen flux data to produce photosynthesis versus irradiance (P-E) graphs (see below). After measuring the oxygen flux, the surface areas of the gorgonian branches were calculated. For gorgonian species with cylindrical branches (E. tourneforti, P. porosa, and P. wagenaari), we measured the length and diameter of each branch and calculated the surface area of a cylinder. For P. anceps, whose branches are blade like, we measured the diameter and length of the blade and calculated the surface area by combining the surface area of all rectangular sides. ## Isolation and oxygen flux of isolated *Symbiodinium* Branches were returned to shaded aquaria for 1 h before Symbiodinium isolation. *Symbiodinium* were isolated by homogenizing, in a mortar and pestle, a 2 cm section of the gorgonian branch, obtained 3 cm below the tip of the branch. The homogenate was diluted in 20 ml of 0.45 µm filtered seawater and centrifuged for 1 min at 500 rpm. The liquid fraction was filtered through 150, 74, and 20 µm nitex meshes. The filtered fraction was then centrifuged for 1 min at 3500 rpm and the Symbiodinium pellet was washed two times with 10 ml of filtered seawater. Following the last wash, the Symbiodinium pellet was resuspended in 5 ml of filtered seawater and divided into aliquots for oxygen flux measurements (0.5 ml), chlorophyll content (1.5 ml), cell density (0.25 ml), and genetic identification (2 ml). To measure isolated Symbiodinium oxygen flux, 0.5 ml of the Symbiodinium slurry was mixed with 2.5 ml filtered seawater containing 4 mM NaHCO₃ and loaded into a water-jacketed glass cell respirometry chamber (StrathKelvin Instruments Ltd). The Symbiodinium samples were stirred with magnetic stir bars and maintained at 29°C with a water recirculator. Oxygen flux measurements were recorded in two chambers simultaneously using Clark-type oxygen electrodes. Respiration was measured in darkness for 10 min before and after a series of irradiance levels. The isolated Symbiodinium were exposed to 13 increasing irradiance levels (0–2200 µmol quanta m⁻² s⁻¹) for 10 minutes per level or until a linear change in oxygen concentration was observed in the chamber. For each sample, the irradiance levels were measured using a 4π quantum sensor (WALZ, Effeltrich, Germany). Oxygen fluxes of isolated Symbiodinium were measured after those of the intact branches, at approximately 15:00 or 16:00 local time. ## Calculation of photosynthetic rates Net and gross rates of oxygen flux were plotted against irradiance to generate photosynthesis versus irradiance (P-E) graphs for both *in hospite* and isolated Symbiodinium. Coefficients from P-E curves (P_s, the maximum photosynthetic rate in the absence of photoinhibition; α, the initial slope of the curve; and β the photoinhibition coefficient) were determined for each sample by fitting the equation of Platt et al. using the nlsList function in the nlme package of the R statistical software. Photosynthetic rates were standardized to the surface area of the gorgonian branch, the total number of Symbiodinium, and the total amount of chlorophyll a. α was standardized to the amount of chlorophyll *a*. To calculate the diurnal balance between gross photosynthesis and respiration, we integrated the P-E curve over the irradiance for a 24 h period. ## *Symbiodinium* density Symbiodinium density in a gorgonian branch was estimated from averaging four replicate hemocytometer counts (0.4 mm³ each) of the Symbiodinium cell density aliquot. Oxygen fluxes of gorgonian branches were standardized to the total number of Symbiodinium cells in a branch, which was estimated using the density of cells in the homogenized piece (cells cm⁻²) multiplied by the surface area of the entire branch. For isolated Symbiodinium, the number of cells in the respirometry chamber was estimated using the cell density determined from the Symbiodinium cell density aliquot. ## Chlorophyll content in Symbiodinium For chlorophyll quantification, the *Symbiodinium* cells in the 1.5 ml aliquot of the *Symbiodinium* slurry were pelleted, and the supernatant was decanted. Chlorophylls were extracted from the *Symbiodinium* cells by adding 950 µl of cold 100% acetone and 50 µl DMSO. After 24 h of extraction at −20°C in the dark, the absorbance of the extract was measured at 630, 660, and 750 nm using an ELYPTICA model ELy-2000 spectrophotometer. Absorbance at 750 nm was subtracted from the absorbance at 630 nm and 660 nm for each sample and chlorophylls a and c₂ were estimated using the equations of Jeffrey and Humphrey. Chlorophyll concentrations were standardized to surface area (µg chl cm⁻²) and to cell density (pg chl cell⁻¹). Oxygen flux of gorgonian branches was standardized to the total amount of chlorophyll *a* in the gorgonian branches, which was calculated using the concentration of chlorophyll per surface area of the homogenized piece multiplied by the surface area of the entire branch. For isolated Symbiodinium, the amount of chlorophyll *a* in the respirometry chamber was estimated using the chlorophyll *a* concentration obtained from the 1.5 ml chlorophyll content aliquot. ## Estimated absorbance and chlorophyll *a* specific absorption Following the oxygen flux measurements, the reflectance spectrum (400–750 nm) of each branch was measured using an Ocean Optics USB 4000 fiber optic spectrophotometer. The fiberoptic cable was held at a 45° angle above a branch, which was illuminated on all sides to produce a homogeneous light field (designed by T. Scheufen, UNAM). A dried gorgonian branch, painted with white acrylic paint, was used to correct for light scattered by the surface of the gorgonian branch. The painted branch reflected ∼90% of PAR compared to a similarly shaped object wrapped in Teflon. Surface-corrected reflectance spectra were standardized to the reflectance value at 750 nm. Estimated absorbance spectra (D_e) were calculated as the negative log of the corrected reflectance. Chlorophyll *a* specific absorption (*a*\*_{chl *a*}), was calculated using the equation *a*\*_{chl *a*} = (D_{e 675}/ρ) x ln(10), where ρ is mg m⁻² of chlorophyll *a*. ## Genetic identification of *Symbiodinium* For each gorgonian branch, 2 ml of the isolated Symbiodinium slurry was centrifuged at 10,000 rpm for 1 min to pellet the Symbiodinium cells. The supernatant was removed and replaced with 2 ml 100% EtOH to preserve the Symbiodinium. DNA was extracted from an aliquot of this solution using the Qiagen DNeasy Plant Mini kit. The internal transcribed spacer 2 (ITS2) region of ribosomal DNA was amplified using the primers ITSintfor2, ITS-Reverse, and ITS2CLAMP following the protocol of LaJeunesse. The PCR products were separated using denaturing gradient gel electrophoresis (DGGE) with a 45–80% denaturing gradient and run at 120 V for 13 h. Dominant ITS2 sequence variants in unique profiles were excised, re-amplified using ITSintfor2 and ITS-Reverse, and then sequenced using an Applied Biosystems 3730XL automated sequencer at the DNA Laboratory at Arizona State University. Since we found that at least one colony of each of the four gorgonian species harbored *Symbiodinium* type B1, we determined whether colonies of the four different species harbored the same B1 lineage. Microsatellite flanking regions from B1 *Symbiodinium* were PCR amplified with the primers B7Sym15 and CA4.86 and directly sequenced using an Applied Biosystems 3730XL automated sequencer. The flanking sequences were concatenated and aligned with 16 published *Symbiodinium* type B1 sequences from Supplementary tables S1 in. Six samples, five from our study (1 *P. anceps*, 2 *P. porosa*, 2 *P. wagenaari*) and one published (B1 lineage 1.4), were only represented by B7Sym15 sequences. For each flanking region, the Jukes Cantor model of sequence evolution was chosen using AIC scores from jModelTest (2.1.4), and substitution rates were assumed to follow a gamma distribution with four categories. In order to determine the phylogenetic relationships among B1 *Symbiodinium*, a Bayesian phylogenetic tree was generated using the MrBayes (3.2.2). Two sets of four independent chains were run for 1,000,000 generations, but the model converged after 916,000 generations. Trees were sampled every 100 generations (916 trees per run) and the first 25% of trees were discarded as burn-in. The average standard deviation of split frequencies was less than 0.01. To confirm the topology generated by MrBayes, 100 bootstrap replicates of a maximum-likelihood phylogenetic tree were generated in Garli 2.0 using the Jukes Cantor model of sequence evolution. *Symbiodinium* type B19, from supplementary table S1 in was used as an outgroup for both analyses. ## Statistical methods In a given host species, different host-symbiont combinations can differ in their physiology. Therefore, we excluded from statistical analyses gorgonian-*Symbiodinium* ITS2 type combinations represented by three or fewer colonies. The sample sizes given in the results reflect the number of colonies used in the statistical analyses. Each parameter was analyzed using a one-way ANOVA with gorgonian species as a factor. Residuals for most parameters were not normally distributed and/or had unequal variance among species. Therefore, most data required transformation using a log, square root, or reciprocal function. When significant differences were found among gorgonian species, Tukey HSD post hoc tests were used to identify significant differences among all pairwise species combinations using Bonferonni-corrected p-values. *a*\*_{chl *a*} data could not be transformed to meet the assumptions of ANOVA and was tested by determining the frequency of obtaining an F statistic greater than or equal to the observed F statistic in 10,000 permutations of the data. # Results ## Photochemical efficiency of photosystem II in Symbiodinium Measuring the effective (Δ*F/F_m'*) and maximum (*F_v/F_m*) quantum yield of PSII throughout the acclimation period enabled us to calculate maximum excitation pressure over photosystem II, *Q_m. Q_m* ranged from 0.25 in *P. wagenaari* to 0.32 in *P. anceps* and did not significantly differ between the four species. The *Q_m* values demonstrated that there was no detrimental tank effect on the gorgonians. ## Oxygen flux in gorgonian branches P-E curves for gorgonian branches did not reach saturation despite being exposed to more than 1800 µmol quanta m⁻² s⁻¹. Therefore, the maximum photosynthetic rate could not be determined as in Platt et al.. Since 1800 µmol quanta m⁻² s⁻¹ was the highest irradiance shared between the two respirometric chambers, the fitted values at 1800 µmol quanta m⁻² s⁻¹ were used as a proxy for the maximum photosynthetic rates. P. porosa and P. wagenaari had the highest average photosynthetic rates per cm² at 1800 µmol quanta m⁻² s⁻¹, with both species having similar gross photosynthetic and respiration rates per cm², although *P*. porosa had significantly higher net photosynthetic rates per cm² than P. wagenaari. *P. anceps* had significantly higher respiration rates per cm² than *E. tourneforti*, but the two species did not differ significantly in their maximum photosynthetic rates per cm². When integrated on a diurnal cycle, the total oxygen produced via photosynthesis (10.3 to 25.9 µmol O₂) was less than the oxygen consumed via respiration (13.4 to 29.7 µmol O₂) for each gorgonian species and therefore the 24 h gross P/R were less than 1. The gross and net photosynthetic rates per *Symbiodinium* cell exhibited a different pattern than the photosynthetic rates per cm². *P. anceps* had significantly higher gross and net photosynthetic rates per cell than all other species. *P. porosa* had the second highest average photosynthetic rates per cell, which were significantly higher than those of *P. wagenaari*, but not *E. tourneforti*. Photosynthetic rates per cell of *E. tourneforti* were similar to both *P. porosa* and *P. wagenaari*. Photosynthetic rates per chlorophyll *a* produced a similar pattern as photosynthetic rates per cell. *P. anceps* had the highest average photosynthetic rates per chlorophyll *a* (gross and net), significantly higher than the two *Pseudoplexaura* species, but similar to *E. tourneforti*. *P. wagenaari* had significantly lower rates than *P*. *porosa*. Differences in the initial slope of the P-E curve, α, mirrored differences in photosynthetic rates per chlorophyll *a*, as α was significantly greater for *P. anceps* than all other species. α of *E. tourneforti* was significantly greater than in both *Pseudoplexaura* species, which had statistically similar α. ## Oxygen flux of isolated *Symbiodinium* Photosynthetic rates of isolated Symbiodinium were lower than the in hospite photosynthetic rates except for *Symbiodinium* from E. tourneforti. With the exception of Symbiodinium from E. tourneforti, isolated Symbiodinium had gross P_max/R less than or equal to one. The initial slopes of the P-E curves (α) were greater for isolated Symbiodinium than for Symbiodinium within gorgonian branches. ## *Symbiodinium* density P. wagenaari had the highest Symbiodinium densities and these differed significantly from the Symbiodinium densities in *P. anceps* and *E. tourneforti*. *P. anceps* had the lowest *Symbiodinium* densities. ## Chlorophyll content *P.* porosa and P. wagenaari had significantly more chlorophyll a and *c*₂ per cm² than P. anceps and E. tourneforti. On the other hand, P. porosa, P. wagenaari, and P. anceps had statistically similar chlorophyll a per Symbiodinium cell, with Symbiodinium in E. tourneforti having significantly lower chlorophyll a per cell than all other Symbiodinium. While chlorophyll *c*₂ per Symbiodinium cell exhibited a similar pattern to chlorophyll *a* per cell, the statistical results were slightly different: Symbiodinium in *P. porosa* had significantly greater chlorophyll *c*₂ per cell than those in *P. anceps*. *Symbiodinium* in *E. tourneforti* had lower chlorophyll *c*₂ per cell than *Symbiodinium* in both *Pseudoplexuara* species, but similar chlorophyll *c*₂ per cell as those in *P. anceps*. *Symbiodinium* in P. porosa and P. wagenaari had statistically similar ratio of chlorophyll a to chlorophyll c₂ to those in *P. anceps* and *E. tourneforti*, but *Symbiodinium* in *E. tourneforti* had a significantly lower ratio of chlorophyll a to chlorophyll c₂ than those in P. anceps. ## Chlorophyll *a* specific absorption P. porosa had the highest estimated absorbance at 675 nm (D_{c 675}), which was significantly greater than that in *P. anceps* and E. tourneforti, but similar to *P. wagenaari*. Consistent with chlorophyll density data, chlorophyll *a* specific absorption (*a*\*_{chl *a*}), was significantly higher in P. anceps and E. tourneforti than in the two Pseudoplexaura species. P. wagenaari had significantly lower *a*\*_{chl *a*} than all other species. ## Genetic identification of *Symbiodinium* All P. anceps colonies harbored B1 Symbiodinium, matching accession AF333511. Eight of the nine E. tourneforti colonies sampled harbored type B1L, matching accession GU907639, while one colony harbored type B1 matching accession AF333511. Six P. porosa colonies harbored Symbiodinium B1i matching accession GU907636. The remaining three P. porosa colonies harbored one of two distinct DGGE profiles with dominant ITS2 sequences identical to type B1. All P. wagenaari colonies harbored type B1 Symbiodinium, albeit with a distinct DGGE profile from the B1 symbiont in P. anceps as discussed in. Since at least one colony from all the four studied gorgonian species hosted *Symbiodinium* type B1, we compared these B1 *Symbiodinium* amongst the four gorgonian species in our study and to published B1 sequences from other cnidarians. Analysis of microsatellite flanking region sequences revealed that *Symbiodinium* B1 from 16 of 19 gorgonian colonies, representing all four gorgonian species, formed a phylogenetic group with high posterior probability. The B1 *Symbiodinium* in this group included *Symbiodinium* from eight of nine *P. anceps* colonies, one colony each of *E. tourneforti* and *P. porosa* and all six *P. wagenaari* colonies. B1 *Symbiodinium* from three gorgonian colonies (from *P. anceps* and *P. porosa*) placed outside this group, but did not cluster with the examined *Symbiodinium* B1 lineages from scleractinian corals. # Discussion Gorgonian corals blanket the landscape of Caribbean coral reefs, yet few data exist on the physiology of their mutualism with *Symbiodinium*. Unlike scleractinian corals, whose abundance has dramatically declined in the Caribbean, gorgonians have withstood recent environmental perturbations. Learning about gorgonian symbioses, at current ambient conditions, enhances our understanding and our ability to predict the future of Caribbean coral reefs. We therefore employed multiple tools to characterize aspects of *Symbiodinium* photosynthesis in four common Caribbean gorgonian species. Oxygen flux data demonstrated differences in the *Symbiodinium* in the four studied gorgonians. For example, the two *Pseudoplexaura* species had the highest average photosynthetic rates per cm², probably due to the higher *Symbiodinium* and chlorophyll densities compared to *P. anceps* and *E. tourneforti*. On the other hand, the two *Pseudoplexaura* species had lower initial slopes of the P-E curves, low photosynthetic rates per *Symbiodinium* cell and per chlorophyll *a*, and lower chlorophyll *a* specific absorption. Taken together, these data suggest that *Symbiodinium* in *P. porosa* and *P. wagenaari* are less efficient in light absorbtion and utilization than *Symbiodinium* in *P. anceps* and *E. tourneforti*. The light levels available for *Symbiodinium* could differ due to *Symbiodinium* self-shading or to host tissue characteristics such as tissue thickness, or pigmentation. The possibility that *Symbiodinium* in *P. porosa* and *P. wagenaari* are less efficient in light absorbtion and utilization is corroborated by looking at changes in chlorophyll a specific absorption coefficient as a function of chlorophyll *a* density. In the two *Pseudoplexaura* species, the *a*\*_{chl *a*} values as a function of chlorophyll *a* density were very low, comparable to those values reported for phytoplankton and freshly isolated *Symbiodinium* from *Porites banneri*. On the other hand, *E. tourneforti a*\*_{chl *a*} values fell within those previously reported for scleractinian corals. Lastly, the *a*\*_{chl *a*} values of *Symbiodinium* in *P. anceps* demonstrated a very high pigment light absorption efficiency, comparable to that of *Symbiodinium* in the scleractinian coral *Porites banneri*. Of the four gorgonian species, *Symbiodinium* in *P. anceps* exhibited twice the photosynthetic rate per *Symbiodinium* cell than that of the next gorgonian species (*P. porosa*) and the highest average photosynthetic rates per chlorophyll *a*. The relatively high photosynthesis per *Symbiodinium* in *P. anceps* may be related to the low density of *Symbiodinium* and chlorophyll *a* per cm². In addition, the thin, angular branches and low polyp density may aid *Symbiodinium* photosynthesis in *P. anceps* by maximizing gas exchange and/or reducing self-shading of *Symbiodinium*. The four Caribbean gorgonian species produced comparable photosynthetic and respiration rates per cm² to the average rates of eight shallow scleractinian coral species (P = 2.0, R = 0.64). Conversely, the Mediterranean gorgonian *Eunicella singularis* at 15 m depth had lower average *Symbiodinium* photosynthetic (∼1) and respiration (∼0.55) rates per cm². The differences could be due to *E. singularis* in deeper waters being exposed to lower irradiance levels than those in the current study. In our study, the four gorgonian species did not exhibit photoinhibition, similar to what has been observed in other symbioses between *Symbiodinium* and cnidarians. The lack of photoinhibition may be due to branch tissue thickness, as was seen in the octocoral *Capnella gaboensis*. The gross P/R ratios in the four gorgonian symbioses, ranging from 2 to 4, were comparable to ratios for other octocorals, anemones, and shallow water scleractinian corals. On the other hand, the ratios of diurnal integrated gross photosynthesis to respiration were below 1. It remains to be determined the extent of the contribution of the *Symbiodinium* autotrophic production to the energy budget of these Caribbean gorgonians. *Symbiodinium* photosynthesis within a host may also be affected by host characteristics. For example, the scleractinian coral skeleton enhances light absorption by *Symbiodinium*, and *Symbiodinium* chlorophyll *a* specific absorption is higher in symbiosis than in isolation. Chlorophyll *a* specific absorption of *Symbiodinium* in the studied gorgonian corals was comparable to that in scleractinian corals , even though gorgonian skeletal structure (sclerites and an axial rod) substantially differs from the calyx structure in scleractinian corals. The calcite sclerites within gorgonian tissues may produce the same effect as the light scattering of the scleractinian coral skeleton, perhaps similar to the influence of the siliceous spicules of sponges on light transmission. Furthermore, in symbioses between cnidarians and Symbiodinium, photosynthesis is dependent upon the genetic identities of both the host and symbiont. Therefore, it is imperative to identify the Symbiodinium. In the Caribbean, scleractinian corals host Symbiodinium belonging to clades A, B, C, or D. On the other hand, the majority of Caribbean gorgonian species host only clade B Symbiodinium. The four gorgonian species in this study were no exception; harboring clade B Symbiodinium belonging to types B1, B1i, and B1L. Types B1i and B1L have only been reported in Caribbean gorgonian species. Sequencing of the B1 Symbiodinium from the four gorgonian species revealed that they harbor B1 Symbiodinium with distinct microsatellite flanking region sequences from those in scleractinian coral species. The B1 *Symbiodinium* obtained from most gorgonian colonies in our study formed a distinct group amongst the previously identified B1 lineages from scleractinian corals. Although both *P. anceps* and *P*. *wagenaari* hosted type B1 Symbiodinium, the photosynthetic characteristics differed between these two symbioses. Photosynthetic variability within *Symbiodinium* type B1 has also been observed in cultures, and may be associated with distinct genetic lineages within type B1. In our study, eight of the nine *P. anceps* colonies, and all *P. wagenaari* colonies, harbored symbionts from the same, highly supported, phylogenetic group within B1 *Symbiodinium*. Therefore, the observed photosynthetic variability between *P. anceps* and *P*. *wagenaari* was not due to different B1 lineages but probably due to the physiology of different host-symbiont combinations. In *E. tourneforti*, Symbiodinium had comparable photosynthetic rates per cell in the intact symbiosis and in isolation. On the other hand, maximum photosynthetic rates per cell in *Symbiodinium* isolated from *P. anceps, P. porosa*, and *P. wagenaari* were lower than those measured *in hospite*, although the average α was higher in isolation. Diminished photosynthetic rates in isolated Symbiodinium cells may occur in the absence of host carbon concentrating mechanisms or differences in carbonic anhydrase activity among *Symbiodinium* types. Exposure to the ionic environment of seawater, or bacteria may also reduce photosynthesis in isolated Symbiodinium. In the sea anemone *Aiptasia pallida*, at ambient temperatures, the photosynthetic rates of isolated Symbiodinium were also lower compared to the intact association, although not to the degree measured here. Conversely, secondary metabolites released from homogenized gorgonian corals may impair isolated Symbiodinium, as reported for the homogenate of the soft coral *C. gaboensis* that lysed Symbiodinium cells. Therefore, secondary metabolites produced by many gorgonians may limit the utility of investigating freshly isolated Symbiodinium. In conclusion, our results contribute to consequential data on Symbiodinium physiological performance in their mutualism with four Caribbean gorgonian species at ambient temperature. Given that gorgonian corals are either maintaining or increasing their abundance on Caribbean coral reefs, understanding aspects of their symbiosis is imperative to understanding the future of Caribbean coral reefs. This study demonstrates differences between Symbiodinium photosynthetic characteristics in the four gorgonian species, collected from the same site, maintained under identical conditions, and with two of the gorgonian species containing the same *Symbiodinium* type. The differences observed between the gorgonian symbioses emphasize the influence of the host physiology and architecture on *Symbiodinium* photosynthesis. # Supporting Information We thank A. Banaszak, S. Enríquez, N. Schubert, R. Smith, and the staff and students at the Unidad Académica de Sistemas Arrecifales, Instituto de Ciencias del Mar y Limnología, Universidad Nacional Autónoma de México for their suggestions and assistance. We acknowledge the use of irradiance data collected by F. Ruiz Rentería and E. Escalante Mancera of the Servicio Académico de Monitoreo Meteorológico y Oceanográfico of the UASA del ICML of the UNAM. In addition, we thank D. Goulet, H. Pearson, and L. Camp, from the University of Mississippi, for their field assistance. [^1]: The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: BDR RIP TLG. Performed the experiments: BDR KPS RIP TLG. Analyzed the data: BDR RIP TLG. Contributed reagents/materials/analysis tools: RIP TLG. Contributed to the writing of the manuscript: BDR TLG.

# Introduction Bone remodelling has traditionally been viewed as an endocrine- and paracrine- regulated process. This view is rapidly changing, however, with growing evidence demonstrating that neuronal factors are critical for normal bone homeostasis. Moreover, efferent neural pathways from the hypothalamus potently modify the activity of bone cells. Studies in leptin-deficient (*ob/ob*) mice revealed that the adipokine leptin, in addition to its actions as an important circulating indicator of the level of adiposity, also modulates bone formation through activation of central, hypothalamic relays via efferent sympathetic nervous output which directly modulates osteoblast activity. A major target of leptin in the hypothalamus is the neuropeptide Y (NPY) system, one of the most prominent regulators of appetite and energy homeostasis. Leptin down-regulates NPY expression in the hypothalamus and in this way reduces appetite and normalises energy expenditure. Importantly however, the NPY system has also been shown to regulate bone homeostasis. Amongst the 5 known Y-receptors (Y1, Y2, Y4, Y5 and in certain species also Y6), NPY mediates its effects on energy homeostasis via hypothalamic Y1 and Y2 receptors. Interestingly, these Y-receptors have also been reported to be critical in the regulation of bone homeostasis with the specific deletion of the Y2-receptor in the hypothalamus resulting in a bone anabolic phenotype. However, the actions of the NPY system in bone are more complex than a simple downstream mediator for leptin. Specifically, Y2-mediated changes occur consistently throughout the skeleton in these mice, while alterations in leptin levels induce opposing effects on cortical and cancellous bone, as evident in studies involving Y2^−/−;*ob/ob* double mutant mice. Furthermore, Y1 receptors have been identified on osteoblasts and a lack of Y1 receptors in mice also leads to increased bone mass. The Y1-mediated effect on bone is not dependent on hypothalamic Y1 receptors suggesting an additional direct action of the NPY system on bone homeostasis. Despite the demonstrated actions of NPY, Y receptors in the control of bone homeostasis, the role of NPY in this process is yet to be defined. Although NPY is the predominant ligand in the central nervous system, Y receptors can also be activated by the two other family members, peptide YY (PYY) and pancreatic polypeptide (PP). Possible redundancy in the functions of NPY-like ligands may explain why initial examination of NPY deficient mice suggested no skeletal changes, a finding contradictory to experiments showing significant increases in bone mass following loss of NPY-producing neurons from the arcuate nucleus of the hypothalamus using monosodium glutamate treatment by the same group. Therefore we used a systematic approach to investigate the specific role of NPY signalling on bone homeostasis employing several NPY mutant mouse models including specific re-introduction of NPY into the hypothalamus of otherwise NPY deficient adult mice. # Results ## High Bone Mass and Reduced Fat Mass in NPY ^−/− Mice While there was no effect of NPY ablation on body weight, body fat mass was significantly reduced in *NPY ^−/−* compared to wild type mice, as evidenced by decreased whole body fat mass measured by DXA , with no effect of genotype on lean body mass. Importantly, lack of NPY was associated with a greater whole body BMD as well as a greater whole body BMC apparent in both genders, although the increase in BMC only reached significance in female *NPY ^−/−*. Consistent with the whole body measurement, lumbar BMD and BMC were significantly increased in *NPY ^−/−* mice compared to gender- matched wild type controls. ## Greater Cortical Bone Formation in NPY ^−/− Mice In order to evaluate the DXA results, cortical bone was examined by peripheral quantitative tomography. There was a significant increase in cortical bone size in *NPY ^−/−* mice. Total bone volume and marrow volume of the mid- femur of *NPY ^−/−* mice were both increased, resulting in greater cortical bone volume and cortical thickness in *NPY ^−/−* mice compared to wild type. These changes were associated with enhanced cortical bone formation, as shown by elevated endocortical mineral apposition rate (MAR). Importantly, these changes in cortical bone mass and formation were evident in the absence of any increases in motor activity in either male or female *NPY ^−/−* mice, negating a potential role of activity and weight bearing in the increases in bone mass seen in these mice. Indeed, a trend for reduced physical activity was evident in *NPY ^−/−* mice in the dark phase, but this was not significant, nor was it seen in the overall 24-hour period. ## Generalized Increase in Bone Formation in NPY ^−/− Mice In the distal femoral metaphysis, cancellous bone volume was greater in *NPY ^−/−* mice, coincident with a 2-fold increase in bone formation rate compared to that in wild type mice. This change was associated with an increase in the extent (mineralizing surface) and speed (mineral apposition rate) of bone formation. Bone resorption indices in *NPY ^−/−* mice – i.e. osteoclast surface and number – were not significantly changed, demonstrating that NPY preferentially controls the anabolic aspects of bone homeostasis. Femoral data is shown for male mice, but results were similar in females. The effects of NPY deletion on femoral cancellous bone were also evident in micro computed tomographs of the distal femoral metaphysis and cancellous bone volume of the 4^th lumbar vertebrae, indicating a generalized anabolic effect of NPY deletion throughout the skeleton. ## Elevated Hypothalamic NPY Expression Reduces Bone Formation To investigate whether alterations in central NPY levels are responsible for the anabolic bone phenotype seen in *NPY ^−/−* mice, NPY was over- expressed specifically in the hypothalamus of adult wild type mice. Wild type mice were injected unilaterally with a recombinant adenovirus-associated virus producing a localized increase in NPY expression (rAAV-NPY+) in the hypothalamus, targeting the arcuate nucleus (Arc), the brain region critical in mediating the central effects of Y receptor signaling on both energy balance and bone mass. Four weeks post-injection, body weight was significantly increased in wild type rAAV-NPY+-injected animals (WT NPY+) compared to wild type controls, injected with an empty virus (WT empty). This weight gain was associated with an increase in white adipose tissue mass. Given the strong and positive association between increases in body weight and bone mass, in part due to mechanical effects of weight bearing, it is notable that tibial BMC was reduced by 19% (p\<0.05) in WT NPY+ mice compared to WT empty controls over the 4-week period, with no change in tibial BMD. This NPY- mediated bone loss was due to the specific action of NPY in the hypothalamus, as no changes in bone homeostasis were evident following rAAV-NPY+ injection into the hippocampus. Histological examination of bones revealed that elevation of hypothalamic NPY expression decreased cortical osteoblast activity, with a 7-fold reduction in endosteal MAR and a significant reduction in periosteal MAR in WT NPY+ mice compared to WT empty controls. These changes can be clearly seen on the endosteal surface after injection of dual fluorescent labels, demonstrating the powerful action of hypothalamic-NPY activated pathways on inhibiting osteoblast activity. In order to determine whether the bone anabolic phenotype observed in *NPY ^−/−* mice is wholly due to the lack of hypothalamic NPY signaling, an additional transgenic mouse model was generated where NPY expression was restricted solely to the hypothalamus. This model was achieved by re-introducing NPY specifically into the hypothalamus of *NPY ^−/−* mice via stereotactic injection of rAAV-NPY+ (*NPY ^−/−* NPY+). As shown in, *NPY ^−/−* NPY+ -mice had a greater body weight gain and more than doubled adiposity within 4 weeks compared to *NPY ^−/−* empty mice, confirming the viability of the NPY reintroduction procedure and demonstrating the critical role hypothalamic NPY plays in regulating body weight and fat mass. Interestingly however, the elevated bone mass and bone formation seen in *NPY ^−/−* mice was only partially corrected by hypothalamic re- introduction of NPY in otherwise NPY deficient mice. As shown above, global deletion of NPY led to a 40% increase in MAR in femoral cancellous bone compared to wild type controls. However, restoration of NPY expression in the hypothalamus of *NPY ^−/−* mice resulted only in a partial normalization of MAR to a level approximately 20% lower then *NPY ^−/−* mice, with similar reductions in the change in femoral bone mass and cancellous bone volume. This inability of hypothalamic NPY to completely restore a normal bone phenotype in *NPY ^−/−* mice suggests that non- hypothalamic pathway(s) may also be important in NPY mediated bone homeostasis. ## NPY Signalling on Bone Cells Alters Osteogenesis In Vivo and In Vitro In order to clarify whether NPY also regulates bone homeostasis by direct action on bone cells and whether NPY is expressed in bone tissue itself, bone sections were examined by *in situ* hybridisation. Both