Datasets:

suolyer

/

pile_pubmed-central

like 1

Introduction {#Sec1} ============ Parkinson Disease (PD) is the second most common neurodegenerative disorder, after Alzheimer's Disease^[@CR1]^. PD is clinically characterized by motor symptoms, including resting tremor, muscle rigidity, bradykinesia and postural instability. PD symptoms result mainly from a progressive loss of dopaminergic neurons in the *substantia nigra pars compacta*, associated with the accumulation of intracellular proteinaceous inclusions (Lewy bodies)^[@CR2]^. Most cases of PD are sporadic, with unknown aetiology, while 5--10% have a known underlying genetic basis. Moreover, several genes have been associated with inherited forms of Parkinsonism^[@CR3]--[@CR5]^. Variants in *PARK2* are the most frequent cause of autosomal recessive juvenile-onset parkinsonism (ARJP) worldwide. Exonic deletions in the *PARK2* gene, encoding parkin, were first reported in Japanese families with ARJP^[@CR6]^. Since then, numerous variants have been identified throughout the sequence of this particularly large gene (1.35 Mb), including large rearrangements, small deletions/insertions and missense/nonsense variants^[@CR3],[@CR4],[@CR7]^. Parkin protein is a well-established RBR (RING-between-RING) type of ubiquitin E3 ligase with multiple domains: an N-terminal ubiquitin like domain (UBL), followed by two RING finger domains (RING0 and RING1), an in-between-RING finger domain (IBR), a linker domain termed repressor element of parkin (REP) and a C-terminal RING finger domain (RING2)^[@CR8]--[@CR10]^. Resolution of the crystal structure showed that parkin exists in an autoinhibitory state, thereby needing additional conformational changes, mediated by PINK1 phosphorylation, to be active^[@CR8]--[@CR12]^. Parkin ubiquitinates a wide variety of cytosolic and mitochondrial proteins, being capable of catalyzing different types of ubiquitination (K63, K48, K11 and K6 linkages). Furthermore, parkin also ubiquitinates itself, promoting its own degradation^[@CR13]--[@CR17]^. Thereby, parkin has been characterized as a multifunctional protein involved in many cellular processes, including control of mitochondrial integrity and mitophagy, regulation of apoptosis, transcription and synaptic function^[@CR18]^. Given the complex activation process of parkin and depending on the specific residue affected, disease-associated variants can affect parkin E3 ligase activity through different mechanisms. These variants can directly impair parkin activity or abolish translation of a functional protein, cause reduced solubility and enhanced aggregation, disturb protein folding and stability, and/or affect parkin ability to bind to cofactors and substrates^[@CR19]--[@CR21]^. Here we report the functional characterization of two parkin truncating variants: N52Mfs\*29 that has a high prevalence in the Portuguese and Spanish populations^[@CR7],[@CR22]^, although without a clear biochemical characterization; and L358Rfs\*77 that was recently identified in the Portuguese population but has not been functionally characterized yet. Our study showed that each variant lead to misfolding and mislocalized parkin, being defective in the ubiquitination process and resulting in an apparent loss of parkin function. Results {#Sec2} ======= The N52Mfs\*29 and L358Rfs\*77 variants are degraded by the proteasome {#Sec3} ---------------------------------------------------------------------- The p.N52Mfs\*29 (c.155delA) parkin variant causes alteration of the open reading frame, which results in a premature stop codon, leading to the loss of most of the protein. This variant contains mostly the UBL domain of parkin^[@CR7],[@CR22]^. The p.L358RfsX77 parkin variant (c.1072--1073delCTinsA), located in the IBR domain (Fig. [1A](#Fig1){ref-type="fig"}), is predicted to alter the open reading frame and consequently introduce a premature stop codon, leading to a truncated protein lacking the REP linker and the RING2 domain and part of the IBR domain^[@CR7]^.Figure 1Parkin truncating variants have reduced protein expression. (**A**) Schematic representation of parkin (NP_004553) functional domains and location of the frameshift variants identified in Portuguese patients. (**B**) Analysis of protein expression of EGFP-tagged parkin clones in HEK293T cells by immunoblotting with anti-EGFP antibody. Beta-actin was used as loading control. Original blots are presented in Supplementary Fig. [S2](#MOESM1){ref-type="media"}. Quantification data (graph) are presented as the mean ± SD of three independent experiments; \*\*p \< 0.01, compared with WT parkin (one-way ANOVA/Tukey); variance homogeneity assumption not fulfilled. (**C**) Identification of parkin proteolytic pathways. Cells were incubated with different drugs for 18 hours and cells lysates subjected to immunoblotting with anti-EGFP antibody. Beta-actin was used as loading control. Membranes were cut and incubated separately with either anti-EGFP or anti-Beta-actin antibodies. Quantification data (graph) of significant conditions (MG132 treatment) are presented as the mean ± SD of three independent experiments; \*\*p \< 0.01, compared with DMSO condition (one-way ANOVA/Tukey). In order to understand the impact of these variants on the expression of parkin, we transfected HEK293T cells with plasmids containing N-terminal EGFP-tagged WT, N52Mfs\*29 or L358Rfs\*77 parkin variants, or control empty plasmid, followed by immunoblotting with anti-EGFP antibody or anti-beta-actin antibody (used as loading control). At 24 hours post-transfection, N52Mfs\*29 and L358Rfs\*77 showed significantly reduced protein levels when compared with WT parkin (Fig. [1B](#Fig1){ref-type="fig"}, about 80% and 92% decrease; p = 0.009 and 0.004, respectively). To note that, both N52Mfs\*29 and L358Rfs\*77 proteins presented the expected molecular weight of about 36 kDa (9 kDa from parkin plus 27 kDa from EGFP) and 77 kDa (50 kDa from parkin plus 27 kDa from EGFP). Quantitative real-time PCR showed no significant reduction on either EGFP-tagged N52Mfs\*29 or L358Rfs\*77 mRNA (Supplementary Fig. [1](#MOESM1){ref-type="media"}). Thereby, decreased levels of parkin variants are not the result of reduction or instability of the mRNA, but may result instead from decreased protein stability. Therefore, we treated cells with different drugs to identify the proteolytic pathway responsible for the diminished levels of parkin protein variants. HEK293T cells were transfected with EGFP-tagged parkin (WT, N52Mfs\*29 or L358Rfs\*77), or control plasmid. 24 hours post-transfection, cells were treated for the following 18 hours with either MG132 (5 µM), rapamycin (200 nM), EerI (10 µM) or control vehicle (DMSO), or NH~4~Cl (15 mM) or control vehicle (DMEM). Treatment of cells with the proteasome inhibitor MG132 resulted in an increase in protein levels of WT parkin (Fig. [1C](#Fig1){ref-type="fig"}), but this was not significant (p = 0.117). Nonetheless, we observed a significant increase in protein levels of both N52Mfs\*29 and L358Rfs\*77, promoted by MG132 (p = 0.003 and p = 0.006 respectively). Inhibition of the proteasome with MG132 restored N52Mfs\*29 levels, while L358Rfs\*77 levels did not completely returned to WT parkin levels, possibly indicating that other proteolytic pathway may also be involved in its degradation. Nevertheless, inhibition of autophagy or ERAD, promoted by NH~4~Cl or EerI, respectively, and induction of autophagy by rapamycin did not significantly altered the levels of WT or parkin variants. These results show that N52Mfs\*29 and L358Rfs\*77 are preferentially targeted to the proteasome for degradation. The N52Mfs\*29 and L358Rfs\*77 variants alter the solubility and localization of parkin {#Sec4} --------------------------------------------------------------------------------------- Previous studies reported that different missense/nonsense parkin variants show differential extractability by detergents due to altered localization and/or aggregate formation^[@CR14],[@CR21],[@CR23]^. To address this issue, N52Mfs\*29 or L358Rfs\*77 variants expressed in HEK293T cells, as previously described, were compared with WT parkin for their ability to be extracted by 0.1% Triton X-100 detergent. Equal amounts of soluble and insoluble fractions were loaded on gel followed by immunoblotting with anti-EGFP antibody. The detergent solubility assay showed that, in comparison with WT parkin, there was a significant increase in the detergent-insoluble fraction when N52Mfs\*29 or L358Rfs\*77 variants were analyzed (Fig. [2](#Fig2){ref-type="fig"}; p = 0.012 and 0.008). Moreover, the L358Rfs\*77 variant mostly adopted an insoluble conformation.Figure 2Parkin truncating variants have altered solubility properties. Equal amounts of soluble (S) and insoluble (I) fractions were loaded on gel followed by immunoblotting with anti-EGFP antibody. EGFP immunoblotting figures have different exposure times (original blots are presented in Supplementary Fig. [S3](#MOESM1){ref-type="media"}). Quantification data (graph) are presented as the mean ± SD of three independent experiments; \*p \< 0.05, \*\*p \< 0.01, compared with WT parkin (one-way ANOVA/Tukey). The differential solubility of parkin variants lead us to study the subcellular localization of these proteins by fluorescence microscopy and subcellular fractionation. EGFP-tagged parkin clones were transfected in HEK293T cells for 24 hours and further prepared for fluorescence microscopy. EGFP-tagged proteins were directly visualized and DAPI was used to label the nucleus. WT parkin was homogenously located in the cytoplasm (Fig. [3A](#Fig3){ref-type="fig"}) with occasional parkin-positive aggregates, which were observed in about 10% of the total population of transfected cells. WT parkin seemed to be present also within the nucleus (Fig. [3A](#Fig3){ref-type="fig"}). In contrast, about 93% of cells transfected with N52Mfs\*29 had parkin-positive aggregates in the cytoplasm (Fig. [3A](#Fig3){ref-type="fig"}). Moreover, N52Mfs\*29 also appears to locate within the nucleus (Fig. [3A](#Fig3){ref-type="fig"}). Surprisingly, L358Rfs\*77 seemed to be mainly located in the nucleus (Fig. [3A](#Fig3){ref-type="fig"}) and, only in a minor extent, in the cytoplasm. About 50% of cells transfected with L358Rfs\*77 had parkin-positive aggregates in both the nucleus and cytoplasm.Figure 3Truncating variants change the subcellular localization of parkin. (**A**) Localization of EGFP-tagged parkin clones in HEK293T cells. EGFP was directly visualized. Photographs were acquired using a Zeiss Axio Imager Z1 microscope. Bars, 10 μm. The percentage of cells with aggregates is also shown (graph). Data are presented as the mean ± SD of three independent experiments; \*\*\*p \< 0.001 (ANOVA/Tukey). (**B**) Subcellular fractionation of HEK293T cells expressing parkin clones. Presence of parkin in each fraction was analyzed by immunoblotting with anti-EGFP antibody and normalized to the specific marker of each fraction. T, total fraction; Cp, cytoplasm fraction; Mb, membrane fraction; SN, soluble nuclear fraction; Ch, chromatin-bound nuclear fraction. Original blots are presented in Supplementary Fig. [S4](#MOESM1){ref-type="media"}. Quantification data (graph) are presented as the mean ± SD of three independent experiments; \*p \< 0.05, \*\*p \< 0.01 (one-way ANOVA/Tukey). The subcellular localization of WT parkin and variants was confirmed in HEK293T cells by subcellular protein fractionation. The fractionation procedure efficiency was controlled by immunoblotting using specific antibodies for each fraction isolated: beta-actin (total fraction), caspase 3 (cytoplasm fraction), GM130 (membrane fraction), HDCA2 (soluble nuclear fraction) and histone H3 (chromatin-bound nuclear fraction). Presence of parkin in each fraction was analyzed by immunoblotting with anti-EGFP antibody and normalized to the specific marker of the respective fraction. The enrichment of parkin in each fraction was determined by calculating the ratio between the amount of recovered parkin in each fraction and the amount of recovered parkin in whole cell lysates. WT parkin was mainly collected in the cytoplasm fraction (90.5%), while a smaller amount was detected in the membrane fraction (8.7%) (Fig. [3B](#Fig3){ref-type="fig"}). WT parkin was also present in the soluble nuclear fraction in a reduced amount (0.7%), but was barely detected in the chromatin-bound nuclear fraction (Fig. [3B](#Fig3){ref-type="fig"}). The N52Mfs\*29 variant was collected in similar amount in the cytoplasm, membrane and nuclear fractions (36.2, 33.4, and 29.4%, respectively), while a smaller amount was detected in the chromatin-bound nuclear fraction (0.45%). When compared with WT parkin, the N52Mfs\*29 variant caused a significant increase on the membrane and soluble nuclear fractions (Fig. [3B](#Fig3){ref-type="fig"}, graph; p = 0.049 and 0.032). Surprisingly, the L358Rfs\*77 variant was predominantly collected in the soluble and chromatin-bound nuclear fractions (50.1 and 31.8%, respectively). This variant was also detected in the cytoplasm and membrane fractions in a reduced amount (8.4 and 9.5%, respectively) (Fig. [3B](#Fig3){ref-type="fig"}). When compared with WT parkin, the L358Rfs\*77 variant caused a significant reduction in parkin levels in the cytoplasm and a significant increase on the soluble and chromatin-bound nuclear fractions (Fig. [3B](#Fig3){ref-type="fig"}, graph; p = 0.012, 0.004 and 0.041). The N52Mfs\*29 and L358Rfs\*77 variants impair parkin activity {#Sec5} -------------------------------------------------------------- Parkin can promote its self-ubiquitination and this activity has been used to measure the enzymatic function of parkin^[@CR13],[@CR14],[@CR21],[@CR24]^. Therefore, we have investigated whether the N52Mfs\*29 and L358Rfs\*77 variants impair parkin self-ubiquitination activity in cells. For that, we co-transfected EGFP-tagged parkin (WT, N52Mfs\*29 or L358Rfs\*77), or control plasmid, with HA-tagged ubiquitin in HEK293T cells. Subsequently, cell lysates were co-immunoprecipitated using a GFP-tag antibody and the immunoprecipitates analyzed for ubiquitin-positive smear by immunoblotting with an HA-tag antibody. Given the low expression of parkin variants (Fig. [1](#Fig1){ref-type="fig"}), compared with parkin WT and control plasmid, the ubiquitin (HA) immunoreactivity was quantified through the HA/EGFP immunoreactivity ratio. The immunoprecipitated WT parkin and the L358Rfs\*77 variant showed a significant increase in HA immunoreactivity, compared with control plasmid, and a characteristic ubiquitin smear (Fig. [4A](#Fig4){ref-type="fig"}; p = 0.007 and 0.004, respectively), consistent with parkin self-ubiquitination activity. On the other hand, the immunoprecipitated N52Mfs\*29 variant did not present a significant increase in HA immunoreactivity compared with control plasmid (Fig. [4A](#Fig4){ref-type="fig"}; p = 0.32), suggesting that this variant lacks self-ubiquitination activity.Figure 4Parkin variants have different ubiquitination activity. (**A**) Co-immunoprecipitation of EGFP-tagged parkin clones with HA-tagged ubiquitin in HEK293T cells. Cell lysates were immunoprecipitated with anti-GFP antibody. Immunoprecipitates were subjected to immunoblotting with anti-HA antibody that revealed a characteristic ubiquitin smear corresponding to the amount of ubiquitination of each parkin clone. The blot was stripped and reprobed with anti-EGFP antibody. The same methodology was applied to whole cell lysates. (**B**) Co-immunoprecipitation of EGFP-tagged parkin clones with HA-tagged ubiquitin and endogenous p62 in HEK293T cells. Cell lysates were immunoprecipitated with anti-p62 antibody. Immunoprecipitates were immunoblotted with anti-HA antibody. Blot was stripped and reprobed with anti-EGFP and p62 antibodies. The same methodology was applied to whole cell lysates. p62 immunoblotting figures have different exposure times (original blots are presented in Supplementary Fig. [S5](#MOESM1){ref-type="media"}). Quantification data (graphs, **A** and **B**) are presented as the mean ± SD of three independent experiments; \*\*p \< 0.01, compared with EGFP control plasmid (one-way ANOVA/Tukey); variance homogeneity assumption not fulfilled (graph A). Next, we examined whether the N52Mfs\*29 and L358Rfs\*77 variants affected the ability of parkin to ubiquitinate substrates in cells in the presence of the proteasome inhibitor MG132. We performed co-immunoprecipitations using p62 antibody followed by immunoblotting with HA-tag and EGFP-tag antibodies. As expected, HA immunoreactivity of immunoprecipitated p62 significantly increased in the presence of WT parkin, confirming that p62 is ubiquitinated by parkin (Fig. [4B](#Fig4){ref-type="fig"}; p = 0.003). On the other hand, HA immunoreactivity of immunoprecipitated p62 did not show a significant increase in the presence of neither N52Mfs\*29 or L358Rfs\*77 variants compared with control plasmid (Fig. [4B](#Fig4){ref-type="fig"}; p = 0.535 and 0.962, respectively), thus showing that these variants impair parkin enzymatic activity. Whole cell lysates were also loaded and membranes further immunoblotted (Fig. [4A,B](#Fig4){ref-type="fig"}) to show the levels of transfected EGFP-clones in each sample. Parkin redistribution to depolarized mitochondria {#Sec6} ------------------------------------------------- Mitochondria dysfunction is an important contributor to PD pathogenesis. Particularly, parkin and PINK1 mediate autophagy of damaged mitochondria (mitophagy), a process that is impaired by PD-associated variants^[@CR25],[@CR26]^. Translocation of parkin to mitochondria is essential in this process, thus we went on to test if parkin variants impair mitochondrial localization after mitochondria membrane depolarization using FCCP. HEK293T cells transfected with EGFP-tagged parkin clones were treated for 1 hour with either FCCP (30 µM) or control vehicle DMSO, and further prepared for fluorescence microscopy or mitochondria isolation experiments. WT parkin was diffusely distributed throughout the cytosol and occasionally seemed to overlap with mitochondria (labelled with TOM20) (Fig. [5A](#Fig5){ref-type="fig"}). However, after FCCP treatment, WT parkin appeared to be specifically recruited to mitochondria (Fig. [5A](#Fig5){ref-type="fig"}), as previously reported^[@CR26],[@CR27]^. Redistribution of WT parkin from the cytosol to the mitochondria after FCCP treatment was confirmed by mitochondria isolation experiments (Fig. [5B](#Fig5){ref-type="fig"}). Presence of parkin in cytosol or mitochondria fractions was analyzed by immunoblotting with anti-EGFP antibody and normalized to caspase 3 or TOM20, respectively. The enrichment of parkin was determined by calculating the ratio between the amount of recovered parkin in mitochondria fraction and the amount of recovered parkin in the cytosol fraction. The amount of WT parkin in the mitochondria fraction significantly increased when the mitochondria membrane was depolarized (Fig. [5B](#Fig5){ref-type="fig"}; p = 0.037). Results concerning N52Mfs\*29 indicate that this variant does not completely abolish the redistribution of parkin to depolarized mitochondria. It was possible to observe some overlap of parkin with mitochondria with and without FCCP treatment (Fig. [5A](#Fig5){ref-type="fig"}), and there was an increase of parkin in the mitochondria fraction after FCCP treatment (Fig. [5B](#Fig5){ref-type="fig"}; however, this increase was not significant (p = 0.056). In contrast, the L358Rfs\*77 variant impaired mitochondria redistribution of parkin (Fig. [5A,B](#Fig5){ref-type="fig"}). Distribution of parkin did not change after FCCP treatment (Fig. [5A](#Fig5){ref-type="fig"}), neither did the amount of parkin in the mitochondria fraction (Fig. [5B](#Fig5){ref-type="fig"}).Figure 5Only WT parkin is able to redistribute to depolarized mitochondria. (**A**) Immunofluorescence localization of EGFP-tagged parkin clones in HEK293T cells treated with FCCP or control vehicle DMSO. EGFP was directly visualized. TOM20 labelled mitochondria. Photographs were acquired using a Zeiss Axio Imager Z1 microscope. Bars, 10 μm. (**B**) Mitochondria isolation from HEK293T cells expressing parkin clones. Presence of parkin in cytosol (**C**) or mitochondria (Mt) fractions was analyzed by immunoblotting with anti-EGFP antibody and normalized to caspase 3 or TOM20, respectively. Original blots are presented in Supplementary Fig. [S6](#MOESM1){ref-type="media"}. Quantification data (graph) are presented as the mean ± SD of three independent experiments; \*p \< 0.05, compared with control DMSO (two-tailed Student's t-test). Discussion {#Sec7} ========== The autosomal recessive inheritance of parkin-related PD suggests that protein loss-of-function is at the basis of ARJP pathogenesis. Functional studies reported that even variants that do not affect parkin E3 ligase activity, manifest loss-of-function by causing defects in protein folding, solubility, aggregation, localization or targeting of proteins for proteasome degradation^[@CR19]--[@CR21]^. Furthermore, despite the high number of parkin variants identified to date (about 170), the pathogenic relevance remains unclear for several of the variants^[@CR3],[@CR20]^. In this study, we determined the pathogenic mechanisms underlying two frameshift parkin variants. The N52Mfs\*29 variant was already reported and is highly prevalent in the Portuguese and Spanish population^[@CR7],[@CR28]^, while the L358Rfs\*77 variant was recently identified in the Portuguese population^[@CR7]^. Both variants seemed to reduce protein stability, since both were prematurely degraded by the proteasome (Fig. [1C](#Fig1){ref-type="fig"}). The low expression of these variants supports the theory of a loss-of-function phenotype^[@CR19],[@CR29]^. However, both protein variants were still expressed at moderate levels, detected by immunoblotting and immunofluorescence, when overexpressed in cultured cells. This raised the question whether N52Mfs\*29 and L358Rfs\*77 endogenous proteins could be fully degraded by the proteasome, what would be interesting to study in patients' cells. Nevertheless, since both overexpressed variants were present in protein aggregates (Fig. [3A](#Fig3){ref-type="fig"}), the ability of the proteasome system to fully degrade them could be compromised. Additionally, it is also important to understand if N52Mfs\*29 and L358Rfs\*77 variants impair the molecular properties and enzymatic activity of parkin. The domain structure of parkin regulates its enzymatic activity. Resolution of the crystal structure showed that parkin exists in an autoinhibitory state. The E2 binding site on the RING1 domain is occluded by the UBL domain and the REP linker, while the catalytic site in the RING2 domain is blocked by the RING0 domain^[@CR8]--[@CR10]^. Therefore, activation of parkin requires additional conformational changes to bring the RING2 and E2 active sites closer for ubiquitin transfer. The activation process requires binding of phosphorylated ubiquitin generated by PINK1, which decreases the association of the UBL domain with the RING1 domain. Thus, PINK1 is able to efficiently phosphorylate the UBL domain of parkin, which then binds to the RING0 domain and releases the catalytic RING2 domain^[@CR30]^. The N52Mfs\*29 and L358Rfs\*77 variants lack important parkin domains, which can cause not only disruption of parkin folding, but also misregulation of parkin autoinhibition and enzymatic functions^[@CR20]^. Moreover, the sequence of N52Mfs\*29 lacking the Ser65 residue that is phosphorylated by PINK1^[@CR12]^, implies that this variant confers distinct molecular properties to parkin. Indeed, we found that N52Mfs\*29 caused protein misfolding leading to the formation of protein aggregates and altered subcellular distribution (Fig. [3](#Fig3){ref-type="fig"}). Thereby, decreased solubility of this variant (Fig. [2](#Fig2){ref-type="fig"}) may be explained not only by the presence of parkin in detergent-insoluble aggregates, but also by its higher degree of association with membranes (Fig. [3](#Fig3){ref-type="fig"}). Given the protein sequence of this variant expressing only the UBL domain and resembling ubiquitin, this phenotype is not surprising. The wide distribution throughout the cell and accumulation on aggregates is typical of ubiquitin. Furthermore, N52Mfs\*29 showed reduced self-ubiquitination activity (Fig. [4A](#Fig4){ref-type="fig"}), which could result from either high turnover of the protein or impaired enzymatic activity^[@CR21]^. Analysis of parkin ability to ubiquitinate a substrate in the presence of a proteasome inhibitor showed that the N52Mfs\*29 variant impaired parkin enzymatic activity, as it was not able to significantly ubiquitinate p62 substrate (Fig. [4B](#Fig4){ref-type="fig"}). Nonetheless, this variant was still able to relocalize to depolarized mitochondria (Fig. [5A](#Fig5){ref-type="fig"}), although this was not significant (Fig. [5B](#Fig5){ref-type="fig"}). Once again, it seems that N52Mfs\*29 mimics ubiquitin properties, which also concentrates on depolarized mitochondria^[@CR15]^. Therefore, we believe that N52Mfs\*29 variant also impairs parkin/PINK1-mediated mitophagy, as described for other parkin variants^[@CR25],[@CR26]^. Interestingly, the L358Rfs\*77 variant also affected protein folding since it caused protein aggregation and, thus, decreased solubility (Fig. [2](#Fig2){ref-type="fig"}), and redistribution of parkin mainly to the nucleus (Fig. [3](#Fig3){ref-type="fig"}). Other variants that lack the C-terminal domain of parkin have also been reported to cause misfolding and aggregation, resulting in altered subcellular distribution^[@CR19],[@CR21],[@CR31]^. Indeed the C-terminal domain of parkin is indispensable for proper folding as shown on crystal structure of parkin. Moreover, this variant lacks the catalytic RING2 domain located in the C-terminal^[@CR8]--[@CR10]^. Nonetheless, redistribution of parkin to the nucleus has not been reported for other variants. We checked the additional 77 amino acids of L358Rfs\*77 for the presence of a nuclear localization signal, but we could not find any using bioinformatics analysis. Thereby, it would be interesting to study if any particular loss/gain of protein interactions and/or post-translational modifications are responsible for targeting this protein to the nucleus. Nonetheless, L358Rfs\*77 retained the capacity of self-ubiquitination, as shown by the presence of high molecular weight species of ubiquitinated parkin (Fig. [4A](#Fig4){ref-type="fig"}). This could indicate either maintenance of enzymatic activity or accumulation of the ubiquitinated protein that was not completely degraded by the proteasome. To account for the potential mechanisms by which L358Rfs\*77, with apparent increased ubiquitination activity, showed a loss-of-function phenotype (Figs [2](#Fig2){ref-type="fig"}, [3](#Fig3){ref-type="fig"}), we studied the capacity of this protein to ubiquitinate a substrate and redistribute to depolarized mitochondria. Our results, confirmed that L358Rfs\*77 was not able to significantly ubiquitinate p62, neither redistribute to depolarized mitochondria. Reduced ability to ubiquitinate substrates could also be the result of loss of binding to substrates, coenzymes or adaptor proteins. This does not seem to be the case for L358Rfs\*77, since it was able to bind p62 substrate (Fig. [4B](#Fig4){ref-type="fig"}), thereby showing that this variant impaired parkin enzymatic activity and mediated-mitophagy. Other truncating variant (W453\*) has also been reported to retain self-ubiquitination activity, but not the ability to ubiquitinate substrates^[@CR13],[@CR21],[@CR26]^. Nevertheless, further studies are needed to fully address potential mitophagy defects caused by N52Mfs\*29 and L358Rfs\*77 variants. Introducing parkin variants in cellular models by gene-editing technologies would be of great potential to study mitophagy impairment. Ultimately, despite variable molecular, binding and ubiquitination properties, all parkin variants eventually lead to impaired degradation of substrates, leading to an abnormal accumulation that may be toxic to cells. In addition, intriguing data suggests that some parkin variants may be toxic to cells (Wang, Lu *et al*. 2007). Despite being highly prevalent in the Portuguese and Spanish populations (Munoz, Tolosa *et al*. 2002, Morais, Bastos-Ferreira *et al*. 2016), the biochemical characteristics of the N52Mfs\*29 variant in PD pathology had not been clarified so far. The same happens for the L358Rfs\*77 variant that was recently identified in the Portuguese population (Morais, Bastos-Ferreira *et al*. 2016). The results of our study imply a causative role for the N52Mfs\*29 and L358Rfs\*77 variants in PD pathogenesis. It was shown that these parkin truncating variants have different biochemical properties, are abnormally localized and lack enzymatic activity. Therefore, loss-of-function activity may be explained not only by loss of parkin domains and disruption of protein folding, but also by mislocalization of parkin from its regular cytoplasmic distribution and site of enzymatic activity. Accumulation of parkin substrates and damaged mitochondria have been observed in other models of parkin loss-of-function, which may ultimately lead to neurodegeneration. Methods {#Sec8} ======= Antibodies {#Sec9} ---------- Primary antibodies: mouse monoclonal anti-EGFP antibody (Abnova, MAB1765), mouse monoclonal anti-GFP antibody (Rockland, 600-301-215), mouse monoclonal anti-beta-actin (Sigma-Aldrich, A5441), mouse monoclonal anti-GM130 (BD Biosciences, 610822), mouse monoclonal anti-caspase 3 (Cell Signaling, 9668), mouse monoclonal anti-HDAC2 (Santa Cruz Biotechnology, sc-9959), rabbit polyclonal anti-histone H3 (Abcam, ab1791), mouse monoclonal anti-p62 (Proteintech, 66184-1-Ig), mouse monoclonal anti-TOM20 (BD Biosciences, 612278). Secondary antibodies: horseradish peroxidase (HRP)-conjugated goat anti-mouse IgG (Santa Cruz Biotechnology, sc-2005), HRP-conjugated goat anti-rabbit IgG (Calbiochem, Merck-Millipore, 401393), mouse TrueBlot^TM^ ULTRA (Rockland, 18-8817-33). Expression vectors {#Sec10} ------------------ The pEGFP-C1-Parkin-WT plasmid was kindly provided by Dr. Kawajiri, S. (Juntendo University School of Medicine, Tokyo, Japan). This plasmid was modified by site-directed mutagenesis using the QuikChange II Kit (Agilent) to produce disease-associated parkin plasmids. The following primers pairs were used to introduce N52Mfs\*29 and L358Rfs\*77 variants: forward primer 5′-GGGAAGGAGCTGAGGATGACTGGACTGTGC-3′ and reverse primer 5′-GCACAGTCCAGTCATCCTCAGCTCCTTCCC; and forward primer 5′- CGAAGGGGGCAATGGCAGGGCTGTGG and reverse primer 5′-CCACAGCCCTGCCATTGCCCCCTTCG, respectively. pRK5-HA-ubiquitin was a gift from Ted Dawson (Addgene plasmid 17608)^[@CR17]^. Cell culture and transfection {#Sec11} ----------------------------- HEK293T cells (kindly provided by Dr. Elsa Logarinho, IBMC/i3S, Porto) were grown in DMEM high glucose GlutaMAX™ supplemented with 10% fetal bovine serum (FBS) and 1% antibiotic-antimycotic (Gibco, Life technologies) at 37 °C in a humidified 5% CO~2~ atmosphere. Cells were transiently transfected for 24 h with each plasmid using jetPRIME (Polyplus-transfection) according to the manufacturer's protocol. In order to inhibit/enhance different proteolytic pathways, cells were treated at 24 hours post transfection with the following drugs for 18 hours: the proteasome inhibitor MG132 (5 μM, Calbiochem, EMD Millipore), autophagy inducer rapamycin (200 mM, Calbiochem, EMD Millipore), ERAD inhibitor EerI (10 µM, Calbiochem, EDM Millipore) or control vehicle DMSO (Sigma-Aldrich), and the lysosomotropic agent NH4Cl (15 mM, Sigma-Aldrich) or control vehicle DMEM (Gibco, Life technologies). To induce mitochondria membrane depolarization, cells were treated at 24 hours post transfection with the mitochondria uncoupler carbonyl cyanide p-(tri-fluromethoxy)phenyl-hydrazone (FCCP, 30 µM) or with the control vehicle DMSO for 1 hour. Western blot analysis {#Sec12} --------------------- Cells expressing target proteins were collected in RIPA buffer (Sigma; 150 mM NaCl, 1.0% IGEPAL® CA-630, 0.5% sodium deoxycholate, 0.1% SDS, 50 mM Tris, pH 8.0) supplemented with cOmplete Protease Inhibitor Cocktail (Roche) and then sonicated. Total protein concentration was measured with the Pierce BCA protein assay kit (Thermo Scientific) according to manufacturer's instructions. Samples (30--50 µg of total protein) were separated on SDS-PAGE and electrophoretically transferred onto PVDF membranes (Merck-Milipore) using a wet electroblotting system (Bio-Rad). Membranes were blocked in 5% non-fat dry milk in PBS-T for 1 hour, at room temperature (RT), and subsequently incubated with specific primary antibodies, as indicated (diluted in 3% non-fat dry milk in PBS-T), overnight at 4 °C, except for beta-actin antibody that was incubated for 1 hour at RT. The membranes were washed with PBS-T and then incubated with HRP-conjugated secondary antibody for 1 hour at RT. Following three washes with PBS-T, detection was achieved using WesternBright^TM^ Sirius or WesternBright^TM^ ECL- HRP substrates (Advansta) and chemiluminescence detected with a ChemiDoc^TM^ XRS+ Imaging System (Bio-Rad). Protein bands were quantified using the Image Lab^TM^ 5.2.1 Software (Bio-Rad). Quantitative comparisons between samples of each experiment were always performed on the same blot. When necessary, membranes were stripped by incubation in stripping solution (62.5 mM Tris-HCl pH 6.7, 2% SDS, 100 mM beta-mercaptoethanol) at 50 °C for 30 min, with gentle agitation. Detergent solubility assay {#Sec13} -------------------------- Cells expressing target proteins were washed with PBS and then collected in 0.1% Triton X-100 (Merck-Millipore) in 1x PBS supplemented with cOmplete Protease Inhibitor Cocktail (Roche). Homogenates were centrifuged at 15000 g for 20 min at 4 °C to separate the supernatant (soluble) and pellet (insoluble) fractions. Pellet fractions were solubilized in 1x laemmli buffer and supernatants were brought to the same volume and 1x concentration by adding 6x laemmli buffer, and both fractions were then boiled at 95 °C for 10 min. To compare the relative distribution of parkin, equal volumes pellet and supernatant were loaded on SDS-PAGE \[adapted from^[@CR19]^\]. Subcellular protein fractionation and mitochondria isolation {#Sec14} ------------------------------------------------------------ For subcellular protein fractionation, cells expressing target proteins were washed and scrapped from the plate in ice cold 1x PBS. The whole cell extract (corresponding to total fraction) was centrifuged at 500 g for 3 min and the supernatant discarded. Subcellular protein extraction was then performed, using the Subcellular Protein Fractionation Kit for cultured cells (Thermo Scientific), according to the manufacturer's instructions. Subcellular fractions corresponding to cytoplasm, membranes, soluble nuclear and chromatin-bound nuclear extracts were isolated. For mitochondria isolation, cells expressing target proteins were washed and scrapped from the plate in ice cold 1x PBS. The whole cell extract (corresponding to total fraction) was centrifuged at 850 g for 2 min and the supernatant discarded. Mitochondria isolation was then performed using the Mitochondria Isolation Kit for cultured cells (Thermo Scientific), according to the manufacturer's instructions. For obtaining the mitochondria pellet, centrifugation was performed at 12000 g for 15 min at 4 °C. The mitochondria pellet was lysed in 1% triton X-100 in TBS (25 mM Tris, 0.15 M NaCl, pH 7.2), supplemented with cOmplete Protease Inhibitor Cocktail (Roche). Subcellular fractions corresponding to cytosol and mitochondria-enriched fractions were isolated. Co-immunoprecipitation and ubiquitination assays {#Sec15} ------------------------------------------------ Cells were collected in lysis buffer (50 mM Tris-HCl pH 8.0, 120 mM NaCl, 0.5% Triton X-100) supplemented with the Pierce Protease and Phosphatase Inhibitor tablets (Thermo Scientific) and sonicated. Dynabeads^TM^ M-280 Sheep Anti-Mouse IgG (Life Technologies) were washed in 3% BSA/PBS. GFP or p62 antibodies were coupled to Dynabeads (1 µg antibody/50 µl beads), by incubating with rotation overnight at 4 °C. Cell lysates were precleared with 5 μL Dynabeads for 1 hour at 4 °C and then incubated with antibody-Dynabeads with rotation overnight at 4 °C. The immunoprecipitates were washed in 3% BSA/PBS and then in PBS, and further transferred to a clean tube. The proteins were eluted by boiling in 1x laemmli buffer at 90 °C for 10 min. Immunofluorescence {#Sec16} ------------------ Cells expressing target proteins were fixed using 4% paraformaldehyde/4% sucrose for 20 min and permeabilized with 0.3% triton X-100 for 20 min. After washing with PBS, cells were blocked in 3% BSA/PBS for 45 min and further incubated with the primary antibodies overnight at 4 °C. Then, cells were incubated with the secondary antibodies for 1 hour at RT, washed with PBS and stained with Hoechst 33342 (Life Technologies) for 5 min to label the nucleus. Preparations were mounted with ProLong^TM^ Gold Antifade Mountant (Life Technologies) and visualized using an epifluorescence Zeiss Axio Imager Z1 microscope equipped with an Axiocam MR3.0 camera and Axiovision 4.7 software. Statistical analysis {#Sec17} -------------------- Analyses of data were performed using the IBM SPSS Statistics 25.0 software. All quantitative data are expressed as mean ± standard deviation (SD) of at least three independent experiments. Statistical significance analysis was conducted using one-way ANOVA with Tukey post-hoc test, when comparing means of more than two groups, or two-tailed Student's t-test to compare means between two groups, the level of statistical significance being set at p \< 0.05. Supplementary information ========================= {#Sec18} Additional information **Publisher's note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Supplementary information ========================= is available for this paper at 10.1038/s41598-019-52534-6. This work was funded by FEDER funds through the *Programa Operacional Factores de Competitividade* -- COMPETE 2020 and by Nacional funds through the *Fundação para a Ciência e Tecnologia* - FCT \[COMPETE: POCI-01-0145-FEDER-007440\]. This work was also funded in part by the FCT grant FCT-ANR/BEX-GMG/0008/2013 and the Porto Neurosciences and Neurologic Disease Research Initiative at the i3S (Norte-01-0145-FEDER-000008), supported by *Norte Portugal Regional* Operational Programme (NORTE 2020) under the PORTUGAL 2020 Partnership Agreement, also through FEDER. The authors also acknowledge the support of the i3S Scientific Platform Advanced Light Microscopy, member of the PPBI (PPBI-POCI-01-0145-FEDER-022122). MS, SM and CP were the recipients of fellowships (SFRH/BPD/116046/2016, SFRH/BD/87189/2012 and SFRH/BD/90048/2012) from the FCT supported by POPH/MCTES funding. M.S. and I.A. conceived and designed the experiments; M.S., S.M. and C.P. performed the experiments; all authors analyzed and interpreted the results; M.S. wrote the manuscript; all authors critically reviewed the manuscript and approved the final version. The authors declare no competing interests.