cancellous and cortical osteoblasts from wild type mice stained positively for NPY mRNA with no staining in *NPY ^−/−* bone tissue, providing the first evidence of NPY expression in osteoblasts *in vivo*. To further investigate what role NPY may play in osteoblast function and development, mRNA was isolated from the bones of wild type and *NPY ^−/−* mice and analysed for the expression of key osteogenic markers. In keeping with their high bone mass and anabolic phenotype, mRNA expression of alkaline phosphatase (ALP) was significantly up-regulated in the long bones of *NPY ^−/−* mice. Importantly, there was also an up-regulation of both Runx2, and Osterix mRNA levels in the *NPY ^−/−* mice. Moreover, *in vitro* osteoblast activity was enhanced in the absence of NPY, evident by a greater time-dependent increase in mineralisation by *NPY ^−/−* bone marrow stromal cells under osteogenic conditions than cells isolated from wild type mice. Taken together, these data indicate that global lack of NPY induces effects inherent to osteoblasts in both cortical and cancellous bone and in the apical and appendicular skeletons. Given that NPY is expressed in osteoblasts, these data further support the concept that NPY could be a direct local regulator of osteoblast function, in addition to effects mediated via the hypothalamus, and highlights the fact that NPY may act via neuronal as well as autocrine mechanisms to regulate bone homeostasis. # Discussion This study has identified a powerful inverse relationship between NPY signaling in the hypothalamus and bone formation. A generalized bone anabolic response resulting from loss of NPY signaling was evident throughout the skeleton, including cortical and cancellous bone as well as axial and appendicular sites. In contrast, NPY over expression in the hypothalamus resulted in an inhibition of bone formation. This central action is consistent with signaling via Y2 receptors expressed on neurons of the arcuate nucleus, as is evidenced by the opposing response reported following specific deletion of Y2 receptors from this region of the hypothalamus. Thus NPY acts in the hypothalamus to tonically inhibit bone formation, and likely does so through signals transduced by Y2 receptors. Results from this study suggest that in addition to the well defined actions of loading from body weight on bone mass, the central ‘perception’ of energy status may have an important influence on bone mass. This is particularly evident in rAAV-NPY+ injected mice. The viral promoter controlling NPY production prevents the normal feedback by signals such as increased serum leptin from inhibiting hypothalamic NPY expression. As a result, the “starvation” signal is not attenuated and body weight continues to increase. Importantly, despite the marked increase in body weight caused by this experimental condition, a decrease in bone mass occurred. While the decrease in bone formation in these mice may in part also result from reduced physical activity and thereby altered weight bearing, this result is also consistent with a skeletal response to the starvation signal induced by AAV NPY+. Thus the central perception of body weight may be important to the regulation of bone mass, in addition to the actual mechanical forces in the bone tissue itself. In this manner, during weight loss, increasing NPY ensures calories are not wasted on the production/maintenance of unneeded skeletal tissue, thereby also increasing the supply of minerals and nutrients stored in skeletal tissue for other vital processes. Conversely, during periods of high nutrient intake and resultant weight gain, reduction in NPY expression ensures sufficient bone formation to guarantee mechanical competence of the skeleton during the period of increased weight bearing. Importantly, we showed here that reduced NPY levels in NPY knockout mice are not associated with significant changes in activity levels, indicating the skeletal response to NPY are not activity-dependent. NPY is therefore ideally placed to match bone mass to body weight, a major regulatory influence on bone homeostasis. Central circuits alone, however, do not appear to explain the entire NPY- mediated signaling pathway to bone, as restoration of hypothalamic NPY expression in *NPY ^−/−* mice failed to completely normalize osteoblast activity. This incomplete rescue of the *NPY ^−/−* phenotype by central NPY replacement indicates a potential role for non- hypothalamic NPY pathways. This finding is consistent with a Y1 receptor- mediated action in the periphery, since hypothalamus-specific deletion of Y1 receptors did not replicate the greater bone formation evident in germline *Y1^−/−* mice. This opens the possibility that lack of NPY in osteoblasts may have contributed to the phenoptype of *NPY ^−/−* mice, consistent with our demonstration of NPY expression in osteoblasts, altered expression of osteogenic transcription factors in bone as well as increased *ex vivo* mineralization of osteoblasts from *NPY ^−/−* mice. Indeed, identification of NPY expression in the osteoblast, along with local expression of Y1 receptor, indicates the potential for local paracrine/autocrine control of osteoblast activity by NPY. Interestingly, in terms of NPY's coordination of body weigh and bone mass, NPY expression has been identified in both osteoblasts and osteocytes. Moreover, this expression was reduced following exposure of cultured osteoblasts to fluid sheer stress, indicating the potential for a local, load-responsive regulation of NPY. In this manner, increasing load would decrease NPY expression, thereby reducing the tonic repression of osteoblast activity as body weight increased. Thus central efferent signals may provide a systemic regulatory influence, via Y2 receptor action in the arcuate nucleus of the hypothalamus, whilst osteoblastic Y1 receptor action may enable local fine- tuning of the systemic response. Taken together, these data highlight the critical role of NPY in the co- ordinated regulation of skeletal tissue. We propose that in situations where hypothalamic NPY expression is decreased, as in short-term overfeeding in which serum leptin levels rise and in turn reduces central NPY signaling, bone formation is stimulated as was the case in *NPY ^−/−* mice. Moreover, when central NPY levels are elevated as in negative energy balance, bone formation is inhibited as was the case in rAAV-NPY+ injected mice. While hypothalamic NPY is ideally placed to modify bone mass to match changes in body weight induced by altered energy homeostasis, NPY produced in bone itself may play an important role in the regulation of bone mass. This later circuit offers also an attractive target to modulate and opens the possibility to increase bone mass without affecting central NPY controlled pathways. # Materials and Methods ## Generation of NPY ^−/− Mice The generation of the *NPY ^−/−* mice was described previously. All research and animal care procedures were approved by the Garvan Institute/St Vincent's Hospital Animal Experimentation Ethics Committee and were in agreement with the Australian Code of Practice for the Care and Use of Animals for Scientific Purpose. ## Viral Vector Injection Elevation of NPY expression was achieved following procedures reported previously. Briefly, 10–12 week-old wild type or *NPY ^−/−* mice were anesthetized and injected in the PVN with recombinant adeno-associated virus expressing either NPY (rAAV-NPY+) or an empty cassette (rAAV-empty) using a stereotaxic table (David Kopf, California, USA). Brain injection co-ordinates relative to Bregma were posterior, 2.1 mm; lateral, ±0.4 mm; ventral, 5.3 mm corresponding to the PVN. An additional control group was injected with rAAV- NPY+ or rAAV-empty vector in the hippocampus (Co-ordinates relative to Bregma were posterior, 1.7 mm; lateral, ±0.8 mm; ventral, 2.2 mm). One microlitre of either rAAV-NPY+ or rAAV-empty virus (1×10⁹ Pfu/µl) was injected bilaterally over 10 minutes using a 26 gauge guide cannula and a 33 gauge injector (PlasticsOne, Roanoke, USA) connected to a Hamilton Syringe and a syringe infusion pump (World Precision Instruments, Sarasota, Florida). The recombinant AAV expressing NPY was under the control of a neuron specific enolase promoter (rAAV-NPY+), which confined the expression to nerve cells. Mice were housed individually for the ensuing 28 days, with *ad libitum* access to standard chow and water. The efficacy of viral NPY expression was confirmed by the resultant weight gain, a characteristic response to elevated hypothalamic NPY production. ## Activity Measurements Ambulatory activity of individually housed mice was evaluated within the metabolic chambers using an OPTO-M3 sensor system (Columbus Instruments, Columbus, OH, USA), whereby ambulatory counts were a record of consecutive adjacent photo-beam breaks. Cumulative ambulatory counts of X, Y and Z directions were recorded every minute and summed for 1-h intervals. ## Bone Histomorphometry Mice were injected with the fluorescent compound calcein (Sigma Chemical Company, St Louis, USA) at 20 mg/kg (s.c.) at 10 and 3 days prior to collection. At 14–16 weeks of age, WT and *NPY ^−/−* mice were sacrificed by cervical dislocation between 10.00–14.00 h for collection of trunk blood in heparinized tubes. Both femora and the lumbar spine were excised and fixed in 4% paraformadehyde for 16 h at 4°C. The right femur was bisected transversely at the midpoint of the long axis and the distal half embedded undecalcified in methacrylate resin (Medim-Medizinische Diagnostik, Giessen, Germany). Similarly, the 4^th lumbar vertebrae was defleshed and embedded. Sagittal sections with 5 µm thickness were analysed as previously described. Mineralized bone in the sections was visualized by von Kossa stain. Cancellous bone volume, trabecular thickness and number were calculated. The mineralizing surface (MS, %), mineral apposition rate (MAR, µm/d) and bone formation rate (BFR = MS/BS \* MAR, µm2/ µm/d) were calculated as previously described, following fluorescence microscopy (Leica, Heerbrugg, Switzerland). Osteoclast surface and osteoclast number were estimated in tartrate-resistant acid phosphatase stained sections where only multinucleated, TRAP-positive cells were included in the analysis. Cortical mineral apposition rate was measured on the anterior periosteal surface in a region extending 1000 µm distal from the mid point of the shaft and in an endosteal region extending 1000 mm proximal from the posterior aspect of the growth plate as previously described. ## Bone Densitometry Whole body bone mineral content (BMC), bone mineral density (BMD), lean mass and fat mass were measured on mice ventral side down with head exclusion and tail inclusion, using a dedicated mouse dual X-ray absorpiometer (DXA) (Lunar Piximus II, GE Medical Systems, Madison WI). Whole femoral BMC and BMD were measured in excised left femora. Femora were scanned with tibiae attached and the knee joint in flexion to ninety degrees to ensure consistent placement and scan of the sagittal profile. ## Micro Computed Tomography Cancellous bone architecture was evaluated at the distal femoral metaphysis, using micro-computed tomography (Micro-CT40, Scanco Medical AG, Basserdorf Switzerland), employing a 12-µm isotropic voxel size. Bone volume fraction (BV/TV, %) was computed without assumptions regarding the underlying bone architecture. ## Quantitative Computed Tomography (QCT) QCT was used to isolate cortical bone for analysis in male mice, using a Stratec XCT Research SA (Stratec Medizintechnik, Pforzheim, Germany). Scans were conducted on excised left femurs as previously described with following settings: voxel size of 70 µm, scan speed of 5 mm/sec and slice width of 0.2 mm. Bones were scanned in 2 consecutive slices, 7 and 7.5 mm from the distal margin of the femur, representing a mid femoral aspect. ## In-Situ Hybridisation of Bone and Hypothalamic Tissue Antisense and sense riboprobes for mouse NPY were generated from a cDNA region corresponding to bases −72 to +89 of the coding sequence and subcloned into pGEM-T Easy plasmid vector (Promega). Probes were labelled with digoxigenin-UTP by *in vitro* transcription with either SP6 (antisense probe) or T7 (sense probe) RNA polymerase using a DIG RNA labelling kit (Roche) according to the manufacturer's instructions and.purified using Sephadex G25 spin columns and their yield was assessed by dot blot. *In situ* hybridisation was carried out as previously described with minor modifications on 5 µm sections of decalcified sagittal section of the distal femur. All procedures were carried out at room temperature unless specified otherwise. Sections were dewaxed in xylene, rehydrated through decreasing concentrations of ethanol (100% to 70%), and then fixed in 2% paraformaldehyde in PBS on ice for 10 minutes. After washing in PBS, sections were incubated at 37°C with proteinase K (Roche) at a concentration of 5 µg/mL in 50 mM Tris-HCl (pH 7.5) and 5 mM EDTA for 20 minutes, followed by 0.1 M glycine in PBS for 5 minutes. Sections were acetylated with 1X triethanolamine containing 0.25% (v/v) acetic anhydride for 10 minutes and washed in PBS before hybridisation. Hybridisation was carried out overnight at 45°C in a humid environment. Sections were incubated with hybridisation solution containing 50% formamide, 5X SSC, 250 µg/ml salmon sperm DNA, 250 µg/ml yeast tRNA, 1x Denhardt's solution, 10% dextran sulphate, 10 mM DTT and 200 ng/mL of digoxigenin-labelled riboprobe. The sections were washed in 2X SSC for 5 minutes followed by 0.2X SSC at 62°C for 30 minutes and 0.1X SSC at 65°C for 30 minutes. They were then incubated with 20 µg/mL RNase A (Roche) in 0.5 M NaCl, 10 mM Tris-HCl, and 1 mM EDTA at 37°C for 15 minutes followed by washes with 2X SSC for 5 minutes, 0.1X SSC at 37°C for 30 minutes. Following the washes, antidigoxigenin Fab antibody fragments conjugated with alkaline phosphatase (Roche) and NBT/BCIP stock solution (Roche) containing 1 mM levamisole were used for colorimetric detection following the manufacturer's instructions. ## Isolation and Differentiation of Bone Marrow Stromal Cells The isolation and differentiation of plastic adherent bone marrow stromal cells (BMSCs) from 5 to 9 week old male mice was carried out as previously described with minor modifications. Following sacrifice by cervical dislocation, marrow was flushed from femurs and tibias with control media and cells were plated at a density of 1.9×10⁶ cells/cm² in 50 cm² plastic tissue culture plates. Control media consisted of α-minimum essential medium containing 10% fetal bovine serum, 2 mM L-glutamine, 2.2 g/L sodium bicarbonate, 0.017 M HEPES, 100 µg/mL penicillin/streptomycin and 34 mg/L gentamycin. The majority of non-adherent cells were removed by medium changes at 3 and 5 days later and discarded. After 7 days in culture, cells were trypsinised, counted and re-plated at 3×10⁴ cells/cm² in 24-well plates in control media. From this time onwards media was changed 3 times per week. Differentiation into mineral-producing osteoblasts was achieved by culturing the cells in osteogenic media consisting of control media supplemented with 50 mg/L ascorbic acid and 10 mM β-glycerophosphate. Mineralisation of extracellular matrix was visualised by von Kossa staining with 2% silver nitrate under UV light for 30 minutes. The extent of mineralisation was quantified using the Leica QWin imaging system (Leica Micro-systems, Heerbrugg, Switzerland). ## RNA Extraction and Quantitative Real-Time PCR Following sacrifice by cervical dislocation, femurs and tibias from 6–10 week old male mice were cleaned and snap frozen in liquid nitrogen. Subsequently femurs and tibias from each leg were homogenised separately in TRIzol® reagent using a polytron. RNA extractions were carried out using TRIzol® reagent according to the manufacturer's instructions. RNA samples were checked for quality and quantified using the Agilent 2100 Bioanalyser (Agilent Technologies) according to the manufacturer's instructions. One microgram of total RNA was taken for cDNA synthesis with oligo(dT)₂₀ and random hexamers using the SuperScript III First-Strand Synthesis System for reverse transcription-PCR (Invitrogen). Quantitative real-time PCR was then carried out using the TaqMan Universal PCR master mix (Applied Biosystems), ABI Prism 7900 HT Sequence Detection System and Software and inventoried kits containing primers and probes from Applied Biosystems. To control for variability in amplification due to differences in starting mRNA concentrations, GAPDH was used as an internal standard. The relative expression of target mRNA was computed from the target Ct values and the GAPDH Ct value using the standard curve method (Sequence Detection Systems Chemistry Guide, Applied Biosystems). ## Statistical Analyses Dual genotype comparisons were assessed using two-tailed students T-test. Multiple genotype comparisons were assessed by factorial ANOVA followed by Fisher's or Contrasts post-hoc tests, using StatView version 4.5 or Super-ANOVA (Abacus Concepts Inc, CA, USA). For all statistical analyses, *P*\<0.05 was accepted as being statistically significant. [^1]: Conceived and designed the experiments: PB NJL HH. Performed the experiments: PB NJL FD SL SA BS EJDL LZ RFE IPW KS YCS EY AA AS. Contributed reagents/materials/analysis tools: MMM MJD DDP DGL SF. Wrote the paper: AS JE HH. [^2]: The authors have declared that no competing interests exist.

# 1 Introduction String similarity join that finds similar string pairs in a given string set or between two given string sets is a fundamental operation in many fields, such as pattern matching, computational linguistics, bioinformatics, and database integration.It is widely used for detection of duplicate web pages in web crawling, collaborative filtering, and entity resolution. For example, given two string sets R = {Mi Li, Qi Wan, …} and S = {M. Li, Qin Wan, …}, we can find all similar pairs \<*r*,*s*\> ∈ *R* × *S* such as \<Mi Li, M Li\> according to a certain similarity function. For string similarity join, fundamental techniques include partitioning techniques (e.g. Pass-Join and PartEnum), prefix-filtering methods (e.g. TrieJoin and PEARL), and other methods (e.g. MTree, SSI, LSH, and FASTSS). Research in this field has been carried out in various scientific disciplines and related methods often are tuned for specific ranges of allowed error thresholds or query lengths, specific hardware properties, specific alphabet sizes, or specific distributions of errors. The big data era is the inevitable consequence of our ability to generate, collect, and store digital data at an unprecedented scale. When there are a large number of sources and a large volume of data, the traditional string join methods become inefficient and ineffective to practice. To address the volume dimension, new techniques have been proposed to enable parallel string join using MapReduce. These include techniques for adaptive blocking and techniques that balance load among different nodes. However, MapReduce is not adapted for the application of data join. In this paper, we propose a parallel string join framework to address the efficiency problem by utilizing the multi-threading technique and distributed computing technique separately. Availability of high-performance CPU and the large memory make the framework very practical. The contributions of this paper are as follows: 1. We propose a parallel processing framework for string similarity join and a related pruning strategy to obtain the high efficiency. The partition- based method and the parallel processing techniques are used to improve the computational performance. 2. We propose a parallel string join algorithm Para-Join to implement the framework in the multi-core system. The multi-threading technique is used to improve the processing performance. We demonstrate that Para-Join method can not only avoid reduplicate computation but also ensure the completeness of the result. 3. We propose a parallel string join algorithm Pada-Join to implement the framework in the cluster environment based on Spark platform to obtain the high efficiency. 4. We have implemented and tested Para-Join and Pada-Join algorithms on real data sets. The experimental results show that our algorithm achieves high performance and outperforms the existing methods. The rest of this paper is organized as follows: Section 2 gives the notations and discussion of some existing techniques. In section 3, we introduce our proposed parallel string similarity join framework and related strategies. Then, section 4 and 5 give the details of Para-Join algorithm and Pada-Join algorithm separately. We present the experimental results in section 6. Related works are introduced in section 7 and we conclude the paper in section 8. # 2. Background ## 2.1 Formal problem statement ### Definition 1: String and string set A string *s* is a finite sequence of symbols over an alphabet *Σ*. The length of *s* is denoted by \|*s*\| and the substring starting at position *i* with length *n* is denoted by *s*(*i*,*n*). All positions in a sequence are zero-based, i.e., the first character of *s* is *s*(0). A string set *S* is a collection of strings. The size of *S* is denoted by \|*S*\|. ### Definition 2: String similarity Given strings *s* and *r*, *s* is similar to *r*, denoted *s* ∼sim *r*, if and only if *Sim*(*s*,*r*)≥*δ*. *Sim(s*,*r)* is a certain similarity function and *δ* is a threshold. If the edit distance is used as the similarity function, *s* is *k*-approximately similar to *r*, denoted *Ed(s*,*r) = k*, if and only if *s* can be transformed into *r* by at most *k* edit operations. The edit operations include replacing one symbol in *s*, deleting one symbol from *s*, and inserting one symbol into *s*. <img src="info:doi/10.1371/journal.pone.0172526.e001" id="pone.0172526.e001g" /> S i m ( s , r ) = 1 − E d ( s , r ) M a x ( \| s \| , \| r \| ) ≥ δ <img src="info:doi/10.1371/journal.pone.0172526.e002" id="pone.0172526.e002g" /> E d ( s , r ) ≤ ( 1 − δ ) × M a x ( \| s \| , \| r \| ) <img src="info:doi/10.1371/journal.pone.0172526.e003" id="pone.0172526.e003g" /> τ = ( 1 − δ ) × M a x ( \| s \| , \| r \| ) ### Definition 3: String similarity (self) join Given two string sets *S* and *R*, a similarity function *Sim* (or an edit distance function *Ed*(·)) and a similarity threshold *δ* (or a distance threshold *τ*), a similarity join finds all string pairs *\<s*,*r\>* (*s*∈*S*, *r*∈*R*) such that *Sim*(*s*,*r*) ≥ *δ* (or *Ed*(*s*,*r*) ≤ *τ*). When *R* is equal to *S*, it refers to string similarity self-join of *S*. In this paper we adopt the edit distance to measure the similarity, so we use *τ* as the similarity threshold directly. ## 2.2 Filtering and verification framework ### Length filtering Length filtering is proposed in reference. The basic idea is that if two strings are similar, the difference of their length cannot be large. For example, if a string s is similar to r, the length of s should be in the range between *\|r\|-τ* and *\|r\|+τ* (*τ* is a given similarity threshold). By using the length filtering method, a string set *s* can be partitioned into subsets and in each subset, the strings share the same length. Then it is not necessary to match the strings in different subsets. Length filtering method is used to reduce the amount of candidate pairs for string similarity matching. ### Position filtering Considering two strings *s* and *r*, *s* is split into *τ+1* disjoint segments. If *r* is similar to *s*, there must exist a substring of *r* which can match one segment of *s* based on the pigeonhole principle. For any segment *s(i*,*n)*, let *W*_*n* denote the set which includes all the segments of *r* with length *n*. We need to check whether *s(i*,*n)* appears in *W*_*n*. Length filtering method doesn’t consider the position of the segments so that it cannot deal with this matter. Position filtering method provides an effective substring selection strategy to generate *W*_*n*. Because the element number in *W*_*n* is smaller than that of all the substrings, a lot of unnecessary calculation will be avoided. For example, if the similarity threshold is *τ* and string *r* has a substring *r(i*,*n)* (*1≤i≤4*) which matches *s(j*,*n)*, position filtering method will give the possible start positions of *r(i*,*n)* in string *r*. The strings *r* and *s* will be split into three parts, the same part *r(i*,*n)* and *s(j*,*n)*, the left part *r(0*,*i)* and *s(0*,*j)*, the right part *r(i+n*,*\|r\|-i-n)* and *s(j+n*,*\|s\|-j-n)*. If *s* is similar to *r*, we can transform *s* to *r* with *Ed*(*r*,*s*) ≤ *τ*. A straightforward method is that we first transform *r(0*,*i)* to *s(0*,*j)*, then we transform r*(i*,*n)* to *s(j*,*n)*, finally we transform *r(i+n*,*\|r\|-i-n)* to *s(j+n*,*\|r\|-j-n)*. The total transformation distance is *Ed*(*r*(0,*i*), *s*(0,*i*)) + *Ed*(*r*(*i*,*n*), *s*(*i*,*n*)) + *Ed*(*r*(*i* + *n*,\|*r*\| −*i* − *n*), *s*(*i* + *n*,\|*s*\| −*i* − *n*)) ≤ *τ*. Based on the above conclusions, we can get the range of the start position of r(i,n) \[*P*_min,*P*_*max*\]. According to reference, we can get *P*_min = max (1, *p*_*i* − ⌊(*τ* − Δ)/2⌋) and *P*_max = min (\|*s*\| − \|*s*(*i*,*n*)\|, *p*_*i* − ⌊(*τ* − Δ)/2⌋), where *p*_*i* is the start positions of *s(i*,*n)* in string *s*, and Δ = \|\|*s*\| − \|*r*\|\|. So we do not need to visit all the substrings of *r*. Reference improved position filtering by using a new substring selection technology called multi-match-aware substring selection which obtained the more accurate values for *P*_*min* and *P*_*max*. ### Extension-based verification Given a candidate pair *\<s*,*r\>*, a simple method to verify whether they are similar is to calculate the edit distance between them. If the edit distance is not larger than *τ*, we can say that s is similar to *r*. A dynamic-programming algorithm can be used to get the edit distance. The time complexity is O(\|*s*\| × \|*r*\|). Actually, we only need to check whether the edit distance is less than *τ* instead of getting the absolute value of edit distance. Length-aware verification was proposed to verify the candidate pair based on length pruning. In order to further reduce the computation, Wang et al. extended this method through putting forward the extension-based verification to calculate the edit distance. This method can reduce the time complexity from O(\|*s*\| × \|*r*\|) to O((*τ* + 1) × min(\|*s*\|,\|*r*\|)). In this paper, we apply the filter-verification framework to implement the string similarity join. The length filter method and the position filter method are used for the first phase, and the extension-based verification method is used for the second phase. # 3. Parallel string join framework ## 3.1 Parallel processing framework Applications continue to become more data-intensive. We assume applications may be pulled apart across the threads in the multi-core system or the nodes in the distributed systems. Although this may complicate data placement and transport, it improves the processing efficiency. shows the proposed parallel string join framework which can benefit from the data parallelism, task parallelism, and resilience. The input includes a string set *S* and a similarity threshold τ. In the second phase, *S* is divided into several partitions. For each partition, a thread or task is created to process it separately. The filter-verification technique is used for string join in each thread or task. Both string partitioning and string matching can be done in parallel. The output is the combination of all the similarity pairs in *S* and is stored in the disk. In this framework, solutions of three issues are taken into consideration: - How to split the dataset into subsets. Since the size of the dataset is not determined and the capability of the system is unknown in advance, it is hard to determine the number of the partitions. - How to calculate the similarity of two strings efficiently. The similarity computation is the core of the framework so we need a more efficient algorithm to deal with it. - How to implement parallel string join algorithm to obtain high efficiency without affecting the accuracy of string matching. The parallel algorithm must guarantee that the accuracy will not be affected. It will be better to improve the accuracy of the parallel algorithms since the multi-threading and the multi-tasking techniques can reduce the time and space complexity. ## 3.2 String similarity computation Resume a token set *Σ*, a string set *S*, a string *s*, a string r, *r*∈*S*, and *s*∈*S*. The function *f*^*c*(⋅) represents the joint frequency vector and function *v*(⋅) represents the interval vector. So *f*^*c*(*s*) is the joint frequency vector of *s* and *v*(*s*) is the interval vector of *s*. $\left\| f^{c}(s) - f^{c}(r) \right\|_{L_{1}}$ is the distance of joint- frequency vector between *f*^*c*(*s*) and *f*^*c*(*r*) about length *L*_*1*. The function *dis*(*v*(*r*),*v*(*s*)) is the distance between *v(s)* and *v(r)*. If the equation *v*(*s*) = *v*(*r*) exists, we can say that *s* and *r* are similar. According to the above rule, we can split *S* into some disjoint subsets *S*₁,*S*₂,⋯,*S*_*n*, *S*₁ ⊂ *S*, *S*₂ ⊂ *S*, *S*_*n* ⊂ *S*. Then we can get the following conclusion. 1. *S*₁∪*S*₂∪⋯∪*S*_*n* = *S*. 2. For each subset *S*_*i* and *S*_*j*, *S*_*i* ∩ *S*_*j* = Φ, *S*_*i* ⊂ *S* and *S*_*j* ⊂ *S*. 3. For each subset *S*_*i* (*S*_*i* ⊂ *S*), if there exist two string *s*_*1* and *s*_*2* (*s*₁,*s*₂ ∈ *S*_*i*), we can get *v*(*s*₁) = *v*(*s*₂). shows an instance of a string set *S*. The size of *S* is 12. Alphabet *Σ* is composed of 26 lowercase letters, i.e., *Σ = {a*, *b*, *…*, *z}*. We divide *Σ* into three subsets, *Σ*_*1* *= {g*,*e*,*b*,*n*,*j*,*h*,*w*,*t*,*x}*, *Σ*_*2* *= {f*,*a*,*o*,*m*,*k*,*i*,*u*,*q*,*z}*, and *Σ*_*3* *= {d*,*c*,*l*,*v*,*s*,*r*,*p*,*y}*. We then further divide *Σ*_*1* into four intervals \[1,2\], \[3,3\], \[4,4\], \[5,6\], divide *Σ*_*2* into three intervals \[2,2\],\[3,3\],\[4,7\], and divide *Σ*_*3* into four intervals \[1,1\], \[2,2\], \[3,3\], \[4,4\], \[5,6\]. First, for each string in *S* we get the joint-frequency vector and the interval-vector by functions *f*^*c*(⋅) and *v*(⋅). *f*^*c*(*s*₁) = (3,2,5), *f*^*c*(*s*₂) = (3,2,6), *f*^*c*(*s*₃) = (3,2,3), *f*^*c*(*s*₄) = (2,3,4), … *v*(*s*₁) = (2,1,5), *v*(*s*₂) = (2,1,5), *v*(*s*₃) = (2,1,3), *v*(*s*₄) = (1,2,4), … From the above computation results, we can get that string *s*₁ and string *s*_*2* belong to the same partition while string *s*₃ and string *s*₄ belong to different partitions. # 4. Parallel processing in multi-core systems In order to make full use of the capability of the multi-core system, we design and implement a parallel string join algorithm called Para-Join. ## 4.1 String join algorithm para-join Algorithm 1 shows the pseudo code of Para-Join. Consider a string set *S*, it is firstly split into *S*' (*S*' = *S*₁∪*S*₂∪⋯∪*S*_*n*) in parallel based by the frequency distribution function called *fqSplit*(·). Then a parallel cycling alternation method is used to deal with *S*'. Theorem 1 shows that our algorithm can eliminate the redundant computation and guarantee the completeness of the result. The major flow can be described as follows. Firstly the set S is split into n subsets and the corresponding threads are created. In each thread, function *para-RR*(·) is invoked to seek the similar string pairs of *S*_*j*. For any subset *S*_*j*, function *para-RS*(·) helps to search all the similar pairs between *S*_*i* and *S*_*j* (i\<j). ### Algorithm 1:  Para-Join **Input:** S     //A set of strings       τ     //A given similarity threshold **Output:** *ψ*  //*ψ* = {(*s*,*r* ∈ *S*)\|*Sim*(*s*,*r*) ≤ *τ*} **1**  **begin** **2**    Main thread: **3**      *S*'←fqsplit(S);  //split the set S into n subsets **4**      *ψ*←Φ; **5**      SubThread threads = new SubThread\[*S*'.size\];  //create a thread array **6**      for (j = 0; j\<*S*'.size; j++) **7**        SubThread threads\[j\] = new SubThread(j);  //create n threads and run them **8**        threads\[j\].start; **9**      for (j = 0; j\<*S*'.size; j++) **10**       threads\[j\].join;  //main thread waits for all of the processing threads to finish **11**     for (j = 0; j\<*S*'.size; j++) **12**       *ψ*←*ψ*∪threads\[j\].get;  // the union of the result produced by the n processing threads **13**   Processing thread with parameter j: **14**      *ψ*1←Φ; *ψ*2←Φ; **15**      *ψ*1←*ψ*1∪para-RR(*S*'.get(j)); **16**       for (i = 0; i\<j; i++) do **17**         *ψ*2←*ψ*2∪para-RS(*S*'.get(i), *S*'.get(j)); **18**       return *ψ*1∪*ψ*2;     **19  end** ### Theorem 1 Para-Join can not only avoid repetitive computation but also ensure the completeness of the result. ### Proof Given a collection of strings *S*, we split it into n small subsets, *S*₁,*S*₂,⋯,*S*_*n*. According to algorithm Para-Join, for any subset *S*_*j* (*j = 1*,*2*,*…*, *n*), we need to find the similar pairs between *S*_*i* and *S*_*j* (*i\<j*). Because the value of *j* ranges from 1 to n, for any *S*_*i* and *S*_*j* (*i≠j*), the search processing will be executed only once. For each *S*_*i*, the algorithm will search the similar pairs in *S*_*i* at first. So Para-Join will not miss any similar pairs, i.e., it can ensure the completeness of the result. Furthermore, there is no redundant similarity computation between any two strings in the algorithm. So we can see that Para-Join can also avoid repetitive computation. # 4.2 Data partition and similarity computation The function *fqSplit*(·) is designed for the data partition. Given a collection of strings S, there exist a lot of methods to split it into some small subsets. In this paper, we propose a parallel strategy which can achieve data partition in a shorter period of time. The pseudo-code is illustrated in algorithm 2. Firstly the frequency variance of each token in *Σ* is calculated. Then Σ is split into multiple subsets in parallel by Z-Collapsing algorithm. Each subset *Σ*_*i* is called a joint-token. For each string, its joint frequency vector is calculated, and for each joint-token, the range of the frequency distribution called range-frequency is also calculated. Finally, the function splits the string set *S* into subsets. ## Algorithm 2:  fqSplit **Input:** *S*    //A given set of strings **Output:** *S*'  //*S*' = {*S*_*i*\|∀*s*,*r* ∈ *S*_*i*, *S*_*i* ⊂ *S*, *v*(*s*) = *v*(*r*)} **1**  **begin** **2**    for each token, calculate the variance of frequency in parallel; **3**    Σ'←Paralleled divide *Σ* into several sets; **4**    for each *s* (*s* ∈ *S*), paralleled calculates *f*^*c*(*s*); **5**    split the range of frequency distribution in parallel; **6**    for each *s* (*s* ∈ *S*), paralleled compute *v(s)*; **7**    *S*'←Partition *S* by using the multi-threading technique in parallel; **8**    return *S*'; **9**  **end** In section 2 the position filtering and the extension-based verification methods have been explained in detail. We design function *posFilter*(·) and function *verification*(·) to implement these two methods. If *posFilter(s*,*r*,*τ)* returns false, string *s* and string *r* are dissimilar. If *posFilter(s*,*r*,*τ)* returns true, pair *\<s*,*r\>* is added into the candidate set. Function *verification(s*,*r*,*τ)* returns the similarity of string *s* and string *r*. In this paper, we develop a pruning strategy by extending the position filtering to remove the dissimilar pairs. By utilizing this pruning strategy, we can get a smaller candidate set. Suppose *s* and *r* denote two different strings, *v(s)* and *v(r)* denote their interval-vector respectively. The following is the description of the process in two different cases. 