(J Am Heart Assoc. 2017;6:e005506 DOI: [10.1161/JAHA.117.005506](10.1161/JAHA.117.005506).)28778941 {#jah32416-sec-0004} Clinical PerspectiveWhat Is New?Pharmacological inhibition of the mechanistic target of rapamycin complex 1 (mTORC1) pathway in late gestation by rapamycin treatment of pregnant mice causes intrauterine growth restriction, thereby reducing body and organ size in the offspring at birth.Rapamycin‐sensitive mTORC1 functions are required for perinatal cardiac growth, primarily impacting cardiomyocyte size and survival but not proliferation in the neonatal heart.Prenatal mTORC1 inhibition reduces cardiac output at birth due to diminished left ventricular dimensions, but contractility is not affected.After prenatal mTORC1 inhibition body and heart weight partially normalize until early adulthood, and cardiac output fully recovers despite a reduced number of cardiomyocytes in the adult heart.What Are the Clinical Implications?Prenatal mTORC1 inhibition could play a role in intrauterine growth restriction caused by various maternal or environmental conditions, such as maternal mal‐ or undernutrition, placental insufficiency, or fetal hypoxia.mTORC1 function is required for fetal cardiac growth, and its inhibition might be involved in developmental programming of heart disease in adulthood. Introduction {#jah32416-sec-0008} ============ The intrauterine environment is a major determinant of embryonic and fetal development. Various maternal or environmental conditions can impair fetal growth, resulting in intrauterine growth restriction (IUGR), which often presents with low birth weight and reduced organ size.[1](#jah32416-bib-0001){ref-type="ref"} Maternal under‐ or malnutrition, placental insufficiency, fetal hypoxia, or drugs (eg, glucocorticoids) are among the factors causing IUGR in humans and animal models.[2](#jah32416-bib-0002){ref-type="ref"} IUGR is considered a major risk factor for various chronic diseases later in life and thereby contributes to developmental (or fetal) programming.[3](#jah32416-bib-0003){ref-type="ref"} Hence, it appears imperative to uncover cellular and molecular mechanisms that regulate intrauterine growth. Relatively little is known about the underlying mechanisms, however, but signaling pathways regulating cell growth and proliferation are likely to be involved. IUGR can be induced in animal models by caloric or protein restriction in the maternal diet during pregnancy,[2](#jah32416-bib-0002){ref-type="ref"} which suggests that a sufficient nutritional supply to the embryo and fetus is required for normal intrauterine growth. The mechanistic (or previously named "mammalian") target of rapamycin (mTOR) pathway is an important sensor of the metabolic and nutritional state of a cell and thereby integrates energy homeostasis and amino acid availability with cell size and proliferation.[4](#jah32416-bib-0004){ref-type="ref"} mTOR is a serine/threonine kinase present in 2 multiprotein complexes, mTORC1 and mTORC2, which differ in the composition of regulatory and adaptor protein binding partners. Whereas mTORC2 is best characterized for its involvement in cell survival, growth, cytoskeletal organization, and cell polarity, mTORC1 is a master regulator of cell growth by activating protein, lipid, and nucleic acid biosynthesis while inhibiting catabolic mechanisms such as autophagy.[4](#jah32416-bib-0004){ref-type="ref"} One of the best characterized mTORC1 functions is the activation of cap‐dependent mRNA translation by phosphorylating its downstream targets eukaryotic translation initiation factor 4E binding protein 1 (4E‐BP1) and ribosomal S6 kinase (S6K1). 4E‐BP1 phosphorylation results in its dissociation from eukaryotic translation initiation factor 4E, thereby releasing its inhibitory effect and allowing translation initiation. In addition, phosphorylation and activation of S6K1 and its downstream target S6 ribosomal protein favor mRNA biogenesis as well as translation initiation and elongation. Upstream inputs that regulate mTORC1 activity include cellular energy status (ie, glucose and ATP levels), oxygen, growth factors, and amino acid availability.[4](#jah32416-bib-0004){ref-type="ref"} Importantly, some (but not all) downstream effects of mTORC1 (in contrast to mTORC2) can be efficiently inhibited by the immunosuppressive drug rapamycin. In various model organisms the mTOR pathway has been shown to control organ size,[5](#jah32416-bib-0005){ref-type="ref"} which also includes the mammalian heart. Although its role during postnatal physiological as well as pathological cardiac hypertrophic growth has been studied extensively,[6](#jah32416-bib-0006){ref-type="ref"} much less is known about the involvement of mTOR signaling in prenatal cardiac development. The heart conditional knockout of the gene encoding mTOR in mice using Cre/loxP recombination mediated by different cardiac Cre drivers has been shown to be lethal during late gestation or at early postnatal stages.[7](#jah32416-bib-0007){ref-type="ref"}, [8](#jah32416-bib-0008){ref-type="ref"}, [9](#jah32416-bib-0009){ref-type="ref"} Efficient inhibition of mTORC1 signaling in the heart, however, appears to be either transiently restricted to a period shortly after midgestation[9](#jah32416-bib-0009){ref-type="ref"} or is only achieved after birth,[7](#jah32416-bib-0007){ref-type="ref"}, [8](#jah32416-bib-0008){ref-type="ref"} which hampers final conclusions about its role in the fetal heart. In addition, inactivating the mTOR gene affects both mTORC1 and mTORC2, such that precisely defining the role of one versus the other is not feasible. Similarly, the heart conditional knockout of the upstream mTORC1 activator Ras homolog enriched in brain (Rheb) causes mTORC1 inhibition only after birth, subsequently leading to postnatal lethality.[10](#jah32416-bib-0010){ref-type="ref"} Thus, the role of mTORC1 during embryonic and fetal cardiac growth is incompletely understood. Restricting amino acid availability in the maternal diet or lowering ambient oxygen concentration during pregnancy as well as reducing blood supply to the placenta cause IUGR in animal models.[1](#jah32416-bib-0001){ref-type="ref"}, [2](#jah32416-bib-0002){ref-type="ref"} Given that mTORC1 is inhibited by amino acid starvation, hypoxia, and low cellular energy levels, thereby impairing cell growth and proliferation,[4](#jah32416-bib-0004){ref-type="ref"} we hypothesized that it might be involved in the IUGR phenotype. In addition, most IUGR conditions reduce heart size at birth and cause neonatal cardiac hypoplasia, which is considered a major cardiovascular risk factor in adulthood.[3](#jah32416-bib-0003){ref-type="ref"} In this regard we have recently described a mouse model of embryonic heart regeneration based on inactivation of the X‐linked gene encoding holocytochrome c synthase (*Hccs*) specifically in the developing heart.[11](#jah32416-bib-0011){ref-type="ref"} The HCCS enzyme is required for electron transport along the mitochondrial respiratory chain. Heterozygous heart conditional *Hccs* knock‐out (KO) females (hereafter referred to as *cHccs* ^*+/−*^) develop a tissue mosaic in the ventricular myocardium composed of 50% cardiomyocytes harboring mitochondrial dysfunction and 50% healthy cells at midgestation. Compensatory proliferation of the healthy cardiomyocyte population allows embryonic heart regeneration, however, such that the neonatal *cHccs* ^*+/−*^ heart is composed of 90% healthy cells and contains only 10% diseased cells.[11](#jah32416-bib-0011){ref-type="ref"} Nevertheless, embryonic heart regeneration is not sufficient to completely build up the myocardium, resulting in *cHccs* ^*+/−*^ hearts being hypoplastic at birth due to a reduced number of cardiomyocytes.[12](#jah32416-bib-0012){ref-type="ref"} This reduction in cell number is postnatally compensated for by accelerated and augmented cardiomyocyte hypertrophic growth, such that heart size normalizes by early adulthood.[12](#jah32416-bib-0012){ref-type="ref"} Given the established role of mTORC1 in heart and organ size control,[5](#jah32416-bib-0005){ref-type="ref"}, [6](#jah32416-bib-0006){ref-type="ref"} we speculated that mTORC1 activity might be important for fetal cardiac growth and regulation of neonatal heart size in general as well as for compensatory growth of *cHccs* ^*+/−*^ hearts in particular. Here we show that inhibiting mTORC1 in the final quarter of gestation by rapamycin treatment of pregnant mice causes IUGR and reduces heart size and cardiac output at birth. Body and heart size partially normalize during postnatal life, and despite a reduction in cardiomyocyte number, cardiac function is not compromised in adult mice after prenatal mTORC1 inhibition. *cHccs* ^*+/−*^ mice after prenatal rapamycin treatment exhibit cellular differences in the myocardium at birth compared with controls but no major alterations of postnatal heart size or function. Methods {#jah32416-sec-0009} ======= Mice {#jah32416-sec-0010} ---- The generation and characterization of heart conditional *Hccs* KO mice have been described previously.[11](#jah32416-bib-0011){ref-type="ref"} Briefly, floxed (fl) *Hccs* mice were bred to mice expressing *Cre* recombinase under the control of the *Nkx2.5* promoter. All mice were maintained on a mixed 129Sv/C57Bl6 genetic background, and all experiments throughout the study were performed on heterozygous *Hccs* KO females (*Hccs* ^*fl/+*^ */Nkx2.5Cre*, referred to as *cHccs* ^*+/−*^) and their respective *Cre* positive female littermate controls (*Hccs* ^*+/+*^ */Nkx2.5Cre,* referred to as *Hccs* ^*+/+*^). The latter were furthermore used to compare the effects of rapamycin versus vehicle treatment in control animals, implying that all mice used in the study were carrying the *Nkx2.5Cre* knock‐in.[13](#jah32416-bib-0013){ref-type="ref"} Unless specifically annotated, all results refer to control mice, whereas data of *cHccs* ^*+/−*^ mice are explicitly labeled in figures or figure legends. The total number of mice included in the different experiments of this study is as follows: vehicle‐treated neonates, *Hccs* ^*+/+*^ n=31, *cHccs* ^*+/−*^ n=30; rapamycin‐treated neonates, *Hccs* ^*+/+*^ n=39, *cHccs* ^*+/−*^ n=41; vehicle‐treated adults, *Hccs* ^*+/+*^ n=9, *cHccs* ^*+/−*^ n=15; rapamycin‐treated adults, *Hccs* ^*+/+*^ n=9, *cHccs* ^*+/−*^ n=11. All animal procedures were performed following institutional guidelines and had previously been approved by the responsible authorities (Landesamt für Gesundheit und Soziales Berlin, approval number G 0027/10). Rapamycin Injection {#jah32416-sec-0011} ------------------- Female mice were mated to the respective males for 1 night and separated the next morning (ie, at 0.5 days postconception \[dpc\]). Dams with visible pregnancy at 14.5 dpc were randomly assigned to vehicle or rapamycin treatment starting in the morning at 15.5 dpc. Rapamycin (Cayman Chemical Co, Ann Arbor, MI) was dissolved as stock solution (20 mg/mL) in dimethylacetamide (DMA, Sigma‐Aldrich, St. Louis, MO). Pregnant dams were treated with rapamycin or vehicle by subcutaneous injections every 12 hours from 15.5 dpc until delivery. Each injection contained 5 mg rapamycin/kg body weight diluted in 200 μL vehicle (10% PEG 300 \[Sigma\], 17% Tween 80 \[Sigma\] in 0.9% NaCl solution) or DMA‐containing vehicle only. Injections were administered alternating into the loose skin of the interscapular or the right and left femoral area. Echocardiography {#jah32416-sec-0012} ---------------- Adult mice (11 weeks old) were anesthetized by inhalation of a 2.5% isoflurane/oxygen mixture using the Vevo compact dual anesthesia system (VisualSonics, Toronto, ON, Canada). Body temperature was kept constant at 37°C using a heat lamp and a rectal temperature probe. Measurements on neonates were performed on awake pups. Echocardiography was recorded using the Vevo 2100 high‐frequency ultrasound system (VisualSonics, Toronto, ON, Canada) with a MicroScanTM transducer MS400 set to 30 MHz for adult mice or MS700 set to 50 MHz for neonatal mice. Operators were blinded for mouse genotypes and treatment groups during echocardiographic recordings as well as data analyses. Organ Preparation and Histology {#jah32416-sec-0013} ------------------------------- Hearts, kidneys, and lungs from neonatal mice were prepared on postnatal day 1, and hearts, livers, kidneys, and spleens from adult mice were prepared at the age of 11 weeks. Adult mice were euthanized by cervical dislocation, and neonates were decapitated. Organs were excised, rinsed in cold PBS, weighed, snap‐frozen in liquid nitrogen, and stored at −80°C. If organs were used for histological analyses, they were fixed in 4% paraformaldehyde/PBS (Sigma) for 24 to 48 hours. The tissue was subsequently dehydrated through an increasing ethanol series, cleared in toluol, and embedded in paraffin. Five‐micrometer paraffin sections were stained with hematoxylin and eosin (Carl Roth, Karlsruhe, Germany) to assess overall cardiac morphology and tissue composition or with Sirius Red (Direct Red 80, Sigma) to visualize myocardial fibrosis. Quantification of Fibrosis in Adult Hearts {#jah32416-sec-0014} ------------------------------------------ Light microscopy images of Sirius Red--stained heart sections from 11‐week‐old adult mice were taken using the Biozero BZ‐8100 microscope (Keyence, Osaka, Japan). Under ×5 optical magnification, 6 random fields of the left ventricular (LV) myocardium, including free wall and interventricular septum, were imaged, thereby covering the entire LV tissue. Two nonadjacent cross sections per heart were used. The percentage of interstitial fibrotic tissue was quantified using the color‐threshold plugin of ImageJ software (<https://imagej.nih.gov/ij/>), which measures the red‐stained area in relation to the total LV myocardial area. Values of all 12 images were averaged to give the mean fibrotic tissue content for each heart. Perivascular fibrosis was excluded and deleted from the images before analysis. Evaluation of Cell Proliferation {#jah32416-sec-0015} -------------------------------- To assess proliferation rates in neonatal hearts, immunofluorescence staining for Ki67 was performed. Paraffin sections were deparaffinized and rehydrated, and heat‐mediated antigen retrieval was performed in sodium citrate buffer (10 mmol/L, pH 6.0) for 20 minutes. After blocking in antibody solution containing 5% normal goat serum (Jackson ImmunoResearch, West Grove, PA) for 1 hour, sections were incubated overnight with a rabbit monoclonal anti‐Ki67 antibody (RM‐9106, Thermo Scientific, Waltham, MA) at 4°C. Secondary antibody detection was performed at room temperature for 1 hour using a goat anti‐rabbit Alexa Fluor 555‐conjugated secondary antibody (Invitrogen, Carlsbad, CA). Nuclei were stained with 4′,6‐diamidino‐2‐phenylindole (DAPI) (Invitrogen). Longitudinal sections were imaged with ×40 optical magnification using the Biozero BZ‐8100 fluorescence microscope (Keyence, Osaka, Japan), and 10 random fields per section were taken within the LV myocardium, including the free wall and interventricular septum. Cells that exhibited colocalization of Ki67 and DAPI were considered to be cycling. Ki67‐positive nuclei and the total number of DAPI‐stained nuclei were manually quantified using the cell counter plugin of the ImageJ software. A total number of ≈5000 nuclei per heart were evaluated, and data from all 10 images were averaged to give the mean proliferation rate for each heart. Evaluation of Cell Death {#jah32416-sec-0016} ------------------------ Apoptotic cells were detected in cardiac paraffin sections using a terminal deoxynucleotidyl transferase dUTP nick end labeling (TUNEL) assay (ApopTag^®^ Fluorescein Apoptosis Detection Kit, Merck Millipore, Billerica, MA) according to the manufacturer\'s instructions. Nuclei were stained with DAPI (Invitrogen). Two longitudinal nonadjacent TUNEL‐stained sections per heart and 6 random fields per section were imaged using the Biozero BZ‐8100 fluorescence microscope (Keyence) with ×15 optical magnification. Cells that exhibited colocalization of TUNEL and DAPI staining were considered apoptotic. TUNEL staining as well as the total number of DAPI‐stained nuclei were manually counted using ImageJ. A total number of ≈15 000 nuclei per neonatal heart were analyzed, and data from all 12 images were averaged to give the mean apoptosis rate for each heart. Cardiomyocyte Cross‐Sectional Area, Length, and Volume {#jah32416-sec-0017} ------------------------------------------------------ To evaluate cardiomyocyte cross‐sectional area (CSA), cardiac paraffin sections were stained with fluorochrome‐conjugated wheat germ agglutinin (WGA, Alexa Fluor 555; Invitrogen) to visualize cell membranes. Nuclei were stained with DAPI or TO‐PRO‐3 (Invitrogen). For neonates, images were taken with the confocal laser scanning microscope TSC SPE (Leica Microsystems, Wetzlar, Germany) applying ×120 optical magnification, whereas adult hearts were imaged with the Biozero BZ‐8100 fluorescence microscope (Keyence) applying ×20 optical magnification. Fifteen random fields of the LV myocardium including free wall and interventricular septum were imaged per section, and 2 nonadjacent WGA‐stained sections per heart were used. At least 200 cardiomyocytes per heart were measured using the area measurement tool of the Biozero BZ image analysis application software (Keyence). Only cardiomyocytes that were cut along their transverse axis and matched the following selection criteria were included: visible central nucleus, cell shape close to circular, clear cell borders discernible, and visible cytoplasm. Values of all cardiomyocytes were averaged to give a mean estimate of cardiomyocyte CSA for each heart. Cardiomyocyte length was measured in adult hearts following immunofluorescence staining of paraffin sections with WGA and an antibody against N‐cadherin (sc‐7939, Santa Cruz Biotechnology, Dallas, TX) to identify intercalated disks (general staining procedure same as for evaluation of proliferation described above). Two nonadjacent sections per heart were imaged at ×40 magnification using a Zeiss Axio Scope.A1 fluorescence microscope. Ten to 15 images per section were taken from areas of the LV myocardium (free wall and interventricular septum) that showed cardiomyocytes in their longitudinal orientation. Rod‐shaped cardiomyocytes with clearly visible cell borders, nuclei and intercalated disks were included. The distance between intercalated disks of ≈70 to 100 cardiomyocytes per heart was measured using the ZEN blue software (Zeiss), and the average cardiomyocyte length was calculated. Calculation of cardiomyocyte volume was based on the simplified assumption that cardiomyocytes are of cylindrical shape (as previously described[14](#jah32416-bib-0014){ref-type="ref"}), such that mean CSA was multiplied by mean cardiomyocyte length for each adult heart. Calculation of Cardiomyocyte Number Per Heart {#jah32416-sec-0018} --------------------------------------------- Cardiomyocyte number per adult heart was estimated based on previously published protocols[15](#jah32416-bib-0015){ref-type="ref"} with minor modifications. Heart volume was calculated by multiplying cardiac wet weight or LV mass (determined by echocardiography) with the tissue density of rat myocardium, ie, 1.048 g/mm^3^.[15](#jah32416-bib-0015){ref-type="ref"} The relative contribution of cardiomyocytes to the cardiac tissue was estimated by determining cardiomyocyte area fraction on WGA‐stained paraffin sections (staining as described above for CSA). Briefly, LV myocardium of 2 nonadjacent sections was imaged at ×40 magnification, resulting in 20 to 30 images per heart. For each image the WGA‐positive area (including nonmyocytes, extracellular matrix, and cell membranes) was measured using the color threshold plug‐in in ImageJ and related to the overall tissue area. The remaining WGA‐negative tissue area was considered to represent cardiomyocytes. Cardiomyocyte area fraction was averaged from all images per heart. Cardiomyocyte (CM) number was calculated according to the following equation:$$\text{CM\ number} = \frac{\text{heart\ volume} \times \text{CM\ area\ fraction}}{\text{CM\ volume}}$$ Western Blot Analyses {#jah32416-sec-0019} --------------------- For isolation of total protein extracts fresh or snap‐frozen tissue samples (heart, lung, kidney) were homogenized in RIPA buffer supplemented with protease (Complete Protease Inhibitor Cocktail Tablets, Roche Diagnostics, Indianapolis, IN) and phosphatase inhibitors (PhosSTOP Phosphatase Inhibitor Cocktail Tablets, Roche Diagnostics) and incubated at 4°C for 2 hours with gentle agitation. For samples on the same gel, equal protein amounts were loaded (20 to 50 μg), separated using denaturing SDS‐PAGE and blotted onto nitrocellulose (GE Healthcare, Chicago, IL) or polyvinylidene fluoride (Merck Millipore) membranes. Membranes were blocked for 1 hour in 5% nonfat dry milk (Carl Roth) in Tris‐buffered saline with Tween 20 and incubated with the following primary antibodies at 4°C overnight: phospho‐S6K1 Thr389 (\#9234), total S6K1 (\#2708), phospho‐S6 Ser235/236 (\#4858), phospho‐S6 Ser240/244 (\#4838), total S6 (\#2217), phospho‐4E‐BP1 Thr37/46 (\#2855), phospho‐4E‐BP1 Ser65 (\#9451), total 4E‐BP1 (\#9644), phospho‐mTOR Ser2448 (\#2971), phospho‐mTOR Ser2481 (\#2974), total mTOR (\#2983), RICTOR (\#2114), RAPTOR (\#2280), phospho‐AMPKα Thr172 (\#4188), total AMPKα (\#2532), phospho‐ULK1 Ser555 (\#5869), phospho‐ULK1 Ser757 (\#6888), total ULK1 (\#8054), LC3B (\#3868), p62 (\#5114), phospho‐Akt Thr308 (\#13038), phospho‐Akt Ser473 (\#4060), total Akt (\#4691) (all from Cell Signaling Technology, Danvers, MA). Antibodies against GAPDH (MA1‐22670, Thermo Scientific, Waltham, MA), α‐tubulin (T9026, Sigma‐Aldrich, St. Louis, MO), and vinculin (V9131, Sigma‐Aldrich) were used for loading control. Secondary detection was performed using horseradish peroxidase--linked secondary antibodies (Cell Signaling Technology). Enhanced chemiluminescence reaction was performed and detected with the imaging system Odyssey Fc (LI‐COR Biosciences, Lincoln, NE) or by exposure of the membrane to chemiluminescence‐sensitive CL‐XPosure films (Thermo Scientific). Intensity of detected protein bands was quantified by densitometry using ImageJ. Quantitative Real‐Time Polymerase Chain Reaction {#jah32416-sec-0020} ------------------------------------------------ Fresh or snap frozen cardiac tissue samples were homogenized in TRIzol reagent (Invitrogen), and total RNA was isolated according to the manufacturer\'s instructions. RNA was subsequently purified using RNeasy spin columns (Qiagen, Hilden, Germany), including digestion of genomic DNA on the column (RNase‐free DNase set, Qiagen). Isolated cardiac RNA was reversely transcribed into cDNA using M‐MuLV reverse transcriptase (New England BioLabs, Ipswich, MA) and random hexamer primers. Quantitative real‐time polymerase chain reaction (PCR) was performed using the Power SYBR^®^ Green PCR Master Mix (Applied Biosystems, Foster City, CA) on the ViiA™ 7 real‐time PCR system (Applied Biosystems). Primers were obtained from BioTeZ (Berlin, Germany), and sequences are as follows: *Nppa* forward 5′‐CAGCATGGGCTCCTTCTCCAT‐3′, *Nppa* reverse 5′‐ TGTACACAGGATTTGCTCCAATATG‐3′, *Myh7* forward 5′‐CTAGAGTCAAAGTGGGCAACG‐3′, *Myh7* reverse 5′‐GTGTCACCATCCAGTTGAACA‐3′, *Nppb* forward 5′‐AGGACCAAGGCCTCACAAAA‐3′, *Nppb* reverse 5′‐TTGAGATATGTGTCACCTTGGAATTT‐3′. Target gene expression was normalized against *Gapdh*:*Gapdh* forward 5′‐AGGTTGTCTCCTGCGACTTCA‐3′, *Gapdh* reverse 5′‐CCAGGAAATGAGCTTGACAAAGTT‐3′. All primers and PCR conditions were optimized to PCR efficiencies between 90% and 110% and a correlation coefficient ≥0.990 using cDNA dilution series. All samples were analyzed in triplicate. Relative expression differences between groups were determined using the ΔΔCT method. Statistical Analyses {#jah32416-sec-0021} -------------------- All data are presented as mean±standard error of the mean. Data sets were tested for normal distribution by the Kolmogorov‐Smirnov test, and homogeneity of variances between groups was assessed by the Levene test using SPSS (IBM, Armonk, NY). If these criteria were met, differences between 2 groups were evaluated with an unpaired, 2‐sided Student t test using Excel 2010 (Microsoft, Redmond, WA) and those among multiple groups with 1‐way ANOVA followed by Bonferroni post hoc test using SPSS (IBM). Differences among multiple groups with unequal variances were evaluated with nonparametric Kruskal‐Wallis 1‐way analysis of variance followed by Mann‐Whitney post hoc test using SPSS (IBM). Genotype distributions were compared to expected Mendelian distribution by a chi‐squared test using Microsoft Excel. Kaplan‐Meier survival curves were plotted using SPSS, and differences between groups were determined by a Log Rank (Mantel‐Cox) test. A probability (*P*) value less than 0.05 was considered to indicate statistical significance. Results {#jah32416-sec-0022} ======= Efficient Inhibition of mTORC1 in Neonatal Mice by Rapamycin Treatment of Pregnant Dams {#jah32416-sec-0023} --------------------------------------------------------------------------------------- In order to inhibit mTORC1 in fetal mice, we treated pregnant dams during the final quarter of gestation (starting at 15.5 dpc until delivery) by subcutaneous injections of rapamycin (5 mg/kg body weight) or vehicle every 12 hours (Figure [S1](#jah32416-sup-0001){ref-type="supplementary-material"}). mTORC1 activity was tested by Western blot analyses on different tissues harvested from newborn mice on postnatal day 1 (P1). Prenatal rapamycin treatment efficiently reduced phosphorylation of the mTORC1 downstream targets S6K1 and S6 ribosomal protein in the offspring heart, lung, and kidney (Figure [1](#jah32416-fig-0001){ref-type="fig"}A). 4E‐BP1 was unaffected, consistent with previous reports showing different responsiveness of mTORC1 targets to rapamycin.[16](#jah32416-bib-0016){ref-type="ref"}, [17](#jah32416-bib-0017){ref-type="ref"} {#jah32416-fig-0001} mTORC1 regulates autophagy in concert with AMP‐activated protein kinase (AMPK) by phosphorylating ULK1, a kinase important for the initiation of autophagy.[18](#jah32416-bib-0018){ref-type="ref"} Phosphorylation of AMPKα was increased in rapamycin‐ compared with vehicle‐treated neonatal hearts (Figure [S2](#jah32416-sup-0001){ref-type="supplementary-material"}A), as previously described in the mouse liver.[19](#jah32416-bib-0019){ref-type="ref"} Importantly, total ULK1 protein amounts are elevated in the P1 heart by prenatal rapamycin treatment (Figure [S2](#jah32416-sup-0001){ref-type="supplementary-material"}B). In agreement with AMPK activation, ULK1 phosphorylation at the AMPK‐dependent residue Ser555 (which promotes autophagy) was increased in rapamycin‐ compared with vehicle‐treated hearts, even after normalization to total ULK1 (Figure [S2](#jah32416-sup-0001){ref-type="supplementary-material"}B). In contrast, phosphorylation of the mTOR‐sensitive Ser757, which inhibits autophagy, was unaltered when normalized to total ULK1 (Figure [S2](#jah32416-sup-0001){ref-type="supplementary-material"}B). In conclusion, besides opposing phosphorylation by AMPK and mTOR, regulation of ULK1 activity in rapamycin‐treated neonatal hearts seems furthermore to involve altered ULK1 protein stability and/or synthesis (as previously proposed).[20](#jah32416-bib-0020){ref-type="ref"}, [21](#jah32416-bib-0021){ref-type="ref"} To investigate the net effect of these interactions, we tested protein levels of p62 (SQSTM1), which is rapidly degraded if autophagy is activated,[22](#jah32416-bib-0022){ref-type="ref"} and found reduced p62 in hearts of rapamycin‐ compared with vehicle‐treated neonates (Figure [1](#jah32416-fig-0001){ref-type="fig"}B). In addition, we detected reduced LC3B‐I but unaltered LC3B‐II protein amounts in rapamycin‐ compared to vehicle‐treated hearts (Figure [1](#jah32416-fig-0001){ref-type="fig"}B), a pattern also observed in heart conditional Rheb KO mice.[10](#jah32416-bib-0010){ref-type="ref"} Consequently, the LC3B‐II/LC3B‐I ratio is increased in rapamycin‐treated hearts, a parameter widely used to indicate activation of autophagy. The fact that LC3B‐II levels are not elevated, as often reported, could indicate accelerated autophagic flux such that, after conversion of LC3B‐I to LC3B‐II, the latter is rapidly degraded via lysosomal turnover.[22](#jah32416-bib-0022){ref-type="ref"} Based on these data we conclude that autophagy is activated in neonatal hearts after prenatal rapamycin treatment. It has been proposed that rapamycin can inhibit mTORC2 if applied in high doses or for prolonged treatment periods.[23](#jah32416-bib-0023){ref-type="ref"}, [24](#jah32416-bib-0024){ref-type="ref"} We did not detect differences in Akt phosphorylation at Ser473 (Figure [1](#jah32416-fig-0001){ref-type="fig"}C), a well‐established mTORC2 phosphorylation site, thereby excluding major inhibitory effects of rapamycin on mTORC2 in the neonatal heart. Finally, given that newborn pups did not receive further treatment after birth, we wanted to know at what postnatal stage mTORC1 activity is fully restored. Western blot analyses revealed that S6 phosphorylation in the heart is still reduced on postnatal day 2 (P2) but returns to normal levels at P3 (Figure [1](#jah32416-fig-0001){ref-type="fig"}D). In summary, rapamycin treatment of pregnant mice during late gestation results in efficient inhibition of rapamycin‐sensitive mTORC1 functions in the offspring at birth, which in the heart is restored by P3. Prenatal Rapamycin Treatment Causes IUGR and Reduces Heart Size at Birth {#jah32416-sec-0024} ------------------------------------------------------------------------ Prenatal rapamycin treatment in late gestation (starting at 15.5 dpc) does not cause fetal lethality, evident as unaltered litter size in rapamycin‐ compared to vehicle‐treated dams on postnatal day 1 (Figure [2](#jah32416-fig-0002){ref-type="fig"}A). In contrast, initiation of rapamycin treatment at 11.5 dpc resulted in spontaneous abortions around 16.5 dpc and fetal lethality with severe growth restriction and malformations. Consequently, all further analyses were performed with treatment commencing at 15.5 dpc (as depicted in Figure [S1](#jah32416-sup-0001){ref-type="supplementary-material"}). Pups born from rapamycin‐treated females were smaller in size at birth compared to vehicle‐treated pups, resulting in a 16.4% reduction in body weight (BW) (Figure [2](#jah32416-fig-0002){ref-type="fig"}B). Similarly, a reduction in heart size was already evident upon dissection and was furthermore confirmed by histological examination (Figure [2](#jah32416-fig-0002){ref-type="fig"}C). The latter does not reveal any major structural or morphological cardiac defects after prenatal mTORC1 inhibition, however. Strikingly, heart weight (HW) was reduced by 34.5%, resulting in a significantly reduced HW/BW ratio in rapamycin‐ versus vehicle‐treated pups (Figure [2](#jah32416-fig-0002){ref-type="fig"}D). In contrast, kidney weight (KW) was reduced in accordance with BW by 19.7%, resulting in unaltered KW/BW ratios between groups (Figure [2](#jah32416-fig-0002){ref-type="fig"}E). In conclusion, rapamycin treatment during late gestation causes IUGR evident as reduced body and organ size. Heart size is disproportionately affected, suggesting that fetal cardiac growth is specifically sensitive to mTORC1 inhibition. {#jah32416-fig-0002} Prenatal Rapamycin Treatment Reduces Cardiomyocyte Size and Induces Apoptosis but Does Not Affect Proliferation in the Neonatal Heart {#jah32416-sec-0025} ------------------------------------------------------------------------------------------------------------------------------------- The rapamycin‐induced reduction in neonatal heart size can be the result of different cellular mechanisms: a reduction in cardiomyocyte size or number, and the latter can be caused by induction of cell death or by inhibition of proliferation. We determined cell cycle activity in neonatal (P1) hearts of rapamycin‐ compared to vehicle‐treated pups by Ki67 immunostaining but did not find differences between the groups (Figure [3](#jah32416-fig-0003){ref-type="fig"}A). In contrast, when we assayed programmed cell death (apoptosis) by TUNEL staining, a significant increase in the number of TUNEL‐positive cells was observed in rapamycin‐treated P1 hearts (Figure [3](#jah32416-fig-0003){ref-type="fig"}B). Most strikingly, cardiomyocyte CSA was significantly reduced in the latter compared to vehicle‐treated pups, indicating a reduction in cell size (Figure [3](#jah32416-fig-0003){ref-type="fig"}C). Taken together, these data suggest that prenatal rapamycin treatment reduces cardiac organ size primarily by reducing cardiomyocyte size in combination with induction of apoptosis but not by interfering with proliferation. {#jah32416-fig-0003} Impact of Fetal mTORC1 Inhibition on a Mouse Model of Prenatal Compensatory Cardiac Growth {#jah32416-sec-0026} ------------------------------------------------------------------------------------------ Given the importance of mTOR for cardiac growth and organ size control,[5](#jah32416-bib-0005){ref-type="ref"}, [6](#jah32416-bib-0006){ref-type="ref"} it might be specifically required for embryonic heart regeneration in *cHccs* ^*+/−*^ mice.[11](#jah32416-bib-0011){ref-type="ref"} Interestingly, neonatal *cHccs* ^*+/−*^ hearts show increased phosphorylation of S6K1 and S6 ribosomal protein (Figure [4](#jah32416-fig-0004){ref-type="fig"}A) but not of mTORC1 (ie, 4E‐BP1, ULK1) or mTORC2 (ie, Akt) (Figure [S3](#jah32416-sup-0001){ref-type="supplementary-material"}) downstream targets compared to controls. This suggests that perinatal compensatory cardiac growth, maturation, or cell survival in *cHccs* ^*+/−*^ mice might depend on mTORC1 signaling to S6K1, which has been shown to fulfill various functions in the heart.[6](#jah32416-bib-0006){ref-type="ref"} To address this question, we subjected *cHccs* ^*+/−*^ fetuses to the same regime of prenatal rapamycin treatment as described above (ie, starting at 15.5 dpc). Similarly to control hearts, rapamycin efficiently inhibited mTORC1 activity in *cHccs* ^*+/−*^ hearts on postnatal day P1 (Figure [S4](#jah32416-sup-0001){ref-type="supplementary-material"}A). In litters from rapamycin‐treated dams, genotype distribution at P1 was as expected for X‐chromosomal inheritance of *Hccs* and not different from vehicle‐treated litters (Figure [S4](#jah32416-sup-0001){ref-type="supplementary-material"}B), excluding prenatal lethality of *cHccs* ^*+/−*^ females upon mTORC1 inhibition. Importantly, the rapamycin‐induced reduction in BW, HW, and HW/BW ratio was similar in *cHccs* ^*+/−*^ females compared to rapamycin‐treated controls (*Hccs* ^*+/+*^) (Figure [4](#jah32416-fig-0004){ref-type="fig"}B). The reduction in heart weight in neonatal *cHccs* ^*+/−*^ compared to *Hccs* ^*+/+*^ females due to the reduced number of cardiomyocytes described previously[12](#jah32416-bib-0012){ref-type="ref"} was evident in both treatment groups but was not aggravated by rapamycin. At the cellular level, rapamycin does not specifically impair proliferation in *cHccs* ^*+/−*^ hearts at P1 (Figure [4](#jah32416-fig-0004){ref-type="fig"}C) but further induces apoptosis (Figure [4](#jah32416-fig-0004){ref-type="fig"}D) and reduces cardiomyocyte CSA (Figure [4](#jah32416-fig-0004){ref-type="fig"}E) when compared to rapamycin‐treated *Hccs* ^*+/+*^ females. In summary, whereas rapamycin does not seem to affect late‐gestational regulation of overall heart size specifically in *cHccs* ^*+/−*^ females, it does have more pronounced effects on cardiomyocyte size and cell survival. The latter implies that rapamycin‐sensitive mTORC1 functions control certain cellular aspects of perinatal compensatory growth and tissue homeostasis in *cHccs* ^*+/−*^ hearts, which could impact on their morphology and function in adulthood (see [Discussion](#jah32416-sec-0006){ref-type="sec"} in Data [S1](#jah32416-sup-0001){ref-type="supplementary-material"}). {ref-type="ref"} persisted after prenatal mTORC1 inhibition (vehicle groups n=9, rapamycin groups n=10). C, Quantification of Ki67‐positive nuclei revealed unchanged proliferation rates within the LV myocardium of vehicle‐ and rapamycin‐treated hearts and between genotypes (vehicle groups n=6, rapamycin *Hccs* ^+/+^ n=12, rapamycin *cHccs* ^*+/−*^ n=13). D, Quantification of TUNEL‐positive nuclei revealed significantly increased apoptosis in hearts after prenatal mTORC1 inhibition, with apoptosis in rapamycin‐treated *cHccs* ^*+/−*^ hearts being significantly higher than in rapamycin‐treated *Hccs* ^+/+^ controls (vehicle *Hccs* ^+/+^ n=4, all other groups n=6). E, Prenatal rapamycin treatment significantly reduced cardiomyocyte cross‐sectional area (CSA) in both genotypes, and CSA in rapamycin‐treated *cHccs* ^*+/−*^ hearts was significantly smaller compared to rapamycin‐treated *Hccs* ^+/+^ controls (vehicle groups n=5, rapamycin groups n=3). B through E, Note that data used for *Hccs* ^+/+^ animals are the same as depicted in Figures [2](#jah32416-fig-0002){ref-type="fig"} and [3](#jah32416-fig-0003){ref-type="fig"}, respectively. (\**P*\<0.05, \*\**P*\<0.01, \*\*\**P*\<0.001; ^§§§^ *P*\<0.001 vs vehicle *Hccs* ^+/+^; ^\#\#^ *P*\<0.01, ^\#\#\#^ *P*\<0.001 vs vehicle *cHccs* ^*+/−*^). mTORC1 indicates mechanistic target of rapamycin complex 1; TUNEL, terminal deoxynucleotidyl transferase dUTP nick end labeling.](JAH3-6-e005506-g004){#jah32416-fig-0004} Normal Contractility but Reduced Stroke Volume and Cardiac Output in Neonatal Hearts After Rapamycin Treatment {#jah32416-sec-0027} -------------------------------------------------------------------------------------------------------------- Considering the significant effects of prenatal mTORC1 inhibition on neonatal heart size, we performed echocardiography on rapamycin‐ and vehicle‐treated pups at P1 in order to evaluate cardiac function. In agreement with the heart weight and morphology data described above, rapamycin‐treated neonates had a significantly reduced systolic LV wall thickness (Figure [5](#jah32416-fig-0005){ref-type="fig"}A), a reduced LV internal diameter (Figure [5](#jah32416-fig-0005){ref-type="fig"}B), and a reduced calculated LV mass (Figure [5](#jah32416-fig-0005){ref-type="fig"}C). LV contractility, however, was not affected by prenatal rapamycin treatment (Figure [5](#jah32416-fig-0005){ref-type="fig"}D). The smaller LV dimensions result in a significantly reduced LV stroke volume and cardiac output in rapamycin‐ versus vehicle‐treated pups, given that heart rate was not different between groups (Figure [5](#jah32416-fig-0005){ref-type="fig"}E). In addition, prenatal mTORC1 inhibition does not have more severe effects on cardiac dimensions or function in *cHccs* ^*+/−*^ compared to *Hccs* ^*+/+*^ females (Table [S1](#jah32416-sup-0001){ref-type="supplementary-material"}). In summary, prenatal rapamycin treatment reduces LV size and cardiac output at birth. {#jah32416-fig-0005} Body and Organ Weights Partially Normalize After Prenatal Rapamycin Treatment Until Adulthood {#jah32416-sec-0028} --------------------------------------------------------------------------------------------- To monitor the postnatal fate of pups born after prenatal mTORC1 inhibition, we analyzed rapamycin‐ and vehicle‐treated animals at the age of 11 weeks. As shown above, mTORC1 activity is restored in the rapamycin‐treated heart 3 days after birth (Figure [1](#jah32416-fig-0001){ref-type="fig"}D), suggesting that body and organ growth should be uncompromised thereafter. Nevertheless, we noticed postnatal lethality in a subset of rapamycin‐treated pups within the first 12 days after birth, resulting in 67.3% survival compared with 95.8% in vehicle‐treated mice (Figure [6](#jah32416-fig-0006){ref-type="fig"}A). After day 12 no further deaths were recorded, and all animals survived to early adulthood. Comparison of genotype distribution at weaning (ie, at 21 days of age) revealed no evidence for preferential lethality of rapamycin‐treated *cHccs* ^*+/−*^ mice after birth (Figure [S5](#jah32416-sup-0001){ref-type="supplementary-material"}). At the age of 11 weeks, rapamycin‐treated animals still had a significantly reduced BW and tibia length (TL) compared with vehicle‐treated mice (Figure [6](#jah32416-fig-0006){ref-type="fig"}B). BW was reduced by 10% compared with 16% at birth, indicating a certain degree of postnatal catch‐up growth, however. Importantly, HW completely normalizes in rapamycin‐ compared with vehicle‐treated mice, resulting in normal HW/BW and HW/TL ratios after 11 weeks (Figure [6](#jah32416-fig-0006){ref-type="fig"}C). Histological examination confirmed mainly normal cardiac morphology, although hearts appeared slightly smaller, and LV walls slightly thinner, compared with the vehicle group (Figure [6](#jah32416-fig-0006){ref-type="fig"}D). Liver and spleen weights were not different in adult rapamycin‐ versus vehicle‐treated mice, whereas KW was significantly reduced (Figure [6](#jah32416-fig-0006){ref-type="fig"}E). Nevertheless, organ weights normalized to BW or TL were not different between treatment groups (Figure [6](#jah32416-fig-0006){ref-type="fig"}F and Table [S2](#jah32416-sup-0001){ref-type="supplementary-material"}). Moreover, no specific effect of rapamycin treatment on body or organ size was observed in *cHccs* ^*+/−*^ compared to *Hccs* ^*+/+*^ females (Table [S2](#jah32416-sup-0001){ref-type="supplementary-material"}). Taken together, after prenatal rapamycin treatment, pups that survived the early postnatal period showed incomplete catch‐up growth of body and kidney size, whereas heart, liver, and spleen weights all normalized until adulthood. {#jah32416-fig-0006} Reduced LV Mass but Normal Cardiac Contractility and Output in Adult Mice After Prenatal Rapamycin Treatment {#jah32416-sec-0029} ------------------------------------------------------------------------------------------------------------ Long‐term consequences of prenatal rapamycin treatment on postnatal heart morphology and function were evaluated by echocardiography at the age of 11 weeks. As suggested by histological examinations (Figure [6](#jah32416-fig-0006){ref-type="fig"}D), rapamycin‐treated animals had a significantly reduced LV wall thickness compared with vehicle‐treated controls (Figure [7](#jah32416-fig-0007){ref-type="fig"}A). The fact that LV diameter was not different between the groups (Figure [7](#jah32416-fig-0007){ref-type="fig"}B) resulted in a significantly reduced LV mass in rapamycin‐treated mice calculated from echo data, whereas the LV mass/BW ratio was unaffected (Figure [7](#jah32416-fig-0007){ref-type="fig"}C). Similar to neonatal stages, rapamycin‐treated adults did not show impairment of LV contractility (Figure [7](#jah32416-fig-0007){ref-type="fig"}D). Interestingly and in contrast to newborns, LV stroke volume, heart rate, and cardiac output were normal in rapamycin‐ compared with vehicle‐treated adults (Figure [7](#jah32416-fig-0007){ref-type="fig"}E). The latter suggests functional recovery during postnatal life, considering reduced cardiac output in rapamycin‐treated animals at birth (Figure [5](#jah32416-fig-0005){ref-type="fig"}E). No major impairment of cardiac morphology and function was observed in rapamycin‐treated *cHccs* ^*+/−*^ compared to *Hccs* ^*+/+*^ females (Table [S3](#jah32416-sup-0001){ref-type="supplementary-material"}), indicating that prenatal mTORC1 inhibition does not negatively influence the long‐term outcome specifically in *cHccs* ^*+/−*^ females despite their reliance on compensatory cardiac growth. In summary, prenatal inhibition of rapamycin‐sensitive mTORC1 functions results in a sustained reduction of LV mass until early adulthood. Despite unaltered overall cardiac wet weight (Figure [6](#jah32416-fig-0006){ref-type="fig"}C), we conclude that LV dimensions are not completely restored in rapamycin‐treated mice within the first 11 weeks of life, although heart size (ie, wet weight and LV mass) appears largely appropriate when normalized to BW. {#jah32416-fig-0007} Reduced Cardiomyocyte Number but No Maladaptive Remodeling in Adult Hearts After Prenatal Rapamycin Treatment {#jah32416-sec-0030} ------------------------------------------------------------------------------------------------------------- The reduced LV wall thickness and mass in adult hearts after prenatal rapamycin treatment could be caused by a reduction in cardiomyocyte number or size. Surprisingly, cardiomyocyte CSA was increased in rapamycin‐ compared with vehicle‐treated mice (Figure [8](#jah32416-fig-0008){ref-type="fig"}A), suggesting an increase in cell size. In addition, we measured cardiomyocyte length in tissue sections by N‐cadherin immunofluorescence staining of the intercalated disks between longitudinally oriented cells (Figure [8](#jah32416-fig-0008){ref-type="fig"}B). Cardiomyocyte length was not different between rapamycin‐ and vehicle‐treated adult hearts (Figure [8](#jah32416-fig-0008){ref-type="fig"}B). Assuming a simplistic model of cardiomyocytes being of cylindrical shape, multiplying CSA by cell length allowed us to calculate the average cardiomyocyte volume per heart. This revealed a significant increase in cardiomyocyte volume in rapamycin‐ compared with vehicle‐treated adult hearts (Figure [8](#jah32416-fig-0008){ref-type="fig"}C). Considering reduced LV mass in echocardiography and unchanged cardiac wet weight in rapamycin‐treated animals, the latter would argue for compensatory cardiomyocyte hypertrophy to account for a reduction in cell number. To estimate cardiomyocyte number per heart, we first determined the relative contribution of cardiomyocytes to the myocardium in WGA‐stained LV tissue sections. These data revealed a slightly but significantly reduced cardiomyocyte area fraction in rapamycin‐ compared with vehicle‐treated adult hearts (Figure [8](#jah32416-fig-0008){ref-type="fig"}D). Based on LV mass or cardiac wet weight, cardiomyocyte area fraction, and cardiomyocyte volume, we were able to calculate the number of cardiomyocytes per heart (see [Methods](#jah32416-sec-0004){ref-type="sec"}). Although this allows only a rough assessment and likely underestimates the absolute cardiomyocyte number, it should be suitable for relative comparisons between treatment groups. The results indeed revealed a significantly reduced number of cardiomyocytes per heart in rapamycin‐ compared with vehicle‐treated mice in adulthood (Figure [8](#jah32416-fig-0008){ref-type="fig"}E). The reduction in cell number could be compensated for by excessive deposition of extracellular matrix and therefore result in maladaptive myocardial remodeling. We did not detect differences in interstitial fibrosis within the LV myocardium between rapamycin‐ and vehicle‐treated mice, however (Figure [8](#jah32416-fig-0008){ref-type="fig"}F). Furthermore, RNA expression of fetal genes to indicate a molecular signature of pathological conditions was not different between groups (Figure [8](#jah32416-fig-0008){ref-type="fig"}G). Finally, prenatal rapamycin treatment does not have more severe consequences on CSA, fibrosis, or fetal gene expression in adult *cHccs* ^*+/−*^ compared to *Hccs* ^*+/+*^ females (Figure [S6](#jah32416-sup-0001){ref-type="supplementary-material"}), indicating that prenatal mTORC1 inhibition can be compensated in *cHccs* ^*+/−*^ hearts. Taken together, prenatal rapamycin treatment reduces the number of cardiomyocytes in the postnatal heart, which can be partially compensated by increased cell volume to achieve a near normal organ size. {#jah32416-fig-0008} Discussion {#jah32416-sec-0031} ========== Intrauterine growth restriction can be induced by a variety of maternal or environmental conditions, most of which converge on the (single or combined) restriction of nutrients, energy, or oxygen to the fetus.[1](#jah32416-bib-0001){ref-type="ref"} The cellular mechanisms that sense such shortage and slow down intrauterine growth, however, are not well understood. Here we show that rapamycin‐sensitive mTORC1 function regulates fetal growth and determines body and organ size at birth. Given that mTORC1 integrates nutrient, amino acid, and oxygen availability with cell growth and proliferation,[4](#jah32416-bib-0004){ref-type="ref"} it seems likely that inhibition of mTORC1 by different IUGR conditions is causally involved in fetal growth restriction and aberrant organ maturation. Indeed, a low‐protein diet (LPD) fed to female mice throughout pregnancy reduces pancreatic β‐cell mass in the newborn offspring, eventually leading to impaired insulin secretion and glucose intolerance.[25](#jah32416-bib-0025){ref-type="ref"} LPD causes reduced mTORC1 activity in β‐cells at birth, and transiently restoring mTORC1 function rescues β‐cell mass and prevents the diabetic phenotype in the offspring.[25](#jah32416-bib-0025){ref-type="ref"} In addition, food restriction in pregnant baboons or sheep reduces mTORC1 activity in the fetal liver and skeletal muscle, respectively.[26](#jah32416-bib-0026){ref-type="ref"}, [27](#jah32416-bib-0027){ref-type="ref"} Although corresponding data for the growth‐restricted fetal or newborn heart are missing, it is tempting to speculate that mTORC1 inhibition is a general consequence of IUGR in various organs, thereby contributing to developmental programming of adult disease. Rapamycin treatment of pregnant dams represents a new IUGR model in rodents, which could prove useful to study developmental programming in various organ systems. We have shown efficient inhibition of rapamycin‐sensitive (ie, S6K1, S6) but not rapamycin‐insensitive (ie, 4E‐BP1) mTORC1 downstream targets in the newborn heart, kidney, and liver and a concomitant reduction in heart and kidney weight. For animal studies of developmental programming, however, it has been recommended to use only 1 representative male and/or female offspring per litter in order to account for differences in the intrauterine or postnatal environment specific for certain pregnancies or dams. A limitation of the current study in this regard is that we included several *Hccs* ^*+/+*^ and *cHccs* ^*+/−*^ females from the same litter, which is common practice when analyzing genetically modified mice. Consequently, we cannot exclude that the close relationship of littermates in contrast to randomly chosen mice might have an effect on certain study results. The rapamycin treatment protocol applied in this study (ie, starting at 15.5 dpc) results in a quite stable ≈16% reduction in neonatal body weight, the degree of which is comparable to other IUGR studies.[28](#jah32416-bib-0028){ref-type="ref"}, [29](#jah32416-bib-0029){ref-type="ref"}, [30](#jah32416-bib-0030){ref-type="ref"} In contrast, the widely used maternal protein restriction during pregnancy has shown quite remarkable variations in neonatal body weight between different studies and investigators, which are likely to be influenced by the mouse or rat strain used, the exact diet composition, the precise onset of LPD, the duration of LPD after birth, and other variables (reviewed by Zohdi et al[31](#jah32416-bib-0031){ref-type="ref"}). In this regard we believe that rapamycin treatment of pregnant dams is easy to control and can be standardized to achieve consistent degrees of growth restriction and therefore reproducible results when studying fetal programming in adulthood. Furthermore, the rapamycin dose can potentially be adjusted to regulate the degree of IUGR, and an adapted treatment regime might even allow to precisely time the period of mTORC1 inhibition by starting and terminating rapamycin application at the desired gestational stage. Such a protocol could determine time windows during embryonic or fetal development that are most susceptible to growth restriction and therefore most relevant for developmental programming. An important consideration in assessing the new IUGR model described here is the effect of rapamycin on the maternal organism. Because it is applied systemically to pregnant mice, it is likely to interfere with placental growth and function, even though placental development is already well established at the onset of treatment (ie, 15.5 dpc in this study).[32](#jah32416-bib-0032){ref-type="ref"} Nevertheless, mTORC1 signaling has been shown to play an important role in placental nutrient sensing and consequently maternal‐to‐fetal nutrient exchange.[33](#jah32416-bib-0033){ref-type="ref"} Interestingly, however, mTORC1 activity is reduced in the placenta by maternal protein or nutrient restriction in rats and baboons, respectively,[28](#jah32416-bib-0028){ref-type="ref"}, [29](#jah32416-bib-0029){ref-type="ref"} as well as in human placentas upon IUGR.[34](#jah32416-bib-0034){ref-type="ref"}, [35](#jah32416-bib-0035){ref-type="ref"} Systemic rapamycin administration to pregnant dams could furthermore interfere with mammary gland development, thereby impairing lactation and milk intake of newborn offspring. Indeed, treatment of pregnant mice with the mTORC1 inhibitor RAD001 or rapamycin starting on the day before delivery and continuing for 5 or 12 days, respectively, disturbs mammary gland tissue architecture and reduces organ size, milk protein production, and pup weight.[36](#jah32416-bib-0036){ref-type="ref"}, [37](#jah32416-bib-0037){ref-type="ref"} Rapamycin furthermore reduces milk protein production in murine and bovine mammary epithelial cells in vitro, which was also observed upon amino acid starvation.[38](#jah32416-bib-0038){ref-type="ref"} Importantly, maternal LPD during pregnancy in rats impairs late gestation mammary gland development,[39](#jah32416-bib-0039){ref-type="ref"} and uteroplacental insufficiency induced by uterine artery ligation in rats impairs mammary gland development and function as well as milk production and composition, thereby restraining postnatal pup growth.[30](#jah32416-bib-0030){ref-type="ref"} These studies indicate that impaired mammary gland function as well as placental mTORC1 inhibition appear to be general features of various IUGR animal models, such that a potential maternal impact by rapamycin treatment would not be fundamentally different. The degree of the latter will have to be compared to other IUGR models in future studies, however. This appears furthermore warranted considering lethality in ≈30% of rapamycin‐treated pups during the first 12 postnatal days (Figure [6](#jah32416-fig-0006){ref-type="fig"}A).The latter could well be caused by lactation insufficiency or metabolic alterations in the maternal organism, resulting in undernourishment of the offspring. Alternatively, postnatal lethality could be independent of maternal factors and might be the result of variable severity of rapamycin‐induced IUGR and organ dysfunction or impaired postnatal compensatory processes (eg, insufficient catch‐up growth) in some pups. Our data show that full mTORC1 activity is essential for fetal heart growth during late gestation, given that neonates after prenatal rapamycin treatment have small hearts and a reduced HW/BW ratio compared with vehicle‐treated controls. The latter is the result of a disproportional (34.5%) reduction in HW compared with BW (16.4%). These findings are in contrast to maternal LPD or placental insufficiency in rats, in which reduction in offspring HW parallels the reduction in BW, such that HW/BW ratio is normal.[30](#jah32416-bib-0030){ref-type="ref"}, [31](#jah32416-bib-0031){ref-type="ref"} Interestingly, the reduction in KW (19.7%) in rapamycin‐treated pups was proportional to body weight, resulting in normal KW/BW ratios. These data suggest that fetal heart growth more heavily depends on rapamycin‐sensitive mTORC1 activity as compared with other organs and that the prenatal heart might be specifically susceptible to fetal programming involving mTORC1 inhibition. Decreased neonatal heart size after prenatal rapamycin treatment is primarily caused by a reduction in cardiomyocyte size, whereas proliferation is unaffected at birth. This is in agreement with rapamycin inhibiting phosphorylation of S6K1 but not 4E‐BP1 in the neonatal heart, as it has been shown that 4E‐BP1 primarily mediates the proliferative effects downstream of mTORC1, whereas S6K1 regulates cell size.[40](#jah32416-bib-0040){ref-type="ref"}, [41](#jah32416-bib-0041){ref-type="ref"} It is furthermore consistent with the heart conditional knockout of mTOR and the upstream mTORC1 activator Rheb, both of which result in full mTORC1 inhibition shortly after birth with reduced cardiomyocyte size by days 8 and 15, respectively,[7](#jah32416-bib-0007){ref-type="ref"}, [10](#jah32416-bib-0010){ref-type="ref"} whereas proliferation rates were not directly determined. We furthermore detected increased apoptotic cell death within the myocardium of rapamycin‐treated neonates, which is in agreement with previous studies showing cardiomyocyte apoptosis upon mTORC1 inhibition in the embryonic or early postnatal heart.[7](#jah32416-bib-0007){ref-type="ref"}, [8](#jah32416-bib-0008){ref-type="ref"}, [9](#jah32416-bib-0009){ref-type="ref"} It is tempting to speculate that in addition to diminished cell size a reduced cardiomyocyte number due to rapamycin‐induced cell death might also contribute to the reduction in heart size. A limitation of our study, however, is that we did not differentiate between cell types when determining proliferation and apoptosis, so we cannot exclude that primarily a nonmyocyte cell population is undergoing cell death. Similarly, unaltered overall proliferation rates in the myocardium could obscure slight differences between cardiac cell types. Nevertheless, when studying adult hearts after prenatal rapamycin treatment, morphometric calculations revealed a reduced cardiomyocyte number in rapamycin‐ versus vehicle‐treated hearts in adulthood. An intriguing question is whether this cardiomyocyte deficit is solely due to increased cardiomyocyte apoptosis in the perinatal phase or whether impaired proliferation in the prenatal period also contributes. In addition, given that murine cardiomyocytes can proliferate during the first postnatal week,[42](#jah32416-bib-0042){ref-type="ref"} and mTORC1 activity is only restored by day 3, cardiomyocyte proliferation could also be impaired shortly after birth. A limitation of our study in this regard is that we did not investigate fetal hearts or placental morphology and function between commencement of rapamycin treatment at 15.5 dpc and birth. Therefore, we cannot exclude that rapamycin induces metabolic alterations in the mother or molecular changes in the placenta that generally impair nutrient supply to the fetus, thereby potentially inhibiting fetal (cardiac) growth beyond the rapamycin‐sensitive functions of mTORC1. For example, impaired placental amino acid exchange[28](#jah32416-bib-0028){ref-type="ref"}, [29](#jah32416-bib-0029){ref-type="ref"}, [34](#jah32416-bib-0034){ref-type="ref"} could potentially result in full mTOR inhibition, which in turn could inhibit cardiomyocyte proliferation in the fetus. Given that such rapamycin‐independent inhibitory effects would be restored quickly after birth, our current study might be unable to detect the molecular and cellular consequences when investigating hearts on postnatal day 1. So whether rapamycin‐treated neonates are born with a reduced cardiomyocyte number or whether this develops after birth will have to be determined in future studies, in parallel with investigations of placental function. The reduction in body and organ weight in rapamycin‐treated mice at birth is partially normalized until early adulthood. Body weight and tibia length remain slightly reduced at the age of 11 weeks, which is in agreement with persistent weight reduction in other IUGR rodent models.[31](#jah32416-bib-0031){ref-type="ref"}, [43](#jah32416-bib-0043){ref-type="ref"} Interestingly, the ability for postnatal catch‐up growth appears to vary among organs. Although absolute heart, liver, and spleen weights are mainly normal in 11‐week‐old rapamycin‐ compared with vehicle‐treated animals, kidney weight is still reduced, although not different when normalized to body weight or tibia length. Nevertheless, these data suggest that the kidney has an impaired postnatal growth plasticity in response to IUGR (as previously proposed by human as well as animal studies[43](#jah32416-bib-0043){ref-type="ref"}, [44](#jah32416-bib-0044){ref-type="ref"}), which might contribute to developmental programming of hypertension and kidney disease. For the heart, we observed a discrepancy between overall gravimetric wet weight and LV mass calculated from echocardiography data. Whereas the latter was reduced in rapamycin‐ versus vehicle‐treated mice in adulthood, wet weight was unchanged. These differences are likely caused by the impact of blood or fluids in the cardiac cavities as well as right ventricular and atrial myocardium on wet weight but not LV mass calculations. Given that histological examinations and echocardiography data confirmed slightly reduced LV wall thickness in rapamycin‐ versus vehicle‐treated adult mice, we conclude that heart dimensions are not fully normalized until early adulthood after prenatal mTORC1 inhibition. This is in contrast to other IUGR animal models, which mainly revealed unaltered heart size in adulthood, although it was often determined as wet weight at later stages.[31](#jah32416-bib-0031){ref-type="ref"}, [43](#jah32416-bib-0043){ref-type="ref"} Therefore, it is quite possible that LV mass completely normalizes in rapamycin‐treated mice with further aging. Neonates after prenatal rapamycin treatment show reduced cardiac output caused by smaller LV dimensions and stroke volume but not impaired contractility when compared with vehicle‐treated controls. This is in contrast to mice with an inducible, heart conditional knockout of mTOR or Raptor in adulthood, which develop contractile dysfunction and heart failure under baseline conditions,[45](#jah32416-bib-0045){ref-type="ref"}, [46](#jah32416-bib-0046){ref-type="ref"} although over a period of 4 to 6 weeks. Importantly, it is also different from heart conditional Rheb KO mice, which show efficient mTORC1 inhibition by postnatal day 5 and contractile dysfunction by day 8.[10](#jah32416-bib-0010){ref-type="ref"} Similarly, heart conditional mTOR KO mice that exhibit efficient mTOR protein depletion by postnatal day 7 develop contractile dysfunction by day 15.[7](#jah32416-bib-0007){ref-type="ref"} The latter 2 genetic models, however, cause a broad inhibition of mTORC1 (and also mTORC2 in case of the mTOR knockout) in the heart beyond rapamycin‐sensitive functions, evident as decreased phosphorylation of 4E‐BP1, which was unaffected in our current study. These data suggest that LV contractility in fetal and neonatal hearts is relatively insensitive toward rapamycin‐dependent mTORC1 inhibition, and cardiac output is primarily determined by LV volume. Within the first few weeks after birth, however, mTORC1 activity becomes essential to establish and maintain normal LV function. The latter coincides with various milestones of postnatal cardiac maturation, such as changes in cardiomyocyte metabolism, growth pattern, tissue composition, and myocardial workload.[47](#jah32416-bib-0047){ref-type="ref"} Thus, it is tempting to speculate that mTORC1 plays an important role in the transition from fetal to postnatal cardiac growth and function. Despite slightly reduced LV mass and wall thickness and a reduced number of cardiomyocytes, adult hearts after prenatal rapamycin treatment exhibit normal contractility and complete recovery of cardiac output as compared with neonatal stages without signs of maladaptive myocardial remodeling. This is consistent with some rodent studies showing unaltered LV contractility after IUGR under baseline conditions in adulthood,[31](#jah32416-bib-0031){ref-type="ref"}, [48](#jah32416-bib-0048){ref-type="ref"} but others have reported reduced heart function at 10 to 12 weeks of age.[49](#jah32416-bib-0049){ref-type="ref"}, [50](#jah32416-bib-0050){ref-type="ref"} Similarly, most IUGR animal models eventually show increased myocardial fibrosis with age, whereas the onset of tissue remodeling in early adulthood is less clear.[31](#jah32416-bib-0031){ref-type="ref"}, [49](#jah32416-bib-0049){ref-type="ref"}, [51](#jah32416-bib-0051){ref-type="ref"} In addition, maternal LPD during pregnancy has recently been proposed to alter the biochemical composition of myocardial tissue in adult offspring.[51](#jah32416-bib-0051){ref-type="ref"} Whether this represents a general effect of IUGR or whether it is specific to the LPD model needs to be confirmed, such that characterizing the cardiac lipid, proteoglycan, and carbohydrate profile in rapamycin‐treated hearts might add valuable information to this question. So although prenatal rapamycin treatment obviously results in an IUGR phenotype at birth, it will have to be established whether it furthermore represents a suitable model for fetal programming in adulthood. In this regard it will be interesting to determine the long‐term cardiac outcome of adult mice after prenatal rapamycin treatment upon aging as well as their response to various challenges (such as ischemia, pressure overload, or neurohumoral stimulation) in future studies. Such data would have important implications for both fetal programming in general as well as for the specific role of mTORC1 in long‐term cardiovascular health and disease susceptibility. Sources of Funding {#jah32416-sec-0033} ================== This work was supported by institutional funds of the Max‐Delbrück‐Center for Molecular Medicine Berlin and the University Hospital Münster but received no specific grant from other funding agencies. Disclosures {#jah32416-sec-0034} =========== None. Supporting information ====================== ###### **Data S1.** Supplemental Discussion. **Table S1.** Echocardiographic Measurements in Neonatal Mice After Prenatal mTORC1 Inhibition **Table S2.** Body and Organ Weights in Adult Mice After Prenatal Rapamycin or Vehicle Treatment **Table S3.** Echocardiographic Measurements in Adult Mice After Prenatal mTORC1 Inhibition **Figure S1.** Inhibition of mTORC1 in fetal and neonatal mice by rapamycin treatment of pregnant dams. Rapamycin was injected subcutaneously at a dose of 5 mg/kg body weight in pregnant dams every 12 hours from 15.5 days postconception until delivery. As controls, pregnant dams were injected with vehicle only. **Figure S2.** Evaluation of the autophagy‐regulating kinases AMPK and ULK1 in neonatal hearts after prenatal rapamycin treatment. A, Phosphorylation of the AMPK (AMP‐activated protein kinase) subunit α is increased in neonatal hearts after prenatal rapamycin treatment, indicating AMPK activation. B, The kinase ULK1 (Unc‐51‐like autophagy‐activating kinase 1) is phosphorylated by AMPK at Ser555 to initiate autophagy, whereas mTOR phosphorylates ULK1 at Ser757 to inhibit autophagy. Consistent with AMPK activation, ULK1 Ser555 phosphorylation is increased in neonatal hearts after prenatal rapamycin treatment, whereas Ser757 phosphorylation is unchanged when normalized to total ULK1. Note that the latter is confounded by a significant increase in total ULK1 protein levels in rapamycin‐ compared to vehicle‐treated hearts. (Densitometric quantification n=4 per group in A and B, \**P*\<0.05, \*\**P*\<0.01, \*\*\**P*\<0.001). **Figure S3.** Unaltered phosphorylation of mTOR, ULK1, 4E‐BP1, and Akt in neonatal *cHccs* ^*+/−*^ hearts. A, Phosphorylation of mTOR at Ser2448 and Ser2481 is unaltered in neonatal *cHccs* ^*+/−*^ hearts compared to controls. Similarly, no difference in total protein amounts of mTOR or its interacting proteins RAPTOR and RICTOR (specific for mTOR complexes 1 and 2, respectively) is detected between groups (densitometric quantification n=8 per genotype; POI=protein of interest). B, Phosphorylation of the mTORC1 downstream targets 4E‐BP1 and ULK1 (at the mTORC1‐dependent site Ser757) is not different between neonatal *cHccs* ^*+/−*^ and control hearts. The kinase Akt acts as both an upstream regulator of mTORC1 and as an mTORC2 downstream target. Phosphorylation of Akt at Thr308 (via the PI3K pathway) and Ser473 (by mTORC2) is unaltered in *cHccs* ^*+/−*^ hearts (densitometric quantification n=5 to 6 per genotype). **Figure S4.** Prenatal mTORC1 inhibition in *cHccs* ^*+/−*^ fetuses does not cause lethality prior to birth. A, Western blots of heart protein extracts from neonatal (P1) *cHccs* ^*+/−*^ females and their *Hccs* ^+/+^ female littermate controls after prenatal vehicle or rapamycin treatment illustrating the phosphorylation status of the mTORC1 downstream targets S6K1, S6 and 4E‐BP1. Note the increased mTORC1 activity toward S6K1 and S6 in *cHccs* ^*+/−*^ females in vehicle‐treated animals (see main text and Figure [4](#jah32416-fig-0004){ref-type="fig"}) but similar reduction of S6K1 and S6 phosphorylation in hearts of rapamycin‐ compared to vehicle‐treated offspring, indicating successful mTORC1 inhibition in *cHccs* ^*+/−*^ hearts. Phosphorylation of 4E‐BP1 was demonstrated to be unaffected by prenatal rapamycin treatment (see main text and Figure [1](#jah32416-fig-0001){ref-type="fig"}). B, The genotype distribution within vehicle‐ as well as rapamycin‐treated litters was not significantly different from expected genotype frequencies, indicating that *cHccs* ^*+/−*^ females do not exhibit prenatal lethality due to mTORC1 inhibition. Statistical significance between treatment groups and the expected genotype distribution was assessed by chi‐squared test. Statistical significance comparing the frequencies of each genotype between vehicle‐ and rapamycin‐treated litters was assessed by unpaired 2‐tailed Student t tests (n=9 litters per group). **Figure S5.** There was no preferential death of rapamycin‐treated *cHccs* ^*+/−*^ females after birth. A, Breeding of males hemizygous for the "floxed" *Hccs* allele (*Hccs* ^*floxed/y*^ */Nkx2.5* ^*+/+*^) to females homozygous for the *Nkx2.5Cre* allele (*Hccs* ^*+/+*^/*Nkx2.5* ^*Cre/Cre*^) generates 2 possible offspring genotypes, ie, control males (*Hccs* ^*+/y*^ */Nkx2.5* ^*Cre/+*^) and heterozygous heart‐conditional *Hccs* knockout females (*Hccs* ^*floxed/+*^ */Nkx2.5* ^*Cre/+*^) at an expected ratio of 50:50. B, Breeding of males homozygous for the *Nkx2.5Cre* allele (*Hccs* ^*+/y*^/*Nkx2.5* ^*Cre/Cre*^) to females heterozygous for the "floxed" *Hccs* allele (*Hccs* ^*floxed/+*^ */Nkx2.5* ^*+/+*^) generates 3 postnatally viable genotypes, ie, control males (*Hccs* ^*+/y*^ */Nkx2.5* ^*Cre/+*^), control females (*Hccs* ^*+/+*^ */Nkx2.5* ^*Cre/+*^), and heterozygous heart‐conditional *Hccs* knockout females (*Hccs* ^*floxed/+*^ */Nkx2.5* ^*Cre/+*^) at an expected ratio of 33% each. Hemizygous heart‐conditional *Hccs* knockout males (*Hccs* ^*floxed/y*^ */Nkx2.5* ^*Cre/+*^) die in utero at stage 10.5 dpc.[11](#jah32416-bib-0011){ref-type="ref"} Genotype distribution within vehicle‐ as well as rapamycin‐treated litters was evaluated at weaning (ie, at 21 days of age). For both breeding schemes no significant difference compared to expected genotype frequencies was observed, indicating that rapamycin‐treated *cHccs* ^*+/−*^ females do not preferentially die within the first 12 days after birth (see Figure [6](#jah32416-fig-0006){ref-type="fig"}A). Statistical significance between treatment groups and the expected genotype distribution was assessed by chi‐squared test (3 litters per group in A; 5 vehicle‐ and 4 rapamycin‐treated litters in B). **Figure S6.** Prenatal mTORC1 inhibition does not cause significantly different tissue remodeling in adult *cHccs* ^*+/−*^ compared to *Hccs* ^*+/+*^ hearts. A, Fluorescence images of cross‐sectioned cardiomyocytes within the left ventricular (LV) myocardium of adult hearts. Cardiomyocyte membranes were stained in red with wheat germ agglutinin (WGA) and nuclei in blue with DAPI (scale bar=100 μm). Cardiomyocyte cross‐sectional area (CSA) was significantly larger in *Hccs* ^*+/+*^ as well as *cHccs* ^*+/−*^ hearts exposed to prenatal rapamycin compared to vehicle treatment. Note that compensatory cardiomyocyte hypertrophy in c*Hccs* ^*+/−*^ mice reported previously[12](#jah32416-bib-0012){ref-type="ref"} is evident in the vehicle group but lost in rapamycin‐treated hearts (vehicle *Hccs* ^*+/+*^ n=6, rapamycin *Hccs* ^*+/+*^ n=7, vehicle and rapamycin c*Hccs* ^*+/−*^ n=5). B, Representative images of Sirius Red--stained LV myocardium of adult mice. White arrowheads highlight interstitial fibrosis, whereas black arrowheads indicate perivascular fibrosis, which was excluded from quantification (scale bar=300 μm). The contribution of fibrotic tissue to the LV myocardium of adult mice did not reveal differences between treatment groups or genotypes (n=6 for all groups except vehicle c*Hccs* ^*+/−*^, n=8). C, Relative *Nppa*,*Myh7,* and *Nppb* RNA expression in adult hearts was determined using quantitative real‐time polymerase chain reaction to evaluate pathological conditions. No major changes in gene expression were observed in rapamycin (R) compared to vehicle (V) groups in either *Hccs* ^*+/+*^ or *cHccs* ^*+/−*^ hearts (n=8 per group). Note that data used for *Hccs* ^*+/+*^ animals in A through C are the same as depicted in Figure [8](#jah32416-fig-0008){ref-type="fig"}. (\**P*\<0.05, \*\**P*\<0.01, ^§§^ *P*\<0.01 versus vehicle *Hccs* ^*+/+*^, ^\#^ *P*\<0.05 versus vehicle c*Hccs* ^*+/−*^). ###### Click here for additional data file. We thank Martin Taube and Stefanie Schelenz for performing echocardiography, Anja Conrad and Michaela Tirre for technical assistance, and the Microscopy Core Facility at the Max‐Delbrück‐Center for technical support.