1. *v(s) ≠ v(r)*. If function *posFilter(s*,*r*,*τ)* returns true and the inequality *dis(v(s)*, *v(r))≤2τ* is established, pair *\<s*,*r\>* can be added into the candidate set. If pair *\<s*,*r\>* is in the candidate set and inequality *verification(s*,*r*,*τ)≤τ is established*, we can get that *s* is similar to *r*. 2. *v(s) = v(r)*. If function *posFilter(s*,*r*,*τ)* returns true and the inequality *dis(v(s)*, *v(r))≤2τ* is established, pair *\<s*,*r\>* can be added into the candidate set. If pair *\<s*,*r\>* is in the candidate set, and the inequality *verification(s*,*r*,*τ)≤2τ is established*, we can also get that string *s* is similar to string *r*. # 4.3 Join operation Two functions named *para-RR* (·) and *para-RS*(·) are designed to do the join operation. The function *para-RR*(·) extends the partition-based algorithm and implements the self-join in a subset by using the multi-threading technique. There are three main steps in *para-RR*(·): Step 1: *S*_*i* is sorted by the string length in descending order. Step 2: The inverted index *L*^*i*_*l* is built. The variable *l* is the string length and the variable *i* is the index of the string segment. Step 3: For any two strings, their similarity is calculated by adopting the above method. For example, given two strings *s* and *r*, function *para-RR*(·) first computes their joint-frequency vectors *f*^*c*(*s*) and *f*^*c*(*r*). If the *L*_*1* distance of their joint- frequency vector is larger than 2*τ*, we can get that these two strings are dissimilar. Otherwise, it will check the string pair \<*s*,*r*\> by invoking function *verification*(·). The function *para-RS*(·) primarily focuses on how to find the similar pairs between two different collections. Given two different sets *S*_*i* and *S*_*j*; *v(S*_*i**)* and *v(S*_*j**)* denote the IDs of *S*_*i* and *S*_*j* respectively. If *dis(v(S*_*i**)*,*v(S*_*j**))* is larger than *2τ*, it shows that they cannot be matched successfully. If *dis(v(S*_*i**)*, *v(S*_*j**))* is not larger than *2τ*, the function *para-RS*(·) can find the similar pairs by employing the above pruning strategy. For example, given a string *r* in *S*_*i*, for any string *t* in *S*_*j* (*l*_*min* ≤ *length(t)* ≤ *l*_*max*)), the function first checks whether *r* and *t* are similar by function *posFilter*(·), and then it calculates the *L*_*1* distance of joint-frequency vector of *r* and *t*. # 5. Parallel processing in distributed systems A big problem for parallel processing with multi-threading technique is the incapability of the system such as the limited memory and the number of cores. One solution is to add the memory and the other solution is to run the framework in a distributed cluster environment. ## 5.1 String join algorithm pada-join Hadoop as a big data processing technology has been around for 10 years and has proven to be the solution of choice for processing large data sets. MapReduce is a great solution for one-pass computations, but not very efficient for use cases that require multi-pass computations and algorithms. Each step in the data processing workflow has one map phase and one reduce phase and the developers will need to convert any use case into MapReduce pattern to leverage this solution. Spark allows programmers to develop complex, multi-step data pipelines using directed acyclic graph ([DAG](http://en.wikipedia.org/wiki/Directed_acyclic_graph)) pattern. It also supports in-memory data sharing across DAGs, so that different jobs can work with the same data. Spark runs on top of existing [HDFS](http://wiki.apache.org/hadoop/HDFS) infrastructure to provide enhanced and additional functionality. We propose a parallel string join algorithm called Pada-Join based on Spark. Algorithm 3 shows the pseudocode of Pada-Join, where the bold functions or methods are provided by Spark. ### Algorithm 3: Pada-Join **Input:** *S*    // A set of strings        *τ*    // A given similarity threshold **Output:** *ψ*//*ψ* = {(*s*,*r* ∈ *S*)\|*Sim*(*s*,*r*) ≤ *τ*} **1** **begin** **2   Map**(\<rid,string\>); //the filter stage **3**   compute joint-frequency vector f(r) for string r; **4**   emit(\<f(r), rid\>); **5   Groupbykey**(\<f(r), rid\>); **6**   emit(\<f(r),list(rid)\>); **7**   \<f(s),list(sid)\>←broadcast(\<f(r),list(rid)\>); **8   MapPartitions**(\<f(r),list(rid)\>,\<f(s),list(sid)\>); **9**   for \<f(ri),list(rid)\> in \<f(r),list(rid)\> **10**     for \<f(si),list(sid)\>) in \<f(s),list(sid)\> **11**       if dis(f(ri), f(si)) ≤ 2τ then emit(\<list(rid),list(sid)\>); **12   Flatmap**(\<list(rid),list(sid)\>); 13   for rid in list(rid) emit(\<rid,list(sid)\>); **14   Join**(\<rid,list(sid)\>,\<rid,r\>); **15**   emit(\<r,list(sid)\>); **16   Flatmap**(\<r,list(sid)\>); 17   for sid in list(sid) emit(\<sid,r\>); **18   Join**(\<sid,r\>,\<sid,s\>); **19**   emit(\<s,r\>); **20   Filter**(\<s,r\>); **21**   if  Sim(s,r)≤τ then emit(\<\<s,r\>,Sim(s,r)\>);       **22    end** The joint frequency vector *f(r)* for each string of the given dataset is generated in the filter stage. In order to get the joint frequency vector, we need to obtain the token set according to the token counting algorithm. Algorithm 4 shows the pseudocode of how to compute the token set, where the bold functions or methods are provided by Spark. If two strings are similar, the distance of their joint-frequency vectors must be less than 2*τ*. The candidate pairs are produced by utilizing the cartesian product of all the distinct pairs of the distribution nodes. However, this operation takes a huge amount of memory to store all the pairs that are distributed across multiple machines. To minimize the size of pairs, the vectors are taken as keys and the string IDs are taken as values so that the pairs sharing the same joint-frequency vector are assigned to the same group (seen in lines 5–6 of algorithm 3). The lines 7–13 in algorithm 3 illustrate the stage of candidate generation. To reduce data communication and data shuffling among the nodes, we store the joint frequency vector groups \<*f(r)*,*list(rid)*\> in memory by generating a broadcast variable \<f(s),list(sid)\>. Then the candidate groups to meet the filtering condition are matched. In the verification phase, *rid* and *sid* need to be converted into string *r* and *s*, and then to be verified. Line 14 does the job by joining dataset *S* with *\<rid*,*list(sid)\>*. The variable *sid* coming from the broadcast *\<f(s)*,*list(sid)\>* is generated from \<*f(r)*,*list(rid)*\>. Then the candidate pair \<*s*,*r*\> can be obtained by joining the dataset *S* with \<*sid*,*r*\> (seen in lines 18–19), and they are matched by calculating their similarity. The output is the final result. ### Algorithm 4:  token count **Input:** *S*    //A given set of strings **Output:** *S*'  //*S*' = {*S*_*i*\|∀*s*,*r* ∈ *S*_*i*, *S*_*i* ⊂ *S*, *v*(*s*) = *v*(*r*)} **1** **begin** **2   Flatmap**(\<rid,string\>); **3**   for each token in the string r emit(\<token,1\>); **4   Reduce**(\<token,list(1)\>); **5**   emit(\<token, tokenfrequency\>); **6**   *S*'←Z-folder(\<tokenid, tokenfrequency\>); **7   return** *S*'; **8**      **end** Because Pada-Join and Para-Join share the same algorithmic logic, Pada-Join can also avoid repetitive computation and ensure the completeness of the result. ## 5.2 Join operation in spark The following shows the computation flow of join operation in Spark. 1\) To get the token set dynamically and by partitioning. In Para-Join, the token set is obtained in advance. In Pada-Join, the token set is obtained dynamically. is an instance to get the token set. After getting the token set, we need to split it into subsets. The partitioning rule is the same as that in Para-Join algorithm, i.e., calculating the frequency distribution for each token and the frequency variance, and then getting the token set according to the Z-folder algorithm. 2\) To get the candidate string pairs by filtering. Then delete the string pairs that are impossible similar. The way is the same as that in Para-Join algorithm. shows an instance to get the candidate string pairs. 3\) To get the result by verification. The verification process is the same as that in Para-Join algorithm. # 6. Experimental evaluation ## 6.1 Experimental environment In this section, we evaluate the parallel string join algorithms based on the real datasets. Four datasets are used in the experiment. All the datasets can be downloaded from <http://doi.org/10.5281/zenodo.293041>. Dataset Ⅲ and Ⅳ can be downloaded from <http://dbgroup.cs.tsinghua.edu.cn/dd/codes/pivotal.tar.gz> too. The first two datasets are relatively small and used to test the single-machine algorithms. The detailed information of these datasets is shown in. Algorithms Pass-Join, Part-Join, and Para-Join are implemented in Java, and Algorithm Pada-Join is implemented in Scala. These algorithms run on three different systems: a multi-core system, a cluster system with 4 nodes, and a single machine with the same configuration as the node in the cluster. The operating system used is Ubuntu 12.04 LTS and the version of JDK is 1.7.0_71. The detailed information of the systems is shown in where system Ⅱ is consisted of 4 nodes and they are virtual machines. The virtual machines are created on a physical hardware with CPU i7-4770 3.40 GHz \*8, RAM 16GB, and hypervisor VMware. We evaluate our framework in two aspects, efficiency and scalability. In efficiency aspect we evaluate the running time of parallel processing with multi-threads and multi-tasks against the existing algorithms. In scalability aspect we evaluate the influence of the number of threads or tasks attending the computation. ## 6.2 Efficiency analysis In this Section, we compare our algorithms with two existing algorithms Pass- Join and Part-Join. For Para-Join, the number of threads is set to 8. The similarity threshold *τ* ranges from 1 to 8. The experimental results are shown in Figs and. Because similarity thresholds have high influence on the running time, the results are shown in different figures with varying thresholds. When the similarity threshold *τ* is small, there is no big difference for the running time among algorithms. For example, when *τ* is 1, the running times of the three algorithms on dataset I are 22s, 25s, and 23s respectively. When the value of τ increases, algorithm Para-Join can show more advantages. For example, when τ is 8, the running time on dataset I of Para-Join is 49s while the others are 136s and 114s respectively. It maintains the same advantage on dataset II. The main reason is that our algorithm can concurrently find the similar pairs in the dataset by using the multi-threading technique. When we test dataset Ⅱ in system Ⅱ and system Ⅲ, the running time is bigger than in other algorithms.. shows the results. We realize that Pada-Join is not suitable for small datasets. When we test dataset Ⅲ and dataset Ⅳ in system Ⅲ, the memory overflow error occurs. However, Pada-Join completes the work successfully. We also realize that Para-Join, Pass-Join, and Part-Join are not suitable for big datasets. For Para-join algorithm, the implementation is to load the input into memory at first and then to process it. For Pada-join, the Spark framework will divide the input into several blocks and store them in the HDFS (Hadoop distributed file system). The size of a block is limited and it can be loaded into memory at the same time. After Spark finishes the processing of one block, it will load another block. The basic differences between Para-join and Pada-Join are their implementation ways and the platforms they run on. So Para-Join algorithm is unable to handle the larger dataset. ## 6.3 Scalability analysis We have designed two cases to evaluate the scalability of the algorithms Para- Join and Pada-Join. ### Case 1 Under the same dataset, we compare the running times by changing the number of threads from 2 to 8 and changing the similarity threshold *τ* from 1 to 8. The experimental results are shown in Figs and. From the figures, we can observe that the running time increases as the value of *τ* increases. The reason is that for the same dataset, when *τ* increases, candidate pairs in the dataset are also increased, result in more operations in the verification process. However, when the value of *τ* is big enough, e.g., *τ* is 8, the running time remains unchanged or even becomes larger. The reason is that a large number of threads increase the communication overhead. ### Case 2 Under the same system configuration, we compare the running times for dataset Ⅲ and dataset Ⅳ by changing the similarity threshold *τ* from 1 to 8. These two datasets are too large to the extent that the other algorithms cannot handle. As *τ* becomes larger, the algorithm becomes more complicated. Because the number of the candidate pairs increases with the size of datasets, the running time also increases. The experimental results are shown in. The algorithms can perform excellently on larger data sets. # 7. Related work There are many previous studies on the development of efficient solutions to the string similarity join problem. ## String similarity functions The string similarity functions are the key for all the string similarity join algorithms. String similarity functions are used to quantify the similarity of two strings. The existing string similarity functions can be roughly divided into two groups, character-based similarity, and set-based similarity. The character-based similarity considers characters in strings to quantify the similarity, such as Edit distance, Hamming distance, and character n-gram similarity. The set-based similarity quantifies the similarity based on the token sets. These functions include Jaccard similarity, Cosine similarity, and Dice similarity. Besides the above similarity functions, there are also some new functions, such as Jaro-Winkler measure and Hidden Markov Mode-based measure. ## String similarity join methods The existing methods for string similarity join can be broadly separated into two categories, based on the filtering-verification framework and the tire tree. Most of the existing methods adopt the first one. These methods include All- Pairs-Ed, ED-Join, AdaptJoin, Part-Enum, Pass-Join, and Part-Join. All-Pairs-Ed is a q-gram-based method, ED-Join improves All-Pairs-Ed using location-based and content-based mismatch filter by decreasing the number of grams, and AdaptJoin algorithm improves the prefix filtering fundamentally for all similarity metrics. Trie-Join and Bed-Tree use a trie tree to do similarity join. With the improvement of these methods, many filtering techniques are proposed such as count filtering, length filtering, position filtering, prefix filtering, and content filtering. Additionally, some parallel methods have been proposed for string similarity join, such as bit-parallel, MassJoin, V-SMART-Join, et al. ## String similarity search It is similar to string similarity join. Firstly the index of the string collection is built. When a query request is submitted, a large number of dissimilar strings are filtered according to the given query string, and then the candidate strings are matched with the given string according to the similarity function. ## Parallel processing techniques There are a lot of works on implementing string join using Map-Reduce framework. Vernica et al. proposed a similarity join method using MapReduce which utilized the prefix filtering to support set-based similarity functions. They selected a subset of tokens as signatures and proved that two strings are similar only if their signatures share common tokens. Afrati et al. proposed multiple algorithms to perform similarity joins in a single MapReduce stage. They analyzed the map, reduce, and communication cost. However, for long strings, it is rather expensive to transfer the strings using a single MapReduce stage. Kim et al. addressed the top-k similarity join problem using MapReduce. Deng et al proposed Mass-Join which extended the existing partition-based signature scheme to support set-based similarity functions. In this paper, we take the multi- threading technology and the multi-tasking technology into consideration and compare them in the string join field. # 8. Conclusions In this paper, a parallel processing framework for string similarity join is proposed for high efficiency. Algorithm Para-Join based on the framework adopts the multi-threading technique and runs on the multi-core system. Algorithm Pada- Join, also based on the framework, adopts the distributed computing technique and runs on the distributed systems. Some conclusions are given by the experimental results and analysis. For relatively small datasets Para-Join can provides very good scalability and outperforms state-of-the-art algorithms because it completes the computation of string similarity join in one node and avoids the overhead of network communication. However, the availability of single-machine algorithms is limited by the memory. For relatively big data set, Pada-Join shows its advantages because of the good scalability of the distributed systems. In the future, we will adopt larger datasets to test Pada- Join algorithm and improve its performance. [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** CY. **Data curation:** CY. **Formal analysis:** YH. **Funding acquisition:** CY. **Investigation:** CY. **Methodology:** CY. **Project administration:** CY. **Software:** XZ. **Validation:** QZ.