Introduction {#S1} ============ In recent years, there have been extensive studies on the role of plant-derived compounds that regulate interactions between herbivorous insects and their host plants ([@B25]; [@B9]; [@B60]; [@B20]). Several studies have shown that plant volatiles play a significant role in mating and host recognition in phytophagous insects by influencing antennal sensitivities to chemical components ([@B67]; [@B49]; [@B44]; [@B10]; [@B19]). Various works have indicated that host-derived volatiles were successfully deployed in the management of agricultural and forestry pests by the development of botanical attractants, push--pull strategies, and modulation of reproductive behaviors. The botanical attractants based on volatiles from host plants are known to effectively attract both sexes of herbivorous insect pests, which plays an important role in monitoring and control of insect pests including *Bactrocera dorsalis*, *Helicoverpa armigera*, *Cydia pomonella*, and *Frankliniella occidentalis* ([@B33]; [@B32]; [@B55]; [@B56]; [@B1]; [@B18]). The push--pull strategies with botanical volatiles have also been widely utilized in pest control, for instance, *Delia radicum* in cabbage, *Diaphorina citri* Kuwayama in citrus, and Aphididae species in wheat-pea strip ([@B28]; [@B68]; [@B66]). Furthermore, several studies have shown that the chemical composition of host plant plays an integral part in mating and oviposition behaviors in various insects, including Lepidoptera, Hemiptera, and Coleoptera ([@B53]; [@B46]; [@B26]). The fall webworm *Hyphantria cunea* (Durry) (Lepidoptera: Arctiidae) is a devastating invasive insect that is native to North America. *H. cunea* was first recorded in Liaoning Province in China in 1979 and later it has widely expanded its distribution to Beijing, Tianjin, Hebei, Shandong, Shanxi, Jiangsu, and Anhui provinces ([@B72]; [@B12]; [@B75]). It has been reported that the occurrence of *H. cunea* in China reached a 20.9% annual increase, and this pest has been causing massive damage to forests, orchards, and even crops because of its polyphagous trait ([@B72]; [@B57]; [@B76]). Currently, more than 400 plant species are recorded to serve as hosts for *H. cunea* worldwide ([@B52]). Several methods including biopesticides, parasitoid wasps, and sex pheromone traps have been used to monitor and control *H. cunea* ([@B57]; [@B71]; [@B29]; [@B50]). Specifically, due to its TYPE II sex pheromone structures, synthetic pheromone luring of *H. cunea* was applied at a higher cost than other lepidopteran pests ([@B22]; [@B57]; [@B52]). Sex pheromone monitoring of *H. cunea* in China nowadays has been solely dependent on imported luring products, which raised the demands of developing botanical attractants/synergists or related push--pull strategy for large-scale IPM needs ([@B6]). There has been less previous evidence regarding the influences of plant volatiles toward *H. cunea.* Some volatile compounds from host plant (*Morus alba*, *Malus spectabilis*) and area-dependent non-host plant (*Robinta pseudoacacia*) have been identified ([@B59]; [@B5], [@B7]). Given the fact that this polyphagous pest may infest different plant hosts in its habitats depending on local vegetation distributions ([@B69]), screen for universal host cues can help to explore the chemical ecology of this moth. On the other hand, recent research has reported morphology of its antennal sensilla, but information on the distribution of sensillar types has not yet been tackled ([@B74]). To date, a number of questions on olfaction of *H. cunea*, such as its key host volatiles as well as the electrophysiological sensitivity on antennae and how plant volatiles affect its reproductive behaviors, remain to be answered. In the current work, we identified 78 host volatile compounds from five favored host species of *H. cunea* in China and narrowed down to six key components for later tests. The distributions of six sensilla types on antennae of *H. cunea* were investigated by scanning electron microscopy (SEM). Later, the electroantennogram (EAG) test against selected host-plant odorants has been used to investigate the sensitive volatiles that have potential effects on the behaviors. To test the reproductive behavioral responses of *H. cunea* to volatile compounds associated with host plants, we studied the effects of sensitive volatiles on the mating and oviposition of *H. cunea*. The results from the current research can provide insights into a better understanding of the olfactory perception of *H. cunea*. Moreover, selection of the effective botanical components can be used for future utilizations as a botanical attractant and pheromone synergetics in ecological-based control of this devastating pest in the fields. Materials and Methods {#S2} ===================== Insects and Plants {#S2.SS1} ------------------ *H. cunea* were obtained from the Chinese Academy of Forestry, where a laboratory population of *H. cunea* using wild moths was established from Qinghuangdao, Hebei Province, China. The pupae were stored in a climate chamber at 25 ± 1°C and R.H. 50--60% under a 16L:8D photoperiod until emergence. Newly emerged moths were immediately separated by gender and kept at a density of five adults per 500-ml beaker with 5% sugar solution provided. One- to three-day-old virgin moths were fasted for 3--6 h before being used in electrophysiological and behavioral tests. Plant leaves from *M. alba*, *Fraxinus chinensis*, *Populus alba*, *Ailanthus altissima*, and *M. spectabilis* were collected in forestry experimental sites of the Tianjin Institute of Plant Protection in Wuqing, Tianjin (39°25′33.3″N, 116°57′35.7″E). Leaf samples were immediately brought back to the lab where extraction process was carried out within 1 day. Chemical Analysis {#S2.SS2} ----------------- Tested leaf samples were cut into fine pieces and boiled with water with 1:1 w/w for 4--6 h to prepare the water distillations, followed by extraction with hexane. Upper phases of the extracts were then collected and desiccated with anhydrous sodium sulfate drying agent columns. Prepared essential extracts were analyzed at 1 ml/min through an Agilent Technologies 5973MD mass spectrometer coupled with an Agilent Technologies 6890N gas chromatography system (Santa Clara, CA, United States) equipped with a quartz capillary column (HP-5MS, 30 m × 0.25 mm × 0.25 μm; J&W Scientific, Palo Alto, CA, United States). Volatile traces were identified by crosschecking with the mass spectrum fragment database (NIST 2.0) with GC/MSD ChemStation (Agilent) and selected components for later tests were confirmed against standard chemical spectrum patterns. Scanning Electron Microscopy {#S2.SS3} ---------------------------- The antennae from the head capsule of emerged adults (*n* = 12; sex ratio, 1:1) were placed in 2% glutaraldehyde solution fixed overnight at 4°C. After three washes at room temperature with 0.1 M PBS, antennae were incubated for 2 h at 4°C and then washed three times with 0.1 M PBS. The specimens were dehydrated through a series of graded ethanol (30, 50, 70, 80, and 90%), followed by three washings in 100% ethanol. After drying in a critical point drier (Bal-Tel CPD 030), antennae were mounted on aluminum stubs, taking care to place them with different orientations, to obtain views of the ventral and dorsal aspects and of both of the lateral sides. Then, the mounted specimens were sprayed with gold (Bal-Tel SCD 005) and observed with SEM (FEI Quanta 200). The sensilla were classified according to their morphology, shape, and size, following [@B51] and [@B73]. Chemicals {#S2.SS4} --------- The 11 *M. alba*-related volatiles were previously identified ([@B59]), and they were selected in this study as references for the later chemicals; detailed information can be found in [Supplementary Table S1](#DS1){ref-type="supplementary-material"}. The reason we put these 11 chemicals in the comparison was that they were from the favorite plant of *H. cunea* and they compose a full spectrum of bioactive volatiles of this host species toward the pest. We suggested that the amplitudes elicited by these chemicals may provide a baseline response for investigation on the newly tested six volatiles in this work. The six additional components (phytol, n-pentatriacontane, linalyl butyrate, palmitic acid, α-linolenic acid, and dibutyl phthalate, all 95% minimum purity) used in this study were selected by either referring to the GC-MS results or to the studies on host plant *M. spectabilis* and area-dependent non-host plant *R. pseudoacacia* of *H. cunea* ([@B5], [@B7]). All of these six synthetic standard components were obtained from Meryer Chemical Technology Co., Ltd. (Shanghai, China). For EAG experiment, the components (phytol, linalyl butyrate, α-linolenic acid, dibutyl phthalate, and other 11 *M. alba* volatiles) were dissolved in paraffin oil for dosage response trials (10 μl applied to a 0.8 × 2 cm filter paper with concentrations at 0.001, 0.01, 0.1, 1, 10 μg/μl, respectively), and n-pentatriacontane and palmitic acid were dissolved at the same dosages in hexane as they were insoluble in paraffin oil. The other 11 reference volatiles were prepared at four concentrations of 0.1--100 μg/μl (equal to 1--1000 μg in dosages). The dosages of the six components for mating and oviposition experiments were decided according to the results of the EAG experiment. Electrophysiology {#S2.SS5} ----------------- The EAG method was adopted to test the receptivity of the antenna of male and female moths to the volatile compounds. Each antenna was carefully excised at both extremes and immediately placed within the indifferent electrodes with Spectra 360 conductive gel. The signals were passed through an IDAC-2 data acquisition controller unit and recorded by a computer using a software package (Syntech, Kirchzarten, Germany). The stimulus was delivered into a purified air stream (30 ml/s) by a gas control unit (Syntech CS-55). The gas outlet faced the antenna with 1 cm distance from it. For each sample, 10 μl was applied to a 0.8 × 2 cm filter paper in a Pasteur pipette. Antenna from both sexes of *H. cunea* were stimulated for 0.8 s with an interval time of 30 s. The same antenna was used to test all of the concentrations of a single compound. Each concentration was tested with six different antennae per gender. The sequence of the tested compound was provided starting from the weakest concentration and followed by increasing concentrations. The control was added before and after stimulation by each concentration of the tested compounds. Recorded EAG data were fed into IDAC-2 and analyzed with EAG 2000 software (Syntech). Reproductive Behavioral Response {#S2.SS6} -------------------------------- To evaluate the reproductive behavioral responses of moths to plant-derived volatiles, single-pair courtship assays were used. We observed the effect of the electrophysiologically active compounds (at the concentration that elicited the strongest EAG response) on mating and oviposition of *H. cunea*. Green rubber septa (The West Company, Phoenixville, PA, United States) were loaded with 100 μl of the tested volatile mixtures at the selected concentration. Septa containing 100 μl of solvent and a blank septum were used as controls. A pair (one male, one female) of 1- to 3-day-old virgin moths was placed in a beaker (500 ml, *d* = 9.4 cm, *h* = 12.4 cm) together with one septum at 25 ± 1°C, R.H. 50--60%, and 16L:8D. The number of pairs that mated within 24 h was observed and counted. Mating rate was defined as the proportion of the mated pairs in 10--15 pairs tested independently, and oviposition rate indicates proportion of mated females that performed oviposition behaviors. Finally, egg numbers were counted after 48 h in oviposition female pairs as female fecundity. A total of four replicates were conducted for each treatment. Statistical Analysis {#S2.SS7} -------------------- The EAG data were standardized to include background influences of solvents before means ± SEM were calculated for parametric tests. Specifically, the relative EAGs to a test stimulus (Sr) were calculated as Sr = 2 Sc/(R′ + R″), where Sc is the absolute amplitude of the stimuli and R′ and R″ are the mean responses to the reference substances before and after stimulation ([@B23]; [@B77]). The same data of each chemical at 10 μg were used in both baseline comparison and dosage comparison. Means ± SEM were calculated for measurement data (sensilla parameters) and proportion data (behavioral tests) before testing in a parametric model. Counts data for sensilla were tested with chi-square tests. Statistics for means were analyzed with either two-tailed *t*-test (two treatments) or two-tailed GLM (\>2 treatments) followed by *Dunnett* against controls or *Tukey* HSD at *P* = 0.05. SPSS version 17.0 (IBM, Chicago, IL, United States) software was used for statistical analysis. Results {#S3} ======= Common Volatiles in Five Favored Host Plant {#S3.SS1} ------------------------------------------- A total 78 components were identified through GC-MS approach in the extracts of host plants for *H. cunea* ([Figure 1](#F1){ref-type="fig"}). Most of the components existed in only one to two blends, but phytol and α-linolenic acid were observed in the GC-MS traces of all five blends from *M. alba*, *F. chinensis*, *P. alba*, *A. altissima*, and *M. spectabilis*. α-Linolenic acid occupied the highest proportion of 17.73 ± 5.6% (mean ± SEM) in the volatiles, and phytol was the second highest at 10.6 ± 4.9%. Palmitic acid existed in four of five plant volatile blends including *M. alba*, *F. chinensis*, *P. alba*, and *A. altissima*, occupying overall 9.8 ± 4.5% of the blends. 2-Methoxy-4-vinylphenol was also observed in four plants of *M. alba*, *P. alba*, *A. altissima*, and *M. spectabilis* with covering 7.73 ± 3.16% of the volatiles. Diacetone alcohol was also found in *M. alba*, *F. chinensis*, *P. alba*, and *A. altissima* with a proportion of 1.96 ± 0.80%. Based on these results, we suggested that the top three components including phytol, α-linolenic acid, and palmitic acid may have vital roles in host communications of *H. cunea* as they were the most common and abundant volatiles in the tested favored host plants. We then decided to include these three chemicals in the successive tests. The main reason for selecting the other n-pentatriacontane, linalyl butyrate, and dibutyl phthalate for testing was based on our previous works which revealed their potentially high electrophysiological activities toward *H. cunea* ([@B5], [@B7]). We sought to find whether some behavioral/ecological significance of these six chemicals can be otherwise exhibited in the later trials. {ref-type="supplementary-material"}.](fphys-11-00486-g001){#F1} Sensillar Types and Distribution {#S3.SS2} -------------------------------- The antenna of *H. cunea* moth is composed of the scape, pedicel, and flagellum; the flagellum of the female is composed of 34--42 sub-segments (mean ± SEM = 40 ± 0.95) and that of the male is 35--42 (38.6 ± 1.24) ([Supplementary Figures S1A--D](#DS1){ref-type="supplementary-material"}). The estimated length of the flagellum was 5467 ± 247.0 μm for males and 5644.3 ± 151.3 μm for females. No difference was observed between genders in terms of length of flagellum ([Supplementary Figure S1E](#DS1){ref-type="supplementary-material"}) (*t*-test, *t*~10~ = 0.61, *P* = 0.55). The SEM investigations of the antenna revealed six types of sensilla, including sensilla chaetica, sensilla trichodea, sensilla basiconica, sensilla coeloconica, sensilla squamiformia, and sensilla böhm bristles ([Supplementary Figure S2](#DS1){ref-type="supplementary-material"}). The antennal morphology and types of sensilla as well as identifiers in *H. cunea* are identical with recently reported works on the same species ([@B74]). Distributions of sensilla types were not even between genders, and males bear significantly more sensilla trichodea per antenna than females ([Figure 2A](#F2){ref-type="fig"}) (chi-square test, χ~4~ = 405, *P* \< 0.0001), indicating its potential role in sex pheromone olfactory reception. Males had significant longer sensilla trichodea (*t*-test, *t*~10~ = 2.57, ^∗^*P* = 0.028) and shorter sensilla chaetica than female adults ([Supplementary Figure S2N](#DS1){ref-type="supplementary-material"}) (*t*-test, *t*~10~ = 2.8, ^∗^*P* = 0.018). Basal width of sensillar types were the same between genders expected for sensilla chaetica ([Supplementary Figure S2N](#DS1){ref-type="supplementary-material"}) (*t*-test, *t*~10~ = 2.9, ^∗^*P* = 0.016). {#F2} Electrophysiological Responses of *H. cunea* to Host Volatiles {#S3.SS3} -------------------------------------------------------------- All chemicals were compared at the concentration of 1 μg/μl, and gender bias for each chemical was shown in the heatmap ([Figure 2B](#F2){ref-type="fig"}). Female moths showed significant higher responses than males when tested with dibutyl phthalate, β-ocimene, hexanal, cis-2-penten-1-ol, and cyclohexanene ([Figure 2B](#F2){ref-type="fig"}) (GLM, dibutyl phthalate: *t* = 2.95, *P* = 0.014; β-ocimene: *t* = 3.78, *P* = 0.0026; hexanal: *t* = 4.30, *P* = 0.001; cis-2-penten-1-ol: *t* = 6.6, *P* \< 0.0001; cyclohexanene: *t* = 3.22, *P* = 0.0073). While no male bias response was observed among all tested chemicals. Within the six new chemicals, linalyl butyrate, dibutyl phthalate, and palmitic acid elicited significantly higher EAG responses in both genders of adults when compared with most other chemicals ([Figure 2B](#F2){ref-type="fig"}) \[GLM and Tukey HSD, females: *F*~(16,85)~ = 10.8, *P* \< 0.0001; males: *F*~(16,107)~ = 14.5, *P* \< 0.0001\]. In the later dosage response tests, significant differences in the EAG responses were observed among the doses for all components ([Figures 3A--F](#F3){ref-type="fig"} and [Supplementary Figure S3](#DS1){ref-type="supplementary-material"}). Strong and significant positive correlation was observed between EAG responses and dosages in males against α-linolenic acid in males \[Linear regression, *r*^2^ = 0.84, *F*~(4,22)~ = 15.99, *P* = 0.028\], n-pentatriacontane \[Linear regression, *r*^2^ = 0.94, *F*~(4,22)~ = 53.6, *P* = 0.0053\], and palmitic acid \[Linear regression, *r*^2^ = 0.85, *F*~(4,22)~ = 17.38, *P* = 0.0251\] ([Figures 3D--F](#F3){ref-type="fig"}). Female EAG responses positively correlated with dosages when tested with palmitic acid (Linear regression, *r*^2^ = 0.89, *F*~4,22~ = 23.51, *P* = 0.0167) ([Figure 3F](#F3){ref-type="fig"}). EAG responses of dibutyl phthalate \[*F*~(4,25)~ = 5.6, *P* = 0.002\] and n-pentatriacontane in females \[*F*~(4,25)~ = 17.9, *P* \< 0.0001\] were decreased along with increase of dosages ([Figures 3C,E](#F3){ref-type="fig"}). The highest responses for phytol were at the dosage of 0.1 μg/μl, and a decrease was observed afterward ([Figure 3A](#F3){ref-type="fig"}) \[GLM and Tukey HSD, female: *F*~(4,25)~ = 11.6, *P* \< 0.0001; male: *F*~(4,25)~ = 3.79, *P* = 0.0153\]. {#F3} When comparing the standardized EAG responses of both sexes of *H. cunea* to all six volatile compounds at the same concentration, male bias was observed for linalyl butyrate at 0.1--10 μg/μl, and α-linolenic acid at 0.001 and 1 μg/μl ([Figures 3B,D](#F3){ref-type="fig"}). In contrast, female bias was observed for phytol at 0.001 μg/μl and dibutyl phthalate at 0.01--0.1 μg/μl ([Figures 3A,C](#F3){ref-type="fig"}). For n-pentatriacontane, female bias was observed at 0.001 μg/μl, while male bias was observed during 0.1, 1, and 10 μg/μl ([Figure 3E](#F3){ref-type="fig"}). Last, female and male moths did not show any significant sex-specific difference in their EAG responses to palmitic acid \[two-way ANOVA, *F*~(4,50)~ = 1.157, *P* = 0.93\]. Mating Rates Are Increased by the Volatiles {#S3.SS4} ------------------------------------------- In the reproductive behavioral assays, background mating rate was 22.5 ± 3.69% (mean ± SEM) in blank control. Solvent as paraffin oil had elicited mating rates at 22.86 ± 2.19% and hexane was at 59.07 ± 3.60% ([Figure 4A](#F4){ref-type="fig"}). Compared to solvent, moth adults were more inclined to mate after they have been exposed to host-plant volatiles including dibutyl phthalate (*t-*test, *t*~6~ = 9.48, *P* \< 0.0001) and phytol (*t-*test, *t*~6~ = 6.33, *P* = 0.0007), but no difference was observed between n-pentatriacontane and its solvent hexane (*t-*test, *t*~6~ = 0.88, *P* = 0.41) ([Figure 4A](#F4){ref-type="fig"}). The mating rates of moths exposed to dibutyl phthalate and phytol have reached 55.83 ± 15.11% and 71.25 ± 18.07%, respectively, at 24 h, more than twofold than the control and the paraffin oil solvent. On the other hand, palmitic acid decreased mating rates of adults compared with its solvent hexane (*t-*test, *t*~6~ = 3.05, *P* = 0.0226) ([Figure 4A](#F4){ref-type="fig"}). {#F4} Over 80% of the mated females conducted oviposition afterward when no chemical was applied ([Figure 4B](#F4){ref-type="fig"}). However, all volatiles including two solvents did not influence the oviposition rates of *H. cunea* females \[GLM and Dunnett test, *F*~(8,27)~ = 0.49, *P* = 0.85\] ([Figure 4B](#F4){ref-type="fig"}). Background fecundity of *H. cunea* was 516.6 ± 45.87 eggs (mean ± SEM) per female in blank control treatment. When comparing the total number of eggs produced in different pairs, no significant difference was observed between the pairs exposed to volatiles and the control or the solvent \[GLM and Dunnett test, *F*~(8,27)~ = 1.57, *P* = 0.15\] ([Figure 4C](#F4){ref-type="fig"}). Discussion {#S4} ========== Plant volatiles serve as essential cues to insects searching for resources, mates, and oviposition sites ([@B26]; [@B21]; [@B24]; [@B65]). The reception of host-plant volatiles in Lepidoptera is accomplished through a highly sensitive chemosensory system on the antenna, which is able to recognize and discriminate many different volatile chemicals ([@B16]; [@B17]; [@B43]). The present electron microscopy study demonstrated that both sexes of *H. cunea* contain six different types of sensilla on their antenna, including sensilla trichodea, sensilla chaetica, sensilla basiconica, sensilla coeloconica, sensilla squamiformia, and sensilla böhm bristles, which revealed identical observations with previous studies ([@B74]). Compared with other Lepidoptera, the morphology of the sensilla on the antenna of *H. cunea* is similar to related species, such as *H. armigera*, *Phauda flammans* Walker, *Dogwood Borer*, *Holcocerus hippophaecolus* Hua, and *Tuta absoluta* ([@B14]; [@B62]; [@B11]; [@B8]; [@B35]). Antennal sensilla sexual dimorphism can be reflected by different length and width diameters of corresponding sensillar types ([@B74]). Furthermore, we have also observed that distribution of sensillar types may vary between genders, which is intimately connected to feeding, mate location, oviposition, and other functions, indicating an especially ubiquitous form of intraspecific variation in moths ([@B2]). The function of sensilla has been reported to be primarily sensory, responsible for olfactory detection and perception of tactile signals. This involved various behaviors including habitat searching, host recognition, host location, host acceptance, copulation, and oviposition ([@B3]; [@B42]; [@B78]). The sensilla trichodea are the most conspicuous and the most numerically abundant type of sensilla on the antenna of *H. cunea*. It is frequently shown that sex pheromone components and plant volatiles are usually detected by sensilla trichodea ([@B27]; [@B42]; [@B11]; [@B36], [@B37]), suggesting that sensilla trichodea may be responsible for distinguishing pheromone components and plant volatiles. There is also ample evidence that sensilla basiconica participate in recognition for plant volatiles ([@B39]; [@B70]). The information of the literature on the function of sensilla is instructional significant for peripheral coding of plant-associated compounds for *H. cunea*. Electrophysiological recordings have been applied to the study of peripheral coding of host-plant volatiles in *H. cunea* ([@B59], [@B58]). Both sexes of *H. cunea* moths showed high sensitivity to the compounds tested that are potentially involved in host--moth interactions. These results are relatively consistent with a preliminary study; volatiles and their dosages elicited electro-physiological responses in the antenna of male and female moths ([@B59], [@B58]). On the other hand, responses to olfactory cues in behavioral assays are not always consistent with EAG tests ([@B13]; [@B64]). Our mating tests also revealed that linalyl butyrate and α-linolenic acid did not influence mating rates in *H. cunea*, although they exhibited significant EAG amplitudes. While we cannot exclude the possible role of these two chemicals in terms of feeding or host location, we still need to be cautious when talking about bioactive compounds without substantial evidence from behavioral tests. Furthermore, significant differences in the EAG responses were observed among the doses, which is consistent with other studies ([@B40]; [@B45]). For instance, the highest EAG responses were found with linalyl butyrate, α-linolenic acid, and palmitic acid at the highest dose; in contrast, the compounds dibutyl phthalate and n-pentatriacontane elicited a more robust response at lower doses. This differential odor sensitivity could be attributed to the presence of specific receptors that respond to particular categories of chemicals ([@B54]). The higher EAG responses were found with linalyl butyrate than all other compounds in both sexes of the moths. Plant-associated ester compounds are more efficient for other moths, such as pear ester in *C. pomonella*, *Rhagoletis*, and *Dysaphis plantaginea* ([@B33]; [@B34]; [@B61]). The results obtained in our electrophysiological experiment prompted us to investigate the function of the compounds derived from plants on the reproduction system. In single-pair courtship assays, we found that the mating rate of *H. cunea* exposed to host-derived volatiles dibutyl phthalate and phytol were more than twice that from the control and paraffin oil. This is consistent with what has been found in the previous study regarding the effects of β-ocimene (host volatile) on mating rates of *H. cunea* moth pairs ([@B58]). These results suggested that host-derived volatile cues could promote mating behaviors in *H. cunea* moths. The same is true for females of *Plutella xylostella*, which rely heavily on host-plant volatiles for mating ([@B63]). Similarly, females of certain tephritid fruit fly species (Diptera: Tephritidae) show increased receptivity to mating after exposure to host fruit volatiles ([@B15]; [@B4]). One of the most important factors is that females could emit more sex pheromone while exposing to plant volatiles, such as in cabbage looper moth *Trichoplusia* ([@B30]), the corn earworm moth *Helicoverpa zea* ([@B48]), and the tobacco budworm *Heliothis virescens* ([@B47]). Also, the plant-provided signals deliver spatial information for insects to find mates. For example, plant volatiles are likely to serve as attractants to insects over a larger spatial range and pheromones are more likely to help guide the insects over a relatively short distance to mate in parasitoids ([@B65]). Host-derived volatiles play an important role as oviposition stimulant for Lepidopterans, such as tobacco hornworm *Manduca sexta*, *Ostrinia latipennis*, *C. pomonella*, and the stem borer *Busseola fusca* ([@B41]; [@B31]; [@B38]; [@B26]). However, our results showed that female oviposition rates and fecundity of *H. cunea* did not show sensitivity toward the volatiles from host plants. In all, potential ecological-based trapping method can be developed by utilizing these common volatiles from host plants of *H. cunea*, and an olfactory peripheral coding mechanism and the molecular basis of this pest are worth looking into by future studies. Data Availability Statement {#S5} =========================== All datasets generated for this study are included in the article/[Supplementary Material](#DS2){ref-type="supplementary-material"}. Author Contributions {#S6} ==================== P-HB and RT designed the study. P-HB, H-MW, and ML conducted the SEM tests. P-HB, H-MW, and RT conducted the electroantennogram experiments. B-SL, B-ML, and X-SG conducted the indoor bioassays. RT, P-HB, and H-MW analyzed the data and drafted the manuscript with contributions from all authors. Conflict of Interest {#conf1} ==================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** This research was funded by the Creative Research for Young Scientists of Tianjin Academy of Agricultural Sciences (China) (2020008), the Central Public-interest Scientific Institution Basal Research Fund (No. Y2020GH21-1) of China, the Agricultural Transformation of Scientific and Technological Achievements of Tianjin: Introduction and Demonstration of sex pheromone Synergist of *Hyphantria cunea* (201901070), the President fund of Tianjin Academy of Agricultural Sciences (China) (16006), and Tianjin Fruits and Forest Research System -- Quality Control and Safety (ITTFPRS2018008). We thank Prof. Zhong-Ning Zhang for providing comments and suggestions on improving our manuscript. We also thank B.F.A. Xiao-Qian Bao for technical support on development of graphics. Supplementary Material {#S9} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fphys.2020.00486/full#supplementary-material> ###### Click here for additional data file. ###### Click here for additional data file. [^1]: Edited by: Peng He, Guizhou University, China [^2]: Reviewed by: Ping Wen, Xishuangbanna Tropical Botanical Garden (CAS), China; Hong-bin Wang, Research Institute of Forest Ecology, Environment and Protection (CAF), China [^3]: ^†^These authors have contributed equally to this work [^4]: ^‡^ORCID: Rui Tang, [orcid.org/0000-0002-9313-0802](http://orcid.org/0000-0002-9313-0802) [^5]: This article was submitted to Invertebrate Physiology, a section of the journal Frontiers in Physiology

Despite major advances in biomedical research, exceptional difficulties remain arising from the complexity of various diseases and their variability. Development of many disorders and their therapeutic responses are associated with disturbances in cell death pathways. Although for \>40 years our knowledge about cell death was restricted to apoptosis and pathological necrosis, nowadays based on Recommendations of the Nomenclature Committee of Cell Death, there are \>10 different cell death modalities recognized.^[@bib1]^ Cell death was classified according to its morphological appearance, enzymological criteria, functional aspects or immunological characteristics.^2^ Intensive work in the field, for example, by the beginning of 2014 \>400 000 publications in PubMed are related to this area of research, led to understanding the biochemical features of various modes of cell death and some of molecular mechanisms of their activation/development/execution. To get the insight into the complexity of the cell death networks, the upcoming field of systems biology has been successfully employed over the past decade. Systems biology is an interdisciplinary field of research that focuses on complex interactions within biological structures using a holistic perspective approach to biological and biomedical investigations. Systems biology combines theoretical and computational approaches with quantitative experimental data. On the theoretical side, a wide spectrum of mathematical formalisms is used. Their choice is based on the question to be answered by the modeling, available experimental data sets and the intricacy of the signaling network under consideration. Boolean models are effectively used to characterize large cell death signaling networks. In Boolean modeling, protein activities are presented by nodes that can be either off or on, and no knowledge is required for the quantitative characteristics of the individual reactions. In contrast, ordinary differential equations (ODEs) describe temporal dynamics of signaling networks and require the knowledge of kinetic parameters of the system as well as a set of temporally solved experimental data. ODE-based modeling is one of the most common approaches used in the analysis of the cell death networks. ODEs might not be sufficient for modeling spatiotemporal processes within the cell, for example, translocations within different compartments that involve spatiotemporal gradients.^[@bib3]^ In this case modeling is conducted using partial differential equations (PDEs). Modeling cell-to-cell variations arising from single-cell measurements requires stochastic simulations. In addition, Petri nets, agent-based models (ABMs) and Bayesian models have been employed for the analysis of the cell death networks. The combination of various mathematical tools allowed to quantitatively describe the major cell death processes and to identify biologically relevant systems\' properties that will be highlighted in this issue. Computational models require the exact knowledge about the numbers and interaction constants of the molecules in the pathway that allows making unique quantitative assessments upon molecular mechanisms of the complex signaling network regulation. These vigorous quantifications require state-of-the-art experimental methodology that includes quantitative biochemistry, cell biology and mass spectrometry techniques. The classical western blot and immunoprecipitation approaches were recently developed in the systems biology studies to generate time-resolved population data on the semiquantitative and quantitative levels. Single-cell analysis has enabled valuable insights into cell death using a number of special tools, including FRET-based and localization-based caspase activity probes. Finally, progress in mass spectrometry field is coming up with AQUA- and SILAC-based technologies, and development toward single-cell mass spectrometry analysis provides yet another major technological advance essential for the quantitative data generation. During the past decade, the powerful methodology of systems biology combining high-level mathematics with the state-of-the-art quantitative experimental work helped us to understand many aspects of cell\'s decision to live or to die, in particular, death receptor- and mitochondria-mediated cell death pathways were elucidated and understood on a systems level with the unprecedented level of detail.^[@bib4],\ [@bib5],\ [@bib6]^ Extrinsic apoptosis is triggered by activation of the death receptors. Modeling death receptor network provided the first example of systems biology study of cell death supported by experimental data.^[@bib7]^ This pioneer study was followed by a number of models uncovering dynamics of death receptor signaling and death-inducing signaling complex (DISC), the latter is essential for the initiation of CD95-mediated apoptotic and non-apoptotic responses.^[@bib8],\ [@bib9],\ [@bib10]^ Furthermore, recently, the stoichiometry of CD95 and TRAIL-R DISC complexes was revealed by quantitative and systems biology approaches.^[@bib4],\ [@bib5]^ Thus, there is a significant progress in this field, which is essential in the context of normal cell physiology and disease. The term 'apoptosome\' was introduced in 1997 by Hengartner^[@bib11]^ and used to describe the second molecular complex, which is activated during mitochondria-mediated apoptosis. In the presence of dATP, cytochrome *c*, released from the mitochondria, interacts with Apaf-1 and procaspase-9 leading to activation of this high-molecular weight complex. Release of cytochrome *c* is regulated by various mechanisms, which involve function of several pro- and anti-apoptotic members from Bcl-2 family proteins. Interplay within these proteins is complex and mathematical modeling of this complexity was able to dissect spatiotemporal activation of apoptosome-dependent pathway.^[@bib3]^ Importantly, there exists a direct connection between functions of DISC and apoptosome complexes, which requires BH3-only proteins of Bcl-2 family.^[@bib12]^ Accumulating evidence revealed the crosstalk not only within apoptosis, but also between various cell death modalities. Natural (mutations) or artificial (chemical inhibitors) inactivation of caspases might cause the shift from apoptosis to necroptosis, or to the mixture of these two cell death modes. Cleavage of Atg5 (essential for development of autophagy) causes shift from autophagy to apoptosis.^[@bib13]^ Depending on the type of lethal stimulus, the cell death process can be initiated in different intracellular compartments, and their crosstalk is essential for all cell death modalities. The inter-organelle crosstalk involves several molecular switches within the signaling networks.^[@bib14],\ [@bib15]^ Among proteins regulating this type of switches is p53, which, depending on localization, can be involved in regulation of apoptosis, autophagy or necroptosis. Interestingly, Bcl-2 family proteins are essential for the regulation of a majority of programmed cell death modalities. The complexity of balance between different cell death modalities is supported by the observation that nature and severity of stimulus might influence the inter-organelle crosstalk. In some cases suppression of the function of a particular intracellular compartment might switch one mode of cell death to another. For example, inhibition of mitochondrial energy metabolism can shift from apoptosis to necrosis. It seems that the point of no return in many cell death modalities is similar and associated with mitochondria. Systems biology can be applied not only to learn more about the molecular mechanisms regulating various cell death modalities, but also their crosstalk. As mentioned above, disturbances in cell death pathways result in pathogenesis of various disorders. Thus, too much cell death is associated with neurological (Parkinson\'s, Alzheimer\'s and Huntington\'s) diseases and immunological disorders (AIDS). On the other hand, too little cell death is associated with development of malignancies. Importantly, there are many pathological situations when crosstalk between different cell death subroutines has a significant role, and treatment of all of these disorders is based on the attempt to either suppress or induce cell death. Thus, suppression of autophagy is followed by the accumulation of ROS, which makes tumor cells more sensitive to apoptosis.^[@bib16]^ Pharmacological inhibition of autophagy does not prevent Atg5-dependent mitotic catastrophe, but shifts the balance to caspase-dependent death of tumor cells.^[@bib17]^ To understand the mechanism of complex crosstalks, especially, in multifactorial disorders, an integrated systems biology approach can be applied. In the following several review articles, published under the title of 'Systems Biology approach to investigate cell death pathways in diseases\' different angles of systems biology requirement for cell death research and medical applications are discussed. Thus, Lavrik^[@bib18]^ summarizes knowledge of how systems biology of death receptor networks is operating in making decision between life and death. On the other hand, Rehm *et al.*^[@bib19]^ discuss achievement, perspectives and challenges in systems medicine that can arrive from computational modeling of the intrinsic apoptosis pathway essential for analysis of chemotherapy resistance. Natural and acquired resistance of tumor cells to well-known therapies in many cases is also associated with dysregulation of apoptosis. The mechanisms of this resistance are different and complex. Cisplatin is often employed for the clinical treatment of patients with various tumors; however, in spite of high rates of clinical responses in many cases tumor cells activate an adaptive response to cisplatin that makes them less susceptible to the antiproliferative and cytotoxic effects to this drug. Recently, many systems biology studies took attempts to resolve the complex problem of cisplatin resistance. These studies are summarized to a great level of detail by Galuzzi *et al.*^[@bib20]^ who are trying to understand how obtained knowledge can be used in the development of rational approaches to tackle the clinically relevant problem of cisplatin resistance. Although major advances have been made in the diagnosis of different tumor types, it is still important to understand the precise functioning of signaling pathways associated with resistance to current treatments and develop biomarkers essential for better diagnostics and prediction of tumor response. Using lung cancer as an example, Viktorsson *et al.*^[@bib21]^ are trying to explain how systems biology-based approaches can help in generation of new biomarkers and novel therapeutic targets in this devastating disorder, or even contribution to personalized treatment of this malignancy. Addressing the question of responsiveness of tumor cells to therapy, one should keep in mind the problem of not simultaneous/uniform response. Many factors can explain this diversity of cellular reactions. Analysis of systems biology models enables to take a general view on the cell death process to identify multifactorial determinants of the cell death decision. Xia *et al.*^[@bib22]^ discuss how 'systems pharmacology\' could help in the development of combinatory treatment strategies to either manipulate the state of cancer cells in order to sensitize them to apoptosis-inducing drugs or manipulate the dynamics of signaling that encode specific cellular responses, compelling cancer cells toward apoptosis. It is known that the main goal of systems biology is to find the most rational route; therefore, a unified 'language\' (model) should be used by bioinformaticians to include all available information essential for understanding the proper functioning of the biological system. One of the aims for researchers in cell death field is to build and validate models to define proper functioning of cell death machinery and systemic understanding of cell death pathways to refine the molecular diagnosis of various types of disease associated with dysfunction of cell death. It will help in optimizing the calculation of prognostic and predictive parameters of guiding new strategies to ameliorate existing treatments and identification of novel targets for therapeutic modulation of cell death pathways. We hope that some questions related to that are addressed in the following reviews. The work in the authors group was supported by grants from the Swedish and the Stockholm Cancer Societies, the Swedish Childhood Cancer Foundation, the Swedish Research Council and Russian Research Foundation (to BZ) and the Ministry of Sciences and Economic Affairs of Saxony-Anhalt (MW-21LMS 5), BMBF (eBIO project "ImmunoQuant" -- TPU -- 0316170G), the Helmholtz-Russia Joint Research Groups -- 2008-2 and RFFI 14-04-00699 (to INL). The authors declare no conflict of interest.