# Introduction Neuroblastoma (NB) is the main cause of tumor-related deaths in children worldwide. Diagnosis and treatment have made great progress in the past 20 years and the average 5-year relative survival rate of NB has reached 50%. Currently, the diagnosis of NB mainly relies on histopathological examination, imaging results, and molecular biomarkers. Early detection of NB is difficult. This may be the most important factor affecting the mortality of patients with NB. Therefore, further study of the molecular mechanisms underlying NB and identification of effective molecular markers for early cancer screening are essential to enhance the therapeutic outcomes and quality of life of children. RNA binding proteins (RBPs) are pleiotropic proteins that can regulate gene expression by interacting with various types of RNA, including rRNA, ncRNA, snRNA, miRNA, mRNA, tRNA, and snoRNA. Currently, over 1500 different types of RBPs have been identified in the human genome through whole-genome sequencing. Specifically the ribonucleoprotein complex formed by the binding of RBP and target RNA regulates the stability of mRNA at the post-transcriptional level, thereby affecting RNA processing, splicing, localization, export, and translation. Ultimately, post-transcriptional modification of mRNA leads to the regulation of various important physiological processes of cells. Current research has found that RBPs play an important role in many human diseases and are key regulators of the development and progression of cardiovascular diseases, myotonic muscular dystrophy, neurological diseases, and cancer. In recent years, new bioinformatics approaches, such as bipartite network projection, IRWNRLPI (Integrating Random Walk and Neighborhood Regularized Logistic Matrix Factorization for lncRNA-Protein Interaction), and HLPI- Ensemble, have greatly improved the research and prediction of RBP function. Herein, we used high-throughput bioinformatics analysis to identify RBPs that are differentially expressed in cancer samples and normal tissue samples, and systematically investigated their expression profiles, functional effects, and potential mechanisms to understand their role in tumors. This study will deepen our understanding of the molecular mechanism of NB and provide potential diagnostic or prognostic biomarkers for NB. # Materials and methods ## Data sets and preprocessing We obtained RNA expression datasets and corresponding clinical data of NB patients from the Therapeutically Applicable Research To Generate Effective Treatments project database (TARGET, <https://target- data.nci.nih.gov/Public/NBL/mRNA-seq/L3/>), and normal neural tissues samples datasets from Genotype-Tissue Expression Database (GTEx, <https://gtexportal.org/>), respectively. All data derived from an open access data platform, and thus, this study did not require ethics committee approval. To determine differentially expressed genes between NB tissue and normal samples, the Limma software package was used for analysis. The GSE85047 dataset was downloaded from Gene Expression Omnibus (GEO) (<https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=gse85047>) and was used as a validation cohort. ## Gene ontology enrichment and KEGG pathway analysis Gene enrichment analysis and pathway analysis was carried out using the R package “clusterProfiler”. ## Protein-–protein interaction network building and subnet detection Differential protein–protein interaction (PPI) information for RBP was evaluated using the STRING database (<http://www.string-db.org/>) and further building and visualization of the PPI network was performed using Cytoscape 3.7.0 software. We used the molecular complexity detection (MCODE) plug-in to cluster genes in the PPI network and to build functional modules. A P-value\<0.05 was considered a statistically significant difference. ## Prognostic model construction The R package “survival” was applied to carry out univariate Cox regression analysis using all differentially expressed RBPs to identify prognostic genes, and Lasso regression was performed to further screen important key genes. Finally, based on the preliminary screening of the above key candidate genes, we built a multivariate Cox proportional hazard regression model and evaluated the survival of patients using the following risk score formula: $$Risk\ score = \beta 1*\mathit{Exp}1 + \beta 2\ Exp2 + \ldots + \beta i\ Expi$$ where, β was the value of the risk coefficient, and Exp represented the value of the expression of a certain gene. Based on the median value of the risk score, NB patients were divided into two groups: low-risk group and high-risk group, and the survival differences between the two subgroups were compared through survival analysis. In addition, the prognostic ability of the above model was estimated through receiver operating characteristic curve (ROC) analysis. A sample of 276 NB patients with reliable follow-up information from the GSE85047 data set was used as the validation group to evaluate the predictive power of the prognostic model. P\<0.05 was considered a statistically significant difference. ### Constructing the lncRNA-miRNA-mRNA network of key RBPs To study the relationship of lncRNA, miRNA and mRNA, based on the interactions of miRNA-lncRNA and miRNA-mRNA, the online database such as starBase (<http://starbase.sysu.edu.cn/agoClipRNA.php?>), targetscan(<http://www.targetscan.org/vert_72/>), and LncBase (<http://carolina. imis.athenannovation.gr/diana_tools/web/index.php?r=lncbasev2%2Findex>) were used to build the lncRNA-miRNA-mRNA network. # Results ## Identification of differently expressed RBPs in NB patients We performed a methodical analysis of the key role and prognostic value of RBP in NB. <u>The study workflow is illustrated in.</u> The NB data was downloaded from TARGET, which contained 144 tumor samples, and the normal nerve tissue data was downloaded from the GTEx database and contained 278 samples. After analyzing the currently known 1542 RBPs, 348 RBPs with significant differences in expression (P-adjusted \<0.05, \|log₂ fold change \[FC\]\|\>1.0) were identified, and comprised 166 up-regulated RBPs and 182 down-regulated RBPs. ## Enrichment analysis of the differently expressed RBPs In order to study the functions and mechanisms of the selected RBP, we used the R package clusterProfiler for enrichment analysis. The results showed that enriched biological processes mainly involved mRNA processing, RNA splicing; ncRNA metabolic processes; RNA phosphodiester bond hydrolysis; RNA splicing, via transesterification reactions with bulged adenosine as a nucleophile; mRNA splicing, via spliceosome; RNA splicing, via transesterification reactions; nucleic acid phosphodiester bond hydrolysis; and RNA catabolic processes. Molecular functions included catalytic activity, acting on RNA ribonuclease activity; nuclease activity; mRNA 3’-UTR binding; endonuclease activity; translation regulator activity; catalytic activity, acting on a tRNA; mRNA binding; double-stranded RNA binding; endoribonuclease activity; and single- stranded RNA binding. Cellular components involved mainly the ribonucleoprotein granule, cytoplasmic ribonucleoprotein granule, ribosome, ribosomal subunit, organellar ribosome, mitochondrial ribosome, P-body, mitochondrial matrix, P granule, and pole plasm. The KEGG analysis indicated mainly enrichment in RNA transport, the mRNA surveillance pathway, ribosome biogenesis in eukaryotes, RNA degradation, ribosome, aminoacyl-tRNA biosynthesis, spliceosome, and the RNA polymerase, Influenza A pathway (Tables). ## PPI network building and subnet detection To further study the function of differentially expressed RBPs and their role in the development of NB, we used Cytoscape software to create a PPI network that contained 311 nodes and 1766 edges. The co-expression network was analyzed using MCODE to recognize potential key modules. The RBPs in subgroup 1 were mainly enriched in ribosome biogenesis in the eukaryote pathway, ribosome biogenesis, rRNA processing, ncRNA processing, maturation of SSU-rRNA, ribosomal small subunit biogenesis, rRNA metabolic process, maturation of SSU-rRNA from tricistronic rRNA transcript (SSU-rRNA, 5.8S rRNA, LSU-rRNA), and in ribosomal large subunit biogenesis. ## Selection of prognosis-related RBP Difference analysis identified a total of 348 key RBPs. To identify the prognostic significance of these RBPs and their effects on clinical outcome and survival, we performed univariate Cox regression analysis and obtained four candidate RBPs related to prognosis. Subsequently, using Lasso regression, the prognostic risk score including multi-factor Cox regression values was constructed. ## Prognosis-related RBPs model building and analysis CPEB3 and CTU1 were identified as the key prognostic genes using the multivariate Cox regression analysis. We used these two hub genes to construct the predictive model. The risk score for each child was calculated based on the following formula: $$Risk\ score = (‐0.60901*expCPEB3) + (0.851637*expCTU1)$$ Then, based on the median value of the individual risk scores, 144 NB patients were stratified into two groups: the low-risk and high-risk groups. The results showed that compared with patients in the low-risk group, patients in the high- risk group had significantly poorer survival (P = 2.152e-04). The value of the area under the curve (AUC) in the TARGET model was 0.720 (Figs,). ## Validation of hub RBPs To evaluate the prognostic value of the RBP prediction model, we used the GSE85047 patient cohort to verify the relationship between risk score and survival. In the GSE85047 cohort, groups were also grouped based on the median value of risk score in the TARGET model. The survival of patients with high-risk scores was poorer than for patients having lower risk scores (P = 0.1237e-08), and the AUC was 0.730 (Figs, and). ## The RBP risk score was an independent prognostic factor We assessed the prognostic value of risk scores of RBPs. For the NB TARGET cohort, the risk scores on univariate analysis were significantly correlated with overall survival (OS) (HR = 1.535, 95% CI = 1.368–1.722, P = 2.69E-13). The multivariate analysis showed that the risk score was an independent prognostic indicator (HR = 1.518, 95% CI = 1.344–1.715, P = 1.91E-11). We constructed a nomogram that integrated multiple risk factors to quantify individual risks to be used in the clinical setting to predict the OS probability at 1, 2, and 3 years. ## Constructing the lncRNA-miRNA-mRNA network of key RBPs Using online databases, we constructed the lncRNA-miRNA-mRNA network of key RBPs. # Discussion The prognosis of different NB patients varies greatly, given the extensive tumor heterogeneity of NB. For low-risk NB patients (most commonly infants), simple observation or surgical treatment can often achieve good results; but for high- risk NB patients, even if a variety of intensive treatment options are combined, the prognosis is still not ideal. The true cause of NB is still unclear. In recent years, with the emergence of immunotherapy and new drugs, the survival of patients in the high-risk group has improved to a certain extent. RBPs have always played a role in the life of RNA. It is not an exaggeration to say that without RBP, RNA cannot achieve anything. The main role of RBPs is to mediate RNA maturation, transport, localization, and translation; one RBP may have multiple target RNAs, and defect in expression may cause multiple diseases. Recently, the importance of RBP in tumor occurrence, development, and metastasis has increasingly been revealed. In our study, we identified 348 RBPs based on NB datasets from the TARGET NB dataset. We systematically analyzed the related biological functions and built RBP PPI networks and the relative subnets. In addition, we performed univariate Cox regression analysis, survival analysis, Lasso regression analysis, and multivariate Cox regression analysis of differential RBPs to further investigate their biological function and prognostic value. The GO enrichment and KEGG pathway analysis of these differentially expressed RBPs indicated that RBP were used in mRNA monitoring pathways, RNA transport, ribosomal biogenesis in eukaryotes, RNA degradation, ribosomes, aminoacyl-tRNA biosynthesis, and spliceosomes. RBPs were significantly enriched in the RNA polymerase pathway, and play a critical role in mRNA processing, RNA splicing, ncRNA metabolism, RNA phosphodiester bond hydrolysis, catalytic activity, and act on RNA ribonuclease and nuclease activity. The PPI network revealed the relationship between RBPs. At present, with the development of computational biology, it is possible to predict the complex regulatory relationship between proteins, which provides a powerful tool for further research. At present, many studies have described the role of RBPs role in metabolism and disease. RBPs plays a dual and opposite role in tumorigenesis, regulating the proliferation of early tumor cells and promoting tumor progression and metastasis in advanced cancer. According to these reports, abnormal expression of multiple RBPs had been found in many malignant tumors. However, the impact of RBP on the occurrence and development of cancer is still poorly understood. In our study, two RBPs were identified as hub RBPs related to NB prognosis: *CPEB3* and *CTU1*. *CPEB3*, also known as the Cytoplasmic Polyadenylation Element Binding Protein is an RBP that shuttles through the cytoplasm. It mainly exists in the cytoplasm and can inhibit the translation of target RNA. GO analysis showed that *CPEB3* was associated with mRNA processing, RNA catabolic processes, mRNA 3’-UTR binding, translation regulator activity, and mRNA binding. *CPEB3* can disrupt the crosstalk between cancer cells and tumor- associated macrophages through the IL-6R/STAT3 signaling pathway, and thereby inhibit epithelial-mesenchymal transition. Studies have found that *CPEB3* is related to the tumorigenesis and development of glioma, high-grade serous ovarian cancer, colorectal cancer, hepatocellular carcinoma, and cervical cancer. *CPEB3* is expressed in both the brain and heart and is involved in the regulation of synaptic plasticity by down-regulating the expression of several plasticity-related proteins (PRPs), such as N-methyl-D-aspartate receptor (NMDAR) and postsynaptic density protein 95 (PSD95). Down-regulation of CPEB3 expression can affect NMDAR activated CaMKII α, which overcomes the inhibition of synaptic transmission under stress. In addition, *CPEB3* knockout mice exhibited hippocampus-dependent neurological dysfunction, including acquisition and extinction of long-term spatial memory and short-term fear memory. In nerve cells, activation of NMDAR can increase the expression of *CPEB3* in the nucleus and redistributes its content in nucleus and cytoplasm. *CPEB3* has low expression in colon cancer and HPV-positive cervical cancer, which indicates it may be involved in tumorigenesis as a tumor suppressor, but the specific mechanism still needs further study. CPEB protein family members contain multiple miRNA binding sites, and can be regulated by a variety of miRNAs. After up regulating the expression of mir-107 in hepatoma cell lines Huh7 and HepG2, the mir-107 bound to the 3‘-UTR region of the *CPEB3* transcript, and resulted in an increased expression of EGFR and phosphorylated Akt and decreased expression of p21. Mir-107 and CPEB3 interaction plays an important role in the occurrence and development of hepatocellular carcinoma by regulating the EGFR signaling pathway. *CTU1*, or cytoplasmic sulfurylase subunit 1, is a protein-coding gene. Diseases associated with CTU1 include spinal cord septal membrane tumors and spinal gliomas. The related pathways include gene expression and tRNA processing, and its function is related to tRNA binding and nucleotide transferase activity. GO analysis shows *CTU1* was associated with ncRNA metabolic process and ncRNA processing. *CTU1* promotes cancer resistance to targeted therapy, its expression is associated with higher rates of morbidity and mortality in spinal cord gliomas, and up-regulation of *CTU1* is involved in human breast cancer metastasis. In our study, based on the two hubs RBPs identified in the TARGET cohort training set, multi-step Cox regression analysis produced a risk score model that could predict the prognosis of NB. In the NB TARGET and GSE85047 cohorts, the survival outcomes of the high- and low-risk subgroups were significantly different. The ROC values of the risk score model of the training set and validation set were 0.72 and 0.73, respectively, indicating that the 2-gene marker prognostic model applied to evaluate the prognosis of NB patients has a certain value. Both genes were differentially expressed in NB and correlated with survival, which strongly suggests that these two genes are potential tumor- related genes and require further study. In the future, we intend to establish a real-life cohort of NB patients to confirm the validity of these genes. However, the molecular mechanisms involving these two RBPs are unknown, and further study of their underlying function may be valuable. In recent years, more and more studies have confirmed that long non-coding RNA (lncRNA) and micro RNA (miRNA) and their interactions play an important role in the diagnostic biomarkers and therapeutic targets of various diseases. Some theoretical methods have played a role in predicting potential lncRNA-miRNA associations. With the development of bioinformatics, some new prediction methods, such as the lncRNA–miRNA interactions prediction by logistic matrix factorization with neighborhood regularized (LMFNRLMI), enable us to study lncRNA-miRNA interactions more accurately, that is, to predict lncRNA-miRNA interactions. In summary, we systematically studied the function and prognostic value of RBPs differently expressed in NB. These RBPs may be associated with the occurrence, development, invasion, and metastasis of NB. The establishment of a prognostic model of NB based on two RBP coding genes is conducive to clinical application. Our results contribute to a better understanding of the pathogenesis of NB and the development of new therapeutic and prognostic molecular markers. Although our gene signature and nomogram showed excellent performance in the training set and validation set, both inevitably had some limitations. First, although our risk score performed well in predicting the survival rate of NB patients, it lacks confirmation from large-scale prospective trials. Secondly, the validation data derived only from the GSE 85047 dataset; thus, the predictive value of our model requires further verification. Third, the molecular mechanisms involving *CPEB3* and *CTU1* have not been verified in NB cells. Thus, our follow-up studies will verify the conclusions reached in this study from the aspect of their clinical application and molecular mechanisms. # Supporting information [^1]: The authors have declared that no competing interests exist.

# Introduction Testicular and epididymal secretions aid the maturation of mammalian spermatozoa to acquire fertilizing ability and this process that involves a series of complex and sequential events involving structural, physiological and biochemical changes. A comparison of the proteomes of testes, epididymis and spermatozoa revealed that 47% of the proteins in the sperm are intrinsic and are acquired from testes. 23% of the proteins are extrinsic that are acquired from the environment, clearly suggesting that secretory proteins in the lumen are added on to the sperm surface. It is also reported that acrosomal protein content of caput and caudal sperm are different, suggesting that sperm undergoes changes during the transit and this is due to addition of a wide variety of proteins added on to its surface. Examples include HongrES1, HE4, cystatin 11 (CST11), lactoferrin, human cathelicidin antimicrobial peptide (hCAP18), ESP13.2, members of the SPAG11 family, members of the PATE family and defensins. Some of the members of defensin, SPAG11 and PATE families are shown to have role in fertilization, suggesting bifunctional role for these proteins in epididymal innate immunity and sperm maturation. Further, amyloidogenic proteins such as cystatin-related epididymal spermatogenic (CRES) protein in the acrosomal matrix of the spermatozoa form amyloids *in vitro* and *in vivo*, besides being antimicrobial. Thus, proteins secreted on to the sperm surface exert diverse function and have been drawing attention in the last two decades. Among the different types of lysozymes, namely, g-type (goose-type), i-type (invertebrate-type), c-type (chicken or conventional-type), plant, phage and bacterial, the c-type is the predominant type and widely expressed in many species. A common feature of most lysozymes is that they possess antibacterial activity. Recently lysozyme- like (*Lyzl*) genes that belong to c-type lysozyme family were identified. Zhang et al., reported the mRNA expression of *Lyzl*2, *Lyzl3*, *Lyzl4* and *Lyzl6* in humans tissues. *Lyzl*2, *Lyzl4* and *Lyzl6* mRNA were found to be expressed only in the testes and *Lyzl*3 was found to be expressed in testes and pancreas. Sperm lysozyme-like protein 1 (SLLP1 or LYZL3) was found to be an intra-acrosomal and non-bacteriolytic c-type lysozyme-like protein in human spermatozoa. LYZL4 is a sperm bound protein with a role in fertilization and is expressed in testes and epididymis. Along with cystatin-c, cystatin 8 and premelanosome protein, presence of lysozyme-like 1, 3, 4 and 5 (LYZL1, LYZL3, LYZL4 and LYZL5) proteins in the acrosomal matrix of the mouse spermatozoa was reported. The LYZL proteins that are similar to lysozyme also possess the amyloidogenic domains suggesting that they may play important role like CRES proteins. Alpha lactalbumin, otherwise called as LYZL7 belongs to glycoside hydrolase family and is a homologous to lysozyme. Recently, a modified form of lactalbumin HAMLET (human lactalbumin made lethal to tumors) was reported to have apoptotic activity against tumor cells. *Lyzl3/Sllp1* was found to be expressed specifically in the male reproductive tract. Spermatozoa incubated with antibodies to human SLLP1 failed to fertilize eggs, thereby demonstrating a role in male reproductive function. Microscopic studies revealed that LYZL3 is located in the acrosomal region before capacitation and moves towards the equatorial segment after capacitation, suggesting that LYZL3 may be an intra-acrosomal protein that is tightly bound to the sperm membrane. Further, interaction studies performed using LYZL3 and oocyte lysate revealed that it may interact with SAS1B, an oolemmal protein and the same was confirmed by surface plasmon resonance. Similarly, incubation of spermatozoa with the mouse LYZL4 antibodies resulted in loss of fertilizing ability. In the mouse, LYZL6 was reported to be present in testes, epididymis and spermatozoa and is antimicrobial in nature. We previously reported the gene expression pattern of some rat *Lyzl* genes with emphasis on *Lyzl4*. However in-depth analysis of their expression pattern and biochemical functions in general physiology and in the male reproductive function in particular are lacking. Hence, in this part of the study we attempted to characterize the expression of rat *Lyzl* genes and their protein products. Biochemical characterization was also undertaken to understand their role in general physiology. # Materials and Methods ## *In silico* analyses Using chicken lysozyme gene sequence, the rat genome (build RGSC v3.4) at NCBI was searched to identify the *Lyzl* genes. The sequences were then analyzed using *in silico* tools to predict various properties of the genes and their proteins. Genomic neighborhood analysis was performed based on the genome assemblies deposited in NCBI and Ensembl. Phylogenetic tree was constructed using neighborhood joining method to understand the evolution of rat LYZL proteins and their conservation among various organisms. Unrooted tree was constructed with the maximum sequence difference of 0.85. Domain / motif prediction was performed for rat LYZL proteins using Multiple Em for Motif Elicitation program (<http://meme.nbcr.net/meme/cgi-bin/meme.cgi>). The secondary structure of rat LYZL proteins was analyzed by Self-Optimized Prediction Method with Alignment (SOPMA) to predict the possible secondary structures like α-helix, β-turn, extended strand and random coil. The program gives the output in the form of percentage of each secondary structure present in the given protein sequence. ## Molecular modeling SWISS MODEL, accessible via the ExPASy web server, or from the program Deep View (Swiss Pdb-Viewer) was used to predict the three dimensional models of LYZL proteins. Full length sequence (without the signal peptide) of the LYZL proteins in FASTA/Raw format or SwissProt / UniProt accession code that are retrieved from NCBI were given as an input for the program. The output was generated as a 3D model using nearest homolog as a template with higher sequence identity using BLOSUM 45, 62, 80 at different Pfam value. The models generated were validated using PROCHECK. Proteins showing more than 90% in the core region and not more than 5% in the disallowed region are considered as good model. In cases where the models does not meet the requirements of quality structure, modeling was repeated after loop refinement and energy minimization and the structures were validated by analyzing the stereochemical features using PROCHECK. PYMOL was used to visualize the structure of the modelled rat LYZL proteins. ## Docking analysis GOLD (Genetic Optimization for Ligand Docking) program was used to analyze the binding ability of LYZL proteins to N-acetyl glucosamine (NAG) trisaccharide. The flow chart of analysis is given in the. To the 3D models generated in this study, hydrogen molecules were added to the receptor and docking (rigid and flexible) performed with tautomers of the ligand and a particular atom number was given from the identified active site with a radius of 10.0 Å. All the other fitness function parameters and the genetic algorithm parameters were kept in default mode. The GOLD was run and the output was viewed using Goldmine and Silver. The output was produced as GOLD Fitness scores and different energy functions. The GOLD fitness scores of 40 and above generated using Goldmine and Silver were mainly considered. The output of these protein-ligand complexes were exported as PDB files using GoldMine. These complexes were then analyzed using molecular graphics viewers Discovery studio or PyMOL. The output was analyzed for the properties such as hydrogen bonding and Vander Waal’s interaction using LigPlot. ## Polymerase chain reaction Wistar rats aged 90 days were obtained from National Institute of Nutrition, Hyderabad. 2 μg of total RNA isolated from brain, heart, lungs, liver, kidney, spleen, ovary, uterus, cervix, caput, corpus, cauda, testis, seminal vesicles and prostrate was reverse transcribed and using gene specific primers, the mRNA expression of rat *Lyzl5* and *Lyzl7* was analyzed using standard PCR conditions (94°C for 1 min followed by 30 cycles at 94°C for 30 sec, 58°C for 30 sec and 72°C for 30 sec, and with a final round of extension at 72°C for 10 min). The internal control gene glyceraldehyde 3 phosphate dehydrogenase (*Gapdh*) was amplified with the same conditions. PCR amplified gene products were analysed by electrophoresis on 2% agarose gels. To study the androgen regulation of *Lyzl* transcripts, epididymides were obtained from sham operated, castrated and testosterone supplemented Wistar rats (n = 5 in each group). Testosterone supplementation was by a 20 mg dihydrotestosterone pellet implanted subcutaneously immediately after castration. The animals were sacrificed 14 days after castration. All procedures involving animals were conducted using the guidelines for the care and use of laboratory animals and this study was specifically approved by the Institutional Animal Ethics Committee of University of Hyderabad (UH/IAEC/2012/AUG/2). For studies on the developmental regulation of *Lyzl* genes, testes and epididymides were obtained from 10- to 60-day old Wistar rats. ## Recombinant protein production The full length cDNA of rat *Lyzl 1*, *3*, *4*, *5*, *6* and *7* without the signal peptide were amplified and cloned into pQE80m vector. Plasmid containing one of the *Lyzl* coding region was transformed into *E*.*coli BL21* and the recombinant protein expression was induced with 1mM IPTG for 3 hr. Recombinant protein was purified using Ni-NTA agarose affinity purification system as per the manufacturer’s instructions (Qiagen, Valencia, CA, USA). The purified recombinant protein with a tag (MRGSHHHHHHGS) at the N-terminus was confirmed by Western blotting using anti-His tag antibody (Santa Cruz Biotechnology, Dallas, USA; sc-57598 RRID: AB_831408). The fractions that contained the protein of interest were dialyzed extensively at 4°C against 10 mM sodium phosphate buffer, pH 7.4. ## Generation of polyclonal antibodies Antibodies to LYZL proteins were raised as described earlier. New Zealand white rabbits aged four months were obtained from National Institute of Nutrition, Hyderabad and immunized intradermally with 600 μg of the recombinant protein mixed with equal volume of Freund’s complete adjuvant. Booster doses containing 600 μg protein was mixed in Freund’s incomplete adjuvant were administered on 3^rd and 28^th day. Antibody titer was checked on 35^th day and final bleeding was done on 42^nd day. Each antibody was checked for cross reactivity against each and every LYZL protein to ensure there is no cross reactivity within recombinant LYZL proteins. We observed no cross reactivity for all the antibodies. The antibodies generated are registered at the Resource Identification Portal (<http://antibodyregistry.org/>) with the following RRID numbers: LYZL1 –AB_2616571; LYZL3 –AB_2616572; LYZL4 –AB_2616573; LYZL5 –AB_2616574; LYZL6 –AB_2616575; LYZL7 –AB_2616576. ## Immunoblotting Testes, caput, cauda, seminal vesicles and prostate tissues were collected from 90 day old Wistar rats and 10% homogenates were prepared in RIPA buffer (25mM Tris-HCL, pH 7.6; 150 mM NaCl; 1% each of NP-40, sodium deoxycholate and sodium dodecyl sulfate) containing protease inhibitors. For sperm lysate preparation, 5 X 10⁶ spermatozoa were used. The homogenates were then centrifuged at 10,000 rpm for 10 min to remove the debris and concentration of the protein in the supernatant was quantified by Lowry’s method. 100 μg of the total protein was separated by electrophoresis on 15% SDS PAGE and transferred to nitrocellulose membrane at 25V for 16 h. The membrane was then blocked with 5% skim milk for 2h at room temperature followed by probing with primary antibody (immune serum) for 1h, subsequently washed with TBS (Tris-buffered saline) and TBS-T (Tris-buffered saline, 0.1% Tween-20). It was then incubated with Goat anti-Rabbit IgG-HRP antibody (Santa Cruz Biotechnology, Dallas, USA; sc-2054 RRID: AB_631748) followed by washes with TBS and TBS-T each for 10 min. At the end of washing, the membrane was developed using enhanced chemiluminescence (GE Healthcare, Buckinghamshire, UK) kit. ## Immunofluorescence Wistar rats aged 90 days were dissected to collect the testes, spermatozoa and epididymides. Testes and epididymides were fixed by immersing in Bouin’s solution and 4% paraformaldehyde solution respectively. Serial sections (five microns) of testes and epididymides were preheated to 60°C for 5 min followed by washings with xylene, gradient alcohol (70–100%) and PBS for 10 min each. Following antigen retrieval (by heating in 10 mM citrate buffer, pH 6.5 for 12 min), permeabilization with PBS containing 1% triton X-100 (PBST) for 15 min and blocking with 10% goat serum for 45 min, the sections were incubated with immune serum and Goat anti-Rabbit IgG tagged with TRITC or FITC (Santa Cruz Biotechnology, Dallas, USA; sc-2091 RRID: AB_649008; sc-2012 RRID: AB_631744) for 1 h each with washing with PBS in between incubations. 4', 6-diamidino-2-phenylindole (DAPI; Sigma Aldrich, St. Louis, USA)) was used to stain the nucleus and images taken using confocal microscope. In the case of spermatozoa, smears on glass slides were prepared and air dried. They were then permeabilized with PBST, blocked with 10% goat serum and processed in a similar way as that of tissue sections. ## Circular Dichroism Circular dichroism was performed for recombinant LYZL proteins using Jasco J-810 spectropolarimeter. Quartz cell with a path length of 0.2 cm was loaded with 200 μl of 0.1μg/μl protein and scanned in the far UV region (180–260 nm). Three scans at a scan speed of 50 nm/min were accumulated and the polarimetry data was collected for every 1 nm. The data thus collected was used for calculating mean residue ellipticity (MRE) and the spectrum was plotted. In addition, the spectra of appropriate blank solution, 10 mM phosphate buffer was subtracted from the spectrum of the protein. Percentage of secondary structure elements were calculated using K2D3 tool of Dichroweb. ## Muramidase Assay The muramidase assay is based on the cleavage of β-glycosidic bond between N-acetyl muramide and N-acetyl glucosamine. Reduction in O.D at 450 nm is observed as the glycosidic bonds are broken. Recombinant rat LYZL protein was incubated with 2 ml of *M*. *lysodeikticus* cells (Sigma Aldrich, St. Louis, USA) in 50 mM KH₂PO₄-NaOH buffer, pH 7.0, and the decrease in turbidity was monitored at 450 nm in a spectrophotometer for every 60 min until 6 h. Change in O.D (Δ O.D) was calculated by subtracting the final O.D from initial O.D. Muramidase activity was expressed as Δ O.D. Lysozyme (Sigma Aldrich, St. Louis, USA) was used as a positive control. ## Isopeptidase Assay The isopeptidase activity assay is based on the cleavage of L-γ-glutamine-p- nitroanilide (L-γ-Glu-pNA) to produce p-nitroanilide (pNA), which exhibits absorbance at 405 nm. Recombinant rat LYZL proteins were added to the reaction mixture containing 1.75 mM L-γ-Glu-pNA (Sigma Aldrich, St. Louis, USA) in 0.05 M 3-(N-morpholino) propane sulfonic acid (MOPS) buffer, pH 7, containing 0.01M NaCl and the formation of pNA was monitored spectrophotometrically at 405 nm for every 1 h until 6 h. Activity was expressed as Δ O.D. Lysozyme was used as positive control. ## Antibacterial Assay Colony forming units (CFU) assay was employed to test the antibacterial activity. Briefly, overnight cultures of *E*. *coli* XL-1 blue were grown to mid-log phase (A₆₀₀ = 0.4–0.5) and diluted with 10 mM sodium phosphate buffer (pH 7.4). Approximately 2 X 10⁶ CFU/ml of bacteria was incubated at 37°C with 10–100 μg/ml of the recombinant LYZL protein and aliquots of the assay mixture were taken at 30, 60, 90 and 120 min after the start of incubation. The assay mixtures were serially diluted with 10 mM sodium phosphate buffer (pH 7.4) and 100 μl of each was spread on a Luria–Bertani agar plate and incubated at 37°C overnight to allow full colony development. The resulting colonies were hand counted and plotted as log CFU/ml. Lysozyme was used as a positive control. ## Scanning Electron Microscopy *E*.*coli* were treated with recombinant LYZL protein for 120 min and the bacterial cells were fixed in Karnovks’s fixative solution (2% paraformaldehyde, 2.5% glutraldehyde in 0.1M phosphate buffer) for overnight at 4°C. They were then serially washed with graded alcohol (30% to 100%) for dehydration and finally suspended in acetone before embedding on carbon tape. The samples were then coated with gold and then observed under scanning electron microscope. Lysozyme was used as a positive control. ## Membrane potential measurement The effect of LYZL protein on the bacterial membrane potential and permeability was determined by using DiOC₂(3) and TO-PRO-3 respectively (Sigma Aldrich, St. Louis, USA). DiOC₂(3) emits green fluorescence as a single molecule, which varies with cell size and is independent of membrane potential. The red fluorescence of DiOC₂(3) is due to dye aggregation and depends on both size and membrane potential. Therefore the ratio of red to green is attained to measure the membrane potential. TO-PRO-3 is a DNA binding dye and is impermeable to live cells. Due to its ability to enter into membrane compromised cells it is used as dead cell indicator. The dead cells are eliminated when measuring the fluorescence of DiOC₂(3). *E*. *coli* grown to mid log phase were diluted in 10 mM sodium phosphate buffer pH 7.4 to a final concentration of 10⁶ to 10⁷cells / ml and then treated with 100 μg of recombinant LYZL protein or 15 μM of the bacterial membrane potential disruptor carbonyl cyanide m-chlorophenyl hydrazine (CCCP; Sigma Aldrich, St. Louis, USA) for 2 h at 37°C in orbital shaker. The cells were then washed in 10 mM phosphate buffer, pH 7.4 followed by incubation with 30 μM DiOC₂(3) and 100 nM TO-PRO-3 for 5 min at room temperature. At the end of incubation, bacterial cells were washed and analyzed in a flow cytometer (BD LSR Fortessa). The far red fluorescence of TO-PRO-3 is measured in the PerCP-Cy5-5A. The green and red fluorescence of dye DiOC₂(3) was measured in FITC and PE-Texas Red-A channel respectively. ## Peptidoglycan binding assay 96 well plates coated with 40 μg/ml peptidoglycan (PGN; Sigma Aldrich, St. Louis, USA) or hyaluronan (Sigma Aldrich, St. Louis, USA) were incubated at 37°C and at 60°C for overnight and 30 min respectively. The plates were then blocked with 1 mg/ml BSA for 2 h and varying concentrations of the recombinant LYZL protein was added to the wells and incubated for 3 h. The wells were then washed 4 times with PBS-T (PBS with 0.1% Tween-20), followed by sequential incubation with primary antibody against the LYZL protein being tested and Goat anti-Rabbit IgG-HRP conjugated antibody (Santa Cruz Biotechnology, Dallas, USA; sc-2054 RRID: AB_631748). After thorough washing, O-Phenylenediamine (OPD; Sigma Aldrich, St. Louis, USA) was used to measure the amount of antibody bound to the protein complex and the binding efficiency is measured in terms of ELISA index (EI). ELISA index is calculated by dividing the average O.D of test samples with average O.D of control samples. Lysozyme was used as a positive control. ## Hyaluronidase activity assay 0.8% hyaluronan in 300 mM phosphate buffer (pH 7.4) was mixed with melted agarose (0.8%). 100 μl of the gel was dispersed into each well of a microtitre plate. After solidification 50 μl phosphate buffer (pH 7.4) containing varying concentrations of recombinant LYZL protein was added to each well and incubated for 17 h. At the end of the incubation, solutions were removed and the hyaluronan was precipitated by adding 100 μl of 10% cetyl pyridinium chloride and incubating for 30 min at 37°C. The turbidity developed was read at 595 nm. Hyaluronidase activity was measured in terms of decrease in turbidity (or O.D). Hyaluronidase was used as a positive control. ## Free radical scavenging assay This assay is based on the ability of 1,1-diphenyl-2-picrylhydrazyl (DPPH; Sigma Aldrich, St. Louis, USA)) to undergo reduction due to the presence of an odd electron, and thus exhibits a strong absorption maximum at 517 nm. As this electron becomes paired off in the presence of a hydrogen donor, a free radical scavenging antioxidant, the absorption capacity decreases, resulting in decolorization. 0.04% of 1, 1 diphenyl-2-picryl hydrazyl (DPPH) was dissolved in methanol. To 100 μl of DPPH solution, varying concentrations of recombinant LYZL protein was added and incubated at room temperature for 30 min. Free radical scavenging capacity was assessed by measuring the discoloration of DPPH at 517 nm. Lysozyme was used as a positive control. ## Statistical Analysis Statistical analysis was performed using ANOVA and Student’s t-test available in Sigma Plot software (SPSS Inc., Chicago, IL, USA). Values shown are mean ± SD. # Results ## *In silico* characterization We previously characterized the general features of some rat *Lyzl* genes and LYZL proteins. In this study, the newly identified *Lyzl* genes and proteins along with those reported earlier were extensively characterized. *In silico* analysis of *Lyzl* genes were carried out and the attributes of the proteins were analyzed using different computational tools. *Lyzl* genes are located on different chromosomes and are not clustered, suggesting that they are not isoforms coded from a single gene. The LYZL proteins are hydrophilic, with molecular weight ranging between 16–18 kDa, with pI values between 4 to 8, possessed signal peptide and the conserved four disulfide bridges formed by 8 cysteine residues characteristic to lysozyme. LYZL1 and LYZL6 conserved the chicken lysozyme active site residues namely Glu54 and Asp71, whereas LYZL4 and 5 conserved only Glu54 and none were conserved in LYZL3 and LYZL7. Except for LYZL7, none of the LYZL proteins have calcium binding site though, they have the calcium binding domain. Glycosylation sites are present only in LYZL5 and LYZL7. Phosphorylation sites were found in all the LYZL proteins. Multiple sequence alignment using T-COFFEE program indicated high similarity of the amino acid sequence among the LYZL proteins. It is interesting to note that though there is high similarity, the amino acid sequences are not identical. CLUSTALW2 and BLAST program were used to determine the pairwise, similarity and comparison scores within the corresponding rat genes and the same are indicated in cyan, green and yellow colored boxes respectively in. Rat LYZL proteins seem to be more similar to their mouse counterparts than with the human counterparts. Highest identity was exhibited by LYZL4 and LYZL6 in all the three species. The similarity score among the rat LYZL proteins was found to be in the range of 45–65 percent, whereas the identity lies between the 27–43 percent. Genomic neighborhood analysis revealed that the rat *Lyzl* genes in general are positioned among other genes similar to that of their mouse and human counterparts except for *Lyzl5* and *Lyzl6*. Neighborhood genes downstream of human *Lyzl5* gene are different from mouse and rat. A gene duplication event may have occurred due to which *Spaca5b*/*Lyzl5b* gene is present in case of human, but not in rat and mouse. Genomic neighborhood of *Lyzl6* for rat and mouse is almost similar. However, in humans, this gene is located in an entirely different neighborhood, suggesting that recombination may have taken place in the recent past after the evolution of mouse. ## Phylogenetic analysis of LYZL proteins The presence of multiple *Lyzl* genes in all mammalian genomes as well as in the genomes of several other vertebrate species raises the possibility that the *Lyzl* gene family may have amplified