1. Introduction {#sec1} =============== There are an increasing number of findings supporting the fact that cerebellum, apart from its significance in movement coordination (sensomotoric cerebellum), plays an important role in cognitive and emotional regulation (cognitive and limbic cerebellum) \[[@B1]\]. In comparison with malformations of other parts of cerebellum and the acquired lesions, congenital vermis deficit presents as pronounced slow psychomotor development and worse cognitive functioning (lower IQ) and language skills, as well as affective dysregulation \[[@B2]\]. Rhombencephalosynapsis (RS) is a rare congenital posterior fossa malformation characterised by hypogenesis or agenesis of the vermis, dorsal fusion of cerebellar hemispheres, and fusion of the dentate nuclei and superior cerebellar peduncles \[[@B3]\]. This paper presents a case of a patient diagnosed with borderline personality disorder (BPD). Brain MRI, which was performed because of severe clinical picture and neurological and cognitive deficits, showed partially developed vermis and rhombencephalosynapsis. This sheds a new light on the severity of and nonresponsiveness to medication. 2. Case Report {#sec2} ============== Unemployed, single, 26-year-old woman, with a qualification in catering industry, living with parents, was admitted for treatment to the Institute of Mental Health in Belgrade after one of her multiple suicide attempts by self-poisoning with benzodiazepines. She had been repeatedly hospitalized since she was 16, at first because of polysubstance abuse and later because of self-destructive behaviour, affective instability, and impulsiveness, and had been diagnosed with BPD. Over the past few years psychotropic medications were discontinued. However, she was unable to get a job and establish an emotional relationship. One year prior to admission to the the latest hospitalization, she started abusing benzodiazepines and alcohol again, culminating in a fall and short hospitalization for brain commotion with normal computerised tomography of brain. Afterwards, there were no signs of acute or chronic complications of craniocerebral injury nor emotional and behavioural changes regarding prior state. We acquired data that her mother had a virus infection in the first trimester of pregnancy with scanty hemorrhage. Delivery was vaginal with labour induction, and the baby girl had a pronounced physiological jaundice. Because of hypotonia, rehabilitation was performed. She was able to stand on her own at 15 months and to walk at 18 months and was not manually dexterous. Convergent strabismus was operated when she was two years old. Her attachment pattern was ambivalent. There is a positive heredity on her father\'s side: one uncle is being treated for depressive disorder and another one for bipolar affective disorder. Upon admission, the patient showed difficulty in fitting in and it was impossible to establish a therapeutic alliance. She manifested affective instability, was irritable, and verbalised suicidal ideas. During therapeutic weekends she stereotypically inflicted self-harm by cutting her forearm skin and by pressing cigarette butts in her hand dorsum. Physical examination has not shown any signs of illnesses and common laboratory test values were within reference values. Thyroid function tests were normal. The dominant finding in neurological status was truncal ataxia. Total score on the scale for the assesment and rating of ataxia \[[@B4]\] was 14 out of 40, with dominant findings on stance (5 out of 6), gait (3/8), and sitting (2/4). Convergent strabismus and bilaterally absent corneal reflexes were also registered as well as a small field of temporoparietal left-sided alopecia. Psychological examination confirmed borderline personality organization, with borderline depression and persistent suicidal risk associated with primary impulse discontrol. Intellectual capacities were at the border of low average: overall IQ was 80, verbal IQ was 84, and manual IQ was 73. The examination emphasized the fact that protrusive negativism was activated regardless of the trigger and that difficulties in attaining aims were possibly also caused by neuropsychologically founded experiential learning capacity reduction. Various combinations of mood stabilisers and antidepressants together with the second-generation antipsychotics were administered, without mood stabilization and receding of suicidal ideation. A partial response was noted to administration of fluoxetine (40 mg daily) together with olanzapine (15 mg daily). All this led us to a decision to perform neuropsychological examination, electroencephalography (EEG), and MRI. Neuropsychological examination showed disturbances in complex forms of attention with unfavourable reflection on phonemic fluency and verbal declarative memory, with the presence of retroactive inhibition and dysexecutive syndrome which affect working memory, impede orientation in initial tasks, and reduce benefits from previous experience. A reduction in categorical fluency and dysnomia was evidenced. Constructional praxis was distorted by parietal type; dynamic dyspraxia was evidenced to the left as well as graphesthesia contralaterally ([Table 1](#tab1){ref-type="table"}). The EEG was normal. The MRI has shown partially developed vermis and fusion of cerebellar hemispheres, which was a characteristic of RS ([Figure 1](#fig1){ref-type="fig"}). There were no signs of hydrocephalus nor any other central nervous system and extracentral nervous system malformations. The MRI results could explain neurological and neuropsychological findings and presence of "cerebellar cognitive affective syndrome" \[[@B5]\] as well as the persistance of mental health problems. Taking into consideration the pharmacoresistance and the high sucidal risk, clozapine (150 mg daily), clomipramine (100 mg daily), and lithium carbonate (900 mg daily) were introduced with careful titrating because of potential neurotoxicity and worsening of ataxia. This resulted in receding of suicidal ideation, partial affective stabilization, better impulse control, and abandoning of self-destructive behaviour whereupon the patient was transferred to the out-patient treatment. At the followup her condition is still characterized by affective instability and impulsiveness but without self-harming and suicidal attempts several months later. 3. Disscussion {#sec3} ============== Limbic cerebellum is represented by vermis and fastigial nucleus and cognitive cerebellum by lateral hemispheres of posterior cerebellum \[[@B1]\]. Acquired and congenital lesions of these regions lead to development of nonmotor deficits termed "cerebellar cognitive affective syndrome" (CCAS) \[[@B5]\]. It is characterized by disturbances of executive functions, visual-spatial disorganization, emotional dysregulation (blunting of affect and disinhibited and inadequate behavior), and language deficits (agrammatism and aprosodia) \[[@B5]\]. Impaired social interaction, aggressiveness, pervasive disturbance of behaviour, and self-harming are also more often noted in patients with vermis agenesis \[[@B2]\]. It should be stressed that there are studies which do not show association of acquired lesions of cerebellum with CCAS but only minor cognitive and affective changes. However, they also confirm that CCAS is a consistent finding for congenital malformations \[[@B6]\]. Rhombencephalosynapsis was first described in 1916 after the autopsy of a young man who committed suicide \[[@B7]\]. It can occur as an isolated anomaly or together with different syndromes \[[@B3]\]. Apart from hypotonia, stereotypical head movements, and strabismus, a clinical picture often includes attention disturbance, hyperactivity, and impulsiveness \[[@B7]\]. Verri et al. \[[@B8]\] have described a patient with mild mental retardation, obsessive-compulsive personality disorder, and oral self-mutilations. Nonsyndromic rhombencephalosynapsis can rarely be asymptomatic and associated with normal neuropsychological findings \[[@B7]\]. Our patient manifested trunk ataxia, and in the early development she had hypotonia and developmental dyspraxia. Pronounced affective instability, dysphoria, impulsiveness, and self-harmful behaviour, as well as the dysexecutive syndrome and decrease of working memory together with phonemic fluency reduction, match the description of CCAS as described by Schmahmann \[[@B5]\]. Bilaterally absent corneal reflexes and a small field of temporoparietal left-sided alopecia were also interesting findings, but there were not enough criteria for diagnosing Gomez-Lopez-Hernandez syndrome, which includes rhombencephalosynapsis, trigeminal anesthesia, and bilateral alopecia \[[@B9]\]. Numerous neuropsychiatric manifestations, which can appear together with cerebellar lesions, have been described and can be classified in several domains: attention disturbances, emotional control disturbances, and disturbances which belong to autistic spectrum and psychotic disorders \[[@B10]\]. These disturbances could be explained both by connections with limbic system and prefrontal cortex \[[@B1]\] and the supposed role of cerebellum in social cognition and forming a theory of mind, that is, the ability to attribute mental states to others. Positron emission tomography in healthy volunteers showed a pronounced cerebellar activation while performing tasks which require activation of brain areas in charge of theory of mind \[[@B11]\]. Taking into consideration the fact that the concept of mentalization was formed by applying the concept of the theory of mind in patients with borderline personality disorder, by adding to the objective and cognitive concept of theory of mind a subjective and affective component \[[@B12]\], we could assume that cerebellar dysfunction could play a role in the mentalization problem as well. On the other hand, fMRI studies showed that processing of negative emotions in patient with BPD was associated with greater activation within insula and posterior cingulate cortex, as well as anterior culmen and posterior declive of the cerebellum, and reduced activation of region that extended from the amygdala to the subgenual anterior cingulate cortex and dorsolateral prefrontal cortex \[[@B13]\]. Two questions remain: whether activation of vermis is primary or compensatory mechanism in BDP patients and how vermal hypoplasia can influence BPD phenotype expression. We should also take into consideration the possibility that some subradiographic cerebral hemisphere or limbic abnormalities could be present and contribute to the clinical picture in this case. The presented case shows an overlap between BPD and CCAS and suggests the importance of neurological and neuropsychological evaluation of patients with severe personality disorders. The authors thank the patient and her family for giving permission to publish this report. Conflict of Interests ===================== The authors declare that there is no conflict of interests regarding the publication of this paper. {#fig1} ###### Neuropsychological assessment. Test Patient Normative Data Deviations -------------------------------------------------- ---------------- ---------------- ------------------------------- WMS-R Verbal memory 7 7 ± 2 / WMS-R Visual memory 3 7 ± 2 −2.00 SD WMS-R Attention/Concentration Index 52 \>70 sec. \<10 Pr TMT-A 82 \<45 sec. \<10 Pr TMT-B 197 \<98 sec. \<10 Pr RAVLT *t* 26 55.3 ± 6.6 −4.44 SD RAVLT *e* 2 14.0 ± 2.0 −6.00 SD RAVLT *r* 12 14.4 ± 0.8 −3.00 SD Phonemic fluency tests for divergent thinking S/8, K/11, L/7 Min. 8 / Categorical fluency tests for divergent thinking 11 19.58 ± 4.0 −2.15 SD RCF C 27 35.1 ± 1.5 −5.40 SD RCF 40′ 4.0 22.7 ± 7.0 −2.67 SD HVOT 21.5/56--60 41--55 Low possibility of impairment WCST CA 0 5.6 ± 1.0 −5.60 SD WCST PR 30 13 ± 9.1 +1.86 SD WCST FMS 0 0.8 ± 1.3 N/A BNT 50 55.86 ± 2.86 −2.04 SD BDAE auditory comprehension 11 11.2 ± 1.1 −0.18 SD BDAE total sentence repetition LP 7/8 7.7 ± 0.6 −1.17 SD Ideomotor praxia 8/8 7/8   Spatial aspects of praxia 10/10 8/10   Dynamic praxia D 8 8--10 Left side impairment L 5 (3 errors) Tactile gnosia D 3/3 2/3 L 3/3 Graphesthesia D 1/5 4/5 Right side impairment L 5/5 VITI-IQ        VIQ 84 99.86 ± 14.98 −1.05 SD  PIQ 73 99.37 ± 14.66 −1.80 SD  FSIQ 80 99.19 ± 15.23 −1.26 SD Wechsler Memory Scale-Revised (WMS-R**)**; Trail Making Test A and B (TMT-A and -B); Rey Auditive Verbal Learning Test (RAVLT): RAVLT *t*: total number of repeated words in five attempts in the RAVLT, RAVLT *e*: number of repeated words after 30 min (evocation) in the RAVLT and RAVLT *r*: number of correctly recognized words; (recognition) in the RAVLT; Rey-Osterrieth Complex Figure Test (RCF): RCF C: copying of the RCF and RCF 40′: 40-minute delayed recall trial; Hooper Visual Organization Test (HVOT); Wisconsin Card Sorting Test (WCST): WCST CA: categories achieved in the WCST, WCST PR: perseverative responses in the WCST and WCST FMS: failures to maintain set in the WCST; Boston Naming Test (BNT); Boston Diagnostic Aphasia Battery (BDAE); Serbian version of Wechsler Adult Intelligence Scale (WAIS)---"Vekslerov Individualni Test Inteligencije" (VITI): verbal IQ (VIQ), performance IQ (PIQ) and full scale IQ (FSIQ). [^1]: Academic Editor: Marie-Cécile Nassogne

1 INTRODUCTION {#SEC1} ============== Large RNA molecules perform diverse functions within cells and have complex 3D structures. For example, the 3D structures of RNA enzymes (ribozymes) allow them to catalyze RNA cleavage (Guerrier-Takada *et al*., [@B11]; Kruger *et al*., [@B13]; Stark *et al*., [@B28]). Other structured functional RNA molecules include riboswitches (Nahvi *et al*., [@B22]) and ribosomal RNA (Ban *et al*., [@B1]; Yusupov *et al*., [@B33]). Although knowing the 3D structures of these molecules is critical in understanding their functions, the protein databank contains few high-resolution RNA crystal structures. For these reasons, computational modeling of RNA 3D structures is an important goal. Both manual and automated methods for building full atomic RNA structures have had success but continue to be challenged by significant limitations. Manual approaches have successfully modeled a number of molecules, including several group I introns (Lehnert *et al*., [@B14]; Michel and Westhof, [@B21]) and the yeast phenylalanine tRNA (Levitt, [@B15]), but require expert knowledge of RNA structure. Semi-automated methods, such as MANIP (Massire and Westhof, [@B20]) and ERNA-3D (Tanaka *et al*., [@B30]) use a fragment-based approach to build full atomic 3D RNA models. However, these methods are limited by the need for significant user interaction as well as expert knowledge. Several automated tools model RNA structure in 3D. The MC-Fold/MC-Sym pipeline uses constraint satisfaction algorithms to build full atomic models of structures (Major *et al*., [@B17]; Parisien and Major, [@B23]). FARNA (Das and Baker, [@B4]) is an automated fragment-based tool that has been used to model molecules as large as 158 nt when combined with multiplexed hydroxyl radical cleavage analysis (MOHCA; Das *et al*., [@B5]). Coarse-grain approaches to modeling RNA 3D structures are fully-automated, can model very large structures (\>160 nt) and can be used in multiscale approaches for modeling large systems. These methods include YAMMP (Malhotra *et al*., [@B18]), YUP (Tan *et al*., [@B31]), DMD (Ding *et al*., [@B8]), RNA2D3D (Martinez *et al*., [@B19]) and NAST (Jonikas *et al*., [@B12]). Although coarse-grain topological models of RNA molecules provide significantly more structural information than secondary structure alone, full atomic models are preferable for studying structure function relationships, and are a prerequisite for most energy-based dynamics and refinement methods. As a result, several of these methods include tool-specific protocols for adding atomic resolution to their coarse-grain models. For example, DMD and its related web-based tool iFoldRNA (Sharma *et al*., [@B25]) use a coarse-grained approach to generate predictions for structures\<50 nt in size, which are then refined to atomic resolution by an unpublished reconstruction protocol. RNA2D3D starts with a nucleotide-resolution first approximation of the molecule 3D structure based on secondary structure information, and adds atomic resolution early in its protocol, using nucleotide geometries observed in crystal structures. Neither of these tools can be used on independently generated coarse-grain structures. In some cases, users of these tools develop their own application-specific approach to adding atomic detail. A recently reported all-atomic model of Pariacoto virus included coarse-grained modeling of RNA followed by addition of atomic detail (Devkota *et al*., [@B7]). The authors extended a crystal structure that contained 35% of the RNA structure by initially modeling the RNA with YAMMP, which uses a coarse-grained one-point-per-residue representation centered on phosphate atoms. To add full atomic detail to the coarse-grain structure, the authors used a fragment-based approach to match existing full atomic structures in the PDB to both base-paired and non-base-paired regions. The authors searched for compatible pieces to generate plausible structures, which they minimized and annealed. Such multiscale approaches are becoming increasingly promising, but are limited by the need for each research group to develop their own protocols for adding atomic detail to coarse-grain models. Several methods in both RNA and protein structure prediction use knowledge-based fragment approaches, most notably FARNA and Rosetta (Simons *et al*., [@B26]), these methods search for similar fragments at the sequence level to inform the structure prediction. Our approach to instantiating full atomic detail is inspired by these methods and based on the observation that fragments in RNA molecules that are similar at the coarse-grain level are often also similar at the full atomic level. Therefore, if a ribosomal RNA fragment is similar to a model fragment at the coarse-grain level, the full atomic detail from the ribosomal crystal structure may contain useful geometric information about how full atomic detail should be instantiated in the coarse-grain model. None of the methods for adding full atomic detail to coarse-grain structures mentioned above are (i) generalized for many types of independently generated coarse-grain structures, (ii) validated on a range of structures sizes and (iii) publicly available. We present a fully automated fragment- and knowledge-based method, called Coarse to Atomic (C2A) for instantiating full atomic detail into coarse-grain structures of RNA molecules. We evaluate the full atomic structures generated by our method, and make the method freely available (along with a manual). Our method can use any atom-based coarse-grain structure template as input, and provides a full atomic, GROMACS energy-minimized structure as output. We use geometric information from RNA crystal structures to reconstruct full atomic detail. We have tested our method on both ideal coarse-grain structures for molecules ranging in size from 70 to 244 residues, as well as on coarse-grain one-point-per-residue models of tRNA generated by NAST. In addition to improving the usefulness of coarse-grain modeling methods, this tool has the potential to bridge the computationally expensive but precise full atomic modeling approach and the fast but detail-poor coarse-grain modeling approach. 2 METHODS {#SEC2} ========= In brief, C2A searches within a reference full atomic structure for coarse-grain matches to fragments of a target molecule, and combines the full atomic versions of the matches to generate a full atomic structure. We then perform a minimization step to reduce any collisions and gaps. We need the following three inputs to perform this method: A coarse-grain template for the target molecule (e.g. a structure with one point per residues representing the C3′ or P atom) ([Fig. 1](#F1){ref-type="fig"}A).A fragment definition for the target molecule (e.g. the secondary structure) ([Fig. 1](#F1){ref-type="fig"}B).A reference full atomic RNA 3D structure database (e.g. ribosomal RNA structures) ([Fig. 1](#F1){ref-type="fig"}C). Fig. 1.C2A method overview. The C2A method uses three pieces of information as input: (**A**) a coarse-grain template for the target molecule; (**B**) a fragment definition for the target molecule, such as the secondary structure; and (**C**) a reference database of full atomic RNA structures. (**D**) In Step 1, the coarse-grain template is divided into structural subsets based on the fragment definition, here we show as an example a double- and a single-stranded fragment. (**E**) Step 2 searches for coarse-grain matches to each fragment in the reference structure and extracts the full atomic detail of each match, we show here the full atomic detail for two matches found in ribosomal RNA for each of the two example fragments. (**F**) Step 3 searches for combinations of matches free of major collisions, we show here how matches for the two example fragments align to the coarse-grain template. (**G**) Finally, Step 4 minimizes the resulting full atomic structure to remove unrealistic gaps and collisions. Based on this input, we generate a full atomic structure by following these steps: Use the fragment definition to break the target molecule into structural subsets (e.g. helices, loops and junctions) ([Fig. 1](#F1){ref-type="fig"}D).Find coarse-grain matches for each fragment in a reference full atomic RNA structure ([Fig. 1](#F1){ref-type="fig"}E).Combine matches to generate a full atomic structure free of major atomic collisions ([Fig. 1](#F1){ref-type="fig"}F).Minimize the structure to eliminate any chemically unrealistic gaps or collisions ([Fig. 1](#F1){ref-type="fig"}G). 2.1 Defining structure fragments {#SEC2.1} -------------------------------- For our method, we define substructures of the target molecule, which we call fragments. Although we use the secondary structure of a target molecule to define these fragments in this article, other fragment definitions that use both single- and double-stranded regions can be easily implemented as well. The fragment definition is provided by the user, and C2A parses the definition to determine the substructures. We use the secondary structure to define two types of fragments within the target molecule: single and double stranded. Helices are double-stranded base-pairing regions and may include bulges of any length. Regions between helices are single stranded and overlap by one residue in the primary sequence with adjacent helices. Fragments may not contain fewer than four residues, and more than one residue of overlap is allowed in cases where there would otherwise be fewer than four residues in a fragment. The result is a ensemble of structural coarse-grain subsets, either double or single stranded, of the target molecule (we show examples in [Fig. 1](#F1){ref-type="fig"}D). 2.2 Searching for fragment matches {#SEC2.2} ---------------------------------- We search within the reference full atomic database of structures for coarse-grain matches to the fragments we have defined ([Fig. 1](#F1){ref-type="fig"}E). We then use the full atomic detail of the matches to build complete full atomic structures. In this article, we use the *Thermus thermophilus* 16S ribosomal RNA solved at 3.00 Å resolution (PDB ID 1N32 chain A) as the primary database of full atomic structure. This database does not contain any structures upon which we tested the method. In our search for coarse-grain matches, we only consider the level of coarse-graining that is represented in the template structure. Thus, for NAST structures, we use template structures where each nucleotide is represented by the C3′ atom position. When searching for matches, we will only consider the C3′ positions of each residue within the reference database. However, our method allows other coarse-graining schemes such as residues represented with the position of the P atom or by more than one atom. Since using P atoms is a more common coarse-grained representation, we have provided an example that uses a P-centered coarse-grain scheme in the examples directory of the download associated with this article. For single-stranded fragments such as loops, junctions and ends, we determine the length of the fragment (in number of residues) and the distance between the defining points of the first and last residues in the template structure (in Angstroms). We search through the reference database for continuous runs of residues of the same length as the fragment. Of these, we keep the ones that have distances between the first and last residue defining atoms (in our case C3′) within a certain user-specified error (typically 10%) of the distance observed in the template fragment. This results in an ensemble of matches within the full atomic structure to the coarse-grain template fragment. We then calculate the coarse-grain RMSD (Root Mean Square Deviation) of each potential match to the coarse-grain template fragment and rank the matches. We keep the 10 best matches by RMSD for each fragment as part of the working set for assembling a full atomic structure. For double-stranded fragments such as helices, we first find all the possible matches for each single-stranded element individually. From each possible combination of matches, we keep those that are within a certain error (we use 10%) of the distances observed in the template fragment between the 5′ and 3′ ends of each fragment. As with the single-stranded fragments, we calculate the coarse-grain RMSD to the coarse-grain template fragment, rank the matches and keep the 10 best for the working set. 2.3 Processing fragment matches {#SEC2.3} ------------------------------- Since we do not consider the residue sequence in our search for matches, many of the matches will have incorrect base atoms. To remedy this, we replace the incorrect nucleoside base atoms with the correct ones while maintaining the orientation of the base plane. We keep the sugar group and phosphate group atoms in the same positions, and replace the base atoms with the correct geometry. We use three atoms to define the plane and orientation of the original base (N9, N1 and C5 for Adenine and Guanine, and N1, N3 and C5 for Cytosine and Uracil). We insert the correct base atoms into the appropriate plane and orientation, resulting in coordinates for the correct residue. This results in fragment matches with the correct nucleotide sequence while maintaining the geometry of the match found in the reference structure. 2.4 Assembling full atomic structures {#SEC2.4} ------------------------------------- We use a Metropolis Monte Carlo approach to assemble a full atomic structure using the coarse-grain template and a working set of best matches for each fragment. We start by randomly selecting one match for each fragment from the working set and aligning the matches to the coarse-grain template. Since we defined single-stranded fragments to overlap by one residue with double-stranded fragments, each of these junctions will have two sets of coordinates for that residue. In these cases, we use only the coordinates from the double-stranded fragment. In each step, we search for pairs of atoms within the full atomic structure that are closer than a cutoff distance (user-specified, 1.0 Å here) and check the fragments containing these pairs for collisions. If we observe any collisions, we randomly select one of the colliding fragments to be replaced with another randomly selected match. We accept the replacement with the probability shown in Equation ([1](#M1){ref-type="disp-formula"}), where *N*~old~ is the number of collisions in the old state and *N*~new~ is the number of collisions in the new state. We continue this attempt for a user-specified number of steps, or until we find a combination of matches with no collisions, whichever occurs first. If no such combination is found within a user-defined number of steps and attempts, the output will be the full atomic structure with least number of collisions observed over the course of this search. In this article, we make five attempts of 500 steps each to generate a full atomic structure free of major collisions. 2.5 Minimizing full atomic structures {#SEC2.5} ------------------------------------- The structures resulting from our assembly protocol are likely to contain both gaps (unrealistically long covalent bonds) and collisions (unrealistically short distances between any atoms), and which may prevent the structures from being used in full atomic studies. In particular, the regions at junctions between fragments may contain significant gaps between atoms that should be bonded because the coordinates for those atoms came from different matches. To remedy these two structural issues, we minimize the full atomic structures with GROMACS using steepest descent minimization. We include the scripts we used for running GROMACS, including parameter files, on our project\'s web site. 2.6 Evaluating full atomic structures {#SEC2.6} ------------------------------------- We evaluated the effect of minimization on junction bonds by calculating the lengths of covalent bonds between fragments both before and after minimization. We compared these values with those of non-junction bonds. We also calculated the lengths of these bonds observed in the relevant crystal structures. We submitted our minimized structures to the MolProbity tool, which calculates a Clashscore and assigns a percentile to the structure (Davis *et al*., [@B6]; Lovell *et al*., [@B16]). We also submitted our structures to the RCSB ADIT structure validation tool which outputs a full report on the structure including RMSD values for covalent bonds and angles relative to standard values for nucleotides (Clowney *et al*., [@B3]; Gelbin *et al*., [@B9]). We made the full reports available on our project web site under the documents section. For all minimized full atomic structures, we calculated the RMSD value relative to the known crystal structure. We also calculated RMSD values for helical (double-stranded) and non-helical (single-stranded) fragments separately. We also applied the recently developed metric of interaction network fidelity (INF) to all of our structures, which considers base--base-stacking and base--base-pairing interactions (Parisien *et al*., [@B24]). We calculated the INF for pairings alone, as well as pairings and stackings combined, which is a much stricter measure. INF values range from 0.00 (worst) to 1.00 (best). 2.7 Validation using ideal backbones from crystal structures {#SEC2.7} ------------------------------------------------------------ To validate our approach, we applied it to seven crystal structures of RNA molecules ranging in size from 70 to 244 residues: Yeast ai5g group II intron ('Ai5gamma', PDB ID 1KXK) 70 residues (Zhang and Doudna, [@B34]).Yeast phenylalanine tRNA ('tRNA', PDB ID 6TNA) 76 residues (Sussman *et al*., [@B29]).Aminoacyl-tRNA synthetase ribozyme ('flexizyme', PDB ID 3CUL) 92 residues (Xiao *et al*., [@B32]).P4-P6 RNA ribozyme domain ('P4-P6', PDB ID 1GID) 158 residues (Cate *et al*., [@B2]).Azoarcus group I intron ('Azoarcus', PDB ID 1ZZN) 195 residues (Stahley and Strobel, [@B27]).Twort ribozyme ('Twort', PDB ID 1Y0Q) 244 residues (Golden *et al*., [@B10]). We stripped each crystal structure of atomic detail, leaving only the position of the C3′ atom for each residue. We defined fragments using the known secondary structure of each molecule and the rules described above. The database did not contain any of these structures. We searched for matches to each fragment and created a working set by keeping the 10 best matches by RMSD at the coarse-grain level. We searched for 10 combinations of matches free of major collisions and minimized the full atomic structures. For the two structures we modeled that were missing the residues (the *Twort* and *Azoarcus* ribozymes), we used the complete coarse-grain structures generated using the NAST tool by Jonikas *et al*. as the template input. 2.8 Building full atomic models of NAST tRNA topologies {#SEC2.8} ------------------------------------------------------- We applied the C2A method to the problem of building full atomic structures based on coarse-grain models built by the tool NAST. For each of the three topologies generated by NAST, we used five models as template starting structures and generated 10 full atomic structures which we then minimized. 3 RESULTS {#SEC3} ========= 3.1 Recovering full atomic detail from ideal coarse-grain templates {#SEC3.1} ------------------------------------------------------------------- The full atomic RMSD values of full atomic models built from coarse-grain crystal structure templates ranged from 1.87 to 3.31 Å, we give the average values in [Table 1](#T1){ref-type="table"}. The average RMSD for the 60 full atomic structures we generated from ideal backbones was 2.75 ± 0.37 Å. We report INF scores for all structures in [Table 2](#T2){ref-type="table"}. We calculated junction bond-distances before and after minimization and report the values in [Table 3](#T3){ref-type="table"}. We report MolProbity Clashschore and percentiles, as well as RMSD values for covalent bonds and angles as evaluated by the RCSB ADIT tool in [Table 4](#T4){ref-type="table"}. We compare these values for our models to those calculated on the relevant crystal structures. We show a sample full atomic tRNA molecule built from the ideal backbone template in [Figure 2](#F2){ref-type="fig"}B and show the full atomic crystal structure for comparison ([Fig. 2](#F2){ref-type="fig"}A). The full atomic models we generated for the *Azoarcus* and *Twort* molecules contain full atomic detail for the loop regions that are missing in the crystal structure. We have posted these structures and all code necessary to reproduce these examples on our project web site. Fig. 2.Images of C2A full atomic modeling results. We show both a full atomic stick and a cartoon representation for each of the following tRNA structures: the full atomic crystal structure (PDB ID 6TNA) (**A**), a full atomic model generated by C2A using the ideal backbone from the crystal structure (**B**), a full atomic model based on the NAST tRNA model A (**C**), model B (**D**), and model C (**E**). Table 1.RMSD of minimized full atomic structuresMoleculeOverallHelicalNon-helicalIDRMSD (Å)fragmentsfragmentsRMSD (Å)RMSD (Å)Validation structures (10 models)1KXK2.13 ± 0.211.36 ± 0.442.41 ± 1.396TNA2.81 ± 0.111.10 ± 0.093.02 ± 1.273CUL3.06 ± 0.181.77 ± 0.822.48 ± 1.901GID3.16 ± 0.082.03 ± 1.222.75 ± 1.451ZZN2.79 ± 0.162.10 ± 0.742.56 ± 1.421Y0Q2.76 ± 0.071.92 ± 0.682.50 ± 1.54NAST tRNA models (50 models)Model A8.39 ± 0.272.65 ± 0.714.13 ± 1.91Model B13.30 ± 0.313.08 ± 1.283.82 ± 1.51Model C15.99 ± 0.763.06 ± 1.134.19 ± 1.95[^2] Table 2.Averages and ranges for INF for base pairs alone and base pairs with stackingMolecule IDINF on pairings and stackingINF on pairings aloneAverageRangeAverageRange1KXK0.51 ± 0.100.44 − 0.780.83 ± 0.030.77 − 0.896TNA0.69 ± 0.010.66 − 0.710.69 ± 0.040.59 − 0.733CUL0.67 ± 0.020.64 − 0.700.65 ± 0.040.57 − 0.691GID0.61 ± 0.030.56 − 0.640.51 ± 0.030.46 − 0.551ZZN0.58 ± 0.020.55 − 0.600.67 ± 0.020.64 − 0.701Y0Q0.61 ± 0.010.58 − 0.630.54 ± 0.020.52 − 0.576TNA-A0.46 ± 0.040.38 − 0.550.35 ± 0.070.18 − 0.516TNA-B0.41 ± 0.050.31 − 0.510.27 ± 0.090.07 − 0.466TNA-C0.41 ± 0.040.34 − 0.490.26 ± 0.080.11 − 0.45[^3] Table 3.Covalent bond lengths before and after minimization for junctions and non-junctionsMolecule IDNon-junction bonds (Å)Junction bonds (Å)RangeAverageRangeAverage1KXKCrystal1.58 − 1.611.60 ± 1.601.59 − 1.631.60 ± 0.01Pre-min1.59 − 1.631.61 ± 1.611.12 − 6.243.29 ± 1.22Post-min1.54 − 1.651.59 ± 1.591.56 − 1.751.64 ± 0.046TNACrystal1.54 − 1.641.60 ± 1.601.58 − 1.621.60 ± 0.01Pre-min1.59 − 1.621.60 ± 1.601.01 − 6.492.49 ± 1.21Post-min1.53 − 1.651.59 ± 1.591.51 − 1.781.62 ± 0.053CULCrystal1.48 − 1.621.60 ± 1.601.59 − 1.621.61 ± 0.01Pre-min1.59 − 1.621.61 ± 1.611.14 − 8.264.07 ± 1.69Post-min1.55 − 1.651.59 ± 1.591.56 − 1.861.67 ± 0.071GIDCrystal1.57 − 1.641.60 ± 1.601.58 − 1.621.60 ± 0.01Pre-min1.59 − 1.641.61 ± 1.611.04 − 12.723.82 ± 2.50Post-min1.49 − 1.701.59 ± 1.591.50 − 2.901.67 ± 0.131ZZNCrystal1.57 − 1.631.61 ± 1.611.57 − 1.641.61 ± 0.01Pre-min1.59 − 1.631.60 ± 1.601.00 − 13.363.64 ± 1.67Post-min1.42 − 2.091.59 ± 1.591.44 − 2.471.67 ± 0.111Y0QCrystal1.58 − 1.651.60 ± 1.601.57 − 1.641.60 ± 0.01Pre-min1.59 − 1.631.61 ± 1.610.55 − 10.963.53 ± 1.88Post-min1.46 − 1.871.59 ± 1.591.48 − 2.131.66 ± 0.07[^4] Table 4.Evaluation of minimized full atomic structures with MolProbity and the RCSB ADIT toolMolecule IDMolProbityRCSB ADITClashscore (goal=0)Percentile (*N*=1784)Covalent bondsCovalent anglesCrystal structureBest modelAverage (10 models)Crystal structureBest modelAverage (10 models)RMSD (Å)RMSD (degrees)CrystalModelsCrystalModels1KXK14.033.338.54 ± 5.54549778.30 ± 22.530.0060.0170.92.86TNA26.464.929.11 ± 4.54199476.60 ± 19.100.0790.0173.42.73CUL7.676.118.56 ± 2.30839079.10 ± 10.810.0070.0201.03.21GID35.7813.3123.21 ± 14.67105735.80 ± 18.360.0090.0310.93.41ZZN32.6715.2929.46 ± 9.82134821.11 ± 16.000.0080.0321.13.61Y0Q40.4715.940.82 ± 17.7284613.60 ± 15.030.0070.0361.03.8[^5] 3.2 Building full atomic models from coarse-grain NAST models of tRNA {#SEC3.2} --------------------------------------------------------------------- We show sample full atomic models built from the three NAST topology predictions of tRNA in [Figure 2](#F2){ref-type="fig"}C--E. The full atomic structures that we generated had full atomic RMSD values within 1 Å of the coarse-grain RMSD of their templates. We report RMSD values to the tRNA crystal structure in [Table 1](#T1){ref-type="table"} along with separate values for helical versus non-helical fragments. We also report INF scores for pairings alone, as well as pairings and stackings in [Table 2](#T2){ref-type="table"}. We report MolProbity Clashscores and RCBS ADIT RMSD values for covalent bonds and angles in [Table 5](#T5){ref-type="table"}. The fragment combination search step succeeded in 50, 24 and 20 of the 50 attempts for each model A, B, and C, respectively. For model B, two of the five templates succeeded in 10 of the 10 combination search attempts, one succeeded in 4 of the 10 attempts, and two succeeded in 0 of the 10 attempts. For model C, two templates succeeded in 10 of the 10 attempts and three succeeded in 0 of the 10 attempts. Table 5.MolProbity Clashscores for full atomic tRNA models built from NAST coarse-grained modelsMolecule IDMolProbityRCSB ADITClashscore (goal = 0)Percentile (*N* = 1784, 0--9999 Å )Covalent bondsCovalent anglesBest modelAverageBest modelAverageRMSD (Å)RMSD (degrees)Model A: all structures1.236.93 ± 5.039984.32 ± 16.670.0223.2Structures A-03.087.14 ± 3.369884.2 ± 14.540.0213.2Structures A-11.239.29 ± 8.389978.1 ± 24.590.0243.4Structures A-25.548.80 ± 2.519277.4 ± 11.890.0223.3Structures A-31.236.52 ± 4.349984.6 ± 16.320.0223.2Structures A-41.232.89 ± 1.569997.3 ± 2.500.0203.1Model B: all structures7.3837.77 ± 26.968525.16 ± 24.340.0414.0Structures B-07.3813.29 ± 4.188557.7 ± 18.660.0223.3Structures B-113.5440.41 ± 23.725618.78 ± 20.090.0394.0Structures B-224.0062.83 ± 34.44238.2 ± 10.530.0504.5Structures B-314.7737.17 ± 17.145017.4 ± 17.880.0463.9Structures B-414.7735.39 ± 22.445023.1 ± 20.360.0444.0Model C: all structures3.0819.04 ± 13.789848.42 ± 28.220.0293.6Structures C-09.2318.34 ± 9.967546.4 ± 24.170.0333.7Structures C-115.3822.65 ± 4.364827.8 ± 10.410.0233.5Structures C-23.0812.19 ± 7.569864.5 ± 29.390.0253.4Structures C-36.1511.57 ± 4.399065.8 ± 19.440.0263.6Structures C-45.5430.46 ± 23.769237.6 ± 34.250.0353.8[^6] 3.3 Computational resources {#SEC3.3} --------------------------- We used a 2.2 GHz CPU to run the C2A method. We analyzed the CPU load for the two parts of C2A code: creating a working library of fragment matches (which only needs to be performed once per template structure) and combining fragments into full atomic structures (which we performed 10 times for each template structures). There are two contributing factors to the time needed to generate the working library of fragment matches: the number and types of fragments, and the size of the reference full atomic database. It took 95 s to load the reference database \[*Thermus thermophilus* 16S ribosomal RNA solved at 3.00 Å resolution (PDB ID 1N32 chain A)\]. It took between 3 s and 20 s to find matches for single-stranded fragments and between 40 s and 440 s to find matches for double-stranded fragments. In total, the first part of C2A took 1226, 1550, 1034, 1297, 1836, 2741 and 3847 s to find matches to all fragments for the Ai5gamma, tRNA, flexizyme, SPR19, P4-P6, *Azoarcus* and *Twort* molecules, respectively. When using ideal backbones as the input for the C2A method, finding combinations of matches free of major collisions took an average of 47 ± 30 s, with the maximum being 117 s. For NAST tRNA models A, B and C, finding matches took average times of 1970, 1956 and 2206 s, respectively. The code and examples we have made available are python scripts that have been tested on both unix and Mac OSX platforms. 4 DISCUSSION {#SEC4} ============ We have presented a method for instantiating full atomic detail in coarse-grain RNA structure backbones using geometry knowledge from a reference database of one or more full atomic RNA crystal structures, in this case the *Thermus thermophilus* 16S ribosome. We assume that if a fragment matches well to pieces of known RNA structure at the coarse-grain level, then the full atomic geometry of the match is a good predictor of the fragment\'s full atomic detail. In this article, we used only data from one ribosomal RNA as a reference structure, through which we search for matches to coarse-grain templates. We did not include any geometric data from our template structures in the reference full atomic structure. To validate our method, we applied it to ideal backbones taken from the crystal structures of seven RNA molecules. For each structure, we kept only the C3′ position of each residue, creating ideal backbone structures with one point per residue and recovered the full atomic detail at an average resolution \<3.0 Å. We were also able to generate full atomic structures of two large RNA molecules that were missing residues in the crystal structure solution (the *Azoarcus* and *Twort* ribozymes) by using the NAST generated complete coarse-grain templates. We were able to minimize the full atomic structures with GROMACS, and remove significant clashes and gaps. The resulting minimized full-atomic structures compare well with the known crystal structures in terms of clashes and gaps. Users with expertise in structure minimization and refinement can use these structures as starting points for further studies. For the two ribozymes that are missing loops in the crystal structure (*Azoracus* and *Twort*), we used the coarse-grain loops modeled by NAST as part of the input template and modeled full atomic detail for the entire molecule, resulting in complete full atomic structures. These two examples show that the combination of NAST and C2A can be used to complete crystal structures that are missing pieces of the molecule. However, the quality of the modeling of the missing pieces depends entirely on the modeling choices by the user when building the coarse-grain template. In our case, we used the geometry of existing loops to build coarse-grain models of the missing loops. We then applied our method to fill in atomic resolution in NAST coarse-grain models of the tRNA molecule. For each of the three coarse-grain topology models generated by NAST, we used five representative structures and applied our method to generate full atomic versions of the models. As an aside, we noticed that the amount of time necessary to find a combination of fragments free of major collisions is an indicator of the quality of the topology. Additionally, all five of the templates used for model A successfully generated full atomic structures, while only three and two templates for models B and C, respectively, were able to do so. We were able to minimize the full atomic structures we built using GROMACS which corrected any gaps or collisions. Until this last minimization step, the method is entirely geometric and does not consider any chemical opportunities such as hydrogen bonds. Through this example, we show that coarse-grain topology models generated from limited structural information (in this case, only primary, secondary and limited tertiary structure data) can be used as starting points for building full atomic structures. These full atomic structures can be used as starting points for further refinement using physics-based approaches, or more detailed knowledge-based approaches. The full atomic structures we generated for tRNA, both from the crystal structure template and the NAST model template, did not recover any tertiary contact base pairings with adequate detail. Recovering these interactions is an important goal of model RNA 3D structure and remains a significant challenge. Our need to avoid serious clashes during the assembly step selects against the close interactions needed to recover these non-helical base pairings. However, once a full-atomic structure is built from a coarse-grain template, knowledge of tertiary interactions and finer-grained dynamics tools can be used to recover these interactions. Preliminary results of using the RNAbuilder tool (S.C.Flores *et al*., submitted for publication) to constrain known tertiary contacts in our full atomic models show improved INF scores. RNAbuilder uses a full atomic model as a rigid template onto which it threads another full atomic molecule which attempts to satisfy both the geometry of the template and the specified tertiary contacts. This method maintains most of the original geometry which additionally constrains tertiary contacts that were not previously present. The C2A method for instantiating full atomic detail into coarse-grain models is limited by the quality of the template structure and information in the reference full atomic structure. Although the process of instantiating atomic detail does not decrease the quality of the model, it does not improve it either. Our method is also limited by the geometric information contained in the chosen reference database. C2A cannot predict entirely new fragment geometries, and assumes that known structures are reasonable sources for data for the structures we are attempting to predict. Although we used only one ribosomal RNA structure, users of C2A can include different or more crystal structures as reference full atomic structural information. Full atomic structures generated by C2A inherit the limitations and problems associated with their coarse-grain templates. Additionally, there is no guarantee of convergence on a combination of full atomic fragments free of major collisions. However, we noticed a trend that a lack of convergence tended to correlate with a coarse-grain template model of poor quality. Our code works with any atom-based coarse-graining scheme. The results we presented in the article are for coarse-grain template structures with a one-point-per-residues representation centered on the C3′ atom. We have also tested our code on P, P-C3′ and P-C2−C3′ coarse-grain representations (not presented in this article). We observe that the more points representing a residue, the better the resulting full atomic structures. The method we have proposed connects the coarse-grain and full atomic approaches for modeling RNA 3D structures. Coarse-grain approaches are generally faster and able to handle large molecules but lack in accuracy and detail. Full atomic approaches are limited computationally, but provide accurate and detailed results. Being able to move back and forth between these two approaches has not been possible because no published, validated and publicly available method existed for building full atomic detail into coarse-grain models. Our method will allow models generated by coarse-grain methods to be refined using full atomic tools. Using full atomic models in coarse-grain methods simply requires removal of full atomic detail to the desired coarse-grain level. We make all code and examples from this article, along with instructions, available for free on our project web site: [www.simtk.org/home/c2a](www.simtk.org/home/c2a) under the Documents link. Detailed instructions for running C2A are including in the NAST/C2A manual which is available at the same address. Supplementary Material ====================== ###### \[Supplementary Data\] We thank Alain Laederach, Daniel Herschlag, Rhiju Das and Samuel Flores for helpful discussions. We are grateful to Christopher Bruns for developing the GROMACS interface Zephyr (available free at <https://simtk.org/home/zephyr>). We also thank Marc Parisien for providing us with code to calculate the INF scores of our structures. *Funding:* NIH Roadmap for Medical Research (grant U54 GM072970); National Institutes of Health (grant P01-GM66275). National Library of Medicine Training (grant LM-07033 to M.A.J.). NIH Biotechnology Training (grant 5 T32GM008412-15 to M.A.J.). *Conflict of Interest:* none declared. [^1]: Associate Editor: Ivo Hofacker [^2]: We report both overall RMSD values and separate values for helical and non-helical fragments. [^3]: INF scores range from 0.0 (worst) to 1.0 (best). [^4]: We report ranges as well as average values. We also report the bond lengths in the relevant crystal structures for the same bonds. [^5]: We give MolProbity Clashscores and percentiles for validation structures. Clashscore is the number of serious steric overlaps (\>0.4 Å) per 1000 atoms. ADIT reports the RMSD values for covalent bonds and angles calculated relative to standard values for nucleotide units. [^6]: For each of three groups of coarse-grained models (A--C), we selected five models as templates for building 10 full atomic structures. We list statistics for each of the three groups, as well as for each coarse-grain template we used.

Introduction {#sec0001} ============ Congenital pulmonary venolobar syndrome is a type of partial anomalous pulmonary venous return in which the affected lung is hypoplastic and is drained by an anomalous vein into the systemic venous system [@bib0001]. Pulmonary sequestration is the aberrant formation of lung tissue that has no connection to the bronchial tree or pulmonary arteries. A horseshoe lung represents a band of pulmonary parenchyma extending between the right and left lungs [@bib0002]. Partial pulmonary venous has an incidence of approximately 1-3 in 100,000 live births with pulmonary sequestration manifesting in approximately 1 in 1000 of live births. Horseshoe lung is also uncommon with an unknown incidence. Presented here is a situation in which all 3 conditions manifest in concert [@bib0003]. Case report {#sec0002} =========== A term female infant was born at full term and discharged a day following delivery. Pertinent medical history included an abnormal fetal magnetic resonance imaging, though the magnetic resonance imaging was performed at an outside facility and not available for review. An antenatal echocardiogram suggested a partial anomalous pulmonary venous return, moderate perimembranous ventricular septal defect, and possible major aortopulmonary collateral vessels. A routine well-child check-up shortly after discharge revealed that the patient had regained her birth weight and had been voiding and stooling normally. At 6 days of life, the infant presented to cardiology clinic due to tachypnea. The tachypnea was initially only observed during feedings in the first few days of life, but it worsened and became apparent more regular. At the time of presentation to the cardiology clinic, there was no associated cyanosis or diaphoresis. There were intermittent suprasternal and subcostal retractions. Upon presentation to the cardiology clinic, the patient\'s vital signs were 2.76 kg, height 49 cm, blood pressure 92/63 mm Hg, heart rate 158 bpm, and respiratory rate 100, which decreased to the 50s transiently during the visit. Oxygen saturation was 97%. A chest radiograph was performed that showed a hypoplastic right lung with mediastinal shift to the right and pulmonary vascular congestion of the left lung ([Fig. 1](#fig0001){ref-type="fig"}). No pleural effusion was seen.Fig. 1Chest radiograph. There is right mediastinal shift with diffuse opacity throughout the right hemithorax (black arrow) and an abnormal vascular structure in the lower right lung (white arrow).Fig 1 The patient was treated with furosemide and followed up 2 days later at which time her breathing had improved. She showed improved coordination with breast feeding and did not demonstrate any choking episodes. A cardiac computed tomography performed during the second week of life showed 2 anomalous veins draining into the inferior vena cava ([Fig. 2](#fig0002){ref-type="fig"}). A well-defined pulmonary sequestration was apparent in the right lower lobe ([Fig. 3](#fig0003){ref-type="fig"}). The pulmonary sequestration was fed by an arterial branch from the aorta ([Fig. 2](#fig0002){ref-type="fig"}). Lung tissue was continuous across the inferior mediastinum in a horseshoe configuration ([Fig. 4](#fig0004){ref-type="fig"}).Fig. 2Three-dimensional reformat of the major vascular structures viewed from the posterior to highlight the right lung vascularity. The right lung is on the right. The normal pulmonary venous system is shown in purple. The aorta is red with a prominent aortic branch feeding the right lower lobe sequestration. The anomalous pulmonary veins are shown in blue. (Color version of figure is available online.)Fig 2Fig. 3Coronal reformat of a CT angiogram of the chest shows a pulmonary sequestration in the right lower lobe (black arrow). An anomalous draining pulmonary vein is indicated by the white arrow.Fig 3Fig. 4Coronal CT through the lower chest demonstrates continuous pulmonary tissue across the posterior mediastinum (white arrow).Fig 4 The left lung exhibited normal venous drainage with 2 pulmonary veins connecting to the left atrium. There were 2 anomalous veins draining from the right lung to the intrahepatic inferior vena cava. The medial segment of the right lung was drained by a pulmonary vein connected to the left atrium. The majority of the hypoplastic right lung was drained via the abnormal ("scimitar") veins. The isthmus of the horseshoe was drained into the left atrium through what appeared to be the left inferior pulmonary vein. The trachea was normal in caliber. There was an abnormal right bronchial branching pattern with an anomalous configuration of secondary bronchi as a result of the altered anatomy in the right lung. The left lung exhibited normal bronchial anatomy. Discussion {#sec0003} ========== Congenital pulmonary venolobar syndrome is a type of partial anomalous pulmonary venous return in which a hypoplastic lung is drained by an anomalous vein into the systemic venous system. Most commonly, the vein drains into the inferior vena cava as in the case presented here. This causes an acyanotic left to right shunt which is typically surgically corrected if symptomatic. The correction involves constructing an interatrial baffle to redirect the pulmonary venous return into the left atrium or reimplanting the anomalous vein into the left atrium. Pulmonary sequestration is the aberrant formation of lung tissue which is not connected to the bronchial tree or pulmonary arteries. The condition predisposes the patient to hemoptysis or infections. These are the main reasons for surgical resection [@bib0004], [@bib0005]. Horseshoe lung occurs when a band of pulmonary parenchyma communicates between the right and left lungs with the formation of a single pleural cavity. This means if there were a pneumothorax, it would affect both lungs. Most cases of horseshoe lung published in the literature occur in the setting of additional pulmonary malformations such as either of the comorbid conditions in this patient. The combination of all 3 disorders is rarer. These findings are important for the radiologist to recognize as each individual finding has its own inherent risks. When occurring together, the patient\'s prognosis, treatment, and surgical approach may be affected. Funding: This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. Competing Interests: The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Najafi A, Caspar SM, Meienberg J, Rohrbach M, Steinmann B, Matyas G. Variant filtering, digenic variants, and other challenges in clinical sequencing: a lesson from fibrillinopathies. Clin Genet. 2020;97:235--245. 10.1111/cge.13640 31506931 Arash Najafi and Sylvan M. Caspar contributed equally to this work. **Peer Review** The peer review history for this article is available at <https://publons.com/publon/10.1111/cge.13640>. **Funding information** Ernst Göhner Stiftung; Gebauer Stiftung; Gemeinnützige Stiftung der ehemaligen Sparkasse Limmattal; Hirzel‐Callegari Stiftung; Palatin‐Stiftung; Stiftung \"Perspektiven\" of Swiss Life; Stiftung Suyana 1. INTRODUCTION {#cge13640-sec-0001} =============== Current high‐throughput sequencing (HTS) technologies such as whole‐exome (WES) and whole‐genome sequencing (WGS) enable the detection of ten thousands and millions of sequence variants, respectively. For Mendelian disorders, the identification of typically one or two disease‐causing sequence variants represents a bottleneck in the filtering and interpretation of the large number of sequence variants detected by HTS. As sequence variants causing Mendelian disorders are expected to be rare,[1](#cge13640-bib-0001){ref-type="ref"} filtering for sequence variants absent or infrequent in population‐based reference datasets is one of the main approaches used to discriminate (likely) pathogenic from benign variants. ExAC and its successor gnomAD are currently the largest publicly available population‐based reference datasets, providing the best variant frequency estimates.[2](#cge13640-bib-0002){ref-type="ref"}, [3](#cge13640-bib-0003){ref-type="ref"} Cutoffs of low allele frequencies are often used, including 1%, 0.1%, or 0.01%.[4](#cge13640-bib-0004){ref-type="ref"}, [5](#cge13640-bib-0005){ref-type="ref"} Recently, a gene‐specific framework has been introduced, which accounts for disease prevalence, allelic and genetic heterogeneity, inheritance mode, and penetrance, resulting in cutoff values even below 0.01%.[6](#cge13640-bib-0006){ref-type="ref"} Frequency cutoffs below 1%, however, may not be applicable to all clinical cases. Indeed, in apparently healthy population‐based reference datasets low‐allele‐frequency cutoffs may not appropriately account for common genetic modifiers or for the considerable number of disease‐affected individuals with atypical, late‐onset, and/or unrecognized phenotypes. Moreover, recent reports of pseudo and true digenic inheritance indicate that the identification of a single disease‐causing variant may be insufficient.[7](#cge13640-bib-0007){ref-type="ref"}, [8](#cge13640-bib-0008){ref-type="ref"}, [9](#cge13640-bib-0009){ref-type="ref"} Neglecting these factors may result in the exclusion of clinically relevant sequence variants, leading to incomplete or even missed diagnoses. Here, we address these challenges by assessing the presence and frequency of a priori (likely) pathogenic variants in gnomAD as well as by showing hitherto unreported cases of digenic variants causing clinically relevant complex phenotypes. 2. MATERIALS AND METHODS {#cge13640-sec-0002} ======================== 2.1. Evaluation of gnomAD {#cge13640-sec-0003} ------------------------- We compared the gnomAD‐based relative frequencies of most likely pathogenic variants in the genes *COL1A1*, *COL3A1*, *FBN1*, *FGFR2*, *JAG1*, *KMT2D*, *NSD1*, *SCN1A*, *TSC1*, and *TSC2* with the previously reported prevalence rates of respective autosomal‐dominant disorders. In addition, we compared the carrier frequencies of selected known pathogenic variants causing pediatric and/or adult‐onset disorders in the genes *BRCA1*, *BRCA2*, *CFTR*, *GJB2*, and *HBB* among ExAC, gnomAD, and currently appreciated estimates as described.[10](#cge13640-bib-0010){ref-type="ref"} We hypothesized that such a comparison of observed and expected frequencies enables to assess whether or not (likely) pathogenic variants are overrepresented in gnomAD and, hence, gnomAD can be considered as representative for the general population (incl. individuals with genetic predisposition to adult‐onset autosomal‐dominant disorders). 2.2. Detection of pathogenic variants in gnomAD {#cge13640-sec-0004} ----------------------------------------------- Variant call format files containing all sequence variants in gnomAD were downloaded from <http://gnomad.broadinstitute.com/downloads> (r2.0.1; 123 136 exomes and 15 496 genomes; released 2017, genome build GRCh37/hg19) and parsed using LeftAlignAndTrimVariants contained in GATK 3.5.[11](#cge13640-bib-0011){ref-type="ref"} Using VarSeq 2.0.1 (Golden Helix, Montana), the downloaded gnomAD sequence variants were annotated with several datasets and in silico tools (Supporting Information [Table S1](#cge13640-supitem-0001){ref-type="supplementary-material"}). To restrict the analysis to high‐confidence calls, we excluded sequence variants with a non‐PASS gnomAD filter, a 75‐mer mappability \<1,[12](#cge13640-bib-0012){ref-type="ref"} or a position outside of the canonical transcript (Table "knownCanonical" in the track "UCSC Genes" from the UCSC Table browser; <http://genome.ucsc.edu>/cgi‐bin/hgTables) as well as indels with a length ≥50 bp. To assess the presence and frequency of (likely) pathogenic variants in gnomAD, we performed automated filtering using VarSeq as well as manual evaluation. Only genes exclusively associated with autosomal‐dominant disorders (status "confirmed" in OMIM version 2018.5, <http://omim.org>) and with a DOMINO score ≥0.9 were selected.[13](#cge13640-bib-0013){ref-type="ref"} To restrict our analyses to genes likely intolerant to loss‐of‐function (LOF) variants, only genes with a pLi score ≥0.9 and an "observed/expected" (o/e) metric 90% confidence interval upper bound \<0.35 in gnomAD were further considered (Figure [1](#cge13640-fig-0001){ref-type="fig"}).[2](#cge13640-bib-0002){ref-type="ref"} Genes associated with disorders inherited in an autosomal‐recessive or X‐linked manner were not analyzed, because for the gnomAD dataset (r2.0.1) information on haplotype and sex is not (yet) available. {ref-type="ref"} ¶ According to Lek et al. (2016).[2](#cge13640-bib-0002){ref-type="ref"} \*\* According to Karczewski et al. (2019).[3](#cge13640-bib-0003){ref-type="ref"} †† In *FBN2* the prevalence calculation was limited to the CCA‐mutation‐hotspot region (exons 23‐34) for categories II‐VI, while for category I (nonsense and frameshift) all exons were considered. ‡‡ Sequence variants predicted "damaging" or "deleterious" by all six used in silico prediction tools (FATHMM, FATHMM‐MKL, MutationAssessor, MutationTaster, Polyphen2, SIFT; see also Supporting Information [Table S1](#cge13640-supitem-0001){ref-type="supplementary-material"}). CCA, congenital contractural arachnodactyly; gnomAD, Genome Aggregation Consortium; HGMD, Human Gene Mutation Database; indel, insertion/deletion; MFS, Marfan syndrome](CGE-97-235-g001){#cge13640-fig-0001} As prime examples, we focused on the genes *FBN1* and *FBN2*, which cause the autosomal‐dominant fibrillinopathies Marfan syndrome (MFS, OMIM \#154700) and congenital contractural arachnodactyly (CCA, OMIM \#121050), respectively.[14](#cge13640-bib-0014){ref-type="ref"}, [15](#cge13640-bib-0015){ref-type="ref"} In addition to nonsense, frameshift, and splicing variants, *FBN1* and *FBN2* sequence variants disrupting the consensus calcium‐binding sequence as well as disrupting or creating disulfide‐bond‐forming cysteines are known to be damaging and MFS/CCA‐causing due to increased proteolytic degradation.[16](#cge13640-bib-0016){ref-type="ref"}, [17](#cge13640-bib-0017){ref-type="ref"}, [18](#cge13640-bib-0018){ref-type="ref"}, [19](#cge13640-bib-0019){ref-type="ref"} Thus, such *FBN1* and *FBN2* sequence variants may be considered as a priori (likely) pathogenic, allowing the more elaborate gnomAD‐based predisposition/prevalence assessment of fibrillinopathies. We considered following sequence variants as most likely pathogenic in six (I‐VI) categories (Figure [1](#cge13640-fig-0001){ref-type="fig"}): (I) Nonsense and frameshift variants with or without expected nonsense‐mediated mRNA decay, because both may cause disease but via separate mechanisms;[20](#cge13640-bib-0020){ref-type="ref"}, [21](#cge13640-bib-0021){ref-type="ref"} (II) Single nucleotide variants located at the canonical splice sites (exonic ±1 bp, intronic ±1‐2 bp) and in silico predicted to alter splicing (Supporting Information [Table S1](#cge13640-supitem-0001){ref-type="supplementary-material"}); (III) In‐frame indels;[22](#cge13640-bib-0022){ref-type="ref"}, [23](#cge13640-bib-0023){ref-type="ref"} (IV) Missense variants disrupting functionally critical amino acids in the calcium‐binding epidermal growth factor domains of FBN1 and FBN2 such as disulfide‐bond‐forming cysteines as well as amino acids Asn, Asp, and Glu directly binding calcium (<http://uniprot.org/uniprot/P35555>; <http://uniprot.org/uniprot/P35556>);[16](#cge13640-bib-0016){ref-type="ref"}, [17](#cge13640-bib-0017){ref-type="ref"}, [18](#cge13640-bib-0018){ref-type="ref"}, [19](#cge13640-bib-0019){ref-type="ref"} (V) Sequence variants not included in categories I‐IV but described as disease‐causing (DM) in the Human Gene Mutation Database (HGMD) professional (v2019.1; <http://portal.biobase-international.com>) and/or as pathogenic in ClinVar (v2019.5; <http://ncbi.nlm.nih.gov/clinvar>) in association with MFS or CCA and with clear evidence for pathogenicity (segregation analysis, functional assays); (VI) Missense variants in *FBN1* and *FBN2* not included in categories I‐V but classified as "damaging" or "deleterious" by all six corresponding in silico prediction tools (FATHMM, FATHMM‐MKL, MutationAssessor, MutationTaster, Polyphen2, SIFT) without additional evidence (Supporting Information [Table S1](#cge13640-supitem-0001){ref-type="supplementary-material"}). Subsequently, categories I‐II were applied to the entire gnomAD dataset, whereas categories III‐VI were only applied to *FBN1* (NM_000138.4, all exons) and *FBN2* (NM_001999.3, CCA‐mutation‐hotspot exons 23‐34).[24](#cge13640-bib-0024){ref-type="ref"} Categories I‐V were defined as sensu stricto selected, that is, sequence variants with clear evidence for pathogenicity, while category VI contains sensu lato selected sequence variants, that is, additional, potentially pathogenic variants requiring manual expert review and interpretation. To largely exclude false‐positive results (ie, indeed non‐pathogenic variants), we manually evaluated all gnomAD variants in categories I and II occurring above a low‐frequency cutoff value of 0.01% or affecting genes with a total frequency of \>1:2000.[5](#cge13640-bib-0005){ref-type="ref"} Accordingly, variants were excluded if one of the following criteria was met: (a) ≥1 "benign" classifications in HGMD and/or ClinVar; (b) \>10% allele frequency in any gnomAD subpopulation; (c) variant exclusively affects weakly expressed exon (\<10% of the highest expressed exon in disease‐relevant tissue according to <http://gtexportal.org>). 2.3. Detection of *FBN1*/*FBN2* dual variants in WGS data {#cge13640-sec-0005} --------------------------------------------------------- WGS (PCR‐free, 60× 150PE) of 500 individuals with rare, mainly cardiovascular or connective tissue disorders was performed as described.[25](#cge13640-bib-0025){ref-type="ref"} FASTQ files were aligned using GENALICE MAP (Genalice, Nijkerk, The Netherlands).[26](#cge13640-bib-0026){ref-type="ref"} Variant calling was performed using the Population Calling module of GENALICE MAP to simultaneously extract all *FBN1* and *FBN2* sequence variants in our database.[26](#cge13640-bib-0026){ref-type="ref"} Using VarSeq, called sequence variants were annotated and filtered for individuals harboring (likely) pathogenic variants in both *FBN1* and *FBN2* (ie, dual variants). As a second interpretation platform, the artificial‐intelligence‐driven interpretation software Moon (Diploid, Leuven, Belgium) was used to independently detect *FBN1*/*FBN2* dual variants. For the confirmation of detected *FBN1* and *FBN2* variants and segregation analyses, Sanger sequencing of the corresponding region was performed as described.[27](#cge13640-bib-0027){ref-type="ref"} Data on clinical phenotypes were collected from medical records and/or during physical examination by one of the authors (A.N., M.R., or B.S.). Written informed consent was obtained from patients and family members. 2.4. ACMG/AMP classifications {#cge13640-sec-0006} ----------------------------- Automated classifications using the American College of Medical Genetics and Genomics (ACMG) and the Association for Molecular Pathology (AMP) published guidelines were obtained from InterVar v.201904/hg19.[28](#cge13640-bib-0028){ref-type="ref"} For *FBN1* variants disrupting disulfide‐bond‐forming cysteines, the ACMG/AMP criterion PM1 (moderate evidence for pathogenicity) was manually adjusted to PS3 (strong evidence for pathogenicity) as previously suggested.[29](#cge13640-bib-0029){ref-type="ref"} To avoid a bias caused by the ACMG/AMP criterion PM2 (moderate evidence for pathogenicity) fulfilled for variants absent from control populations, the automated classification was manually adapted by applying the criterion PM2 for extremely rare gnomAD variants as well (allele count ≤2). Likewise, for the two families with *FBN1*/*FBN2* dual variants, segregation data (PP1) were manually adjusted in the classification. 2.5. Statistical analysis {#cge13640-sec-0007} ------------------------- The upper and lower limits of the 95% confidence interval (CI) of a proportion were calculated using the online tool VassarStats with a correction for continuity (<http://vassarstats.net/prop1.html>) or the software GraphPad Prism 8 (GraphPad Software, California). Proportion of genetic predisposition and disease prevalence was calculated under the assumption that carrier individuals in gnomAD harbor no more than one (likely) pathogenic sequence variant in the considered gene(s). 3. RESULTS {#cge13640-sec-0008} ========== 3.1. Detection of pathogenic variants in gnomAD {#cge13640-sec-0009} ----------------------------------------------- Our comparison revealed that the evaluated carrier frequencies in the genes *BRCA1*, *BRCA2*, *CFTR*, *GJB2*, and *HBB* are not overrepresented in gnomAD (Supporting Information [Table S2](#cge13640-supitem-0001){ref-type="supplementary-material"}), confirming ExAC‐based previous results.[10](#cge13640-bib-0010){ref-type="ref"} In categories I and II (Figure [1](#cge13640-fig-0001){ref-type="fig"}), considering all gnomAD sequence variants in genes associated with autosomal‐dominant disorders (pLi ≥0.9, upper bound of the o/e metric 90% CI \<0.35) we identified by software filtering and manual evaluation a total of 2653 a priori (likely) pathogenic variants in 253 genes. Ten of these genes are the major cause of disorders with previously reported prevalence but we found no clear evidence that gnomAD‐based predisposition/prevalence rates are significantly higher than reported estimates (Supporting Information [Figure S1](#cge13640-supitem-0001){ref-type="supplementary-material"}). Per gene, 1 up to 130 gnomAD individuals harbor an a priori (likely) pathogenic variant with relative allele frequencies ranging from 1/246 272 (0.0004%) to 82/183 872 (0.0446%; *ATXN7* c.2673delA; Supporting Information [Figure S2](#cge13640-supitem-0001){ref-type="supplementary-material"}), resulting in disease predisposition/prevalence estimates ranging from approximately 1:100 000 to approximately 110:100 000 not associating with pLi, o/e, or DOMINO values (Supporting Information [Table S3](#cge13640-supitem-0002){ref-type="supplementary-material"}, Supporting Information [Figure S2](#cge13640-supitem-0001){ref-type="supplementary-material"}). During manual evaluation, we detected 16 apparently a priori (likely) pathogenic variants, which we subsequently reclassified as (likely) non‐pathogenic and thus excluded from the analysis. This reclassification was due to the less frequent allele being the reference allele in GRCH37/hg19 or to miscalled deletion/insertion variants or because the sequence variants are exclusively present in a weakly expressed exon (Supporting Information [Table S4](#cge13640-supitem-0001){ref-type="supplementary-material"} and [Table S5](#cge13640-supitem-0001){ref-type="supplementary-material"}). For the more elaborate predisposition/prevalence assessment of fibrillinopathies, we considered sequence variants of categories I‐V (sensu stricto selected, Figure [1](#cge13640-fig-0001){ref-type="fig"}) in the genes *FBN1* (all exons) and *FBN2* (CCA‐mutation‐hotspot exons 23‐34), thereby identifying most likely MFS‐ and CCA‐causing variants in 67 and 39 gnomAD individuals, respectively (Table [1](#cge13640-tbl-0001){ref-type="table"}, Figure [2](#cge13640-fig-0002){ref-type="fig"}, Supporting Information [Table S6](#cge13640-supitem-0003){ref-type="supplementary-material"}). Furthermore, from category VI (sensu lato selected, Figure [1](#cge13640-fig-0001){ref-type="fig"}) we added additional 880 *FBN1* and 321 *FBN2* missense variants that were in silico predicted to be (likely) deleterious, leading to a total of 947 and 360 potentially MFS‐ and CCA‐causing variants, respectively. Accordingly, the gnomAD‐based predisposition to MFS and CCA ranges from 4.8:10 000 (95% CI: 4‐6:10 000) up to 68:10 000 (95% CI: 64‐72:10 000) and 2.8:10 000 (95% CI: 2‐4:10 000) up to 26:10 000 (95% CI: 23‐29:10 000), respectively, depending whether sensu stricto and/or lato selected variants were considered (Table [1](#cge13640-tbl-0001){ref-type="table"}). In gnomAD, the majority of the sensu stricto selected *FBN1* and *FBN2* sequence variants were detected in individuals aged ≥50 years and were rare, with a non‐reference allele count of 1. However, the three most frequent sensu lato selected *FBN1* variants, c.3890A\>G (289 of 299 non‐reference alleles in Latino), c.3896C\>T (70 of 73 non‐reference alleles in Latino), and c.3089A\>G (64 of 69 non‐reference alleles in South Asian) were predominant in one gnomAD subpopulation, indicating a founder effect or selection bias (Supporting Information [Table S6](#cge13640-supitem-0003){ref-type="supplementary-material"}). ###### Overview of likely disease‐causing *FBN1* and *FBN2* variants in gnomAD +--------------------------------------------------------------------------------------------+--------------------+------------------+--------------------+------------------------------------------------+ | | *FBN1* exons 23‐34 | *FBN1* all exons | *FBN2* exons 23‐34 | *FBN2* all exons | +:===========================================================================================+:===================+:=================+:===================+:===============================================+ | Nonsense and frameshift (category I) | 2 | 8 | 5 | 29[a](#cge13640-note-0005){ref-type="fn"} (27) | +--------------------------------------------------------------------------------------------+--------------------+------------------+--------------------+------------------------------------------------+ | Splicing (category II) | 0 | 26 (10) | 0 | 36 (25) | +--------------------------------------------------------------------------------------------+--------------------+------------------+--------------------+------------------------------------------------+ | In‐frame indels (category III) | 1 | 9 (6) | 0 | 8 (5) | +--------------------------------------------------------------------------------------------+--------------------+------------------+--------------------+------------------------------------------------+ | Disulfide bonds (category IV) | 1 | 8 | 4 | 36 (31) | +--------------------------------------------------------------------------------------------+--------------------+------------------+--------------------+------------------------------------------------+ | Calcium binding (category IV) | 4 | 12 (12) | 6 (5) | 49 (23) | +--------------------------------------------------------------------------------------------+--------------------+------------------+--------------------+------------------------------------------------+ | HGMD 2019.1/ClinVar 2019.5 (category V)[b](#cge13640-note-0006){ref-type="fn"} | 1 | 4 (3) | 0 | 0 | +--------------------------------------------------------------------------------------------+--------------------+------------------+--------------------+------------------------------------------------+ | Sensu stricto selected sequence | 9:138 632 / | 67:138 632 / | 39:138 632 / | 158:138 632 / | | | | | | | | variants (categories I‐V) / (Prevalence) | (0.65:10 000) | (4.83:10 000) | (2.81:10 000) | (11.40:10 000) | +--------------------------------------------------------------------------------------------+--------------------+------------------+--------------------+------------------------------------------------+ | Sensu lato selected sequence variants (category VI)[c](#cge13640-note-0007){ref-type="fn"} | 566 (40) | 880 (162) | 321 (71) | 1.999 (301) | +--------------------------------------------------------------------------------------------+--------------------+------------------+--------------------+------------------------------------------------+ | All variants (categories I‐VI) / (Prevalence) | 575:138 632 / | 947:138 632 / | 360:138 632 / | 2,157:138 632 / | | | | | | | | | (41.48:10 000) | (68.31:10 000) | (25.97:10 000) | (155.59:10 000) | +--------------------------------------------------------------------------------------------+--------------------+------------------+--------------------+------------------------------------------------+ | All individuals in gnomAD (exomes and genomes) | 138 632 | | | | +--------------------------------------------------------------------------------------------+--------------------+------------------+--------------------+------------------------------------------------+ *Note*: Numbers indicate the total number of variants, while numbers in parenthesis indicate the number of unique variants. Abbreviations: gnomAD, Genome Aggregation Consortium; HGMD, Human Gene Mutation Database; indel, small insertion/deletion. Nonsense/frameshift variants were counted regardless of their position in *FBN2*. Only sequence variants not already included in categories I‐IV and passing manual evaluation were counted. Sequence variants predicted as \"damaging\" or \"deleterious\" by all of the six used in silico prediction tools (FATHMM, FATHMM‐MKL, MutationAsessor, MutationTaster, PolyPhen2, SIFT) not contained in I‐V were counted; see also Supporting Information [Table S1](#cge13640-supitem-0001){ref-type="supplementary-material"}. ![*FBN1* and *FBN2* a priori (likely) pathogenic variants in gnomAD and *FBN1*/*FBN2* dual variants in Family 1 and Family 2. Lollipops show the type and position of variants in relation to protein domain structure. Red boxes indicate the severe/neonatal region in *FBN1* (exons 23‐34) and the comparable congenital contractural arachnodactyly (CCA)‐mutation‐hotspot region in *FBN2* (exons 23‐34). cbEGF, calcium‐binding epidermal growth factor; EGF, epidermal growth factor; HGMD, Human Gene Mutation Database; indel, small insertion/deletion; TB, transforming growth factor β binding. Information on protein domains was obtained from umd.be/FBN1 and umd.be/FBN2. Lollipop diagrams were generated using the R package "trackViewer," available from <http://bioconductor.org/packages/release/bioc/html/trackViewer.html> \[Colour figure can be viewed at <http://wileyonlinelibrary.com>\]](CGE-97-235-g002){#cge13640-fig-0002} 3.2. Evaluation of in silico prediction tools and ACMG/AMP classifications {#cge13640-sec-0010} -------------------------------------------------------------------------- Splicing (category II) and functionally critical missense (category IV) variants in *FBN1* and *FBN2* served as a set of positive controls for the used in silico prediction tools (Supporting Information [Table S1](#cge13640-supitem-0001){ref-type="supplementary-material"}). All but two missense variants disrupting disulfide‐bond‐forming cysteines were correctly identified as "damaging" or "deleterious" by all six used in silico missense prediction tools, whereas five sequence variants disrupting calcium binding were missed by at least one tool. In contrast, 41, 38, and 21 of the 47 sensu stricto selected *FBN1* variants were classified as variant of unknown significance (VUS) according to the ACMG/AMP guidelines by using automated classification, manually adjusting variants disrupting disulfide‐bond‐forming cysteines from PM1 to PS3,[29](#cge13640-bib-0029){ref-type="ref"} and applying the criterion PM2 for extremely rare (allele count ≤2 in gnomAD) variants, respectively (Supporting Information [Table S6](#cge13640-supitem-0003){ref-type="supplementary-material"}). All splicing (category II) and functionally critical missense (category IV) variants were identified as phylogenetically conserved by at least one of the three used conservation scores (PhastCons, PhyloP, and SiPhy; Supporting Information [Table S1](#cge13640-supitem-0001){ref-type="supplementary-material"}), of which the SiPhy score generated the fewest apparently false‐negative results (Supporting Information [Table S6](#cge13640-supitem-0003){ref-type="supplementary-material"}). The CADD algorithm, capable of predicting the effect of all types of sequence variants, assigned high values to the majority of the sensu stricto selected sequence variants, but it also scored six *FBN1*/*FBN2* sequence variants in categories I‐V below a conservative PHRED‐scaled CADD cutoff score of 20 (Supporting Information [Table S6](#cge13640-supitem-0003){ref-type="supplementary-material"}). 3.3. Evaluation of HGMD and ClinVar entries {#cge13640-sec-0011} ------------------------------------------- Sixty‐six and two entries not contained in categories I‐IV were classified as DM (ie, disease‐causing mutations; not to be confused with the ACMG/AMP classification pathogenic and likely pathogenic) in HGMD (v2019.1), while six and none were classified as pathogenic in ClinVar (v2019.5) for MFS and CCA, respectively. By manual evaluation of corresponding publications/entries, segregation or functional analyses provided proof of pathogenicity for three of the *FBN1* HGMD DM entries, which, but no *FBN2* variant, could be included in category V (Table [1](#cge13640-tbl-0001){ref-type="table"}). 3.4. Families with pathogenic variants in both *FBN1* and *FBN2* {#cge13640-sec-0012} ---------------------------------------------------------------- In our cohort of 500 genomes, screening for individuals harboring (likely) pathogenic variants in both *FBN1* and *FBN2* resulted in two unrelated individuals (index patients with dual variants). Segregation analysis by Sanger sequencing of family members revealed a further dual variant carrier as well as individuals harboring only one or none of the sequence variants (Table [2](#cge13640-tbl-0002){ref-type="table"}). Affected family members were physically examined, revealing overlapping as well as distinguishing MFS and CCA clinical features (Table [2](#cge13640-tbl-0002){ref-type="table"}, Supporting Information [Table S7](#cge13640-supitem-0001){ref-type="supplementary-material"}). ###### Clinical features observed in Family 1 and Family 2 harboring sequence variants in *FBN1* and *FBN2* +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | | Family 1 | Family 2 | | | | | | | +:=============================================================================================================+:==============================================+:================+:================+:===========================+:================+:================+:=================+:==========================================+ | Age at examination (years) / Sex | 70 / M | 15 / F | 12 / F | 45 / M | 46 / F | 58 / F | 21 / M | 53 / M | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Affected gene | *FBN1* | *FBN1* | *FBN1*/*FBN2* | *FBN1*/*FBN2* | *FBN2* | *FBN1* | *FBN1*/*FBN2* | *FBN2* | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | *MFS‐associated features* [55](#cge13640-bib-0055){ref-type="ref"} | | | | | | | | | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Pes valgus | \+ | − | − | \+ | − | − | − | (+) | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Pes planus | (+) | \+ | \+ | \+ | \+ | \+ | \+ | − | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Pneumothorax | − | − | − | − | − | − | − | \+[a](#cge13640-note-0010){ref-type="fn"} | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Reduced elbow extension (r / l) | \+ / + | \>180° / \>180° | \+ / + | \+ / + | -- / -- | \>180° / \>180° | \>180° / \>180° | -- / -- | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | 3 of 5 facial features (dolichocephaly, enophthalmus, malar hypoplasia, retrognathia, downslanting fissures) | − | − | − | − | − | − | \+ | − | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Skin striae | \+ | − | − | − | (+) | − | − | \+ | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Myopia (\>3 diopters) | − | \+ | − | − | − | − | \+ | \+ | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Aortic dilatation (cm) | n/a | − | − | \+ (4.9) | − | − | \+ (4.1) | − | | | | | | | | | | | | | | | | (Z.score: \>2.0) | | | (Z‐score: \>2.0) | | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Ghent nosology systemic score | \<7 | \<5 | 7 | \>7 | \<7 | \<5 | \>7 | \<7 | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | *CCA‐associated features* [24](#cge13640-bib-0024){ref-type="ref"} | | | | | | | | | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Contractures | | | | | | | | | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Fingers (r / l) | -- / -- | -- / -- | \+ / + | \+ / + | \+ / + | -- / -- | -- / -- | \+ / + | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Elbow (r / l) | \+ / +[b](#cge13640-note-0011){ref-type="fn"} | -- / -- | \+ / + | \+ / + | -- / -- | -- / ‐ | ‐ / ‐ | -- / ‐ | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Hip | n/a | − | − | n/a | n/a | − | − | − | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Knee (r / l) | -- / -- | -- / ‐ | ‐ / ‐ | -- / ‐ | ‐ / ‐ | -- / ‐ | ‐ / ‐ | -- / ‐ | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Crumpled ears (remark) | − | − | -- (small ears) | \+ | − | − | − | − | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | *MFS‐ and CCA‐associated features* | | | | | | | | | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Palatal arch | n/a | narrow | high | n/a | n/a | normal | high | high | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Mitral valve prolapse | − | − | − | − | (+) | − | − | − | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | US/LS ratio \< 0.85 (height, cm) | -- (172) | -- (163.5) | -- (159) | \+ (179.5) | \+ (171) | -- (168) | -- (188) | -- (195) | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Armspan/height ratio \> 1.05 (armspan, cm) | -- (180.5) | -- (163.5) | -- (159) | -- (188.5) | -- (179.5) | -- (169) | \+ (205) | -- (198.5) | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Scoliosis or thoracolumbar kyphosis | (+) | − | \+ | − | (+) | (+) | \+ | − | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Wrist sign | − | − | \+ | \+ | − | − | \+ | \+ | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Thumb sign | − | − | \+ | \+ | − | − | \+ | \+ | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ | Pectus abnormity | − | \+ (carinatum) | \+ (excavatum) | \+ (carinatum & excavatum) | (+) (carinatum) | − | \+ (carinatum) | − | +--------------------------------------------------------------------------------------------------------------+-----------------------------------------------+-----------------+-----------------+----------------------------+-----------------+-----------------+------------------+-------------------------------------------+ *Note*: Sequence variants detected in **Family 1**: *FBN1* (NM_000138.4) c.8489A\>G p.(Gln2830Arg) and *FBN2* (NM_001999.3) c.3974‐26T\>G p.(Asn1327_Val1368del), sequence variants detected in **Family 2**: *FBN1* c.6595G\>A p.(Gly2199Ser) and *FBN2* c.3481G\>A p.(Glu1161Lys). Abbreviations: +, present; −, absent; (+), mildly present; CCA, congenital contractural arachnodactyly; F, female; l, left; LS, lower segment; M, male; MFS, Marfan syndrome; n/a, information not available; r, right; US, upper segment. Pneumothorax due to tuberculosis infection during adolescence. Likely explained by his old age of 70 years at examination. The female index patient of Family 1 (Ab2) (Figure [3](#cge13640-fig-0003){ref-type="fig"}A), who was referred with suspected CCA, harbors the heterozygous *FBN2* branch‐point variant c.3974‐26T\>G causing in‐frame exon skipping (p.Asn1327_Val1368del) first described as segregating with CCA in the large family of Ab2.[30](#cge13640-bib-0030){ref-type="ref"} In addition, she harbors the heterozygous *FBN1* variant c.8489A\>G, p.(Gln2830Arg), not detected in gnomAD and in silico predicted as (likely) deleterious. According to the ACMG/AMP guidelines, these *FBN2* and *FBN1* variants can be classified as likely pathogenic (LP) and VUS, respectively (Supporting Information [Table S8](#cge13640-supitem-0001){ref-type="supplementary-material"}). Ab2 presented with CCA‐associated clinical features such as elbow and finger contractures, but no MFS‐associated cardiovascular involvement (likely due to her young age of 12 years at examination). In addition, she presented with MFS/CCA‐overlapping features such as wrist and thumb sign (arachnodactyly), pectus excavatum, and scoliosis (Table [2](#cge13640-tbl-0002){ref-type="table"}, Supporting Information [Figure S3](#cge13640-supitem-0001){ref-type="supplementary-material"}). The father of Ab2 (Ab4; first described as IV‐9)[30](#cge13640-bib-0030){ref-type="ref"} harbors the same *FBN1*/*FBN2* variants and presented with similar clinical features, but also with CCA‐associated crumpled ears and MFS‐associated aortic dilatation (4.9 cm diameter, Z‐score \> 2.0). Ab1, the sister of Ab2, only harbors the *FBN1* variant and, as expected, presented with MFS‐associated features such as elbow hyperextension, myopia, and pectus carinatum. Ab6, the grandfather of Ab2, only harbors the *FBN1* variant as well but presented with mild MFS/CCA‐overlapping scoliosis and CCA‐associated elbow contractures at age of 70 years at examination. Ab3, the aunt of Ab2, only harbors the *FBN2* variant and presented with CCA‐associated finger contractures and mild MFS/CCA‐overlapping mitral valve prolapse and pectus carinatum. Ab7, the grandmother of Ab2, also only harboring the *FBN2* variant was not available for clinical examination. Ab5, the mother of Ab2, likely harbors neither sequence variant but was not available for genetic testing and examination. {#cge13640-fig-0003} The male index patient in Family 2 (Ac1) (Figure [3](#cge13640-fig-0003){ref-type="fig"}B), who was referred to us with suspected MFS, harbors the described (DM? in HGMD)[31](#cge13640-bib-0031){ref-type="ref"} heterozygous *FBN1* variant c.6595G\>A, p.(Gly2199Ser), which is in silico predicted as deleterious but detected three times in gnomAD and thus classified as VUS according to the ACMG/AMP guidelines. In addition, he harbors the reported (DM in HGMD)[32](#cge13640-bib-0032){ref-type="ref"} heterozygous *FBN2* variant c.3481G\>A, p.(Glu1161Lys), which is absent from gnomAD, affects a calcium binding residue, and is predicted as deleterious and classified as LP by in silico algorithms and the ACMG/AMP guidelines, respectively (Table [2](#cge13640-tbl-0002){ref-type="table"}, Supporting Information [Table S8](#cge13640-supitem-0001){ref-type="supplementary-material"}). Ac1 presented with MFS‐associated features such as elbow hyperextension, high myopia (−23/−23 dpt), and mild aortic dilatation (4.1 cm; Z‐score \> 2.0) as well as MFS/CCA‐overlapping features such as severe lumbar scoliosis (surgically corrected), wrist and thumb sign (arachnodactyly), high‐arched palate, and marfanoid habitus. However, he showed no CCA‐associated contractures (Table [2](#cge13640-tbl-0002){ref-type="table"}). Ac2, the father of Ac1, only harbors the *FBN2* variant and presented with CCA‐associated features such as finger contractures, MFS/CCA‐overlapping features like arachnodactyly and high‐arched palate, as well as high myopia (−12/−9 dpt), which is frequent in MFS and may also occur in CCA.[33](#cge13640-bib-0033){ref-type="ref"}, [34](#cge13640-bib-0034){ref-type="ref"} Both Ac3, the mother of Ac1, and Ac4, the aunt of Ac1, only harbor the *FBN1* variant and presented with mild MFS phenotypes and non‐marfanoid habitus. The mother (Ac3) presented with MFS‐associated elbow hyperextension, mild scoliosis, and no cardiovascular involvement (Table [2](#cge13640-tbl-0002){ref-type="table"}), while the aunt (Ac4) reportedly is of short stature with no cardiovascular involvement, but was not available for clinical examination. 4. DISCUSSION {#cge13640-sec-0013} ============= Our data show that, with appropriate consideration, gnomAD, the largest publicly available dataset for population‐based allele frequencies, can be used not only for variant filtering, interpretation, and carrier screening but also for the estimation of disease predisposition/prevalence. Accordingly, we calculated gnomAD‐based predisposition/prevalence estimates for autosomal‐dominant disorders at genome scale, thereby providing a novel estimate for MFS and the first estimate for CCA. Using the example of these fibrillinopathies, we also showed that the genetic predisposition to highly‐penetrant autosomal‐dominant disorders may occur more frequently than previously assumed. Moreover, by presenting two families with hitherto unreported co‐occurrence of *FBN1*/*FBN2* variants causing increased or hidden clinical features of MFS and CCA, we exemplified that segregating (ie, not‐linked) pathogenic variants in more than one gene can underlie the clinical phenotype of patients and the clinical variability of affected family members. Thus, several challenges and conclusions emerge. First, although WGS is the most comprehensive HTS method offering advantages over targeted sequencing and WES,[25](#cge13640-bib-0025){ref-type="ref"}, [35](#cge13640-bib-0035){ref-type="ref"} data interpretation remains a challenge.[12](#cge13640-bib-0012){ref-type="ref"} Currently, there is no generally accepted gold standard software solution for variant interpretation comparable to BWA/GATK for alignment and variant calling.[11](#cge13640-bib-0011){ref-type="ref"}, [12](#cge13640-bib-0012){ref-type="ref"} According to the ACMG/AMP guidelines, allele frequencies higher than expected for the disorder (criterion BS1) or reported \>5% in population‐based reference datasets (criterion BA1) count as strong or stand‐alone evidence of benign impact of a variant.[1](#cge13640-bib-0001){ref-type="ref"} Accordingly, low‐frequency cutoffs are often used, down to \<0.01%.[6](#cge13640-bib-0006){ref-type="ref"} Although such cutoffs help to reduce or minimize the number of candidate variants, they may remove clinically relevant variants from consideration. Examples include the sensu stricto selected *FBN1* variant c.863A\>G, p.(Asp288Gly) (allele frequency: 12/277 018; 0.0043%) and the sensu lato selected *FBN1* variant c.3026C\>G, p.(Pro1009Arg) (allele frequency: 16/246 218; 0.0065%) (Supporting Information [Table S6](#cge13640-supitem-0003){ref-type="supplementary-material"}), which are more frequent in gnomAD than the recently introduced maximum tolerated allele count of 2 in approximately 120 000 alleles (approximately 4/240 000; approximately 0.0017%) for MFS.[6](#cge13640-bib-0006){ref-type="ref"} In addition, low‐frequency cutoffs exclude weaker variants such as the homozygous Ehlers‐Danlos syndrome causing but heterozygous phenotype‐modifying *COL5A1* variant c.1588C\>T, p.(Gly530Ser)[36](#cge13640-bib-0036){ref-type="ref"} and clinically relevant but common genetic modifiers like chr1:203138970A\>G (rs2250509)[37](#cge13640-bib-0037){ref-type="ref"} modifying the expressivity of *MYH7*‐related cardiomyopathy with gnomAD allele frequency of 3.4% and 18.8%, respectively. Moreover, due to a founder effect or selection bias, certain potentially pathogenic variants requiring manual expert review, including the three most frequently occurring *FBN1* variants in category VI (c.3089A\>G, c.3890A\>G, and c.3896C\>T in Supporting Information [Table S6](#cge13640-supitem-0003){ref-type="supplementary-material"}), may occur rather frequently in one ethnicity (up to 0.12% allele frequency) and thus may be filtered out by using too low cutoff frequencies (ie, below 0.12%). Thus, filtering for sequence variants extremely rare or absent in population‐based reference datasets is helpful to identify variants that should not be missed in a first step, while variants with an increased frequency should not be a priori disregarded without an appropriate consideration and literature review, especially in complex cases.[38](#cge13640-bib-0038){ref-type="ref"} Additional rare exonic variants may pose challenges for interpretation, such as variants solely affecting weakly expressed exons or variants miscalled as two independent variants instead of a small insertion/deletion (Supporting Information [Table S5](#cge13640-supitem-0001){ref-type="supplementary-material"}). To avoid missed or incomplete diagnoses, there is an evident demand for more advanced filtering methods that may consider more elaborated algorithms, adapted cutoffs, paralogue annotations,[39](#cge13640-bib-0039){ref-type="ref"} facial phenotyping algorithms,[40](#cge13640-bib-0040){ref-type="ref"} or data sharing platforms.[41](#cge13640-bib-0041){ref-type="ref"} Second, although current widely‐used in silico tools and disease‐associated databases are powerful, they are not without fault. In this study, not all in silico tools correctly classified all of the sensu stricto selected *FBN1* and *FBN2* variants. By individual use, these tools are therefore not capable of depicting all of the underlying disease‐causing mechanisms and, hence, we recommend combining multiple tools for the highest accuracy and the robust application of the ACMG/AMP criteria PP3 and BP4.[38](#cge13640-bib-0038){ref-type="ref"} Moreover, several sequence variants classified as DM in HGMD or pathogenic in ClinVar were considered for category V but were excluded, because manual evaluation of the corresponding publication/entry revealed no convincing evidence of pathogenicity (ie, functional characterization or segregation analyses). Thus, even in curated datasets such as HGMD and ClinVar unreliable variant‐disease associations might be present, which is why they need to be used with care and evaluated by experts.[42](#cge13640-bib-0042){ref-type="ref"}, [43](#cge13640-bib-0043){ref-type="ref"} Third, the ACMG/AMP guidelines provide a framework for clinical variant classification.[1](#cge13640-bib-0001){ref-type="ref"} To support implementation and increase inter‐laboratory concordance,[44](#cge13640-bib-0044){ref-type="ref"} (semi‐)automated open source (eg, InterVar, Genetic Variant Interpretation Tool)[28](#cge13640-bib-0028){ref-type="ref"}, [45](#cge13640-bib-0045){ref-type="ref"} and commercial (eg, <http://varsome.org>; <http://goldenhelix.com>) classification tools have been developed. Furthermore, a refinement of the ACMG/AMP guidelines into 108 criteria[46](#cge13640-bib-0046){ref-type="ref"} as well as several gene‐ and disease‐specific adaptations have been introduced.[29](#cge13640-bib-0029){ref-type="ref"}, [47](#cge13640-bib-0047){ref-type="ref"}, [48](#cge13640-bib-0048){ref-type="ref"} However, because 10 of the 28 ACMG/AMP criteria need manual adjustment (eg, by using functional or segregation data), the result obtained by (semi‐)automated classification is often incomplete, regardless of the applied tool.[28](#cge13640-bib-0028){ref-type="ref"} Unfortunately, information for manual adjustment is unavailable in gnomAD and thus the ACMG classifications in this study are based on automated classification, explaining why the majority of our positive control variants in *FBN1* were classified as VUS. This, however, would likely be resolved using additional information. For instance, the *FBN1*‐specific refinement of the ACMG/AMP criteria led to a more appropriate classification of disulfide‐bond‐disrupting *FBN1* variants as LP instead of VUS (Supporting Information [Table S6](#cge13640-supitem-0003){ref-type="supplementary-material"}).[29](#cge13640-bib-0029){ref-type="ref"} In addition, not only because gnomAD contains pathogenic variants but also because larger population datasets will emerge in the future, the criterion PM2 (moderate evidence for pathogenicity; fulfilled for variants absent from control populations) may require adjustment. As the application of population‐based data is a major source of inter‐laboratory discordances,[38](#cge13640-bib-0038){ref-type="ref"} increased number of population‐based criteria (eg, 5 criteria instead of the existing 3, PM2, BA1, BS1),[46](#cge13640-bib-0046){ref-type="ref"} or gene‐specific frequency thresholds[6](#cge13640-bib-0006){ref-type="ref"}, [48](#cge13640-bib-0048){ref-type="ref"} could help to appropriately implement the PM2 criterion. Indeed, in the absence of additional information, application of the PM2 criterion to extremely rare variants (allele count ≤2 in gnomAD) may lead to a reclassification, for example, from VUS (PVS1, PP3) to pathogenic (PVS1, PM2, PP3), representing a major caveat. Moreover, the ACMG/AMP guidelines were developed for highly‐penetrant monogenic disorders, complicating their implementation in cases with (pseudo) digenic inheritance or with genetic modifiers (eg, because of the application of segregation data). As the ACMG/AMP guidelines are widely used, it is crucial to use classification tools prudently and to understand their limitations. Depending on the used tool, the thresholds for several criteria may vary, leading to inconsistent or VUS classifications, which may be pivotal as the ACMG/AMP guidelines are frequently invoked for clinical decision support[47](#cge13640-bib-0047){ref-type="ref"} and because VUS are not recommended to be reported to patients.[49](#cge13640-bib-0049){ref-type="ref"} Fourth, as we detected 2653 a priori (likely) pathogenic gnomAD variants in genes associated with autosomal‐dominant disorders, this dataset should be regarded as apparently, rather than completely, non‐affected. Nevertheless, our data suggest that (likely) pathogenic variants are not overrepresented and we found no clear evidence for considerable bias due to included disease‐specific studies. Thus, gnomAD can be considered as an appropriate representation of the general population, thereby expanding the knowledge of pathogenic variants in ExAC to the much larger gnomAD database,[10](#cge13640-bib-0010){ref-type="ref"} enabling (re‐)estimation of the population prevalence of genomic disorders. In general, gnomAD genes of categories I‐II (Figure [1](#cge13640-fig-0001){ref-type="fig"}) with an increased number of (likely) pathogenic variants are not infrequently associated with highly‐penetrant diseases with late onset and/or unapparent phenotype, which may occur in fibrillinopathies as well. Indeed, without appropriate imaging of the aorta, even old (≥50 years) carriers of pathogenic *FBN1* and/or *FBN2* variant(s) may be considered as apparently non‐affected in gnomAD. Our gnomAD‐based MFS predisposition rate of approximately 5:10 000 (*FBN1* categories I‐V) is higher than the previously reported prevalence of approximately 1‐3:10 000,[33](#cge13640-bib-0033){ref-type="ref"}, [50](#cge13640-bib-0050){ref-type="ref"} suggesting that highly‐penetrant fibrillinopathies may be underdiagnosed, especially in mild cases. However, *FBN1* mutations have been reported in other non‐MFS disorders as well (cf. HGMD professional; <http://portal.biobase-international.com>). As for *FBN2* we only considered the CCA‐mutation‐hotspot exons 23‐34, the real predisposition to CCA might be even higher than our estimate of approximately 3:10 000. In contrast, some of the detected a priori (likely) pathogenic variants in gnomAD, although de facto (likely) pathogenic, might not lead to disease in certain resilient individuals.[51](#cge13640-bib-0051){ref-type="ref"} Resilience may be due to protective genetic modifiers or absence of expression of the affected allele due to random monoallelic expression.[52](#cge13640-bib-0052){ref-type="ref"}, [53](#cge13640-bib-0053){ref-type="ref"} Because we only considered PASS gnomAD variants (ie, genotype quality ≥20, allelic balance \>0.2, and read depth ≥10)[2](#cge13640-bib-0002){ref-type="ref"} with a mappability =1,[12](#cge13640-bib-0012){ref-type="ref"} the variants included in this study are most likely true variants and do not result from mosaicism or alignment difficulties. Fifth, the presented two families with hitherto unreported *FBN1*/*FBN2* dual variants showing mixed phenotypes of MFS and CCA exemplify the importance of considering clinically relevant pathogenic variants in more than one gene. The presented cases neither fulfill the definition of true digenic inheritance nor of classic genetic modifiers, because the sequence variants alone also cause at least a mild phenotype and the dual variant carriers do not show novel but rather a combination of MFS‐ and CCA‐associated features.[9](#cge13640-bib-0009){ref-type="ref"}, [37](#cge13640-bib-0037){ref-type="ref"} Thus, the presented cases are most accurately described as two diseases segregating independently according to classic Mendelian inheritance that underlie the phenotype of dual variant carriers.[9](#cge13640-bib-0009){ref-type="ref"} According to our data, the probability of the co‐occurrence of *FBN1* and *FBN2* variants in one individual might be at least approximately 1.36 × 10^−7^, implying that additional, unrecognized cases may exist. As expected, only individuals harboring an *FBN1* variant (with or without the *FBN2* variant) show MFS‐associated aortic dilatation, whereas only individuals harboring an *FBN2* variant (with or without the *FBN1* variant) show CCA‐associated contractures (Table [2](#cge13640-tbl-0002){ref-type="table"}). Individuals carrying only the familial *FBN1* variant (Ab1, Ab6, and Ac3) show several features of MFS including joint laxity, indicating the clinically relevant effect of both *FBN1* variants classified as VUS according to the ACMG/AMP guidelines (Table [2](#cge13640-tbl-0002){ref-type="table"}). The effect of dual variants appears synergistic for some clinical features, which has also been observed in an *Fbn1*/*Fbn2* double knock‐out mouse model.[54](#cge13640-bib-0054){ref-type="ref"} For instance, severe kyphoscoliosis is exclusively present in the dual‐variant carrier (Ac1) in Family 2 (*FBN1* c.6595G\>A, *FBN2* c.3481G\>A) but in neither of the parents (Ac2, Ac3). In addition, myopia is more pronounced in the dual‐variant carrier Ac1 (−23/−23 dpt) than in the *FBN2* variant carrier Ac2 (−12/−8 dpt), suggesting a synergistic effect of both variants regarding the ocular phenotype as well. In contrast, for other distinguishing clinical features, the effect of one variant appears to be more pronounced, because the dual‐variant carriers (Ab2, Ab4) in Family 1 show CCA‐specific contractures, while the dual‐variant carrier (Ac1) in Family 2 shows MFS‐associated elbow hyperextension. If pathogenic variants in more than one gene underlie the disease phenotype such as in the presented cases, molecular genetic diagnosis is crucial for etiology‐oriented treatment and particularly for genetic counseling regarding family planning. Thus, information on possible digenic inheritance is clinically highly relevant and known cases are listed in the Digenic Diseases Database (<http://dida.ibsquare.be>). The main limitation of our study is that for gnomAD phenotypic data of the participants is (yet) unavailable and thus the clinically relevant effect of detected variants can not be directly assessed. Furthermore, information whether multiple sequence variants are harbored by one individual is also unavailable in gnomAD, which would be necessary to assess digenic inheritance and genetic modifiers. Moreover, as we restricted our genome‐scale analysis to categories I and II containing a priori (likely) pathogenic variants (nonsense, frameshift, and canonical splice sites), we potentially missed other (likely) pathogenic variants. We considered that for some genes (eg, *GFAP*, *TGFBR1*, *TGFBR2*) LOF is not (yet) established as a disease mechanism. Thus, to exclude such genes, we restricted our analyses to genes likely intolerant to LOF as predicted by both the pLi score and the o/e metric.[2](#cge13640-bib-0002){ref-type="ref"}, [3](#cge13640-bib-0003){ref-type="ref"} Indeed, for all but one (*ATXN7*) of seven genes with (likely) pathogenic gnomAD‐based allele frequencies \>1:2000 (Supporting Information [Table S4](#cge13640-supitem-0001){ref-type="supplementary-material"}), our manual evaluation revealed LOF as an established disease mechanism. On the other hand, although our data suggest that variants causing severe (pediatric) diseases are likely not overrepresented in gnomAD, the dataset includes disease‐associated projects, potentially biasing our gnomAD‐based disease prevalence calculations. Thus, considering incomplete penetrance, variable expressivity and late disease onset as well, the genome‐scale prevalence values presented here should be regarded as predisposition estimates, aiming to highlight challenges in allele‐frequency‐based variant filtering and HTS data interpretation. Taken together, in the current genomics era using appropriate short‐ and/or long‐read HTS platforms, the bottleneck on the path to accurate diagnosis is the interpretation of variants rather than the generation of data.[12](#cge13640-bib-0012){ref-type="ref"} Current interpretation approaches are challenged by VUS, inappropriate variant filtering, and unrecognized digenic inheritance/variants or genetic modifiers, warranting improvement. Hence, instead of hard filtering, variant interpretation should include large‐scale deeply geno‐ and phenotyped databases, powerful in silico tools, and data sharing as well as, if available, appropriate segregation and functional analyses to aid expert evaluation and enable accurate diagnosis. CONFLICT OF INTEREST {#cge13640-sec-0015} ==================== The authors declare that they have no conflict of interest. AUTHOR CONTRIBUTIONS {#cge13640-sec-0016} ==================== All authors revised the work critically for important intellectual content and gave approval for the final version of the manuscript. G.M. conceptualized this study. A.N., B.S., and M.R. performed physical examination of the patients. A.N., S.C., and J.M. analyzed clinical data. S.C. performed the analysis of database‐derived data. S.C., A.N., J.M., and G.M. wrote the manuscript and B.S. and M.R. critically reviewed the manuscript. ETHICS APPROVAL {#cge13640-sec-0017} =============== The employed procedures were reviewed and approved by the local institutional review committee, following the principles outlined in the Helsinki Declaration. Informed consent was obtained from all individual participants included in the study or their guardians. Supporting information ====================== ###### **Appendix S1.** Supporting Information. ###### Click here for additional data file. ###### **Appendix S2.** Supporting Information. ###### Click here for additional data file. ###### **Appendix S3.** Supporting Information. ###### Click here for additional data file. The authors would like to thank the patients for their participation in this study. This work was supported by funding from the Ernst Göhner Stiftung, Gebauer Stiftung, Hirzel‐Callegari Stiftung, Stiftung Suyana, Palatin‐Stiftung, Gemeinnützige Stiftung der ehemaligen Sparkasse Limmattal, and the Stiftung "Perspektiven" of Swiss Life. DATA ACCESSIBILITY {#cge13640-sec-0018} ================== The data that support the findings of this study are available on request from the corresponding author.