early in vertebrate evolution. The phylogenetic analysis reveals that the LYZL proteins are present widely among various organisms especially the vertebrates (– Figs). The rat LYZL proteins seem to have orthologs in rodents, placentals and primates. ## Secondary structure prediction The secondary structure prediction for rat LYZL proteins using SOPMA showed that LYZL 1, 5, 6 and 7 contains 42, 44, 44 and 44 percent of α-helix respectively, suggesting that they have predominantly α-helical pattern. LYZL3 and 4 contains 39 and 55 percent of random coil and at the same time α-helix content is also comparably higher. Circular dichroism was performed to understand the folding of the recombinant LYZL proteins. The CD spectra of recombinant LYZL proteins show a peak at 210 nm which is characteristic of α-helical protein. Further, mean residue ellipticity (MRE) values for each recombinant protein when tested using K2D3 (secondary structure analysis program) showed that these proteins contain α-helix pattern in their structure. ## Molecular modelling The three dimensional structures of LYZL proteins were predicted by SWISS MODEL and validated by PROCHECK. The reliability and correlation of the predicted model was measured in terms of G- factor and root mean square deviation (RMSD) between the template and the modeled protein. The templates used for generating these models, homology between the LYZL protein and the corresponding template, G-factor and RMSD values of the predicted 3D structures are detailed in. Ramachandran plot analyses indicate that majority of the amino acids of all the LYZL proteins are in the allowed and generously allowed regions. Further, the G-factor and RMSD values of all the LYZL proteins are in the acceptable limits. ## Molecular docking To gain further insights into the functional aspects of the LYZL proteins, docking of the LYZL proteins with N-acetyl-D-glucosamine (NAG), was carried out with chicken egg lysozyme bound to NAG as the template. Chicken egg lysozyme interacted with 11 amino acids which were primarily side and main chain interactions. The nature and number of amino acids that interacted with NAG (indicated with an underscore) varied among the LYZL proteins. The remaining amino acids that were found to be interacting with NAG in lysozyme are replaced by similar but are not identical amino acids in the LYZL proteins. Further analyses were carried out to obtain docking score by GOLD. LYZL1 and LYZL6 had a score of 50.81 and 45.89 respectively. Such a high score indicates their ability to strongly interact with NAG. LYZL3, 4, 5 and 7 had docking scores of 26.58, 39.40, 18.92 and 22.9 respectively indicating weak interaction and this could be due to absence of the active site amino acids. LYZL proteins docked with NAG were superimposed with the chicken lysozyme bound to NAG trisaccharide. As shown in, LYZL1 and LYZL6 displayed a complete superimposition with chicken lysozyme without any distortion of either the protein or ligand at any location. The remaining rat LYZL proteins did not show a complete overlap in the ligand region although there is complete overlap in the protein part indicating that these proteins may not interact with the substrate. ## *Lyzl* gene expression In a previous study, we analyzed the mRNA expression pattern of *Lyzl1*, *Lyzl3*, *Lyzl4* and *Lyzl6*. In this study, we report the mRNA expression pattern of *Lyzl5* and *Lyzl7* using gene specific primers in various rat tissues. We present the data of all *Lyzl* genes for better understanding. Among the male reproductive tract tissues, *Lyzl* genes were expressed only in the testis. However, *Lyzl7* expression was detected in the cauda along with testis. Further, PCR analysis using cDNA obtained from non-reproductive tissues and female reproductive tissues indicated that *Lyzl* 3, 5 and 6 transcripts are confined only to the testis. *Lyzl1*, *4* and *7* were found to be expressed in non-reproductive tissues as well. In the epididymis, *Lyzl* genes analyzed in this study were not detected at all the ages during development. In the testes, the *Lyzl* transcripts are expressed at all the age groups starting from 30 days. ## Expression of LYZL proteins Since mRNA expression of the *Lyzl* genes was found in the rat male reproductive tissues, their translation products (proteins) were also analyzed by immunoblotting. LYZL1, 3, 4 and 5 are expressed only in testes. LYZL6 is observed in both epididymis and testes. Only LYZL3, 4 and 6 were detected on sperm. LYZL7 expression was not detected in any of the tissues analyzed. ## Immunolocalization of LYZL proteins LYZL1 protein was localized only in the testes, especially in the germinal epithelium. It was also detected in the head region of spermatozoa obtained from adult rats. Similarly, LYZL3, 4 and 5 (Figs ,) were found to be localized in the testes and on the spermatozoa. LYZL3 and 5 are localized to head region of the spermatozoa, whereas LYZL4 expression was restricted to tail region. The expression of LYZL4 on the spermatozoa was previously reported by us and the same is being included in for a comprehensive analysis. LYZL6 expression was detected in in the epididymis and testes and also in the head region of the spermatozoa. LYZL7 expression was undetectable in all the tissues analyzed. ## Muramidase assay Only LYZL1 and LYZL6 proteins exhibited a concentration dependent muramidase activity and was comparable to the positive control, lysozyme. The time course activity of LYZL6 was similar at all the concentrations used. This could be due to a very high activity of this protein even at the lowest concentration used. The remaining LYZL proteins did not show any muramidase activity The activity of LYZL4 was previously reported by us and the same result is included in for comparison. ## Isopeptidase assay Among the LYZL proteins that were tested, only LYZL1 and LYZL6 displayed isopeptidase activity in a concentration dependent manner, whereas LYZL3, 4, 5 and 7 did not exhibit any isopeptidase activity. We previously reported the activity of LYZL4 and the same result is included in this figure for comparison. ## Antibacterial assay Colony forming units (CFU) assay was employed to test the antibacterial activity of LYZL proteins. LYZL1 and LYZL6 exhibited bacterial killing activity, whereas the remaining proteins failed to decrease bacterial count. This may be due to absence of the active site in these proteins. Antibacterial activity of LYZL4 was reported and the same result is included in this figure for comparison. ## Measurement of membrane potential and Permeability The membrane potential and permeability of the bacterial cells treated with recombinant LYZL proteins was measured using DiOC₂(3). shows the measurement of DiOC₂(3) fluorescence in FITC-A (green) and PE-Texas Red-A (red) channel. The mean fluorescence intensity on PE Texas Red-A channels denotes the aggregation of the dye due to increased membrane potential. CCCP treatment caused decrease in membrane potential thereby decreased mean fluorescence intensity in PE-Texas Red-A channel. Green fluorescence is the measure of cell size to detect aggregation. CCCP treatment did not cause aggregation of bacterial cells which is indicated by the mean fluorescence intensity in FITC-A channel. Treatment of cells with recombinant LYZL1 and 6 caused increase in green fluorescence showing that they possibly tend to aggregate bacterial cells. Normalized ratio between the red and green fluorescence shows the membrane potential independent of cell size. Addition of recombinant LYZL1 or 6 protein to *E*. *coli* resulted in decreased ratio of red/green fluorescence in comparison to phosphate buffer treated bacterial cells, suggesting clump formation and also change in membrane potential due to addition of these proteins. The bacterial cells treated with lysozyme also showed a change in the membrane potential similar to CCCP. TOPRO-3fluorescence is measured in PerCP-Cy5-5A channel. Increase in mean fluorescence intensity denotes increase in the membrane permeability. Treatment with LYZL proteins resulted in increased TOPRO-3 fluorescence indicating the ability of these proteins to cause membrane permeabilization. *E*. *coli* treated with recombinant LYZL1 and LYZL6 were observed under electron microscope to study the morphological changes caused by these proteins. PBS treated *E*. *coli* cells show normal smooth surface whereas the LYZL1 and LYZL6 treated cells display rough cell surface with membrane blebbing. In addition, release of cytosolic content of the bacterial cells was observed. The actions of these proteins are similar to that exhibited by lysozyme. ## Peptidoglycan binding ability LYZL domain, which has the catalytic cleft and is responsible for binding to cell wall component, was found to be present in all the rat LYZL proteins. We observed that though all LYZL proteins possess the domain, only LYZL1 and LYZL6 show antimicrobial activity. Hence, analyzing the binding efficiency of the LYZL proteins with the bacterial cell wall components may help in understanding the differential antibacterial ability. As anticipated, lysozyme displayed a concentration dependent peptidoglycan binding ability. LYZL1 and LYZL6 had higher peptidoglycan binding ability than LYZL3, 4, 5 and 7, which may be due to presence of active site. However, the binding ability of all the LYZL proteins was significantly less than lysozyme at all the concentrations tested. ## Hyaluronan binding ability Lysozyme exhibited hyaluronan binding in a dose dependent manner which may be due to chemical similarity between hyaluronan and peptidoglycan. Among the LYZL proteins tested, LYZL3 had the highest hyaluronan binding ability followed by LYZL4, LYZL5, LYZL1 and LYZL7. LYZL6 had the least hyaluronan binding ability. ## Hyaluronidase activity Hyaluronidase used as a positive control caused the clearance of cetyl pyridinium chloride resulting in decrease of O.D at 595 nm. Surprisingly, though LYZL proteins had hyaluronan binding ability, none of them exhibited hyaluronidase activity at all the concentrations tested. ## Free radical scavenging activity Lysozyme exhibited potent free radical scavenging activity, which is evident by the discoloration of DPPH and there by decrease in the O.D at 517 nm. Except for LYZL5 and LYZL6, all other LYZL proteins caused a decrease in the O.D of DPPH at 517 nm in a dose dependent manner. Among them LYZL4 had the highest antioxidant potential. # Discussion In this study we report the molecular characterization of rat LYZL proteins using *in silico* methods. The rat *Lyzl* genes were found to be located on different chromosomes suggesting that they are independent genes and are not isoforms coded by a single gene. All the LYZL proteins display similar physical properties such as molecular weight, pI and post translational modifications. Post translational modifications are crucial for proteins involved in the male reproductive tract function. The specific posttranslational modifications that occur in LYZL proteins and their relevance to spermatogenesis, sperm maturation and function are yet to be investigated. Similarity search using BLAST shows that rat LYZL proteins are similar to c-type lysozyme and conserve the characteristic lysozyme-like super family domain and the 8 cysteines that form four disulfide bonds. The catalytic mechanism of c-type lysozymes involves the interaction of Glu35 and Asp52 of the active site with beta-1,4 glycosidic bond of the substrate. Both the active site amino acid residues are conserved only in LYZL1 and LYZL6, whereas they are partially conserved in LYZL4 and LYZL5 and completely absent in LYZL3 and LYZL7. However, all the LYZL proteins contain the additional substrate binding sites similar to c-type lysozyme. These observations suggest that LYZL proteins could have arisen from single gene and diverged at a later time point. LYZL7 is slightly different from the remaining LYZL proteins, in terms of similarity and identity with lysozyme and the presence of calcium binding site, which is a not feature of lactalbumin. At the secondary and tertiary structure levels, LYZL proteins predominantly contain α-helical pattern, a feature similar to lysozyme. Docking studies indicate substrate interactions of LYZL 1 and 6 and the amino acids present in the binding site of all LYZL proteins are characteristic to lysozyme. Amphipathicity is an important feature of antimicrobial proteins. All the rat LYZL proteins identified in this study are amphipathic in nature. However, the antimicrobial nature of rat LYZL proteins seem to be predominantly determined by the presence of the two active site amino acid residues. The conservation of active site amino acids in LYZL1 and LYZL6 confers them antimicrobial activity and may play a role in reproductive tract immunity. On the other hand, the partial conservation or absence of active site amino acid residues and the dissimilarity of LYZL7 indicates a divergence in the functional role in the reproductive tract of these proteins belonging to the same family. Identification and functional characterization of testicular and epididymal proteins provided insights in to the molecular and physiological mechanisms in male reproduction and continues to be an active area of research. In order to determine the role of LYZL proteins in male reproduction in this study, we report the expression pattern of two additional rat *Lyzl* transcripts and proteins and present the results in combination with those observed for other *Lyzl* genes characterized by us in a previously. Some of the *Lyzl* mRNA transcripts were found to be expressed only in testes in the male reproductive tract. Such an exclusive expression in the testis suggests a role for these proteins in spermatogenesis. In addition to their expression in testis LYZL1, 4 and 7 are expressed in other non-reproductive tissues suggesting that these proteins may have roles beyond reproduction. Human LYZL4 transcripts were detected in the testes and pancreas similar to the expression pattern observed in this study. The expression of mouse *Lyzl* genes in testes and epididymis was also reported. Further LYZL4 was detected in brain and lungs in addition to testes and epididymis in mouse. Our results indicate the expression of *Lyzl*4 in brain, lungs, kidney, ovary and uterus in addition of testes and epididymis. These results are more or less similar to that observed in earlier reports. However there are variations in the tissue expression pattern of these genes in different species indicating a possible variation in functional role in different species. On the other hand, an ortholog of the human LYZL2 gene is not found in the rats. The *Lysc1* gene that codes for calcium binding lysozyme is present only in higher mammals and not in rodents and this was due to deletion of the genomic region during evolution. It is possible that the genomic region that contains the *Lyzl2* gene in the rat could have been deleted during evolution. Developmental regulation of a wide variety of genes due to the fluctuations of androgens at various stages in the male reproductive system has been studied extensively. Androgen levels in the testis, epididymis and blood vary during development in rodents. *Lyzl* mRNA transcripts were not detected in the epididymides obtained from 20–60 day old rats suggesting that their expression pattern is not androgen dependent in this organ system. The presence of *Lyzl1*, *3*, *4*, *5*, *6* and *7* mRNA transcripts was observed in the testes starting from 30 day post-natal development and seem to correlate with the minimal androgen levels suggesting that *Lyzl* gene expression may be androgen dependent during development in the testis. Androgen dependent expression of *Lyzl4* during development was reported in the mouse testis. Further studies are required to determine the molecular mechanisms that operate in controlling the expression of *Lyzl* transcripts during development. Though the mRNA expression pattern of *Lyzl* genes are reported, this is the first study to report the LYZL protein expression pattern in male reproductive tract by immunolocalization and Western blotting. Localization of LYZL proteins in growing spermatids and in the germinal epithelium indicates that they may be secreted and added on to the sperm surface during spermatogenesis. LYZL4 was found to be localized in tail portion of mouse spermatozoa. Similar observation was made in case of LYZL6. Our results are in consistent with earlier reports. Though some of the Lyzl gene expression was testis specific, their protein products were detected in epididymis also. This could be due to the movement of these proteins along with luminal fluid from testis to epididymis. The localization of LYZL proteins on the sperm surface indicates their possible role in spermatogenesis, sperm maturation, capacitation and acrosome reaction, sperm- egg fusion and fertilization. Studies using active immunization or knock out models will pin point the role of each of these proteins in the male reproductive tract. Some of the proteins in the testicular and epididymal mileu bind to the spermatozoa and influence their function at multiple levels. Further, generation and maturation of male germ cells are regulated by the epididymal and testicular proteins. Besides their role in sperm maturation and function, some of these proteins are known to exhibit potent antimicrobial activity, thereby forming important components of male reproductive tract innate immunity as well. Proteins belonging to the HE2 and PATE families exhibit potent antimicrobial activity, besides their role in sperm maturation and function. Lysozyme is one of the abundant proteins in reproductive tract tissue secretions. Because of its ability to cleave the glycosidic bond of peptidoglycan, it displays potent antimicrobial activity and also influences the microenvironment by regulating the viscosity of the semen. Since LYZL proteins are similar to lysozyme, they may also play a role in male reproductive immunity. We observed that only LYZL1 and 6 display potent antibacterial activity against *E*. *coli*. The antibacterial activity of LYZL proteins was demonstrated in other species. For example, human LYZL6 was found to be a potent antibacterial protein. LYZL3, 4, 5 and 7 did not display antibacterial activity. This could be due to the lack of essential amino acids in the active site. The human c-type lysozyme SLLP1, was non-bacteriolytic similar to rat LYZL3, 4, 5 and 7. Substrate binding assays also indicate that only LYZL1 and 6 exhibit higher affinity to bind peptidoglycan in comparison with the remaining proteins. Similar trend was observed in muramidase and isopeptidase assays. These properties can be attributed to the presence of active site residues in LYZL1 and 6. The antibacterial activity of LYZL1 and LYZL6 may play a significant role in male reproductive tract innate immunity. The lack of antibacterial activity of LYZL3, 4, 5 and 7 indicate that their role may be confined only to sperm function, whereas a broader role for LYZL1 and 6 may be expected. LYZL1 and 6 exhibited peptidoglycan binding ability, whereas the other LYZL proteins did not. The ability of LYZL1 and 6 to bind peptidoglycan could also contribute to their antibacterial activity. It is very interesting to speculate that LYZL proteins that do not conserve the active site amino acids and lack peptidoglycan binding ability may have evolved to perform another distinct set of functions that govern sperm function. On the other hand, LYZL3, 4, 5 and 7 proteins displayed potent hyaluronan binding ability than LYZL1 and LYZL6. Surprisingly, none of the LYZL proteins had hyaluronidase activity. Hyaluronan binding proteins play a crucial role in fertilization. For example, the hyaluronan-binding protein modulates sperm-egg interaction and undergoes extensive phosphorylation, but lacks hyaluronidase activity. The cellular events that occur during fertilization due to hyaluronin binding ability of LYZL proteins remains to be investigated. Oxidative stress is a common factor in testicular dysfunction. Antioxidant systems such as enzymes, metals and molecules and proteins exist in the testes to maintain the oxidant balance. The free radical scavenging ability of LYZL proteins indicate a possible function for them in maintaining oxidative stress in the male reproductive tract. The free radical scavenging activity of lysozymes is attributed to the disulfide bonds in these proteins. Though all LYZL proteins have four disulfide bonds, they vary in their ability of scavenge free radicals. The variation observed strengthens the possibility of diverse roles of these proteins though they belong to same family. In conclusion, we report that rat LYZL proteins are predominantly expressed in the male reproductive tract. Their antibacterial activity, hyaluronin binding ability and free radical scanvenging activity indicate a diverse role for these proteins in male reproduction and immunity. # Supporting Information [^1]: The authors have declared that no competing interests exist. [^2]: **Conceptualization:** SY. **Formal analysis:** SY. **Funding acquisition:** SY. **Investigation:** GN. **Methodology:** GN. **Project administration:** GN. **Resources:** SY. **Supervision:** SY. **Validation:** GN. **Visualization:** SY. **Writing – original draft:** SY. **Writing – review & editing:** SY.