1. Background {#sec0005} ============= Lower respiratory tract infections (LRTI) are a common cause of hospitalization and are caused by various bacterial or viral pathogens.[@bib0005] To date, more than 20 viruses have been shown to cause respiratory tract infections and every year new viruses are associated with LRTI.[@bib0010], [@bib0015], [@bib0020], [@bib0025], [@bib0030], [@bib0035] Studies suggest that an early and accurate identification of a viral etiologic agent during emergency department evaluation or in hospitalized patients has been associated with a decrease in antibiotic use, a decreased number of additional tests performed, a decreased length of stay in the emergency department and a decreased length of hospital stay.[@bib0040], [@bib0045], [@bib0050], [@bib0055] Clinical diagnosis of a respiratory tract infection is difficult in patients with atypical illness characteristics and in low prevalence setting. Identification of the etiological agent can be useful if there is a need for a specific therapeutic approach, if hospital hygiene measures are necessary and for epidemiological reasons.[@bib0060] Laboratory techniques such as antigen immunoassays, viral culture and serology have several shortcomings such as a limited sensitivity, long turnaround times and limited spectrum of detectable pathogens.[@bib0060], [@bib0065] A combination of methods is often necessary to improve diagnosis. The use of nucleic acid amplification tests (NAAT) such as the polymerase chain reaction (PCR) enables same day detection of causal pathogens and therefore the prompt initiation of appropriate interventions. Multi parameter NAAT allow broad spectrum detection of pathogens, thereby saving time and work.[@bib0070] In addition, a study by Brunstein and colleagues reports preliminary clinical evidence that the diagnosis of co-infections is medically relevant and that an effective treatment for severe RTI requires the detection of all involved pathogens.[@bib0075] 2. Objectives {#sec0010} ============= This study compares the performance of the xTAG Respiratory Viral Panel (RVP) assay (Luminex Molecular Diagnostics, Toronto, Canada) with the Respifinder assay (Respifinder) (Pathofinder, Maastricht, Netherlands) for the detection of viral respiratory pathogens. The RVP assay comprises a multiplex reverse transcription polymerase chain reaction (RT-PCR), followed by a multiplex Target-Specific Primer Extension. The target-specific primers are chimeric primers juxtaposed to a Universal Tag sequence. This allows sorting on a Luminex xMAP instrument.[@bib0080] The Respifinder assay is based on the multiplex ligation-dependent probe amplification (MLPA) technology. The MLPA reaction is preceded by a pre-amplification step to ensure the detection of DNA and RNA viruses with the same specificity and sensitivity as individual singleplex real-time RT-PCR.[@bib0085] Both assays can detect and differentiate following respiratory viruses: influenza A (InfA) and B (InfB), respiratory syncytial virus A (RSVA) and B (RSVB), parainfluenza 1--4 (PIV-1 to PIV-4), coronavirus NL63 (CoV-NL63), coronavirus OC43 (CoV-OC43), coronavirus 229E (CoV-229E), human metapneumovirus (hMPV) and adenovirus (adeno). Additionally, in each assay a probe for InfA H5N1 is included. The RVP assay can also subtype for InfA H1 and InfA H3. The Respifinder assay can detect rhinovirus (rhino), which is also detected by the RVP, but this assay cannot differentiate rhinovirus from enterovirus (rhino-entero). Moreover, the detection of coronavirus HKU1 (CoV-HKU1) and coronavirus SARS (SARS-CoV) is also included in the RVP assay. 3. Study design {#sec0015} =============== A total of 9 external quality assurance (EQA) panels (QCMD, Scotland) containing 106 EQA samples, of which 95 samples were expected to be positive for influenza, respiratory syncytial virus, parainfluenza, coronavirus, rhinovirus, human metapneumovirus and adenovirus were analyzed once. \[MV.RS 2007/2008, ADV 2007, PINF 2006/2008, RV.CV 2007/2008, INF 2006/2008\]. Extraction of nucleic acids (NA) was performed with easyMag (bioMérieux, Lyon, France), using the generic 2.0.1 protocol. An extraction and amplification control was included for every sample according to the manufacturer\'s instructions. Sample volume for extraction was 200 μl, elution volume was 50 μl for RVP and 100 μl for Respifinder. All panels were analyzed with the RVP and Respifinder assay according to the manufacturer\'s instructions. Respectively 5 μl for RVP and 10 μl for Respifinder of NA were added to the PCR mixture resulting in an end concentration of 2% of the total sample concentration for both assays. Amplification, hybridization and ligation were performed on a PTC200 (Bio-Rad Laboratories, Hercules, CA, US). For RVP, data acquisition was performed on the LX200 (Luminex Molecular Diagnostics), using the TDAS RVP-1 software. For Respifinder, fragment analysis was performed on the CEQ 8000 genetic analysis system (Beckman Coulter, Brea, CA, US), followed by analysis with the Fragment Analysis software. 4. Results {#sec0020} ========== For all negative samples, a negative result was obtained with both assays. A positive result was found with the Respifinder assay in 74 (78%) of 95 positive samples. Most of the false negative samples (15 of 21) were weak positive, except for 1 adenovirus type 31 sample. For 1 weak positive sample an inconclusive result (extraction and amplification control negative) was obtained. Weak positive samples were samples with a stock dilution of 10^−6^, *Ct*-value ≥35 by independent testing or a concentration ≤100 copies/ml. For 4 samples, a negative or an inconclusive result was obtained. These samples had an expected *Ct*-value of 33--34. Additionally, an adenovirus was detected in 3 RSV positive samples, which was confirmed with an adenovirus specific real-time PCR.[@bib0090] In 1 InfA H1 sample, the H5N1 variant was also detected. The RVP was positive in 31 (33%) samples. All adenovirus, CoV-NL63, CoV-OC43, and CoV-229E samples were false negative. Furthermore, for PIV-1, 7 of 8 samples were false negative and 1 of 8 samples resulted in an inconclusive result (signal for viral target within the equivocal zone as mentioned in the kit insert). For InfA, 8 of 14 samples were false negative and 5 of 14 samples showed an inconclusive result. Almost all weak positive samples were false negative (25 of 30) or inconclusive (1 of 30). The sensitivity for weak positive samples was 47% for Respifinder and 13% for RVP. Results are shown in [Table 1](#tbl0005){ref-type="table"} ([Supplementary data in Table 2](#sec0040){ref-type="sec"}).Table 1Pathogen results after analysis of EQA samples with Respifinder and RVP.PathogenSensitivity (%)Inconclusive[b](#tblfn0010){ref-type="table-fn"} (%)RespifinderRVPRespifinderRVPhMPV703000RSVA10043029RSVB1006000Adeno type 1100000Adeno type 3100000Adeno type 4100000Adeno type 310000PIV-16302513PIV-210010000PIV-310010000PIV-4838300Rhino 16804000Rhino 727510000Rhino 90806000CoV-NL6380000CoV-OC43100000CoV-229E50000InfA H14314029InfA H37101460InfB835000Weak positive samples[a](#tblfn0005){ref-type="table-fn"}471333All783338[^1][^2] 5. Discussion {#sec0025} ============= Both assays have an excellent specificity which was demonstrated by the concordant results for all negative samples. RVP was positive in no more than 31 of 95 samples, which results in an overall sensitivity of 33%. The limited sensitivity can be explained by a limited sensitivity for adenovirus, CoV-NL63, CoV-OC43, CoV-229E, PIV-1, InfA and for weak positive samples. A positive result was found with the Respifinder assay in 74 of 95 samples, which is equivalent with an overall sensitivity of 78%. In general, the false negative results can be explained by the limited sensitivity for weak positive samples. The false positive result for InfA H5N1 could probably be explained by cross-hybridization of a variant of InfA H1N1 with the H5N1 probe. According to the kit insert, the diagnosis of a H5N1 infection may not be based on a positive finding with the Respifinder assay alone. For weak positive samples, sensitivity was low for both assays. It should be noticed that less than 60% of the participants in the EQA round reported a positive result in 14 of 30 weak positive samples. Furthermore, the clinical relevance of these weak positive samples remains unclear. A recent study, published by Utokaparch et al.[@bib0095] suggests that the total viral load for children with LRTI is significantly increased compared to children with non-LRTI. This study was performed on quality assurance samples only and the obtained results have to be confirmed with analysis of clinical samples. However, these analytical evaluation results can give an indication on the technical performance of both assays.[@bib0100] A study on the clinical performance of RVP by Mahony et al.[@bib0105] seems to indicate that the RVP has a higher sensitivity (98.5%) compared to direct fluorescent-antibody assay (DFA) and culture (68.8%). The sensitivity of the Respifinder compared to cell culture was 100%, except for PIV-3 (80%).[@bib0085] Data on the performance characteristics of other commercial multiplex assays show that these assays are more sensitive than viral culture.[@bib0110], [@bib0115] Only few studies compared the performance of commercially available multiplex assays. A study by Balada-Llasat et al.,[@bib0115] evaluating the performance of 3 commercial assays for the diagnosis of respiratory viral infections in adults showed that, when compared to culture, the RVP had a sensitivity and specificity of 100% and 91%, compared to MultiCode-PLx (EraGen Biosciences, Madison, WI, US) with 89% and 87% and Resplex II (Qiagen, Hilden, Germany) with 89% and 94%, respectively. However, the Resplex II offered the broadest virus detection range and the MultiCode-PLx offered the greater ease of use. Another study[@bib0110] seems to indicate that the Respifinder has a better sensitivity than the Seeplex RV12 detection kit (Seegen, Rockville, MD, US). To conclude, multi parameter assays can be a useful tool for broad spectrum detection of respiratory pathogens, although time-to result could be improved. Negative results should be interpreted with care because of the limited sensitivity for some pathogens. Conflict of interest statement {#sec0030} ============================== Funding: None. Competing interests: None declared. Ethical approval: Not required. Appendix A. Supplementary data {#sec0040} ============================== Supplementary data associated with this article can be found, in the online version, at [doi:10.1016/j.jcv.2011.08.017](10.1016/j.jcv.2011.08.017). [^1]: Samples with an expected result: dilution of 10^−6^, *Ct*-value ≥35 or concentration ≤100 copies/ml. [^2]: Inconclusive result: for RVP: signal for viral target within the equivocal zone as mentioned in the kit insert, for Respifinder: extraction and amplification control negative.

All relevant data are within the manuscript and its Supporting Information files. Introduction {#sec001} ============ The population around the world has been growing rapidly and has a corresponding increase in food demand. The improvement in environmental efficiency of beef production systems seems to be, at least for the foreseeable future, part of the solution for the issue of global food security \[[@pone.0220247.ref001]\]. Notwithstanding, ruminant livestock systems are under continued political pressure to reduce their greenhouse gas (GHG) outputs. Cattle production is an important driver for Brazil's economy, and ranks second worldwide, with approximately 212 million head \[[@pone.0220247.ref002]\]. Additionally, Brazil is the largest beef exporter, maintaining trade relations with 180 countries. Traditionally, the national herd is maintained in an extensive pasture-based production system. However, more recently, there has been a notable shift in Brazilian beef production, with livestock farming gradually occupying less land with increased production and productivity gains \[[@pone.0220247.ref003]\]. The modern, intensive livestock systems, like beef production in grain-finishing systems, offer both substantially lower land requirements and greenhouse gases (GHG) emissions per kilogram of meat than traditional, extensive ones \[[@pone.0220247.ref004]\]. However, the GHG emissions by ruminants in adaptive grazing systems has been shown in some studies \[[@pone.0220247.ref005], [@pone.0220247.ref006]\], to be considerably lower than previously thought. This decrease was attributed to the quality and productivity of the pastures, potentially increasing soil carbon sequestration thereby negating atmospheric emissions \[[@pone.0220247.ref007]\]. Therefore, the best option could be a system that incorporates well-managed grass systems complemented by grain-fed components as cattle reach compositional maturity. On the other hand, genetic improvement in beef cattle has a potential for reducing CH~4~ emissions \[[@pone.0220247.ref008], [@pone.0220247.ref009]\]. The Zebu (*Bos indicus*) animals, for example, are quite resistant and adaptable to tropical climates and, because of that, the Nellore is the most prevalent breed in Brazil. However, *Bos taurus* animals demonstrate greater yield potential, especially under appropriate conditions \[[@pone.0220247.ref010]\]. Thus, crossing breeds could be a viable alternative to improve the production rates of purebred cattle in tropical climates. Faster-growing animals can be more efficient in quantity of product produced, because they should theoretically partition relatively more feed nutrients into production. Thus the output of polluting excretion products on a per unit product basis should be less for these animals \[[@pone.0220247.ref011]\]. Due to the contribution of livestock in GHG, there is a strong motivation for the measurement of enteric CH~4~ to be accurately performed. Besides this, methane emission inventories are based on models developed in temperate climates and, therefore, precise methane measurements of tropical region production systems are crucial to reduce the uncertainties of these inventories and evaluate GHG mitigation strategies \[[@pone.0220247.ref012]\]. The objective of this trial was to examine the animal performance and enteric CH~4~ production, yield and intensity from two breed compositions in a Brazilian beef cattle production system--rearing in an integrated crop-livestock system but finished in a feedlot. Our hypothesis was that: *(i)* Performance of crossbred animals would be superior than Nellore in a Brazilian beef cattle production system; and (*ii*) CH~4~ yield and intensity would be lower for crossbred animals compared to Nellore. Materials and methods {#sec002} ===================== Treatments and experimental design {#sec003} ---------------------------------- The experiment was conducted at Brazilian Agricultural Research Corporation--Embrapa Maize & Sorghum (Sete Lagoas, Minas Gerais, Brazil; 19°28′S; 44°15′W, at 732 m altitude). Climate data for the experimental period was obtained at the meteorological station located at Embrapa and are presented in [S1 Fig](#pone.0220247.s001){ref-type="supplementary-material"}. All experimental procedures used in this experiment were approved by the Ethics Committee for Animal Use of Universidade Federal de Minas Gerais (UFMG, protocol number 326/2014). At trial onset, 10 mo old steers were divided into two groups according to their breed composition as follows: Nellore (171.5 ± 19.47 kg, n = 10), Angus x Nellore crossbred (214.2 ± 26.41 kg, n = 10) in the first year and Nellore (215.8 ± 32.34 kg, n = 25), Angus x Nellore crossbred (242.5 ± 32.26 kg, n = 25) in Year 2. Grazing management {#sec004} ------------------ The animals were evaluated in the rearing period, with initial age of 10 months, in the integrated crop-livestock (ICL) system under no-tillage system adopted since 2005. The pasture consisted of *Megathyrsus maximus* cv. Mombaça and the total pasture area of 5.5 hectares (ha) was subdivided into five sub-paddocks of approximately 1.1 ha each, used as a rotational grazing system with seven days grazing period and 28 days of rest. After each grazing cycle, the paddocks were fertilized with 150 kg of Nitrogen/ha. The experimental grazing period lasted 230 and 216 days in the first and second year, respectively. All animals were drenched with an anthelmintic prior to the start of grazing. The energetic-protein supplement ([Table 1](#pone.0220247.t001){ref-type="table"}) was offered *ad libitum* throughout the grazing period in a collective feeder. Supplement daily intake was estimated by dividing the total supplement consumed by the number of animals for each day in each period. As the supplement was offered *ad libitum* in a collective feeder, consumption might be different among animals related to self-intake regulations. 10.1371/journal.pone.0220247.t001 ###### Percentage of ingredients of the energy-protein mineral supplement used in pasture test and TMR diet used in feedlot. {#pone.0220247.t001g} ------------------------------------------------------------------------------------------------- Ingredients (%DM) Pasture Supplement Feedlot\ TMR diet ------------------------------------------------------------ -------------------- ---------- ---- **Corn Silage** \- \- 35 **Corn grain ground** \- \- 54 **Corn gluten meal** 84 86 \- **Soybean meal** 5 7 5 **Mineral Salt**[^**a**^](#t001fn001){ref-type="table-fn"} 11 7 6 ------------------------------------------------------------------------------------------------- ^a^Amounts of minerals (per kg of supplement): Year 1: phosphorus (P), 9 g; calcium (Ca), 20 g; sulfur (S), 16 g; magnesium (Mg), 2 g; sodium (Na), 37 g; zinc (Zn), 600 mg; copper (Cu), 150 mg; manganese (Mn), 140 mg; cobalt (Co), 20 mg; iodine (I), 17 mg; selenium (Se), 3 mg; iron (Fe), 100 mg. Year 2: P, 6 g; Ca, 20 g; S, 16 g; Mg, 1.4 g; Na, 9 g; Zn, 450 mg; Cu, 100 mg; Mn, 100 mg; Co, 14 mg; I, 12 mg; Se, 2 mg; Fe, 100 mg. Feedlot: P, 18 g; Ca, 50 g; S, 10 g; Mg, 20 g; Na, 30 g; Zn, 1303 mg; Cu, 375 mg; Fe, 500 mg; Mn, 520 mg; Co, 50 mg; I, 50 mg; Se, 9 mg; Fe, 500 mg; lasalocid sodium, 450 mg. Available herbage mass (AHM) was sampled within each paddock by cutting 5 randomly selected quadrats (1.0 m × 1.0 m) to ground level (5-cm stubble height) using hand shears before grazing. All collected herbage from each strip was collected, weighed and subsampled. A subsample (fresh weight) of the herbage sample from each quadrats was dried for 72 h at 65°C and was taken for subsequent chemical analysis. A further subsample was manually separated in leaf, stem, and dead content, and was dried for 72 h at 65°C. Leaves were used to characterize the composition of the forage ingested by the animals. It was decided to evaluate only leaf, since it represented almost all the forage sampled (above 60%). The forage allowance (kg dry matter \[DM\]/100 kg BW/day) was calculated by the ratio of forage production (kg DM/day) to total body weight of animals. In year one, there were 20 additional testers animals, that did not belong to the evaluated genetic groups and remained on pasture throughout all the experimental period. Feedlot management {#sec005} ------------------ In the feedlot, the animals were divided into groups according to the breed composition. The feedlot period began in June of each year, and the animals were allocated to collective pens measuring 20 x 12 m each and equipped with feed lanes and drinkers. The pens had enough space to ensure adequate animal well-being, with the minimal 18.5 m^2^ area per animal, observed in pens with 13 animals (year 2). All animals were drenched with an anthelmintic prior to the start of feedlot. The cattle were fed three times per day--at 0700, 1100 and 1600 h. The ration was adjusted daily to maintain 5 to 10% refusals. The amount of feed offered was recorded per pen, and refusals were weighed daily. Feed samples were taken monthly for chemical analysis. The animals were adapted to the experimental diets for 21 days. Initially, 60% corn silage and 40% concentrate diet were supplied, the amount of concentrate was increased until the ratio of roughage: concentrate was 35:65 (DM base). The diet was formulated to allow for 1.4 kg average daily weight gain \[[@pone.0220247.ref013]\] and consisted of corn silage, ground corn, soybean meal, and trace mineral mixture ([Table 1](#pone.0220247.t001){ref-type="table"}). A gain of 200 kg BW during the feedlot period was stipulated as the slaughter criterion. Animals remained in feedlot for 111 and 105 days (AN) and 138 and 127 days (NEL) in the first and second year, respectively. Animal performance was determined monthly by recording body weight (BW) following a fast of feed and water for 16 hours. The average daily gain (ADG) was calculated as the difference between the final body weight (FBW) and the initial body weight (IBW) of each period (grazing and feedlot), divided by the total number of days. On the day of slaughter, animals were weighed in the morning, before being sent to the slaughterhouse, where they were kept fasting for 24 hours with only *ad libitum* water intake. All the animals were slaughtered in a commercial slaughterhouse, according to the humanitarian procedures required by Brazilian legislation. The weight of hot carcass (WHC) was recorded immediately after the carcass was cleaned. Carcass yield (CY) was calculated by the ratio of WHC to FBW. The mean daily weight gain of carcass (ADGc) was calculated according to Eq ([1](#pone.0220247.e001){ref-type="disp-formula"}): $$ADGc = \frac{\left\lbrack {WHC - \left( {IBW \times 50\%} \right)} \right\rbrack}{days\ in\ feedlot}$$ Methane production measurement {#sec006} ------------------------------ Enteric CH~4~ emissions were measured using the sulfur hexafluoride (SF~6~) tracer technique reported by \[[@pone.0220247.ref014]\] and modified by \[[@pone.0220247.ref015]\] during three periods---feedlot in first year, grazing and feedlot in second year. Technical problems prevented the measurement of methane in the first year of grazing. Eight animals from both breed composition were evaluated in each period. Enteric CH~4~ emissions were measured for at least 3 days per animal. According to \[[@pone.0220247.ref016]\] a 3-days period is necessary to achieve an R of 0.70 for CH~4~ emissions by SF~6~ technique and the number of required animals to detect a difference of 20% in CH~4~ emissions among treatments is 6--8 animals per group. Ten days before the beginning of each measurement, a SF~6~ permeation tube was introduced directly into the rumen of each animal via the esophagus. The permeation rates were 4.44 ± 0.28; 4.60 ± 0.39 and 4.29 ± 0.06 mg/d (mean ± SD) in feedlot first year, grazing second year, and feedlot second year, respectively, as given by an 8-weeks calibration assay in a controlled environment at 39°C. Expired gases were collected with a sampling apparatus containing a collection canister made of polyvinyl chloride (PVC) equipped with a capillary tube (0.127 mm diameter). The capillary was calibrated to allow the vacuum inside the canister remaining at 40--60% of the initial vacuum after 24 h of measurement. If the pressure inside the canisters was below or above the 40--60% range, gas samples were not collected. Additionally, an identical set was used to collect background air samples at two points at the same time canisters were collected from animals. Canisters were removed daily at 0900 h, evacuated, and replaced then the contents were sampled. Animals were moved to a chute area for each canister evacuation, and total time to sample and replace canisters for all animals in both breed compositions groups was approximately 1 h. To collect enteric CH~4~ and SF~6~ samples, the canisters were vacuumed to approximately −12 PSI with vacuum pump. After the collection period, canisters were individually connected to a dilution system, and the final pressure was recorded. Nitrogen was then added slowly until canister pressure reached +13 PSI. Pressure readings were recorded to calculate the dilution factor \[[@pone.0220247.ref017]\]. After pressurization, the contents of the canisters were transferred under positive pressure to four pre-evacuated 20 mL Exetainers vials (Labco Limited, Lampeter, UK) for each animal. The collected respired air were analyzed immediately after the end of the experimental period. Analysis of CH~4~ and SF~6~ concentrations were determined by gas chromatography at the Laboratory of Gas Chromatography, Embrapa Dairy Cattle, in Juiz de Fora, Minas Gerais, Brazil. The SF~6~ (ppt) and CH~4~ (ppm) concentrations in the sampling canisters were determined using two separate gas chromatographs; models 6890 N plus and 7820A, respectively (Agilent Technologies, Santa Clara, CA). Both chromatographs were equipped with a split-splitless [injector](https://www.sciencedirect.com/topics/agricultural-and-biological-sciences/injectors), but a μECD detector (electron capture) was used to measure SF~6~ and a FID detector (flame ionization) was used to measure CH~4~ concentration. For SF~6~ analysis, a column (HP-Molsieve, Agilent Technologies, Santa Clara, CA) was used with N~2~ as carrier gas at a flow rate of 5.0 mL/min with N~2~ as the makeup gas at 40 mL/min, with μECD detector. The gas chromatograph was calibrated weekly using SF~6~ (White Martins, São Cristóvão, RJ) standards ranging in concentrations from 30, 100, 500, 1500, 3000 ppt. The CH~4~ was analyzed using two columns, (HP-Plot/Q and HP-Molsieve, Agilent Technologies, Santa Clara, CA) with H~2~ as carrier gas at a flow rate of 7.0 mL/min, with FID detector. The gas chromatograph was calibrated using CH~4~ (Linde AG, Rio de Janeiro, RJ) at 4.8, 9.7, 19.6, 102, 203 ppm. The CH~4~ emission rate (RCH~4~, g/d) for each animal was calculated using the SF~6~ and CH~4~ mixing ratio (μmol/mol) sampled by the canisters on the animals (SF~6~ and CH~4~ canister, respectively) and those used for background (SF~6~ and CH~4~ background, respectively), and the predetermined SF~6~ release rate (RSF~6~, g/d) from the permeation tubes, where molecular weights (MW) of the gases is MWCH~4~ = 16 and MWSF~6~ = 146, as described by \[[@pone.0220247.ref018]\], using Eq ([2](#pone.0220247.e002){ref-type="disp-formula"}): $${RCH}_{4} = {RSF}_{6} \times \left\lbrack \frac{\left( {{CH}_{4}canister - {CH}_{4}background} \right)}{\left( {{SF}_{6}canister - {CH}_{6}background} \right)} \right\rbrack \times \left\lbrack \frac{{MWCH}_{4}}{{MWSF}_{6}} \right\rbrack \times 100$$ Individual animal methane emissions were expressed as methane production (g CH~4~/animal/day, kg CH~4~/year, and kg CH~4~/period), methane yield (g CH~4~/kg DMI) and methane intensity (g CH~4~/kg BW), besides g CH~4~/kg ADG. Intake measurement {#sec007} ------------------ Individual DMI was determined for eight animals from each group in each period (grazing or feedlot, year 1 and 2), the same animals used for the methane measurement. Titanium dioxide (TiO~2~) was used as intake marker, and 10 g were administered to the animals once daily for 12 days during each period. TiO~2~ was stored in paper cartridges and introduced directly into the esophagus of the animals at 0900 h with the aid of a PVC applicator. Fecal samples were collected once daily during the last 5 days of the dosage period. Samples of feces corresponding to the different collection times composed a sample for each animal. Feces were dried at 65°C until constant weight. Dried feces were ground through a 1mm screen with a Wiley mill and analyzed by atomic absorption spectrophotometry. The TiO~2~ content was determined according to \[[@pone.0220247.ref019]\]. The standard curve was prepared using 2, 4, 6, 8 and 10 mg TiO~2~ and the spectrophotometer readings were recorded at a wavelength of 410 nm. For the calculation of fecal production (FP) estimated by TiO~2~, the following formula was used: $$FP = \frac{{TiO}_{2}supplied}{{{TiO}_{2}in\ feces}/{DM\ 105{^\circ}C}}$$ where FP = fecal production obtained by TiO~2~, g DM/day; TiO~2~ supplied = amount of TiO~2~ supplied to the animals per day (10 g); TiO~2~ in feces = percentage of titanium in feces, %; DM 105°C = the dry matter of feces at 105°C. Fecal Production and indigestible NDF (iNDF) were used to estimate dry matter intake (DMI, kg/day) for each animal. Indigestible NDF was used as the internal marker and obtained after *in situ* incubation of a diet (iNDF diet) and feces (iNDF feces) samples for 288 hours in the rumen of a cannulated bovine \[[@pone.0220247.ref013]\]. Follow equation ([Eq 4](#pone.0220247.e004){ref-type="disp-formula"}) was used for DMI: $$DMI = FP \times \left( \frac{iNDF\ feces}{iNDF\ diet} \right)$$ Average daily DMI during the methane measurement period and CH~4~ emission rate were used to calculate methane yield (g CH~4~/kg DMI). The average BW, ADG, DMI and feed and conversion efficiency were calculated over the same CH~4~ measurement period in both grazing and feedlot. Chemical analysis {#sec008} ----------------- Forage samples, supplements, diets, and refusals of foods were collected, oven-dried in a forced-ventilation oven at 65°C, for at least 72 hours, and ground in a Wiley mill (Alpax, Diadema, SP, Brazil) through a 1-mm sieve. The constituents were determined as described by \[[@pone.0220247.ref020]\], according to the following methods: dry matter (DM), 934.01; crude protein (CP), 984.13 (Leco FP-428, Australia Pty Ltd., Castle Hill, New South Wales, Australia); neutral detergent fiber (NDF), 2002.04; acid detergent fiber (ADF), 973.18; ether extract, 920.85; and ash (500°C furnace for 6 h), 938.08. Statistical analysis {#sec009} -------------------- To evaluate animal performance, a completely randomized design was used. Data for daily DMI were averaged per animal per 5-d period. The methane production data was averaged per animal per 3-d period minimum. Breed composition, year and the interaction between year and breed were included in the model, as a fixed effect. The distribution of model residuals was tested for normality using Shapiro-Wilk W test and for uniformity using the Cochran test. The mathematical model used was: Y~ijk~ = μ + B~i~ + Y~j~ (BY)~ij~ + ε~ijk~, in which: Y~ij~ is the observation of the animal k, from the breed i, in year j, μ is the mean effect; B~i~ is the fixed effect of the breed composition i, (i = 1, 2); Y~j~ is the fixed effect of the year j, (j = 1, 2); (BY)~ij~ is the interaction effect breed i and year j and ε~ijk~ is the random error associated with each animal. Statistical analysis was performed using PROC GLM from SAS software (version 9.2; SAS Inst. Inc., Cary, NC). Means were compared using the Fisher's test. Treatment differences were considered significant at *P*\<0.05. Results and discussion {#sec010} ====================== Grazing and feedlot diet characteristics {#sec011} ---------------------------------------- Forage production during the grazing period was satisfactory and corresponded with average herbage mass (AHM) of approximately 3,884 kg DM/ha. Stocking rate was higher in the second year (2880 versus 2025 kg BW/ha in the first year), and forage allowance (kg DM/100 kg BW) was 6.9 and 4.9 in the first and second year, respectively ([Table 2](#pone.0220247.t002){ref-type="table"}). 10.1371/journal.pone.0220247.t002 ###### Forage characteristics and productivity for grazing and feedlot system for each year in an intensive beef cattle production system. {#pone.0220247.t002g} System Item Year 1 Year 2 ----------------------------------- ----------------- -------- -------- **Grazing** N° animals 40 50 Days in grazing 230 216 Herbage Mass, kg DM/ha 3,824 3,944 Stocking Rate, kg BW/ha 2025 2880 Forage Allowance, kg DM/100 kg BW 6.9 4.9 Total Gain, kg/animal 166.7 156.3 Total Gain, kg BW 6660 7800 **Feedlot** Days in feedlot 125 116 Total Gain, kg/animal 175.8 189.2 Total Gain, kg BW 7020 9450 DM, dry matter; BW, body weight. Forage production and quality were not limiting allowing animals to achieve high gains during the grazing period (Tables [2](#pone.0220247.t002){ref-type="table"} and [3](#pone.0220247.t003){ref-type="table"}). Forage analysis was performed on the leaves only. Leaves are preferentially grazed by the cattle when the availability of forage was not limiting \[[@pone.0220247.ref021]\]. We assumed that the animals had the opportunity to select and eat high quality plant material with nutritional composition more similar to that found in the leaves which justifies the use of this type of forage sampling for analysis. 10.1371/journal.pone.0220247.t003 ###### Chemical composition of *Megathyrsus maximus* \'Mombaça\' pasture, of the supplement and of the TMR diet offered in the feedlot for the two breed compositions during experimental period. {#pone.0220247.t003g} Item Grazing Period[^a^](#t003fn001){ref-type="table-fn"} Feedlot Period ------------------------------------------------------ ------------------------------------------------------ ---------------- ------- ------- ------- ------- ------- ------- **DM, %** 25.49 86.78 27.8 90.29 59.94 60.09 57.94 58.61 **Ash**[^**b**^](#t003fn002){ref-type="table-fn"} 8.06 26.27 7.24 23.95 3.60 3.50 4.23 4.34 **OM**[^**b**^](#t003fn002){ref-type="table-fn"} 88.44 65.77 86.04 72.75 92.28 92.46 86.81 86.75 **CP**[^**b**^](#t003fn002){ref-type="table-fn"} 12.7 20.68 13.24 20.87 15.31 15.52 16.03 16.02 **EE**[^**b**^](#t003fn002){ref-type="table-fn"} 1.78 4.09 2.05 3.41 3.75 3.71 4.09 4.28 **NDF**[^**b**^](#t003fn002){ref-type="table-fn"} 64.34 27.23 67.01 28.74 26.40 25.97 27.48 27.30 **ADF**[^**b**^](#t003fn002){ref-type="table-fn"} 44.79 8.40 35.51 8.02 12.19 12.06 11.81 11.65 **Hem**[^**b**^](#t003fn002){ref-type="table-fn"} 34.48 18.83 35.56 20.72 14.21 13.99 15.47 15.66 **Cel**[^**b**^](#t003fn002){ref-type="table-fn"} 41.36 7.41 32.54 7.13 10.39 10.32 11.22 11.13 **Lignin**[^**b**^](#t003fn002){ref-type="table-fn"} 3.43 0.99 2.97 0.89 1.80 1.74 0.59 0.52 **CC**[^**b**^](#t003fn002){ref-type="table-fn"} 28.30 72.76 28.91 71.26 73.60 75.55 72.51 72.69 **P**[^**b**^](#t003fn002){ref-type="table-fn"} 0.22 0.88 0.20 0.84 0.33 0.33 0.37 0.37 **Ca**[^**b**^](#t003fn002){ref-type="table-fn"} 0.64 4.18 0.69 3.43 0.39 0.40 0.55 0.53 **TDN, %** 56.95 70.00 55.84 74.00 75.93 76.18 75.31 75.42 ^a^The grazing period was 1^st^ year-- 10/29/2015 to 06/15/2016 and 2^nd^ year-- 11/16/2016 to 06/20/2017. ^b^%DM; DM, dry matter; OM, Organic matter; CP, Crude protein; EE, Ethereal extract; NDF, Neutral detergent fiber; ADF, Acid detergent fiber; Hem, Hemicellulose; Cel, Celulose; CC, Cell content; P, Phosphorous; Ca, Calcium; TDN, Total digestible nutrients. The TDN was estimated using the formula recommended by \[[@pone.0220247.ref022]\]: TDN (%) = 83.790--4171 x NDF (forage) and TDN (%) = 91.0246 − 0.571588 x NDF (FL diet); NEL: Nellore; AN: Angus x Nellore crossbred. The high herbage availability and CP during the experimental period may have resulted from nitrogen fertilization and from the use of the ICL system. As the ICL system has been improved over the years, the stocking rate was higher than that obtained in previous study executed in the same area during 2013/2014 \[[@pone.0220247.ref023]\]. The stocking rate was 1093.5 and 1431.0 kg BW/ha in the dry and rainy seasons, respectively \[[@pone.0220247.ref023]\], which was lower than the present study during the rainy period (2880 and 2025 0 kg BW/ha in the first and second year, respectively). This difference was attributed to the greater number of animals used in the current study. The greater stocking rate maintained by our grazing strategy led to a more efficient forage utilization resulting in more and higher quality forage throughout the grazing season. \[[@pone.0220247.ref024]\] simulated scenarios for beef production in Brazil and indicated the best scenario was similar to the system in this study (Nellore and Nellore crosses animals in rearing phase in rotational grazing), and resulted in a lower stocking rate (1237.5 kg BW/ha), which attests the potential of ICL systems to increase the animals' production. Animal performance {#sec012} ------------------ There was an effect of the breed composition on the performance variables in both the grazing and feedlot period ([Table 4](#pone.0220247.t004){ref-type="table"}). Due to the inherent properties of hybrid vigor, the AN sired cattle began the trial with a greater BW in relation to NEL, despite them being comparably aged. 10.1371/journal.pone.0220247.t004 ###### Effects of breed composition on animal performance of beef cattle in grazing and feedlot tests (where NEL = Nellore, AN = Angus x Nellore crossbred). {#pone.0220247.t004g} ---------------------------------------------------------------------------------- NEL\ AN\ SEM P Value (n = 35) (n = 35) ------------------------- ---------- ---------- ------ --------- -------- -------- **Grazing** Initial Weight, kg 203.13 234.44 5.55 \<0.01 \<0.01 0.28 Final Weight, kg 351.71 404.41 7.94 \<0.01 \<0.05 0.15 Total Gain, kg 148.58 169.97 4.19 \<0.01 0.09 0.14 ADG, kg/animal/d 0.675 0.772 0.01 \<0.01 \>0.10 0.19 **Feedlot** Initial Weight, kg 337.74 418.38 6.40 \<0.01 \<0.01 \>0.10 Final Weight, kg 509.41 617.45 9.72 \<0.01 \<0.01 0.16 Total Gain, kg 171.67 199.07 4.88 \<0.01 \<0.05 \<0.05 ADG, kg/d 1.320 1.869 0.04 \<0.01 \<0.01 \<0.05 Carcass Weight, kg 284.23 352.43 5.79 \<0.01 \<0.05 0.10 Carcass Yield, % 55.79 57.08 0.27 \<0.01 \<0.01 0.18 Carcass ADG FL, kg/d 0.886 1.344 0.03 \<0.01 \<0.01 \<0.05 Carcass ADG Total, kg/d 0.521 0.721 0.01 \<0.01 0.06 \<0.05 ---------------------------------------------------------------------------------- ADG, average daily gain; FL, feedlot; SEM, standard error of the mean. Total gain and ADG in the grazing period were higher for AN (*P*\<0.01) and, consequently, they presented greater weight at the end of this period (*P*\<0.01) ([Table 4](#pone.0220247.t004){ref-type="table"}). The weight gains obtained in the current study were higher than those reported by \[[@pone.0220247.ref025]\] with similar levels of DMI. These authors evaluated Nellore animals (initial weight of 373 kg) in continuous grazing put-and-take stocking of *Urochloa brizantha* Stapf cv. Marandu and the animals obtained DMI of 5.93 kg/day and ADG of 0.447 kg/day, compared to 0.675 and 0.772 kg/d for Nel and AN cattle in the current study. Voluntary intake of forage was estimated by use of external and internal markers. The estimation of feed intake in pasture-raised animals continues to be costly and highly variable, despite advances in the experimental and analytical procedures over time \[[@pone.0220247.ref025]\]. However, in this study, the DMI values obtained for grazing animals are in accordance with \[[@pone.0220247.ref026]\] (5.90 kg DMI/d for NEL vs. 6.23 kg DMI/d for AN). These results show the capacity for improved animal production per area in ICL systems. Although the beef cattle sector in Brazil is still characterized by regions with low efficiency indexes \[[@pone.0220247.ref027]\], ICL systems could improve animal production and reduce environmental impacts from livestock in pasture-based beef production systems in the tropical regions. In the feedlot period, there was a significant difference between the two breed compositions for all variables evaluated. The AN animals had higher ADG and feed conversion than NEL but did enter the feedlot at a significantly higher weight. The AN animals reached the desired finishing weight in 111 and 105 d in the first and second year, respectively. The NEL animals, although they remained in the feedlot longer (138 and 127 days in feedlot in first and second year, respectively) had lower total weight gain (172 kg) compared to AN. Higher carcass weight were observed in AN animals when compared to NEL. The differences observed for carcass weight in this current study were related to differences in slaughter weight of the animals. Average finishing weights in the feedlot were similar to those reported in previous experiments using Angus cross and Nellore cattle \[[@pone.0220247.ref028], [@pone.0220247.ref029]\]. Breed composition had significant effect on carcass yield and carcass ADG (*P*\<0.01), with AN animals being greater than NEL. Carcass ADG in feedlot was 35% higher for AN than NEL, while carcass ADG total (considered throughout the experiment period) was 28% higher for AN. This observed increase in productivity results in fewer finished animals needed to produce a given quantity of meat \[[@pone.0220247.ref030]\], which may contribute to reducing the environmental impacts of beef production. Crossbred animals showed greater performance throughout the experimental period (total gain of 383 kg versus 306 kg for Nellore animals), but the growth rates reached by both breeds were satisfactory. High gains can be explained by the animals' physiological conditions (non-castrated) and age (up to 24 months old) \[[@pone.0220247.ref031], [@pone.0220247.ref032]\], beyond the effect of cross breeding animals alone \[[@pone.0220247.ref033], [@pone.0220247.ref034]\], in addition to the high concentrate diet in the finishing phase. Animal performance is not only a direct effect of the quality and quantity of the diet but also animal genetic potential \[[@pone.0220247.ref035], [@pone.0220247.ref036]\]. We observed that in appropriate conditions of feeding, AN animals obtain greater performance. Methane emissions {#sec013} ----------------- The effects of breed composition on methane emissions are presented in [Table 5](#pone.0220247.t005){ref-type="table"}. It is important to note that due to technical issues with the methane measurement, methane emission measurements during the grazing period were only performed in year 2. Despite this, the focus of our study is not the comparison between years and the design of the study and the statistical analysis allowed us to discuss these data without leading us to partially misleading conclusions. 10.1371/journal.pone.0220247.t005 ###### Effects of breed composition on methane emissions of beef cattle in grazing and feedlot tests (where NEL = Nellore, AN = Angus x Nellore crossbred). {#pone.0220247.t005g} --------------------------------------------------------------------------- NEL\ AN\ SEM P Value (n = 8) (n = 8) ------------------- --------- --------- ------- --------- -------- -------- **Grazing** Total DMI, kg/day 5.90 6.23 0.31 0.66 \- \- Forage DMI 4.78 5.11 1.44 0.66 \- \- Supplement DMI 1.12 1.12 \- \- \- \- BW average, kg 314.6 336.6 9.33 0.07 \- \- ADG, kg/day 0.680 0.729 0.03 0.22 \- \- Feed Conversion 8.98 8.81 0.50 \>0.10 \- \- Feed Efficiency 0.119 0.122 0.007 \>0.10 \- \- CH~4~, g/day 79.69 98.05 4.45 \<0.01 \- \- CH~4~, kg/year 29.08 35.78 1.62 \<0.01 \- \- CH~4~, kg/period 17.21 21.17 0.85 \<0.01 \- \- CH~4~, g/kg DMI 14.31 16.76 1.32 0.17 \- \- CH~4~, g/kg BW 0.24 0.28 0.05 \<0.01 \- \- CH~4~, g/kg ADG 119.53 140.03 8.09 0.07 \- \- **Feedlot** Total DMI, kg/day 9.29 12.44 0.39 \<0.01 0.10 \<0.01 BW average, kg 386.2 488.6 4.87 \<0.01 \<0.01 0.25 ADG kg/day 1.49 2.26 0.07 \<0.01 0.13 \<0.05 Feed Conversion 7.17 5.93 0.36 0.06 0.05 \<0.01 Feed Efficiency 0.167 0.193 0.009 0.09 \>0.10 \<0.01 CH~4~, g/day 168.72 209.84 7.78 \<0.01 \<0.01 \>0.10 CH~4~, kg/year 61.58 76.59 2.84 \<0.01 \<0.01 \>0.10 CH~4~, kg/period 22.34 22.67 0.98 \>0.10 0.05 \>0.10 CH~4~, g/kg DMI 18.52 17.83 0.89 \>0.10 \<0.05 \<0.05 CH~4~, g/kg BW 0.43 0.42 0.11 0.73 0.08 0.66 CH~4~, g/kg ADG 122.76 97.49 6.86 \<0.01 \>0.10 0.06 CH~4~, g/kg CW 0.079 0.067 0.10 \<0.01 0.16 0.52 CH~4~, g/kg ADGc 192.34 174.54 7.67 \<0.05 \<0.05 0.28 --------------------------------------------------------------------------- DMI, dry matter intake; BW, body weight; ADG, average daily gain; CH~4~, methane; CW, carcass weight; ADGc, ADG of carcass; SEM, standard error of the mean. The CH~4~ production (g/day and kg/year) were lower in NEL than AN animals in both grazing and feedlot systems (*P*\<0.01). Methane production emitted per period was calculated, according to grazing and feedlot days. The NEL emitted 19% less CH~4~ than AN in grazing, but no differences between breed composition in feedlot were observed ([Table 5](#pone.0220247.t005){ref-type="table"}). Even though AN animals showed higher methane emission (g/day), the total methane emission during the finishing phase was the same for both breed compositions, because these animals spent less time in feedlot. The intensification of beef cattle production systems leads to a reduction in emissions of GHGs per unit of product, and greater reductions may theoretically be possible if animals of higher performance were utilized \[[@pone.0220247.ref024]\], as confirmed in this study by the Angus x Nellore cattle. Methane production (g/day and kg/year) measured in grazing period were lower than those reported by \[[@pone.0220247.ref025]\]. The higher methane emission reported by these authors compared to the values obtained in this study may be attributed to the continuous grazing system used, where forage presents greater fiber content and therefore provides higher production of CH~4~. Compared to continuous grazing systems, multi paddock (MP) grazing can improve forage quality as well as forage production; thus, MP grazing is potentially a good option to reduce GHG emission \[[@pone.0220247.ref006]\]. According to \[[@pone.0220247.ref006]\], total GHG emissions could be reduced by as much as 30%, only by increasing forage quality and digestibility. Although differences were observed for methane production (g/day and/or kg/year) measured in feedlot between the breed composition, the mean of methane emission were similar to those reported by other studies \[[@pone.0220247.ref037]--[@pone.0220247.ref039]\]. Feedlot diets generally do not exhibit many discrepancies in nutritional composition, and therefore lower methane emissions variations are observed at that stage. It was found that there was no difference in DMI between breed compositions in pasture, but in the feedlot AN presented higher DMI than NEL (*P*\<0.01). Despite the difference in DMI, breed composition did not influence the CH~4~ yield (g CH~4~ per unit of DMI) neither in pasture nor in feedlot. The AN consumed more feed in feedlot, however when CH~4~ volumes were compensated for feed intake, there were no significant differences between breeds. Methane production expressed as g/kg DMI in the current study was similar to previously observed production rates in Nellore animals by \[[@pone.0220247.ref037]\] (17.1 g de CH~4~/kg DMI). DMI in our study was 2.4 and 2.5% BW for NEL and AN, respectively, and similarities in methane yield could be due to the similarity in intakes between the two breed compositions. Larger and fast-growing cattle will generally eat more and produce more enteric methane than smaller, slower-growing cattle under the same feeding regimen \[[@pone.0220247.ref040]\]. However, as shown in this trial, these animal makeup for higher daily CH~4~ production with improved performance. In the grazing period, no difference in BW was observed. As the ADG of the AN animals was higher than NEL in the grazing and feedlot period ([Table 4](#pone.0220247.t004){ref-type="table"}) the difference between the BW of the two breed compositions was higher in the feedlot. It was observed that there was no difference for BW during the grazing season, but significant differences for CH~4~ intensity (g CH~4~/kg BW) were detected with AN emitting more. However, methane intensity was similar between the breed compositions in the feedlot even with differences in BW (*P*\<0.01) for the animals in this finishing stage. Regarding CH~4~ per unit of ADG, no difference was observed between the two breed compositions in pasture (*P* = 0.07). In contrast, in feedlot the CH~4~/ADG or CH~4~/carcass ADG was significantly lower (*P*\<0.01) in AN than NEL animals. The AN animals were more efficient and obtained lower CH~4~ per ADG compare to NEL in feedlot. Previous studies did not support the hypothesis that an increase in feed efficiency decreases CH~4~ production \[[@pone.0220247.ref041], [@pone.0220247.ref042]\]. However, \[[@pone.0220247.ref043]\] showed that more efficient animals produce less enteric CH~4~ production than less efficient animals, especially when these animals are fed a high concentrate diet, which agrees with our results. Previous research has focused on the use of feedlots as a strategy to reduce CH~4~ emissions per kg of meat produced compared with grazing system. However, the majority of studies evaluated the continuous grazing management system and assumed steady-state soil carbon (C) to model the grass-finishing environmental impact \[[@pone.0220247.ref007]\]. Additionally, in these studies the ADG is generally below what can be achieved in well managed intensive pasture systems and because of this a substantial reduction in net GHG emissions may occur in pasture systems even when requiring double the land of feedlot systems, as a consequence of increased animal performance and sequestration. Important caveats to this hypothesis include potential soil carbon saturation and the permanence of the carbon sequestered. For instance, \[[@pone.0220247.ref044]\] estimated a 20-year window of improved soil carbon sequestration before reaching a dynamic equilibrium in the soil. However others have indicated periods of time from 90 to 300 years before reaching saturation \[[@pone.0220247.ref006], [@pone.0220247.ref045]\]. Concerning total methane production (adding the methane emitted in both grazing and feedlot), it was observed that the CH~4~ production was higher for AN animals compared to NEL (43.84 kg versus 39.55 kg). However, methane production per kg of carcass was 0.124 versus 0.139 for AN and NEL, respectively. These results suggest that the methane production of crossbred animals is compensated by better performance, resulting in lower CH~4~ per kg of meat produced, when this intensive production system is used in tropical climate conditions. Identifying efficient cattle breeds and adopting appropriate production systems is an important challenge for meat production worldwide with the growing concern about beef productions impact on the environment \[[@pone.0220247.ref046]\]. In addition to reducing enteric CH~4~ emissions/kg of meat produced, another advantage of intensification is associated with the reduction of the land area required to produce the same amount of product. This has the potential to reduce the degraded area and, in addition, contribute to the non-opening of new areas and mitigate future deforestation \[[@pone.0220247.ref047]\]. Although, the main disadvantage of the intensification is the need for substantial initial investment and working capital. The higher production of meat (kg produced/area) in the intensive tropical pasture system resulted in reduction in the unit cost and increase in the rate of return investment, despite the greater need for investment and cost per animal, when compared to the extensive systems \[[@pone.0220247.ref048]\]. Conclusions and implications {#sec014} ============================ The present study proposed to compare the GHG mitigation potential of two breed compositions in established Brazilian intensive beef cattle production system. Our data shows that CH~4~ intensity and CH~4~ emission per ADG might be altered depending of breed and diet composition, although breed composition did not influence the CH~4~ yield. In grazing, NEL presented less CH~4~ intensity, but AN animals in feedlot contributes to the reduction of methane per ADG. Overall, the AN animals were more efficient and had greater weight gain compared to Nellore, resulting in lower methane per kg of meat produced over the whole experimental period. The data generated could contribute to the development of methane mitigation policies, assuming standard systems that combines pasture use in the rearing phase and grain-based diet for finishing the animals. The integrated systems could enable high gains per unit of land, and feedlot finishing contributes to increased productivity of the whole system. Therefore, associating these two systems for beef cattle breeding in a tropical climate conditions with extensive pasture areas seems to be in line with new GHG reduction policy. Supporting information {#sec015} ====================== ###### Climate data for the experimental period from October 2015 to November 2017, measured at the Embrapa Maize and Sorghum Research Centre meteorological station, Sete Lagoas, MG, Brazil. (PDF) ###### Click here for additional data file. ###### Dataset used for statistical comparison of two breed composition. (XLSX) ###### Click here for additional data file. The authors are grateful to all support staff at Embrapa Maize & Sorghum, to Matsuda Indústria e Comércio Ltda. for providing the supplements, and all the researchers and technicians at Embrapa Dairy Cattle Research Centre for the help in CH~4~ analysis. [^1]: **Competing Interests:**Fabiano Alvim Barbosa is currently employed by De Heus Animal Nutrition. De Heus did not play any role in the study design, data collection and analysis, decision to publish, or preparation of the manuscript. De Heus did not provide any financial support. This does not change our adherence to PLOS ONE policies on data and materials sharing.

Background {#Sec1} ========== Cancer refers to a group of diseases characterized by uncontrolled growth and division of cells in the body, and is caused by environmental as well as genetic factors. Genetic factors include, but are not limited to inherited germline mutations, changing DNA methylation rate and microRNA modifications. Cancer is a leading cause of death in most countries. The number of new cases of cancer is 454.8 per 100,000 incidents per year and the number of cancer deaths is 171.2 per 100,000 incidents per year \[[@CR1]--[@CR4]\]. Accordingly, developing methods for detection and treatment of cancer is a main area of interest as well as a challenge. Several studies have been conducted to find genes that are involved in cancer development \[[@CR5]--[@CR8]\]. Even though there has been some degree of success in identifying genes that are strongly associated with cancer, much is yet to be done for discovering causal genes and variants. In addition, most of those studies disregard the position of those mutations, whereas mutations at different positions of a certain gene may lead to various levels of damage \[[@CR9], [@CR10]\]. Proteins are responsible for most cellular functions and their malfunction may undermine cellular performance \[[@CR11]\]. Only some of the mutations in coding regions, and not all of them lead to cancer. Therefore, distinguishing mutations with drastic impacts on protein functionality may help discriminate driver mutations from less significant ones. To this end, some researches have focused on mapping genomic positions to protein sequences and tried to distinguish mutations that affect the functionality of proteins \[[@CR10], [@CR12]\]. Protein domains are conserved regions of proteins that can fold and act independently \[[@CR13]\]. Therefore, it is plausible that mutations within these regions may cause more damage compared to other mutations \[[@CR13]\]. To this aim, some efforts have been made to study cancer mutations at the protein domain level. Nehrt et al. \[[@CR12]\] mapped non-synonymous somatic mutations of *Breast Invasive Carcinoma* and *Colon Adenocarcinoma Tumor* samples to their corresponding protein domains, in order to extract domains with significant mutation frequency. In another study by Yang et al. \[[@CR10]\], mutational protein domain hotspots for 21 different cancer types were determined by mapping somatic mutations to protein domains. Regions with high numbers of mutation for each cancer type were called hotspot. This study represents a method to explore protein domains with significant mutation frequencies, using whole exome sequencing data. Beside analyzing Pfam protein domains as sequence-based domains, CATH protein domains have also been studied as structure-based domains, which were not included in relevant studies to this date. Moreover, in order to more specifically pinpoint the domains of each cancer type, 29 different cancer types as well as pan-cancer were investigated in this study. In addition, the frequency of mutations in mitochondrial genes, stem cell-specific genes and DNA repair genes were examined. These sets of genes are likely to have important roles in cancer development and progression. Furthermore, the interconnectivity of proteins with mutation on causal domains was investigated. Methods {#Sec2} ======= Data extraction {#Sec3} --------------- Whole-exome sequencing data of 7685 cancer patients from 29 different cancer types containing 2,057,977 somatic mutations are downloaded from the TCGA (The Cancer Genome Atlas) data portal \[[@CR14]\]. The detailed list of cancer types as well as the number of patients for each type is shown in Table [1](#Tab1){ref-type="table"}. The names of downloaded files (in July 2015) for each cancer type is shown in Additional file [1](#MOESM1){ref-type="media"}: Table A1. The data are extracted from non-metastatic patients before giving radiotherapy or chemotherapy and are mapped to the human genome references of the GRCh37 \[[@CR15]\].Table 1Prevalence of patients, mutations and domain-specific mutations in different cancer typesCancer typeAbbrevationpatientsSomatic MutationsMutations on Protein Coding RegionsMutations on Pfam DomainsMutations on CATH DomainsAdrenocortical carcinomaACC45122,67921,35583351090Bladder Urothelial CarcinomaBLCA169169,165154,89369,72711,081Breast invasive carcinomaBRCA98398,88293,40540,3906512CholangiocarcinomaCHOL460859370303025484Colon adenocarcinomaCOAD412125,386123,91755,9138055Esophageal carcinomaESCA20279,53667,56629,2494340Glioblastoma multiformeGBM9220,93220,22599051692Head and Neck squamous cell carcinomaHNSC545151,456131,31459,7689301Kidney ChromophobeKICH178873980233425489Kidney renal clear cell carcinomaKIRC3651,23547,96421,0593404Kidney renal papillary cell carcinomaKIRP26933,24731,43313,5302074Brain Lower Grade GliomaLGG18247,28642,89019,7503489Liver hepatocellular carcinomaLIHC23089,04282,44336,7495470Lung adenocarcinomaLUAD275242,542216,730101,61613,522Lung squamous cell carcinomaLUSC17165,30663,98130,5424206Ovarian serous cystadenocarcinomaOV52412,75112,2145824998Pancreatic adenocarcinomaPAAD17582,87172,46229,2794223Pheochromocytoma and ParagangliomaPCPG425870676293115472Prostate adenocarcinomaPRAD11626,86224,09410,9541657Rectum adenocarcinomaREAD25434,26033,88515,7742313SarcomaSARC7583,01971,65030,0214526Skin Cutaneous MelanomaSKCM38753,08649,64523,3472993Stomach adenocarcinomaSTAD144223,884213,64296,34014,143Testicular Germ Cell TumorsTGCT35630,29326,17010,9021518Thyroid carcinomaTHCA12316,80715,38372451243ThymomaTHYM24839,46735,89215,6552379Uterine Corpus Endometrial CarcinomaUCEC66212,745204,82693,93713,959Uterine CarcinosarcomaUCS5714,21211,8035266876Uveal MelanomaUVM80498844112087313pan-cancer76852,057,9772E + 06852,729126,822 Since we are interested in discovering the role of protein domains in cancer, only protein-coding genes were selected among genes reported in TCGA data. Among 2,057,977 somatic mutations reported in this database, 1,896,875 of them occurred in protein-coding regions. Given that synonymous mutations have no effect on protein sequence and no demonstrable impact on phenotype \[[@CR16], [@CR17]\], in this study, only non-synonymous mutations within protein coding regions are considered. Protein domains can be defined in two different ways, either by their sequences or by their structural characteristics. Both these definitions are considered in this study in order to better understand the role of domains in cancer. Pfam \[[@CR18], [@CR19]\] and CATH \[[@CR20], [@CR21]\] databases are used to extract sequence-based and structure-based domains, respectively, and the UCSC (University of California Santa Cruz) \[[@CR22]\] tables and PDB \[[@CR23]\] database are exploited to extract the start and end positions of coding regions, exons, and more specifically, domains in genome. HUGO (HUman Genome Organization) \[[@CR24]\] standard gene nomenclature is employed to identify protein-coding genes. The number of protein-coding genes in HUGO is 19,011, all except for 22 of which are linked to PDB and Pfam entries, and 10,913 of them have Pfam domains. All predicted Pfam domains, without any constraints on *E*-value and bit-rate, were extracted in this study. A CATH domains was selected if it is represented by the same exact sequence in a UniProt record. To map CATH domains position form PDB residue to UniProt sequence, we used SIFTS \[[@CR25]\], which is a manually curated database to match the positions of PDB entries to UniProt sequences. Identification of candidate domains and genes {#Sec4} --------------------------------------------- Once data have been acquired and evaluated, the next step was to extract candidate regions by use of statistical analysis. Candidate regions (domains or genes) are regions in which mutations occur more frequently than expected. If mutations are mutually independent and uniformly distributed over the combined sequences of coding regions within human genome, then for each mutation, the probability of occurring on the *i* ^th^ coding region is *p* ~*i*~, which is equal to the length of *i* ^th^ coding region, *l* ~*i*~, divided by total length of coding regions, *L*, in the whole genome, that is, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {p}_i=\frac{l_i}{L} $$\end{document}$. To extract candidate regions at domain level, *l* ~*i*~ and *L* are respectively set to be the length of *i* ^th^ domain and the total length of domains in the whole proteome. Suppose that *n* is the number of mutations occurred in all protein-coding regions and *k* ~*i*~ is the number of mutations happened in the *i* ^th^ coding region, then the probability of having *k* ~*i*~ mutations on the *i* ^*th*^ coding region can be modeled by the binomial distribution, as follows \[[@CR26]\].$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathit{\Pr}\left(X={k}_i\right)=\left(\genfrac{}{}{0pt}{}{n}{k_i}\right){p}_i^{k_i}{\left(1-{p}_i\right)}^{\left(n-{k}_i\right)}=\left(\genfrac{}{}{0pt}{}{n}{k_i}\right){\left(\frac{l_i}{L}\right)}^{k_i}{\left(1-\frac{l_i}{L}\right)}^{\left(n-{k}_i\right)} $$\end{document}$$ To determine whether a protein-coding region is a potential candidate for a specific cancer, the number of observed mutations on that region is compared with what would be expected by the binomial distribution model, and a *p*-value threshold of 0.05 is adopted to test the null hypothesis \[[@CR27]\]. To this aim, for *k* observed mutations on each region, the hypothesis is rejected if *p*(*X* \< *k*) \> 0.95, where$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ P\left(X<k\right)=\sum \limits_{j=0}^{k-1}\mathit{\Pr}\left(X=j\right)=\sum \limits_{j=0}^{k-1}\left(\genfrac{}{}{0pt}{}{n}{j}\right){p}^j{\left(1-p\right)}^{n-j} $$\end{document}$$ Since multiple independent hypothesis tests are conducted in all cases, to maintain the family-wise error rate (FWER) \[[@CR27]\], a post hoc Bonferroni \[[@CR28]\] test is applied. Accordingly, when *m* independent hypothesis tests are performed, the criterion for rejecting the null hypothesis is divided by *m*. In other words, when the significance level for the whole family of tests is set to be 0.05, then with Bonferroni correction each individual hypothesis is evaluated at a significance level of$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \frac{0.05}{m} $$\end{document}$. To eliminate the possibility of overflow or underflow in computing values such as$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \left(\genfrac{}{}{0pt}{}{n}{k}\right) $$\end{document}$, $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\left(\frac{l}{L}\right)}^k $$\end{document}$ and $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ {\left(1-\frac{l}{L}\right)}^{\left(n-k\right)} $$\end{document}$, log Pr (*X* = *k*) is calculated instead of Pr(*X* = *k*). Accordingly, computations are performed using eq. [3](#Equ3){ref-type=""} instead of eq. [1](#Equ1){ref-type=""}:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathit{\log}\mathit{\Pr}\left(X=k\right)=\mathit{\log}\left(\genfrac{}{}{0pt}{}{n}{k}\right)+k\mathit{\log}p+\left(n-k\right)\mathit{\log}\left(1-p\right) $$\end{document}$$ In addition, to avoid numerical problems in computing $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \left(\genfrac{}{}{0pt}{}{n}{k}\right) $$\end{document}$ in eq. [3](#Equ3){ref-type=""}, Stirling's approximation \[[@CR29]\] is applied. Aims and objectives of the study {#Sec5} -------------------------------- There are more than 100 types of cancer \[[@CR30]\] and despite their differences, they present underlying biological (genetic) similarities. Pan-cancer study aims to uncover similarities and differences between various cancer types \[[@CR31]\]. With this background, all the data downloaded from different cancer types are assembled together in this study to form a pan-cancer dataset for further investigations. The main focus of this study is to assess the frequency of mutations on domain regions. However, we are also interested in evaluating the frequency of mutations in protein coding regions of three particular sets of genes: mitochondrial genes, stem cell-specific genes and DNA repair genes. Mitochondria are responsible for producing energy in almost all cell types and have their own DNA. Since mitochondrial DNA mutations are known to be highly associated with human cancer \[[@CR32]\], mutations within the mitochondrial genome are investigated in this study. Most of cancerous cells possess the classical characteristics of normal stem cells, including extensive capacity of self-renewal and acquired resistance to apoptosis \[[@CR33], [@CR34]\]. Therefore, genes responsible for the maintenance of stem cells are appropriate candidates for our goal. Mutation in genes that are associated with DNA repair function in a cell may induce partial loss of gene functionality \[[@CR35], [@CR36]\]. In this light, studying the presence of mutations in these genes may also be informative for cancer research. Results and discussion {#Sec6} ====================== This study covers four areas of assessment, namely, mutations in protein coding regions of mitochondrial, stem cell-specific and DNA repair genes, and mutations in protein domain regions. The results of each assessment are described in the following subsections. Mitochondrial genes {#Sec7} ------------------- Several studies have reported the presence of somatic mitochondrial mutations in cancer cells. Even though many of these studies have demonstrated the role of mitochondrial mutations in human cancers such as *Kidney Renal Cell Carcinoma* \[[@CR37]\], *Breast Invasive Carcinoma* \[[@CR38]\], *Gastric Carcinoma* \[[@CR39]\], *Prostate Adenocarcinoma* \[[@CR40]\], *Ovarian Carcinoma* \[[@CR41]\] and *Thyroid Carcinoma* \[[@CR42]\], such an association was not identified in all relevant studies. For instance, studies on *Glioblastoma Multiforme* \[[@CR43]\] and *Liver Hepatocellular Carcinoma* \[[@CR44]\] have not been able to pinpoint the role of mitochondrial mutations in cancer. In this light, it is worthwhile to further investigate the role of mitochondria in cancer development. To examine the role of mitochondria in cancer development, the observed somatic mutations in all mitochondrial genes are studied. There are 37 different genes in mitochondrial DNA, which are assigned to six groups of complexes, based on their roles (shown in Additional file [1](#MOESM1){ref-type="media"}: Table A2). For instance, the genes MT-RNR1 and MT-RNR2, which are responsible for making rRNAs, are assigned to rRNA complex \[[@CR45]\]. Mutations within each group are identified to better understand its role in developing cancer. To identify mitochondrial candidate genes associated with each of the 29 cancer types as well as pan-cancer (30 cancer types in total), the statistical analysis is performed on two levels. In the first level of analysis, each mitochondrial gene is considered individually, while in the second level, genes are analyzed in their group (complex) context. Accordingly, in the Bonferroni correction, the parameter *m* is set to 37 × 30 and 6 × 30 for the first and second level, respectively. With a corrected *p*-value threshold of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \frac{0.05}{m} $$\end{document}$, there are 13 cancer types and pan-cancer (shown in Table [2](#Tab2){ref-type="table"}) for which at least one mitochondrial candidate gene or complex is identified. In Table [2](#Tab2){ref-type="table"}, the number in parentheses next to each gene shows the percentage of patients for which this gene is mutated. All in all, nine mitochondrial genes have been identified as candidate ones: MT-CO2, MT-CYB, MT-ND1, MT-ND5, MT-RNR1, MT-RNR2, MT-TL1, MT-TT and MT-TV. Additional file [1](#MOESM1){ref-type="media"}: Table A3 shows the number of patients with mitochondrial mutations and the number of mutations for each.Table 2Candidate mitochondrial genes and complexes for each cancer typeCancer TypeGenes (Percentage)Complexes (Percentage)ACCMT-TL1 (32.6)-BRCAMT-RNR1 (3.6), MT-RNR2 (5.6), MT-TT (0.8)rRNA (8.4)CHOLMT-RNR2 (33.3)rRNA (33.3)ESCAMT-CYB (13.7)COMPLEX III (13.7)HNSCMT-ND5 (1)-KICHMT-ND1 (13.6)-LIHCMT-ND5 (12.9)COMPLEX I (18.3)LUAD-COMPLEX III (1.1)PRADMT-RNR2 (1.9)-SARCMT-RNR2 (12.6), MT-CO2 (7.1)rRNA (14.2), COMPLEX IV (15.7)TGCT-COMPLEX III (6.3)THYMMT-RNR2 (27.6)rRNA (27.6)UCECMT-RNR1 (11.7), MT-RNR2 (11.7), MT-TV (12)rRNA (0.41.7)Pan CancerMT-RNR2, MT-CYBrRNA, COMPLEX III Among six mitochondrial groups of complexes, ATP synthesis and tRNA complexes have not been chosen as candidate for any cancer type. In particular, no significant mutation was observed in genes MT-ATP6 and MT-ATP8 in any of the cancer types. This result is consistent with the assumption that more energy is required for rapid reproduction in cancerous cells. The results also show that two mitochondrial genes, namely MT-RNR2 and MT-CYB are significantly mutated in pan-cancer. Stem cell-specific genes {#Sec8} ------------------------ Researches have pointed out a number of similarities between stem cells and cancer cells, including their self-renewal potential and their ability to migrate to other regions of the body \[[@CR46]--[@CR48]\]. Moreover, the ability of stem cells to differentiate into various types of cells increases the risk of malignant transformations. Accordingly, stem cell-specific gene analysis is expected to provide a foundation for better understanding of their role in cancer. The stem cell-specific gene set studied in this research (shown in Additional file [1](#MOESM1){ref-type="media"}: Table A4), which is first identified by Palmer et al. \[[@CR49]\], contains 182 protein-coding genes. To extract candidate stem cell-specific genes associated with each of the 29 cancer types as well as with pan-cancer, statistical analysis was performed and subsequently, in the Bonferroni correction, the parameter *m* was set to 182 × 30. With a corrected *p*-value threshold of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \frac{0.05}{182\times 30} $$\end{document}$, 57 stem cell-specific genes were selected as candidates for at least one cancer type. The most significant genes among them are CHEK2 and KMT2C, which are associated with 20 and 18 different cancer types respectively. The other genes are related to at most seven types. Given that some researches have already demonstrated the role of CHEK2 \[[@CR50], [@CR51]\] and KMT2C \[[@CR52]\] in different cancer types, their identified association with a large number of cancer types is not surprising. *Rectum Adenocarcinoma* and *Lung Squamous Cell Carcinoma* cancers are the only cancer types for which no candidate stem cell-specific gene has been identified. In Table [3](#Tab3){ref-type="table"}, the list of candidate stem cell-specific genes for each cancer type is shown. Similar to Table [2](#Tab2){ref-type="table"}, the numbers in this table also show the percentage of patients in which the genes are mutated.Table 3Candidate stem cell-specific genes for each cancer typeCancer Type(Percentage)Genes(Percentage)ACC (56.5)HDAC2 (5.4), ERCC2 (20.7), GARS (38.0), PRR34 (8.7)BLCA (30.3)CHEK2 (6.1), ERCC2 (9.7),KMT2C (20.9)BRCA (8.2)KMT2C (6.9), PILRB (0.8), HLA_DRB5 (0.7)CHOL (33.3)CHEK2 (8.3),KMT2C (25),GIMAP8 (2.8)COAD (1.5)HLA_DPA1 (1.5),ESCA (9.3)NREP (2.2),BRINP1 (7.7)GBM (4.4)CHEK2 (1.8),TSHZ2 (2.5)HNSC (20.4)CHEK2 (3.8), LIN28B (1.5), BRINP1 (3.1), KMT2C (12.0), NPR3 (2.7)KICH.21.2)DIMT1 (1.5),KMT2C (13.6),HLA_DRB5 (7.6), HLA_DQA1 (3)LIHC (5)HTR7 (5)KIRC (11.1)DNMT3B (3.1), CHEK2 (2.2), RRAS2 (1.8), NREP (0.7), TNFSF10 (1.3), FYB (2.9),\ SMARCC2 (3.1), RCSD1 (2), HLA_DRB5 (1.8)KIRP (7.1)CHEK2 (5.9), DPH3(1.2)LGG (9.3)CHEK2(3.9),HDAC2(1.7),ZBTB20(4.6)LUAD (42.0)SPDL1 (1.8), CHEK2 (7.2), TRPC4 (7.2), CDH6 (7.2), GIMAP1 (2.2), KMT2C (17.8), PILRB (2.2), TSHZ2 (6.8), NPR3 (4.6), FYB (5.5)OV (2.6)BOD1 (0.9), HAS2 (1.7)PAAD (57.3)CHEK2 (17.0), BBS9 (9.4), GARS (5.8), SLC24A1 (9.4), KMT2C (17), SMARCC2 (13.5), NPR3 (8.8), AFTPH (13.5)PCPG (14.3)CHEK2 (5.1), NUSAP1 (4.0), KMT2C (5.1),\ HLA_DRB5(1.1)PRAD (8.9)CHEK2(3.5),KMT2C(5.4)SARC (4.3)ZNF788 (2.8), BRINP1 (2.0)SKCM (37.3)GDF3 (8.0), CCDC90B (4.0), CDH6 (10.7), KMT2C (16.0), GIMAP5 (6.7) ,GIMAP7 (6.7),GIMAP1 (6.7),GIMAP8 (12)STAD (32.3)CHEK2 (5.4), SOHLH2 (4.4), BRINP1 (5.9), KMT2C (16.5), TSHZ2(7.2),ZBTB20(9)TGCT (2.8)C10orf128 (2.1), HLA_DRB5 (2.1), HLA_DQA1 (1.4)THCA (4.8)CHEK2 (1.4), GDF3 (0.8), RIOK2 (0.8), HLA_DRB5 (1.7)THYM (5.7)CHEK2 (5.7)UCEC (9.7)ATP11C (9.7)UCS(15.8)CHEK2(7),KMT2C(10.)UVM(13.8)CHEK2(7.5),NUSAP1(5),HLA_DRB5(5)Pan Cancer(24.3)CHEK2 (4.1), SOHLH2 (1), BRINP1 (2.2), TRPC4 (2.4), CDH6 (2.1), KMT2C (10.6), PILRB (0.8)\ ,HLA_DRB5 (1.1), TSHZ2 (2.6), NPR3 (1.7),GIMAP1 (0.9), GIMAP8(1.7),ZBTB20(2.1) DNA repair genes {#Sec9} ---------------- DNA repair genes are responsible for recognizing and correcting damages in the replication of DNA. Hence, mutations in DNA repair genes can be expected to alter the efficiency of repairing mechanism, which in turn can be associated with severe health issues such as cancer. Moreover, it has been reported that DNA repair genes are frequently mutated in cancer \[[@CR53]\]. Accordingly, studying mutations within DNA repair genes may be helpful for revealing their role in cancer. To identify DNA repair genes which are associated with a certain type of cancer, a statistical analysis similar to that performed in previous subsections was applied. 174 known DNA repair genes, reported in \[[@CR54]--[@CR56]\], are shown in Additional file [1](#MOESM1){ref-type="media"}: Table A5. By setting the parameter *m* to 174 × 30 in the Bonferroni correction, 27 DNA repair genes were identified as candidate for at least one cancer type. The results show that the most significant DNA repair gene is TP53, which was identified as candidate for 25 cancer types as well as for pan-cancer. This conforms with the previous findings about the crucial role of TP53 mutations in cancer development \[[@CR57], [@CR58]\]. This further endorses the reliability of the other results in this study. For each cancer type other than *Testicular Germ Cell Tumors*, at least one candidate DNA repair gene was identified. In particular, *Pancreatic Adenocarcinoma* has eight candidate DNA repair genes, and ATM, TCG, TP53 and CHEK2 are the candidate DNA repair genes for pan-cancer. Table [4](#Tab4){ref-type="table"} shows the candidate DNA repair genes for each cancer type and the number next to each gene shows the percentage of patients in which this gene is mutated.Table 4Candidate DNA repair genes for each cancer typeCancer Type(Percentage)Genes(Percentage)ACC (37)MSH3(6.5),TP53(19.6),ERCC2(20.7)BLCA (63.6)ATM (13.6), TP53 (49.8) ,ERCC2 (9.7), CHEK2 (6.1)BRCA (33.4)TP53 (33.4)CHOL (22.2)TP53 (13.9), CHEK2 (8.3)COAD (52)TP53 (52.0)ESCA (87.9)TP53 (87.9)GBM (28.7)TP53 (28.7)HNSC (71.6)TP53 (71.2), CHEK2 (3.8)KICH (33.3)TP53 (33.3)KIRC(10.9)FANCE (4.4), DDB1 (4.9), RPA1 (2.2), TP53 (4.2), CHEK2 (2.2)KIRP (17.2)OGG1 (2.4), MSH3 (4.1), TDG (3.6), TP53 (4.1), CHEK2 (5.9)LGG (50.4)TP53 (48), CHEK2 (3.9)LIHC (32.2)TP53(32.2)LUAD (57.8)ERCC5 (3.3), TP53 (54.7), CHEK2 (7.2)LUSC (79.2)TP53 (79.2)OV (84.8)TP53 (84.8)UCEC (34.7)MSH4 (7.3), TP53 (29)PAAD (84.2)ERCC3 (8.8), XPC (9.9), WRN (14.6), TDG (9.9), FAN1 (9.9), EME2(11.1),\ TP53 (67.3), CHEK2 (17)PCPG (17.7)FANCD2 (5.1), ERCC8 (1.1), TDG (7.4), CHEK2 (5.1)PRAD (19.3)ATM (4.5), TP53 (10.8), POLI (1.4), CHEK2 (3.5)READ (67.2)TP53 (67.2)SARC(37.0)TP53 (37)SKCM(22.7)BLM (6.7), MPG (4.0), TP53 (10.7), CHEK2 (09.3)STAD (57.9)UVSSA (4.4), SLX4 (6.7), TP53 (49.9), CHEK2 (5.4)THCA (4.5)SMUG1 (0.8), TDG (2.2), TP53 (0.8), CHEK2 (1.4)THYM (5.7)CHEK2 (5.7)UCS (91.2)TP53 (91.2)UVM(20)FANCD2 (6.3), CCNH (2.5), TDG (3.7), CHEK2(7.5)Pan Cancer(45.2)ATM(5.5),TDG(1.7),TP53 (39.1),CHEK2 (4.0) To identify cancer-associated genes within mitochondrial, stem cell-specific and DNA repair genes, not only the mutations on domain regions but all those on full protein coding regions are included in the assessment. To be more confident in extracting cancer-associated genes within each biological process, its related candidate genes were restricted to those which also contain at least one candidate domain. Upon studying the mitochondrial genes, we found no candidate domains (defined in the following sections) associated with those genes. Among candidate stem cell-specific genes, 51% and 46% of them contain at least one Pfam and one CATH candidate domain, respectively, as shown in Fig. [1a and b](#Fig1){ref-type="fig"}. For each cancer type, the entire list of stem cell-specific genes with Pfam and CATH candidate domains are presented in Additional file [1](#MOESM1){ref-type="media"}: Tables A6 and A7. Similarly, 25% and 26% of candidate repair genes consist of at least one Pfam and one CATH candidate domain, respectively, as shown in Fig. [1c and d](#Fig1){ref-type="fig"}. More details on repair genes with Pfam and CATH domains are given in Additional file [1](#MOESM1){ref-type="media"}: Tables A8 and A9.Fig. 1Comparison of candidate genes and genes with candidate domains. (**a**) Comparison of stem cell genes and genes with Pfam candidate domains. (**b**) Comparison of stem cell genes and genes with CATH candidate domains. (**c**) Comparison of DNA repair genes and genes with Pfam candidate domains. (**d**) Comparison of DNA repair genes and genes with CATH candidate domains CATH candidate domains {#Sec10} ---------------------- A key objective of this study is to identify CATH candidate domains, which have gone unnoticed in the previous researches conducted in this field. There are 759 CATH-reported domains which are located in 2993 human proteins. Detailed information for each CATH-reported domain can be found in Additional file [1](#MOESM1){ref-type="media"}: Table A10. In addition, the position of each CATH domain on each protein-coding gene is available in Additional file [2](#MOESM2){ref-type="media"}: Table B1. To assess CATH domains, the significance level of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \frac{0.05}{30\times 759} $$\end{document}$ was used. The results indicate that each cancer type has a number of associated CATH candidate domains ranging from 1 to 19, while pan-cancer analysis reveals 93 related CATH candidate domains. Some domains seemed to not be associated with any individual cancer type, yet they were identified as significant candidates in the pan-cancer study. We say a candidate domain "covers" a particular patient, if the patient has at least one mutation in that specific candidate domain. Surveying the results, we realize that each CATH candidate domain of each cancer type covers various percentages of patients in that cancer type, ranging from 0.02% to 95%. Moreover, all CATH candidate domains of each cancer type cover 28% to 98% of patients of that cancer type. The CATH candidate domains identified for *Breast Invasive Carcinoma*, *Ovarian Serous Cyst Adenocarcinoma* and pan-cancer are presented in Table [5](#Tab5){ref-type="table"}. The number next to each domain shows the percentage of patients which are covered by this domain. Additional file [1](#MOESM1){ref-type="media"}: Table A11 shows CATH candidate domains in each cancer type. To assess the statistical significance of an identified candidate domain, the percentage of patients covered by that domain can theoretically be used as a selection attribute.Table 5Candidate domains for Breast Invasive Carcinoma and Ovarian Serous CystadenocarcinomaCancer Type (Percentage)CATH Domains (Percentage)BRCA (77.3)1.10.1070.11 (33.88), 1.10.220.60 (0.31), 1.10.437.10 (1.73), 1.10.510.10 (25.84), 2.170.260.10 (0.71), 2.40.250.10 (2.03), 2.60.200.10 (1.83), 2.60.40.10 (20.24), 2.60.40.1110 (4.27), 2.60.40.60 (4.48), 2.60.40.720 (33.27)4.10.365.10 (0.71)OV (80)1.10.287.650 (0.87), 2.60.40.720 (80.00), 3.30.450.40 (0.87)Pan Cancer (91.5)1.10.10.10 (7.90), 1.10.10.440 (0.92), 1.10.10.60 (4.23), 1.10.101.10 (1.59), 1.10.1070.11 (15.52), 1.10.1300.10 (5.87), 1.10.1380.10 (2.34), 1.10.150.210 (0.78), 1.10.150.50 (3.03), 1.10.150.60 (1.30), 1.10.1520.10 (0.55), 1.10.1540.10 (1.47), 1.10.167.10 (3.70), 1.10.246.10 (2.17), 1.10.287.450 (0.94), 1.10.437.10 (2.25), 1.10.472.10 (4.68), 1.10.490.10 (2.46), 1.10.506.10 (0.78), 1.10.510.10 (44.89), 1.10.555.10 (3.85), 1.10.565.10 (10.70), 1.10.630.10 (10.98), 1.10.640.10 (0.98), 1.10.720.50 (0.64), 1.10.750.10 (3.32), 1.10.800.10 (1.60), 1.20.1050.10 (4.98), 1.20.1250.10 (5.37), 1.20.1260.10 (1.29), 1.20.1280.50 (1.17), 1.20.1340.10 (1.61), 1.20.245.10 (0.95), 1.20.5.100 (1.17), 1.20.5.110 (0.48), 1.20.5.50 (1.86), 1.20.58.60 (2.32), 1.20.82.10 (1.01), 1.20.920.10 (6.57), 1.20.930.40 (4.19), 1.25.10.10 (8.64), 1.25.40.20 (8.26), 2.10.220.10 (7.52), 2.10.25.10 (6.69), 2.10.310.10 (0.46), 2.10.60.10 (1.76), 2.10.70.10 (6.31), 2.130.10.10 (7.43), 2.120.10.80 (1.91), 2.140.10.30 (3.99), 2.130.10.130 (3.15), 2.170.270.10 (4.18), 2.170.8.10 (1.20), 2.30.30.190 (1.21), 2.30.39.10 (8.76), 2.30.42.10 (8.87), 2.40.128.20 (4.23), 2.40.20.10 (1.94), 2.40.250.10 (0.38), 2.40.50.40 (4.42), 2.60.120.200 (4.42), 2.60.120.260 (4.65), 2.60.20.10 (2.49), 2.60.200.10 (3.99), 2.60.210.10 (2.78), 2.60.40.10 (33.88), 2.60.40.1110 (6.69), 2.60.40.1120 (1.57), 2.60.40.60 (2.32), 2.60.40.720 (36.55), 2.60.60.20 (1.59), 2.70.98.20 (2.38), 2.80.10.50 (5.22), 3.10.100.10 (5.87), 3.10.20.230 (0.94), 3.10.200.10 (3.60), 3.10.50.10 (2.64), 3.10.620.10 (0.44), 3.20.20.100 (4.55), 3.20.20.140 (4.49), 3.30.1370.10 (2.34), 3.30.1490.20 (2.13), 3.30.300.30 (1.70), 3.30.450.40 (0.48), 3.30.450.50 (0.87), 3.30.70.1230 (0.88), 3.30.70.1470 (0.91), 3.30.70.330 (12.00), 3.30.800.10 (1.98), 3.30.9.10 (0.77), 3.40.190.10 (3.68), 3.40.50.10140 (1.54), 3.40.470.10 (1.13), 3.40.50.10190 (4.22), 3.40.50.1370 (0.75), 3.40.50.2300 (1.51), 3.40.50.300 (25.67), 3.40.718.10 (5.78), 3.90.1170.10 (0.43), 3.90.1230.10 (1.63), 3.90.190.10 (13.32), 4.10.280.10 (1.08), 4.10.365.10 (0.34), 4.10.75.10 (0.72) Pfam candidate domains {#Sec11} ---------------------- There are 6009 predicted Pfam domains located in 17,722 human proteins. Detailed information for Pfam domains can be found in Additional file [1](#MOESM1){ref-type="media"}: Table A12. In addition, the position of each Pfam domain on each protein-coding gene is available in Additional file [2](#MOESM2){ref-type="media"}: Table B2. The significance level of $\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \frac{0.05}{30\times 6009} $$\end{document}$ was used to perform statistical assessment, the results of which show that each cancer type has a different number of Pfam candidate domains, ranging from 3 to 93. For pan-cancer, the number of identified Pfam candidate domains is 202, which indicates a large number of domains are significant to pan-cancer but not to individual cancer types. The results are consistent with those of CATH domains. Each Pfam Candidate domain of each cancer type covers different percentages of patients with a minimum of 0.2% and a maximum of 98%. Overall, all Pfam candidate domains of each cancer type cover 74% to 100% of patients of that cancer type. Table [6](#Tab6){ref-type="table"} shows Pfam candidate domains of *Breast Invasive Carcinoma* and *Ovarian Serous Cyst Adenocarcinoma* and the number next to each domain shows the percentage of patients, which are covered by this domain. Additional file [1](#MOESM1){ref-type="media"}: Table A13 shows Pfam candidate domains in each cancer type. Similar to CATH candidate domains, the percentage of patients covered by a candidate domain can be used as a proper measure. For instance, P53 and tm_4 cover the first and the second highest percentages (42% and 28%) of *Breast invasive carcinoma* patients, respectively, which shows their significant role in this particular cancer.Table 6Pfam candidate domains for Breast Invasive Carcinoma and Ovarian Serous CystadenocarcinomaCancer Type (Percentage)Pfam Domains (Percentage)BRCA (78.5)7tm_4 (41.51), ATP-synt_A (0.81), Atrophin-1 (2.85), CBF_beta (2.24), COX1 (2.03), COX3 (1.12), Cadherin (23.91), Cytochrom_B\_N_2 (1.12), DUF4647 (1.32), FAM219A (0.51), FRG1 (1.32), GATA (3.97), G_path_suppress (1.12), H-K_ATPase_N (0.31), Histone (7.32), NADH5_C (0.92), NADHdh (1.63), Oxidored_q4 (0.71), Oxidored_q5_N (0.81), P53 (28.48), P53_tetramer (2.03), PI3K_C2 (2.54), PI3K_P85_iSH2 (2.34), PI3K_p85B (1.02), PI3Ka (13.22), PTEN_C2 (2.03), Proton_antipo_M (3.56), Runt (2.44), T-box (4.17), TMEM247 (1.12), Tis11B_N (1.02)OV (88.3)7tm_4 (49.57), DUF2462 (0.43), DUF4552 (1.30), MRP-S32 (0.87), NtCtMGAM_N (2.17), ODAM (1.30), P53 (72.61), P53_tetramer (4.78), PTCRA (0.87), Sam68-YY (1.30), UPF0054 (0.87) The statistical analysis conducted in this study is different to that used by Nehrt et al. \[[@CR12]\]. Moreover, different data sources were exploited in these two studies. Therefore, it is no surprise that the results of the two studies are dissimilar. To further emphasize the difference between these approaches, we remark that the number of Pfam domains examined in our study is much larger than that of Nehrt et al. \[[@CR12]\] due to the cut-off used in that study for minimum protein or domain length (150 amino acids) and due to Pfam *E*-value threshold used for inclusion (0.001). The comprehensive comparison performed over Pfam and CATH regions (discussed in the next section) clearly indicates the high reliability of Pfam-reported domains, regardless of their associated *E*-values. Furthermore, 5918 out of 6009 investigated Pfam domains have *E*-value less than threshold of 0.001. Also, among 769 identified Pfam candidate domains, 754 (98%) satisfy the threshold condition. Accordingly, we decided not to exclude any Pfam-reported domain. In addition, the statistical method used by Nehrt et al. \[[@CR12]\], is extremely sensitive to the number of patients having mutations within the domain region of each protein. This is due to the fact that the number of mutations in each domain is normalized by the cumulative length of all its associated proteins, wherein at least one patient had mutation. Hence, if a new patient with a mutation on an associated protein is added, for which no previous mutation is reported, this would significantly impact the normalizing factor, and subsequently, the statistic used. Moreover, the threshold level of 0.1 is applied in Nehrt et al. \[[@CR12]\] for determining significantly mutated domains, by using local false discovery rate (LFDR). As shown in Fig. [2](#Fig2){ref-type="fig"}, Nehrt et al. \[[@CR12]\] reported 41 and 45 Pfam domains as significantly mutated in *Breast Invasive Carcinoma* and *Colon Adenocarcinoma Tumor*, respectively, while our results identified 31 Pfam candidate domains for *Breast Invasive Carcinoma* and 35 ones for *Colon Adenocarcinoma Tumor*. Comparing the results of the two studies shows that they share nine domains for *Breast Invasive Carcinoma* including CBF_beta, FRG1, GATA, P53, PI3K_p85B, PI3Ka, PTEN_C2, T-box and Tis11B_N. Moreover, the four domains of APC_crr, MH2, P53 and PI3K_p85B are reported by both studies for *Colon Adenocarcinoma Tumor*.Fig. 2The comparison between our study and Nehrt et al. \[[@CR13]\] In another study by Yang et al. \[[@CR10]\] mutations were obtained from COSMIC database \[[@CR59]\] and the analysis was restricted to potentially damaging missense mutations, predicted by IntOGen-mutation platform. To determine significantly mutated domains in a given cancer type, Fisher's exact test was exploited in that study. Accordingly, the results obtained by Yang et al. \[[@CR10]\] are different from those of this study, as expected. The list of cancer types investigated in Yang et al. \[11\] and those considered in this study share 13 in common. For each of these 13 cancer types, significantly mutated domains obtained by both studies are shown in Table [7](#Tab7){ref-type="table"}. Based on these two studies, seven cancer types share P53 as one of their significant domains.Table 7Shared significant domains in our study and Yang et al. \[[@CR11]\]Cancer typeShared domainsBRCACytochrom_B\_CP53COADMH2ESCAP53HNSCP53KIRCBromodomainOxidored_q3VHLLIHCP53LUADP53Pkinase_TyrRasSushiLUSCP53SushiOVP53PAADRasPRADMATHSKCMPkinase_TyrUCECDSPcNADHdhPI3K_p85BPI3Ka CATH vs. Pfam protein domains {#Sec12} ----------------------------- There is a gap between the number of sequenced proteins and that of proteins with known structure, which can also be observed at the level of protein domains. On the other hand, structure-based protein domains are biologically more informative and reliable. Therefore, to benefit from the high number of sequence-based protein domains as well as from the accuracy of structure-based protein domains, both sequence-based and structure-based domains are studied in this research. CATH and Pfam databases are used to extract structure-based and sequence-based domains, respectively. Through further investigation, for each protein which has both Pfam and CATH annotations (2974 proteins), the overlap between its Pfam domain region and CATH domain region is computed. For instance, as it is shown in Fig. [3a](#Fig3){ref-type="fig"}, for gene VPS25 which contains two homologous domain superfamilies with CATH IDs 1.10.10.10 (amino acids 102--176) and 1.10.10.570 (amino acids 1--176) as well as one Pfam domain of ESCRT-II (amino acids 10--145), the computed overlap is from amino acid 10 to 145. This overlap covers 77% of CATH domain region and 100% of Pfam domain region. Overall, for all 2974 proteins with both Pfam and CATH annotations, computed overlaps cover 79% of CATH domain regions and 75% of Pfam domain regions, on average, as shown in Fig. [3b](#Fig3){ref-type="fig"}. This suggests that for a protein with no annotation in CATH database, it is reasonable to study its Pfam domain region as a representative of its functional unit.Fig. 3The overlap between Pfam domain region and CATH domain region. (**a**) The overlap between Pfam domain region and CATH domain region for gene VPS25. (**b**) The average overlap between Pfam domain regions and CATH domain regions In addition, the percentage of patients in each cancer type, which are covered by Pfam candidate domains are compared with the ones covered by CATH candidate domains (shown in Fig. [4](#Fig4){ref-type="fig"}). As it is shown in Fig. [4](#Fig4){ref-type="fig"}, for several cancer types including *Bladder Urothelial Carcinoma*, *Breast Invasive Carcinoma*, *Uterine Corpus Endometrial Carcinoma* and *Uterine Carcinosarcoma*, Pfam and CATH candidate domains cover the same percentage of patients, while in some other types such as *Adrenocortical Carcinoma*, there is a huge gap between the two. The considerably high level of overlap between Pfam and CATH domain regions suggest that wherever CATH candidate domains are incapable of covering patients, Pfam candidate domains are suitable substitutions.Fig. 4CATH vs. Pfam candidate domain coverage for patients Among 6009 investigated Pfam domains, 769 are identified as candidate domains in at least one cancer type. Candidate domains are observed to be significantly mutated in varying numbers of cancer types (more details are given in Additional file [1](#MOESM1){ref-type="media"}: Table A14). To assess the contribution of each candidate domain in different types of cancer, the list of 769 candidate domains were sorted in decreasing order based on the number of associated cancer types. The 17 top-listed domains, presented in Additional file [1](#MOESM1){ref-type="media"}: Table A15, are found to be the least number of candidates that each studied cancer type is associated with at least three candidate domains within them. Given that P53 is one of the most commonly mutated domains in all cancers, it is no surprise that it is placed at the top of the list, above other domains. The second domain in the sorted list, tm_4, is identified as a candidate domain for 22 cancer types and for pan-cancer. The tm_4 domain, which is present in a large number of proteins (376), has not previously been implicated in cancer susceptibility, hence can be seen as a newly found candidate. Proteins of keratin family contain six domains, all except Keratin_assoc are found to be candidate in different numbers of cancer types, ranging from 6 to 17. Interestingly, three of keratin-related domains (Keratin_B2, Keratin_B2_2 and Keratin_2\_tail) are placed in our list of top 17 domains. The great contribution of keratin-related domains to cancer may be due to their role in protecting epithelial cells from damage or stress \[[@CR60]\]. Similar investigations performed on CATH domains show that among 759 CATH domains, 181 are identified as candidate ones. Detailed information on their associated cancer types are given in Additional file [1](#MOESM1){ref-type="media"}: Table A16. Going through the sorted list of CATH candidate domains shows that the 15 top-listed domains, presented in Additional file [1](#MOESM1){ref-type="media"}: Table A17, are found to be the least number of candidates that each studied cancer type is associated with at least one candidate domain within them. Besides, this study sheds some light on the role of domains in cancer. For instance, there are in total 181 CATH and 769 Pfam candidate domains associated to at least one cancer type or to pan-cancer. 94% of Pfam domains and 95% of CATH domains have mutations in more than 95% of their corresponding proteins. However, a high percentage of proteins with mutations on a particular domain does not necessarily imply that domain as a significant candidate. As an example, Pkinase is a domain involved in 348 proteins, for which the number of occurrences on those proteins is 369. Based on the data available, the total number of mutations on this domain in different cancer types is 346, yet it is not identified as a candidate domain for any cancer type. In contrast to Pkinase, Phostensin_N is a domain which is located on two different proteins with mutations on only one of them, yet it is identified as a candidate domain for *Pancreatic Adenocarcinoma*. Further investigation was performed to determine specific domains of each cancer type. As an example, there are 30 Pfam candidate domains for *Breast Invasive Carcinoma* among which four are only specific to this cancer type, including Atrophin-1, CBF_beta, GATA and Runt. The entire list of specific domains for all cancer types is available in Additional file [1](#MOESM1){ref-type="media"}: Table A18. Subsequently, a protein is termed "Pfam specific protein" to a cancer type, if it contains only those Pfam candidate domains which are specific to the cancer of interest. Additional file [1](#MOESM1){ref-type="media"}: Table A19 shows all proteins with mutations on Pfam candidate domains and Additional file [1](#MOESM1){ref-type="media"}: Table A20 presents Pfam specific proteins of all cancer types. Similar tables (Additional file [1](#MOESM1){ref-type="media"}: Tables A21, A22 and A23) are provided for CATH candidate domains. For each cancer type, the number of Pfam candidate domains, proteins with mutations on Pfam candidate domain, Pfam specific domains and specific proteins are summarized in Table [8](#Tab8){ref-type="table"}. Table [9](#Tab9){ref-type="table"} shows similar results for CATH candidate domains. On average, the number of CATH specific domains for a cancer type is 2.4, whereas this number is 15.8 for Pfam specific domains.Table 8Number of Pfam Candidate domains, proteins with candidate domains and specific domains and proteins for each cancer typeCancer typeNumber of Pfam candidate domainsNumber of Proteins with Pfam candidate domainsNumber of Pfam specific candidate domainsNumber of Pfam specific proteinsACC573232347BLCA5110761564BRCA31495411CHOL39451921COAD3595567ESCA2535545GBM2447468HNSC42743542KICH552002123KIRC9049454157KIRP36842244LGG25366315LIHC2543259LUAD115232827248LUSC31112858OV1118866PAAD14379370151PCPG701512044PRAD343531320READ292661017SARC25383711SKCM4011161125STAD5214761035TGCT663662943THCA383041519THYM184155UCEC231094316UCS44571927UVM63992234Pan-cancer222368527315 Table 9Number of CATH Candidate domains, proteins with candidate domains and specific domains and proteins for each cancer typeCancer typeNumber of CATH candidate domainsNumber of proteins with CATH candidate domainsNumber of CATH specific candidate domainsNumber of CATH specific proteinsACC71545BLCA2264911BRCA1231512CHOL93924COAD1637700ESCA1319122GBM92800HNSC1437200KICH101867KIRC194933KIRP11110810LGG72112LIHC1430724LUAD32590324LUSC1115400OV3612PAAD20249627PCPG95633PRAD1311345READ711600SARC828512SKCM17281516STAD2657526TGCT97835THCA59113THYM5146211UCEC2644945UCS112234UVM61322Pan-cancer1044561848 To evaluate the reliability of the proposed methodology, the following process was performed. For each cancer type, the genes that contain identified candidate domains and have mutations in that cancer were compared with experimentally verified cancer genes from COSMIC \[[@CR59]\]. A list of 616 unique genes for which mutations have been causally implicated in cancer, were downloaded from COSMIC database (in May 2017). Since some of the cancer causal genes are associated to more than one cancer type, the cumulative number (counting with repetition) of genes, for all 29 cancer types investigated in this study tallied 967. Considering all 29 cancer types (shown in Fig. [5](#Fig5){ref-type="fig"}), 814 COSMIC genes have mutations on Pfam domains, out of which, 413 genes contain at least one Pfam candidate domain. Therefore, among COSMIC genes with at least one Pfam domain, 51% have Pfam candidate domains. For each cancer type, the list of its specific COSMIC genes which also have Pfam candidate domain is given in Additional file [1](#MOESM1){ref-type="media"}: Table A24.Fig. 5Comparison of causal genes in COSMIC with genes having candidate domains. (**a**) Comparison of COSMIC genes with genes having Pfam candidate domains. (**b**) Comparison of COSMIC genes with genes having CATH candidate domains Similar analysis for CATH domains indicates that among 967 cancer causal genes, there are 446 genes with mutations on their CATH domains, out of which, 289 genes have mutations on CATH candidate domains. Detailed information for each cancer type is presented in Additional file [1](#MOESM1){ref-type="media"}: Table A25. Even though 52% of the cancer causal genes reported by COSMIC have no CATH domains, among those with at least one CATH domain, 65% have the candidate ones. This suggests that CATH domains are superior indicators compared to Pfam domains, in identification of cancer causal genes. Curiously, studying mutations at domain level provides an interesting insight into the role of gene families in cancer development. For instance, in *Uterine Corpus Endometrial Carcinoma*, eight out of nine genes in TCEAL gene family have mutations on identified Pfam candidate domains. Members of this family had already been identified as nuclear phosphoproteins that modulate transcription in a promoter context-dependent manner \[[@CR61]\]. Besides, for 19 out of 22 genes of FGF gene family, mutations occur on Pfam candidate domains in *Stomach adenocarcinoma*. FGF's are known to play a key role in the processes of proliferation and differentiation of a wide variety of cells and tissues \[[@CR62]\]. We also evaluated the impact of domain mutations in cancer using SnpEff \[[@CR63]\], which annotates the effects of variants on genes and classifies them as low, moderate and high impact. The results indicate that 13.4% of mutations on Pfam and CATH domains are classified as high impact mutations and the rest are reported as moderate ones. Restricting the analysis to candidate domains shows that 14.3% and 17.7% of mutations on Pfam and CATH candidate domains are considered as high impact ones, respectively. Moreover, we calculated the frequency of different types of variants, including nucleotide variants, insertions and deletions, on domain regions. Our analysis reveals that all variants has similar frequencies on domains and candidate domains. Protein-protein interactions {#Sec13} ---------------------------- Protein-Protein Interactions (PPIs) commonly refer to physical contact between two or more proteins \[[@CR64]\] and offer a wealth of molecular information, which exists in various molecular pathways. Besides, domains are usually responsible for mediating protein-protein interactions \[[@CR65]\]. Understanding specific interaction map of a disease is becoming increasingly crucial in elucidating its underlying molecular mechanisms. Therefore, it is reasonable to study PPI for proteins with candidate domains. To this end, the specific proteins of each cancer type are exploited to determine specific PPIs of that cancer type. The interactome database STRING \[[@CR66]\] was used to extract all interactions with the highest confidence score (0.9 and above) among specific proteins of each cancer type. The results show that in 20 out of 29 cancer types and also in pan-cancer, specific proteins are significantly connected. The connectivity significance of proteins was determined by the PPI enrichment *p*-value, reported by STRING. This measure quantifies whether the set of input proteins have more interactions among themselves than a random set of proteins of similar size. Subsequently, for each cancer type in which a significant interaction network is not formed on its specific proteins, the same analysis was performed on the entire list of proteins with mutation on Pfam candidate domains of that cancer type. The entire lists of proteins are significantly interconnected for all cancer types other than *Ovarian serous cystadenocarcinoma*. Table [10](#Tab10){ref-type="table"} gives some information about the number of nodes and edges in the interaction networks of each cancer type, as well as their enrichment *p*-values. Furthermore, for each cancer type the interactome analysis was performed on the entire list of proteins with mutation on CATH candidate domains of that cancer type. The results show that in all cancer types except *Uveal Melanoma*, proteins with mutations on CATH candidate domains are significantly connected. Some information including the number of nodes and edges in each interaction network, and its enrichment *p*-value is presented in Table [11](#Tab11){ref-type="table"}. Due to the small number of proteins in the CATH specific lists, interactome analysis was not performed on them.Table 10The result of interactome analysis for Pfam candidate domains in different cancer typesCancer typePfam specific proteinsProteins with Pfam candidate domain\#Nodes\#EdgesExpected number of edgesPPI enrichment *p*-valueSignificant or Not\#Nodes\#EdgesExpected number of edgesPPI enrichment *p*-valueSignificant or NotBRCA161450.002YESCOAD7100.04YESGBM8100.126NO47126875580YESKIRC15640164E-07YESKIRP433030YESLGG14207E-04YESLUAD24728290YESOV6001NO186310.0738NOUCEC16107E-04YESACC46310.069NO319113140YESBLCA641421E-07YESCHOL21402E-04YESESCA5100.002YESHNSC422510YESKICH23120.79NO198105330YESLIHC9100.016YESLUSC8001NO1132924225830YESPAAD15159216E-12YESPCPG43510.01YESPRAD20712E-06YESREAD17401E-04YESSARC11302E-05YESSKCM252247E-11YESSTAD356910YESTGCT42813E-06YESTHCA19100.149NO301132290YESTHYM5001NO403200YESUCS26110.462NO554580YESUVM34330.609NO9822140.039YESPan-cancer312231340YES Table 11The result of interactome analysis for CATH candidate domains in different cancer typesProteins with at least one CATH candidate domainCancer type\#Nods\#EdgesExpected number of edgesPPI enrichment *p*-valueSignificant or notBRCA31511362610*YES*COAD37713253630*YES*GBM286190*YES*KIRC4937131.62E-08*YES*KIRP110199570*YES*LGG212642.55E-13*YES*LUAD58821736520*YES*OV6300.000256*YES*UCEC44921596710*YES*ACC15100.152*YES*BLCA648306714870*YES*CHOL392164.84E-06*YES*ESCA1917031800*YES*HNSC37013223660*YES*KICH181527.81E-10*YES*LIHC30710652920*YES*LUSC154177400*YES*PAAD2496102310*YES*PCPG5635105.99E-10*YES*PRAD113249720*YES*READ116285680*YES*SARC2859022230*YES*SKCM2814981450*YES*STAD57521817620*YES*TGCT789190*YES*THCA912481360*YES*THYM146411830*YES*UCS221560.00156*YES*UVM13310.186***NO***Pan-cancer45718216100*YES* Overall, there are 4968 proteins with mutations on Pfam candidate domains, each of which is associated with 4 cancer types on average. About 31% of them are specific to only one cancer type and more than 7% of them (353 proteins) are linked to at least half of cancer types. Given these figures, one can expect the specific proteins to be not significantly interconnected in some cancer types, even though their entire list of proteins with mutations on candidate domains form highly connected networks. Additional file [1](#MOESM1){ref-type="media"}: Table A26 presents the number of associated cancer types for each protein. To have a more comprehensive picture of CATH candidate domains, similar results for 1379 proteins with mutations on CATH candidate domains are given in Additional file [1](#MOESM1){ref-type="media"}: Table A27. Website {#Sec14} ------- As previously noted, data integration is an essential requirement for this study and related fields of research. Several data sources were used to incorporate different types of information for human protein-coding genes, including gene symbols and protein identifiers from multiple resources, the start and end positions of each CATH and Pfam domain within a given protein, and the positions of exons and introns in the human genome. This comprehensive data integration provides researchers with a unified data source, which can be accessed via <http://www.cancerouspdomains.ir>, as well as from Additional file [2](#MOESM2){ref-type="media"}: Table B3. All previously mentioned tables in Additional file [1](#MOESM1){ref-type="media"} are also downloadable from the website. An example of using the embedded search engine to extract the integrated information for gene TP53 is depicted in Fig. [6](#Fig6){ref-type="fig"}.Fig. 6The result of searching TP53 in website Some of the highly beneficial data provided on the website are the five graph charts, which show the associations between different groups of genes or domains to various cancer types. These graph charts are represented by bipartite graphs in which one set of nodes corresponds to the genes/domains and the other set of nodes corresponds to cancer type. In each bipartite graph, an edge connects a gene/domain to a cancer type, if this gene/domain is identified as a candidate in the cancer type. For instance, one set of nodes represents Pfam candidate domains and the other set represents different cancer types. An edge connects two nodes in these sets, if the corresponding domain is a candidate for the corresponding cancer type. The graph chart of Pfam candidate domains is illustrated in Fig. [7](#Fig7){ref-type="fig"}.Fig. 7The graph chart for Pfam candidate domains A help file is also provided for detailed description of the information embedded in the website. Some tailored interface options such as "moving", "zoom in" and "zoom out" are available to control the size of the display. These options are specifically helpful due to the large number of CATH and Pfam candidate domains. Furthermore, by clicking on each cancer type, all related candidate domain or genes can be distinguished via a change of edge colors. The same option is also provided for each candidate domain or gene. Conclusion {#Sec15} ========== Distinguishing mutations on protein-coding regions which impact the functionality of coded proteins, is one of the main obstacles in the study of cancer genomics. The exome-wide study of somatic mutations for several cancer types and the available structural information of proteins provide invaluable resources for specifically studying cancer genomics at functional level. Both sequence-based and structure-based domains, which are available in Pfam and CATH databases respectively, are promising representatives of functional regions within proteins. Accordingly, extracting domain regions which are significantly mutated in a cancer type reveals critical information required for validating the impact of mutations. In this paper, a comprehensive investigation is conducted on all 29 TCGA cancer types and pan-cancer to identify sequence-based and structure-based domains in which mutations occur with high statistical significance. The domains identified for each cancer type offer an explanation for the functional impact of mutations in that cancer type. It is shown that each cancer type has its specific set of candidate domains, which in turn suggests a specific set of proteins associated to that cancer type. The interactome analysis showed that the specific proteins of each cancer type are significantly connected. This interconnectivity supports the idea that leveraging domain regions can improve accurate identification of functionally related causal proteins. Through further investigation, the frequency of mutations in mitochondrial genes, stem cell-specific genes and DNA repair genes was determined to examine their role in cancer development and progression. Additionally, this study provides other researchers with a comprehensive and unified data repository for studying both CATH and Pfam domain regions on protein-coding genes. Moreover, it has clarified the associations between different groups of genes or domains and various cancer types. All this information is accessible via our website. Additional files ================ {#Sec16} Additional file 1:It contains 27 supplementary tables, Table A1--27 with the data described in the text. (XLSX 1032 kb) Additional file 2:It contains three tables, Tables B1-B3 with the data gathered from different datasets. All these data are used in study. (XLSX 55782 kb) HUGO : HUman Genome Organization TCGA : The Cancer Genome Atlas UCSC : University of California Santa Cruz **Electronic supplementary material** The online version of this article (doi:10.1186/s12859-017-1779-5) contains supplementary material, which is available to authorized users. The authors would like to thank Farshad Jafari for helping to design the website. In addition, the authors wish to express their gratitude to the anonymous reviewers, whose constructive comments improved the manuscript. Funding {#FPar1} ======= Not applicable. Availability of data and materials {#FPar2} ================================== All information is accessible via our well-designed website <http://www.cancerouspdomains.ir>. Initial idea of the research was proposed by SH, AND, AMB, AJ and ZR. The framework was designed, implemented, and tested by SH, AND, AJ, and ZR. All authors participated in designing the structure and organization of the manuscript equally. All authors read and approved the final manuscript. Ethics approval and consent to participate {#FPar3} ========================================== Not applicable. Consent for publication {#FPar4} ======================= Not applicable. Competing interests {#FPar5} =================== The authors declare that they have no competing interests. Publisher's Note {#FPar6} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Cerebral vasculitis (CV) is one important cause of stroke, especially in younger adults, and it remains difficult to differentiate this condition from other causes of ischemic brain injury.^[@R1][@R2][@R3]^ Gold standard for the diagnosis of CV is a biopsy of an affected brain vessel, but a negative result does not rule out the disease. Other diagnostic tools such as MRI, magnetic resonance angiography, or CSF analysis have a high sensitivity (close to 100%) but low specificity (around 30%).^[@R2]^ Due to the different therapy approaches, it is crucial to distinguish CV from other causes of cerebral ischemia or inflammatory brain disease. Proinflammatory cytokines such as interferon-γ (IFN-γ) and interleukin-17 (IL-17) are potent mediators of chemokine release, activation of endothelial surfaces, and recruitment of inflammatory cells. In particular, IL-17 is known to play an important role in the pathogenesis of anti--neutrophil cytoplasmic autoantibody--associated vasculitis (AAV) and giant cell vasculitis.^[@R4]^ Here, we focused on the production of IFN-γ and IL-17 by T lymphocytes in the CSF of patients with a cerebral ischemic event as a potential biomarker for cerebral vasculitis. Our primary research goal was to determine whether IL-17 could serve as a biomarker for cerebral vasculitis within an unselected sample of patients presenting with stroke-like symptoms. METHODS {#s1} ======= Patients and diagnostic criteria. {#s1-1} --------------------------------- Blood and CSF were obtained from patients consecutively admitted to the University Medical Center Hamburg-Eppendorf with stroke-like symptoms between 2013 and 2015 ([table 1](#T1){ref-type="table"}). These participants were divided into patients with a diagnosis of primary angiitis of the CNS (PACNS) according to the Calabrese diagnostic criteria^[@R2]^ (n = 6), secondary angiitis of the CNS (SACNS, n = 3), giant cell vasculitis (GCV, n = 3), and ischemic stroke where vasculitis was ruled out (n = 15). Classification of the patients was done by senior clinicians (C.G., T.M.) blinded to the results of the laboratory test. Finally, 6 patients with noninflammatory neurologic disease from whom CSF was taken during diagnostic procedures were included in the study as controls ([table 1](#T1){ref-type="table"}). ###### Participant characteristics   Patient consents. {#s1-2} ----------------- All procedures were approved by the local ethics committee and informed consent was obtained from all patients or their legal representatives. Expansion of CSF cells. {#s1-3} ----------------------- After centrifugation of the CSF, cells were stimulated with 2.5 μg/mL phytohemagglutinin in RPMI1640 medium containing 10% human serum, antibiotics, and in the presence of irradiated (36 Gy) feeder cells. IL-2 (equivalent to 20 U/mL) was added to the cultures every 3--4 days together with 50% fresh medium. After 2 weeks, cells were harvested and used directly for cytokine determination or restimulated for further expansion. In about 10% of the samples, expansion was not successful. Flow cytometry. {#s1-4} --------------- For intracellular cytokine detection, expanded CSF cells were stimulated with 250 ng/mL PMA and 1 μg/mL ionomycin. A total of 100 μL heparin-blood was stimulated in parallel with 5-fold concentrations of PMA and ionomycin. Cells and blood were cultured for 5 hours in serum-free medium (X-Vivo15) in the presence of 10 μg/mL Brefeldin A, as described.^[@R5]^ Cells were subsequently harvested, permeabilized, and stained for lineage markers CD3, CD4, CD8, and TCRγδ, and with anti-IFN-γ and anti-IL-17. Dead cells were discriminated using LIVE/DEAD stain (Invitrogen, Carlsbad, CA). In approximately 10% of the patients, immunocytochemistry from the peripheral blood failed due to unsuccessful stimulation. Samples were analyzed in a LSR Fortessa (BD Biosciences, East Rutherford, NJ) flow cytometer and results were evaluated using FlowJo software. Cytokine determination in serum and CSF. {#s1-5} ---------------------------------------- Multiplex determination of IL-17, IFN-γ, IL-1β, IL-6, tumor necrosis factor--α, and IL-12/23 was performed using the high-resolution Meso Scale Discovery (MSD) (Rockville, MD) platform. Statistical analysis. {#s1-6} --------------------- The primary research question of this study was the determination of sensitivity and specificity of IL-17 as a biomarker for cerebral vasculitis within an unselected sample of patients presenting with stroke-like symptoms in order to provide Class II evidence for IL-17 as a biomarker for cerebral vasculitis. For classification of the IL-17 values, a cutoff of 7% was used since it was located at the middle of an interval that separated normal and increased values of IL-17. The samples were analyzed retrospectively. We used SPSS (Chicago, IL) version 22.0 for statistical calculations. Paired Student *t* test was used to compare peripheral blood and CSF data. To compare multiple groups, we used analysis of variance. Differences were considered significant if *p* ≤ 0.05. RESULTS {#s2} ======= Considering the inflammatory nature of cerebral vasculitis, and in search of putative biomarkers, we performed a phenotypical characterization of peripheral immune cells in our cohort. We could not find any differences in the frequency of all immune cell subsets or in the expression of activation or maturation markers (not shown). To assess lymphocytes close to the target organ, we expanded cells from the CSF until enough cells were available for cell surface phenotyping and intracellular cytokine analysis, typically 2--4 weeks. After expansion, in all patient groups CD4+ cells were the dominant population, while the rest were CD8+ and Tγδ lymphocytes. We observed a higher percentage of CD4+ cells producing IL-17 exclusively in patients with cerebral vasculitis ([figure 1](#F1){ref-type="fig"}). Interestingly, all samples with CSF cells analyzed of patients with PACNS and SACNS groups showed high levels of IL-17-producing CD4+ cells (average 8.8%, range 7.3%--10.5%, for PACNS; average 10.4%, range 8.9%--11.9%, for SACNS). In both cases, comparison to the IL-17 production by CSF cells in stroke patients was highly significant (*p* \< 0.001). The GCV patients showed rather low IL-17 production (1.2%, range 0.5%--2.1%), comparable to the other groups (2.5%, range 0.14%--5.4% in stroke patients; and 3.6%, range 1.9%--5.7% in the control group). With respect to the primary research question, we set a cutoff at 7% and considered stroke patients as reference cohort and the PACNS, SACNS, and GCV as case cohort ([table 2](#T2){ref-type="table"}). We found that an increased proportion of IL-17-producing CD4+ cells in CSF accurately identifies patients with cerebral vasculitis with a sensitivity of 73% (95% CI 39%--94%) and a specificity of 100% (95% CI 74%--100%). {#F1} ###### Estimates of sensitivity and specificity of IL-17 producing CD4+ cells in the CSF (cutoff 7%) for differentiating stroke due to cerebral vasculitis from other ischemic strokes  We also assessed soluble IL-17 in the CSF of our patients using the highly sensitive MSD platform. All values were under detection levels, except for one patient where CSF was taken shortly before he died of a fulminant vasculitis (PACNS5; data not shown). In this patient, cells were not available for analysis. In contrast to IL-17, IFN-γ production by expanded CSF cells was very high, up to 90% of the cells, but similar in all groups ([figure 1](#F1){ref-type="fig"}). We found no differences in cytokine production by peripheral CD4, CD8, or Tγδ cells ([figure 1](#F1){ref-type="fig"} and not shown). In one case (patient SACNS1), we had enough CSF cells to directly analyze cytokine production ex vivo, without expansion. We found that CD4+ cells also produced high amounts of IL-17 (8.2%) and IFN-γ (51.3%) compared to peripheral blood cells (1.3% and 2.9%, respectively). Of note, we found that the percentages of cytokine-producing T cells in the CSF of the same patient after the 2-week expansion period (7.5% IL-17 and 56.4% IFN-γ) were similar to the ex vivo cells ([figure 2](#F2){ref-type="fig"}). This suggests that our expansion protocol did not substantially alter the cytokine profile of the CSF cells. {#F2} DISCUSSION {#s3} ========== Even with modern imaging techniques and improved detection methods, the diagnosis of cerebral vasculitis remains uncertain. There is no information available on biomarkers associated with CV. Here, we report on intracellular IL-17 in expanded T cells from the CSF of patients with cerebral ischemia as potential diagnostic tool to help distinguish patients with cerebral vasculitis from patients with stroke due to other causes. Cellular immunity plays a major role in the pathogenesis of systemic vasculitis. For example, a chronic CD4 T-cell activation can be found in peripheral blood from patients with AAV with the degree of CD4+ T-cell activation correlating with the disease severity.^[@R6]^ In patients with AAV, T cells are biased towards Th17 cells following in vitro stimulation and increased levels of Th17 cells can persist even during remission,^[@R4],[@R7]^ which argues in favor of Th17 cells being present in all disease stages. In addition, serum IL-17A and IL-23 levels are elevated in acute disease stages of patients with AAV but also in a proportion of convalescent patients, while IFN-γ production returns to control levels.^[@R8]^ In agreement with these observations, we also found persistent Th17 cells in the CSF of patients who were in remission and in patients with normal CSF cell count ([table 1](#T1){ref-type="table"}), while we did not see a difference in the production of IFN-γ in our expanded T-cell population compared to controls. Interestingly, in the 3 patients with a proven GCV, we did not detect an increased IL-17 in the expanded CSF cells. In untreated patients with GCV, Th17 cells can be found in the blood and are part of the cellular infiltrates in the arteries.^[@R9]^ However, arteries from treated patients contain few Th17 cells, which is a clear difference from AAV. We would argue that the reason for the absence of IL-17-producing T cells in the CSF of patients with GCV can be explained by 2 possibilities: (1) GCV only affects large vessels, which do not come in contact with the CSF, or (2) Th17 cells already disappeared after the steroid treatment. The presence of Th17 cells and the role of IL-17 can be generalized to other autoimmune diseases such as Churg-Strauss vasculitis, rheumatoid arthritis, and systemic lupus erythematosus, making it a target for novel therapeutic interventions. Humanized anti-IL-17 antibodies neutralizing the biologic activity of IL-17 induce clinically relevant responses in patients with psoriasis and rheumatoid arthritis.^[@R10]^ Anti-IL-17 is now on trial for AAV and might constitute a future therapeutic option for cerebral vasculitis. The levels of IL-17 in the CSF have proven too low for detection, even using a highly sensitive method. Detection of IL-17 produced by the cellular component is feasible, but due to the low number of T cells in the CSF, it is unavoidable to expand the cells before analysis, and this manipulation could bias the results. However, in 2 patients (one of them with a neurosarcoidosis, and therefore not included in the study), we were able to collect enough cells to stain IL-17 and IFN-γ directly and after the expansion with similar results, arguing that our expansion protocol reflects the in vivo situation. In addition, recent publications demonstrate cytokine memory for T-helper cells.^[@R11],[@R12]^ Taken together, it is likely that in systemic autoimmune disease, particularly vasculitis, persistent Th17 cells can be found. Compartments such as the CSF are spaces with close contact to small vessels and can reflect inflammatory vessel pathology. Thus, CSF Th17 cells might be a much more specific biomarker than cell count or protein elevation. Limitations. {#s3-1} ------------ As the choice of the cutoff was data-driven and sensitivity and specificity resulted from an in-sample evaluation, an optimistic bias of the result cannot be excluded. Therefore, an independent validation of the cutoff and the diagnostic accuracy in a dataset of appropriate size is recommended. The authors thank the patients and healthy blood donors for their cooperation; the FACS Core Facility at the UKE for technical help; and Dr. Vettorazzi, Department of Medical Biometry and Epidemiology, for help with the statistical analysis. AUTHOR CONTRIBUTIONS ==================== Vivien Thom: study concept and design, acquisition of data, analysis and interpretation. Sabrina Schmid: acquisition of data. Mathias Gelderblom: study concept and design, critical revision of the manuscript for important intellectual content. Manuela Kolster: acquisition of data. Romy Hackbusch: acquisition of data. Simon Schuster: critical revision of the manuscript for important intellectual content. Götz Thomalla: critical revision of the manuscript for important intellectual content. Oliver Keminer: acquisition of data. Ole Pless: critical revision of the manuscript for important intellectual content. Christian Bernreuther: acquisition of data, critical revision of the manuscript for important intellectual content. Markus Glatzel: acquisition of data, critical revision of the manuscript for important intellectual content. Karl Wegscheider: analysis and interpretation. Christian Gerloff: critical revision of the manuscript for important intellectual content, study supervision. Tim Magnus: study concept and design, analysis and interpretation, critical revision of the manuscript for important intellectual content, study supervision. Eva Tolosa: study concept and design, analysis and interpretation, critical revision of the manuscript for important intellectual content, study supervision. STUDY FUNDING ============= No targeted funding. DISCLOSURES =========== V. Thom and S. Schmid report no disclosures. M. Gelderblom served on the scientific advisory board for Merz Pharmaceuticals, received travel funding and/or speaker honoraria from Merz Pharmaceuticals, received research support from Merz Pharmaceuticals and Allergan and is on the editorial board for PLOSone. R. Hackbusch, M. Kolster, and S. Schuster report no disclosures. G. Thomalla received travel funding and/or speaker honoraria from Bayer, Boheringer Ingelheim, Daichii Sankyo, and Bristol Myer Squibb/Pfizer, consulted for Acandis and GlaxoSmithKline, and received research support from Bayer, European Union, and Deutsche Forschungsgesellschaft. O. Keminer, O. Pless, and C. Bernreuther report no disclosures. M. Glatzel received research support from Janssen Pharmaceuticals. K. Wegscheider served on the scientific advisory board for Resmed and Biotronik, and received research support from Resmed, Biotronik, BMBF, and DFG. C. Gerloff serves on the scientific advisory board for Bayer Vital, Boehringer Ingelheim, EBS Technologies, and Silk Road Medical, received travel funding and/or speaker honoraria from Bayer Vital, Boehringer Ingelheim, Biogen Idec, ev2/Covidien, GlaxoSmithKline, Grifols, Inomed, Lundbeck, Nexstim, Pfizer, Sanofi Aventis, UCB, and Merck Serono, is on the editorial board for *INFO Neurology Psychatrie*, and *Aktuelle Neurologie*, is editor of the textbook Therapie und Verlauf neurologischer Erkrankungen (Kohlhammer Verlag), has consulted for EBS Technologies and Silk Road Medical, and received research support from Merz Pharmaceuticals, Allergan, Novartis, NeuroConn, DFG, BMBF/DFG, EU FP7, and Wegener Foundation. T. Magnus received travel funding and/or speaker honoraria from Novartis, Merck-Serono, Grifols, Boehringer Ingelheim, and CSL Behring, and received research support from Novartis, ERANET, DFG, and Werner Otto Society. E. Tolosa received research support from DFG, Wissenschaftsstiftung Hamburg, and Werner-Otto Foundation. Go to [Neurology.org/nn](10.1212/NXI.0000000000000214) for full disclosure forms. AAV : anti--neutrophil cytoplasmic autoantibody--associated vasculitis CV : cerebral vasculitis GCV : giant cell vasculitis IFN-γ : interferon-γ IL-17 : interleukin-17 MSD : Meso Scale Discovery PACNS : primary angiitis of the CNS SACNS : secondary angiitis of the CNS [^1]: These authors contributed equally to this work. [^2]: Funding information and disclosures are provided at the end of the article. Go to [Neurology.org/nn](10.1212/NXI.0000000000000214) for full disclosure forms. The Article Processing Charge was paid by the authors.

Introduction {#S0001} ============ Despite advances in prevention, screening, diagnosis, and treatment during the past decade, cervical cancer remains a major issue of public health, representing the fourth most common female malignancy worldwide.[@CIT0001] Approximately 90% of the 270 000 cervical cancer deaths in 2015 occurred in low income and middle-income countries.[@CIT0002] An increasing trend in incidence and mortality of cervical cancer has also been observed in China.[@CIT0003] With advances in the development of instrumentation and surgical expertise, we witnessed a progressive shift from traditional open surgery towards minimally invasive surgery in the treatment of cervical cancer. Minimally invasive surgery is now considered as a widely accepted approach for the management of early-stage gynecological malignancies, which is particularly beneficial in terms of blood loss, pain, hospitalization and recovery, without detrimental effects on the curative or survival outcomes.[@CIT0004]--[@CIT0006] Robotic surgery is the most advanced technology for minimally invasive surgery, which was approved by the US Food and Drug Administration for gynecology in 2005. Sert et al. were the first to report robot-assisted radical hysterectomy (RRH) and lymph node dissection in 2006.[@CIT0007],[@CIT0008] In 2015, our hospital carried out the first RRH operation. In order to ensure better surgical quality and standardization of the entire procedure from pre-operation to post-operation, in this study, a total of 805 cervical cancer patients received RRH. However, this treatment method is somewhat controversial. Several studies revealed that minimally invasive surgery may be associated with shorter overall survival than open surgery[@CIT0009] and increased rates of death and recurrence[@CIT0010] in patients with cervical cancer. This may lead to a paradigm shift in the treatment of cervical cancer, therefore we further evaluate the data from our institution. In previous studies, a great deal of articles focused on the long-term effects after RRH,[@CIT0009]--[@CIT0013] while researches on the short-term effects were rare relatively. In order to better evaluate this therapeutic approach, we concentrated on the short-term perioperative impacts of RRH on patients. Although there were a little of researches about perioperative complications for cervical carcinoma,[@CIT0014],[@CIT0015] most studies were based on retrospective reviews of medical records or were performed without consideration of the severity of each complication or based on their own criteria. Therefore, it is probable that not all complications have been fully documented, and it is difficult to compare complication rates and identify risk factors for complications reliably. The incidence of perioperative complication is an important index reflecting the effect of surgery, and the principle of classification for perioperative complications should be simple, reproducible, flexible, and applicable irrespective of the cultural background.[@CIT0016],[@CIT0017] Such requirements are met by the Clavien--Dindo classification (CDC) proposed in 2004.[@CIT0018] Since then, this classification has been applied to many surgeries including gastrectomy, renal cell cancer resection, colorectal resection, pancreaticoduodenectomy, breast cancer and urological resection.[@CIT0017],[@CIT0019]--[@CIT0022] However, the perioperative complications assessed by CDC in radical hysterectomy of cervical cancer have been described scarcely. Therefore, the purpose of this study is to analyze the severity of perioperative complications of cervical cancer patients undergoing RRH by CDC, evaluate the relationship between NACT and complications, and further confirm the feasibility and safety of RRH of cervical carcinoma after NACT. Materials and Methods {#S0002} ===================== Patients {#S0002-S2001} -------- From January 2016 to April 2019, a total of 805 patients receiving RRH in our hospital were asked to participate in this retrospective observational study. 244 patients underwent NACT, and 561 patients did not receive NACT. After patients received NACT treatment for 3 courses, the eligible patients wanted to receive RRH with Da Vinci Si Surgical System. The surgery was performed by an experienced surgical team proficient in gynecologic oncology, the member of which all have extensive experience in minimally invasive surgery. All pathological diagnosis was confirmed by two experienced pathologists. The study was approved by the institutional review board (IRB) of the Third Xiangya Hospital, Central South University. Radical hysterectomy was performed for all patients with proper consultation and written informed consent was obtained from each subject. Data Collection {#S0002-S2002} --------------- Data of patients' demographics, laboratory examination, clinical manifestation and perioperative complications were obtained from medical records, which include age, stage according to the International Federation of Gynecology and Obstetrics (FIGO), histopathological type, tumor grade, American Association of Anesthesiologists (ASA) classification, classification for HPV and NACT. Intraoperative parameters include mean surgical time, blood loss, number and metastasis condition of pelvic lymph nodes and postoperative hospital stay. According to the definition of perioperation, perioperative complications are divided into intraoperative complications and postoperative complications. NACT Regimen {#S0002-S2003} ------------ Patients with squamous cell carcinoma received 3 courses of TIP (paclitaxel 175 mg/m^2^ + ifosfamide 5 g/m^2^+ cisplatin 75 mg/m^2^), and patients with cervical adenocarcinoma received 3 courses of TEP (paclitaxel 175 mg/m^2^+epirubicin 80 mg/m^2^+ cisplatin 75 mg/m^2^).[@CIT0023]--[@CIT0025] The rest of the chemotherapy regimens were performed with reference to the clinician's experience and the actual situation of the patients. Within 2--3 weeks after the completion of NACT, patients were re-evaluated through imaging and physical examination. After the assessment was passed, Da Vinci robot-assisted hysterectomy was performed for the patient with the patient's knowledge and consent. Perioperative Complications {#S0002-S2004} --------------------------- Perioperative complications were defined according to previously reported references, and subdivided into intraoperative and postoperative complications.[@CIT0018],[@CIT0026] Intraoperative complications included transfusion within 72 h after surgery, ureter or bladder injury and bowel injury. Postoperative complications included fever (\>38°C) for \>24 h postoperatively, urinary retention, short-term abnormal liver and renal function, severe edema of lower extremity, lymphocytic cyst infection, postoperative infection, severe anemia, bowel obstruction, vaginal vault dehiscence, vault bleeding, urinary tract infection, lymphedema, fistula, pelvic infection, remnant drain catheter and deep venous thrombosis. All perioperative complications were classified into 5 grades according to CDC. Grade I complications: no special treatment is required; Grade II complications: medical treatment and parenteral nutrition are required; Grade III complications: surgery, endoscopic or radiological intervention are required to be performed under general anesthesia (Class IIIb) or local anesthesia (Class IIIa); Grade IV complications: intensive care is required due to single or multiple organ failure; Grade V complications: death.[@CIT0017] Statistical Analysis {#S0002-S2005} -------------------- Statistical analyses were performed using IBM SPSS Statistics23.0 (IBM SPSS Inc., Chicago, IL). Descriptive statistics were performed on the distributions of demographic characteristics and perioperative complications. Continuous data were described as mean ± standard deviation and categorical data as number (percentage). Continuous variables were compared using the Student *T*-test or Mann--Whitney *U*-test. Categorical variables were compared using the Chi-square test or Fisher's exact test. Two-level logistic regression model was used to evaluate the risk factors between NACT and perioperative and severe complications. Two-level logistic regression model was used to estimate the odds ratio (OR) and the 95% confidence interval (CI) of the risk of perioperative complications by analysis of multiple clinical indicators. Variables with *p*\<0.05 in the univariate analysis were considered in a multivariate analysis. *P*\<0.05 was considered statistically significant. Results {#S0003} ======= Patient Characteristics {#S0003-S2001} ----------------------- The demographic characteristics of the study population are shown in [Table 1](#T0001){ref-type="table"}. 805 patients receiving RRH were included in the study, among which, 244 patients (30.31%) underwent NACT and 561 patients (69.69%) did not receive NACT. 439 patients (54.53%) were in stage IA-IB and 366 patients (45.47%) were in stage II or above by FIGO staging. Medium differentiation of tumor grade and squamous of histology cell were observed at highest frequency in 489 (60.75%) and 641 (79.63%), respectively. 164 patients (20.37%) had preoperative comorbidities with ASA score ≥ 3. Type 16 and 18 of HPV were regarded as relatively high risk for cervical carcinoma and were detected in 587 patients (79.92%). 137 patients (17.02%) showed lymph node metastasis. The median of surgical time, blood loss, indwelling time of drainage tube and postoperative hospital stay were 140.00, 150.00 and 3.00, respectively. The mean of age, number of pelvic lymph nodes and days of postoperative hospital stay were 49.44, 12.74 and 7.40, respectively. Table 1Summary of Patient CharacteristicsCharacteristicsTotal (n=805)Age (years)49.44±9.19Stage (FIGO) IA1189 (23.49%) IA28 (0.99%) IB1215 (26.71%) IB227 (3.35%) IIA1150 (18.63%) IIA287 (10.81%) IIB and above129 (16.02%)Tumor grade Well differentiated99 (12.30%) Moderately differentiated489 (60.75%) Poorly differentiated217 (26.96%)Histopathological type Adenosquamous carcinoma18 (2.24%) Squamous641 (79.63%) Adenocarcinoma76 (9.44%) Other70 (8.70%)ASA classification Ⅰ38 (4.72%) Ⅱ603 (74.91%) ≥164 (20.37%)Classification for HPV Negative42 (5.22%) Low risk7 (0.87%) High risk169 (20.99%) Extremely high risk587 (72.92%)Lymph node metastasis No668 (82.98%) Yes137 (17.02%)Neoadjuvant chemotherapy No561 (69.69%) Yes244 (30.31%)Surgical time (mins)140.00 (115.00--169.75)Blood loss (mL)150.00 (80.00--250.00)Number of lymph nodes dissection12.74±8.16Indwelling time of drainage tube (day)3.00 (2.00--3.00)Postoperative hospital stay(day)7.40±2.34[^1] Perioperative Complications {#S0003-S2002} --------------------------- Detailed information on perioperative complications in patients are shown in [Table 2](#T0002){ref-type="table"}. In this study, of 805 patients receiving RRH, 363 patients (45.09%) had perioperative complications; 64 patients (7.95%) had intraoperative complications; 328 patients (40.75%) had postoperative complications; and 29 patients had both intraoperative and postoperative complications. According to the CDC, 334 (60.95%), 146 (26.64%), 66 (12.04%), 2 (0.36%) and 0 (0.00%) perioperative complications were classified as Grade I, Grade II, Grade III, Grade IV and Grade V, respectively. Severe complications are defined Grade III and above according to CDC. Our study showed that most of the complications of patients receiving RRH were grades I and II, and the most common complications were Grade I urinary retention and Grade II urinary tract infection, however there were also 63 cases of severe complications. Table 2Perioperative ComplicationsCDCNumberPercent (%)Perioperative complications36345.09Intraoperative complications647.95Postoperative complications32840.75Severe complications637.83CDC classification548100.00Grade I33460.95 Fever (\>38°C)539.67 Urinary retention14125.73 Short-term abnormal liver function519.31 Short-term abnormal renal function91.64 Severe edema of lower extremity40.73 Lymphocytic cyst264.74 Intraoperative blood transfusion509.12Grade Ⅱ14626.64 High fever with elevated white blood cells488.76 Urinary tract infection8415.33 Severe anemia20.36 Lymphocytic cyst infection50.91 Bowel obstruction71.28Grade6612.04 Bowel injury61.09 Bladder or ureter injury112.01 Ureteroscopy10.18 Vaginal vault dehiscence61.09 Vaginal fistula366.57 Bladder vaginal fistula40.73 Lymphatic fistula20.36Grade Ⅳ20.36 Deep vein thrombosis20.36 Risk Factors for Perioperative and Severe Complications {#S0003-S2003} ------------------------------------------------------- The univariate analysis showed that perioperative complications correlated significantly with age (*p*=0.008), clinical staging by FIGO (*p*\<0.001), tumor grade (*p*\<0.001), histopathological type (*p*\<0.001), lymph node metastasis (*p*\<0.001), NACT (*p*\<0.001), surgical time (*p*=0.011), blood loss (*p*\<0.001), number of pelvic lymph node dissection (*p*=0.001) and indwelling time of drainage tube (*p*\<0.001) ([Table 3](#T0003){ref-type="table"}). Factors with *p* value\<0.05 in the univariate analysis were selected as covariables in the two-level logistic regression analysis. It was found that NACT (OR = 9.59, 95% CI: 6.43--14.28, p \<0.001) was an independent risk factor for a higher perioperative complication rate. Besides, compared with well-differentiated tumor, moderately differentiated (OR=1.85, 95% CI: 1.03--3.33, p=0.041) and poorly differentiated tumor (OR=4.63, 95% CI: 3.08--6.96, p\<0.001) ([Table 4](#T0004){ref-type="table"}) are another independent risk factor for a higher perioperative complication rate. For severe complications, the univariate analysis was only significantly correlated with human papillomavirus typing (*P*=0.008), and no further multivariate analysis was conducted. Table 3Univariate Analysis of Risk Factors for Perioperative and Severe Complications After Robot-Assisted Radical HysterectomyVariablesPerioperative Complications (%)*P*-valueSevere Complications (%)*P*-valueAge(years)50.38 ± 8.900.00849.75 ± 9.850.783Stage (FIGO)\<0.0010.971 IA117.36 (63/189)7.41 (14/189) IA20.00 (0/8)0.00 (0/8) IB132.09 (69/215)8.84 (19/215) IB248.15 (13/27)7.41 (2/27) IIA138.67 (58/150)7.33 (11/150) IIA266.67 (58/87)9.20 (8/87) IIB79.07 (102/129)6.98 (9/129)Tumor grade\<0.0010.530 Well differentiated26.27 (28/99)5.05(5/99) Moderately differentiated56.85 (278/489)8.38(41/489) Poorly differentiated28.28 (57/217)7.83(17/217)Histopathological type\<0.0010.941 Adenosquamous carcinoma55.56 (10/18)5.56 (1/18) Squamous47.43 (304/641)8.11 (52/641) Adenocarcinoma44.74 (34/76)6.58 (34/76) Other21.43 (15/70)7.14 (5/70)ASA classification0.5570.167 Ⅰ39.47 (15/38)0.00 (0/38) Ⅱ44.61 (269/603)7.96 (48/603) ≥48.17 (79/164)9.15 (15/164)Classification for HPV0.4260.008 Negative57.14 (14/42)2.38 (1/42) Low risk45.56 (4/7)42.86 (3/7) High risk45.66 (77/169)8.88 (15/169) Extremely high risk33.33 (268/587)7.50 (44/587)Lymph node metastasis0.0010.655 No42.37 (283/668)7.63 (51/668) Yes58.39 (80/137)8.76 (12/137)Neoadjuvant chemotherapy\<0.0010.669 No30.84 (173/561)7.49 (42/561) Yes52.34 (190/244)8.61 (21/244)Surgical time (min)143.00 (121.00--170.00)0.011138.00 (115.00--161.00)0.717Blood loss (mL)200.00 (100.00--300.00)\<0.001200.00 (100.00--200.00)0.566Number of lymph nodes dissection13.81 ± 7.660.00112.21 ± 7.360.591Dwelling time of drainage tube (day)3.00 (3.00--4.00)\<0.0013.00 (2.00--4.00)0.490Postoperative hospital stay (day)7.55 ± 2.160.0957.57 ± 2.510.536[^2] Table 4Multivariate Analysis of Risk Factors for Perioperative Complications After Robot-Assisted Radical HysterectomyVariablesPerioperative ComplicationsOR95% CIP-valueTumor grade Well differentiatedRef Moderately differentiated1.851.03--3.330.041 Poorly differentiated4.633.08--6.96\<0.001Neoadjuvant chemotherapy NoRef Yes9.596.43--14.28\<0.001[^3] Associations Between NACT and Complications {#S0003-S2004} ------------------------------------------- [Table 5](#T0005){ref-type="table"} shows the association between NACT and perioperative and severe complications. The rough estimate demonstrated NACT was associated with perioperative complications, especially postoperative complications, but not with intraoperative or severe complications. After the confounders of model I was adjusted, it was found that NACT was associated with postoperative complications (OR=17.19, 95% CI: 8.49--34.83, p\<0.001) and perioperative complications (OR=10.83, 95% CI: 5.62--20.86, p\<0.001). Even if all confounders were adjusted in model II, NACT was still associated with postoperative complications (OR=17.65, 95% CI: 8.63--36.08, p\<0.001) and perioperative complications (OR=11.08, 95% CI: 5.70--21.54, p\<0.001). However, NACT was not associated with intraoperative complications (OR=0.51, 95% CI:0.18--1.41, *p*=0.194) and severe complications (OR=1.68, 95% CI:0.64--4.41, *p*=0.294), respectively. Therefore, this seems to indicate that NACT in cervical cancer patients can only lead to an increase in mild complications and has certain safety. Table 5Association of Neoadjuvant Chemotherapy and Perioperative ComplicationsComplicationsCrude ModelAdjusted Model I^a^Adjusted Model II^b^OR (95% CI)*P*-valueOR (95% CI)*P*-valueOR (95% CI)*P*-valueSevere complications1.16 (0.67--2.01)0.5871.64 (0.64--4.20)0.3041.68 (0.64--4.41)0.294Intraoperative complications0.82 (0.46--1.46)0.8190.49 (0.18--1.35)0.1650.51 (0.18--1.41)0.194Postoperative complications9.77 (6.87--13.90)\<0.00117.19 (8.49--34.83)\<0.00117.65 (8.63--36.08)\<0.001Perioperative complications7.89 (5.55--11.21)\<0.00110.83 (5.62--20.86)\<0.00111.08 (5.70--21.54)\<0.001[^4][^5] Associations Between Times of NACT and Complications {#S0003-S2005} ---------------------------------------------------- As shown in the previous results, NACT is associated with perioperative complications. This study is expected to verify whether the incidence of complications increase as the times of NACT increase. Therefore, the two-level logistic regression model was applied to clarify the special correlation, and results are shown in [Table 6](#T0006){ref-type="table"}. A step-wise algorithm was used to select factors associated with NACT considering the multicollinearities. In the crude model, the risk of perioperative complications in the group with 1 NACT (OR = 6.77, 95% CI: 3.80--12.06, P \<0.001) and the group with 2 or more NACTs (OR = 8.54, 95% CI: 5.70--12.79, P \<0.001) was higher than the group without NACT and the difference was statistically significant (P\<0.05). Similarly, the postoperative complications in the group with 1 NACT (OR=7.74, 95% CI: 4.41--13.59, p\<0.001) and the group with 2 or more NACTs (OR=10.97, 95% CI: 7.31--16.48, p\<0.001) was higher than the group without NACT.As previously depicted, no significant correlation was found between intraoperative complications and severe complications and the times of NACT. After all confounding factors were adjusted; the upward trend still existed stably. The risk of perioperative complications in the group with 1 NACT (OR=9.39, 95% CI: 4.61--19.14, P\<0.001) and the group with 2 or more NACTs (OR=10.90, 95% CI: 5.12--23.20, P\<0.001) was higher than the group without NACT; the risk of postoperative complications in the group with 1 NACT (OR=14.33, 95% CI:6.80--30.16, P\<0.001) and the group with 2 or more NACTs (OR=21.34, 95% CI:9.40--48.43, P\<0.001) was higher than the group without NACT. These seemed to herald a dose-determined relationship between perioperative complications and times of NACT. Table 6Association Between Times of Neoadjuvant Chemotherapy with ComplicationsTimes of NACTCrude ModelAdjusted Model I ^b^Adjusted Model II ^c^OR (95CI)P-valueOR (95CI)P-valueOR (95CI)P-valueSevere complications NA ^a^RefRefRef One1.23 (0.50--3.00)0.6571.67 (0.59--4.76)0.3341.64 (0.57--4.72)0.357 Two or above1.26 (0.69--2.30)0.4572.49 (0.79--7.78)0.1182.50 (0.78--8.00)0.122Intraoperative complications NA ^a^RefRefRef One0.70 (0.24--2.00)0.5040.54 (0.16--1.85)0.3240.55 (0.16--1.89)0.341 Two or above0.96 (0.51--1.79)0.8970.60 (0.20--1.86)0.3800.64 (0.21--1.98)0.443Postoperative complications NA ^a^RefRefRef One7.74 (4.41--13.59)\<0.00113.95 (6.66--29.23)\<0.00114.33 (6.80--30.16)\<0.001 Two or above10.97 (7.31--16.48)\<0.00121.91 (9.70--49.49)\<0.00121.34 (9.40--48.43)\<0.001Perioperative complications NA ^a^RefRefRef One6.77 (3.80--12.06)\<0.0019.20 (4.53--18.68)\<0.0019.39 (4.61--19.14)\<0.001 Two or above8.54 (5.70--12.79)\<0.00111.39 (5.37--24.16)\<0.00110.90 (5.12--23.20)\<0.001[^6][^7] Discussion {#S0004} ========== Perioperative complications of radical hysterectomy of cervical cancer usually lead to longer hospital stays, higher medical cost, and delayed adjuvant therapy. The incidence of perioperative complications is also an important indicator for measuring the operative quality. However, there is still no consensus on the definition and classification of perioperative complications, which makes it difficult to evaluate the surgical procedure. The principles of complications classification should be simple, reproducible, flexible and convenient, and widely applicable.[@CIT0016] To meet these requirements, we adopted the well-standardized classification, known as the Clavien--Dindo classification system, which has been proven to be a reliable tool for quality assessment for surgery in many fields.[@CIT0017],[@CIT0019]--[@CIT0022] The complication is defined as "any deviation from the normal postoperative course" and the severity is classified according to the type of treatment required, such as surgical intervention or pharmacological treatment. We analyzed and classified the perioperative complications of RRH for cervical cancer. Different clinical trials reported complication rates ranging widely from 4.2% to 58.6%.[@CIT0013]--[@CIT0015],[@CIT0019],[@CIT0027] In the present study, the perioperative and severe complication rates were 45.09% and 7.83%, respectively. However, the complication rates in our previous studies seem to be somewhat higher because most grade I complications, such as asymptomatic fever and transient hepatic and renal function abnormality, were included in the perioperative complications. Meanwhile, the most common complications were Grade I urinary retention and Grade II urinary tract infection, and it is speculated that patients with robot-assisted surgery have a higher incidence of nerve damage.[@CIT0028] Therefore, it is recommended that nursing staff should pay more attention to patients after RRH to improve the quality of surgery. Then, univariate and multivariate analyses were performed in whole patients to investigate risk factors correlating with perioperative and severe complications. Poorly differentiated tumor and NACT were independent risk factors for perioperative complications, while the only classification of HPV was correlated with severe complications by univariate analysis. Multivariate analysis indicated that a poorly differentiated tumor was identified as an independent risk factor related to perioperative complications. Previous studies have shown the degree of differentiation was associated with a recurrence rate of cervical cancer. Wang et al suggested that moderately and highly differentiated tumor could indicate a high recurrence rate of cervical cancer,[@CIT0029] while Gong et al found that low levels of tumor differentiation were one of the independent risk factors for recurrent cervical cancer.[@CIT0030] Researches on differentiated tumor levels related to the short-term effect were rare relatively. In the present study, we found that a poorly differentiated tumor was significantly associated with perioperative complications. Patients with poorly differentiated tumor were always in bad nutritional status, with common symptoms such as anemia, weight loss, and hypoproteinemia. Although we adjusted their nutritional status before the operation, it might still influence the vulnerability of surgical stress and the occurrence of perioperative complications. However, the association between tumor grade and complications still needs further studies. NACT was identified as another predictor for perioperative complications in multivariate analysis. NACT was performed in radical surgery for cervical cancer over 20 years.[@CIT0031] Although NACT has chemotherapy toxicity such as gastrointestinal reactions and bone marrow inhibitory reactions,[@CIT0032] possible advantages include the potential for decreasing tumor size, reducing lymph nodes metastasis and distant metastasis, which may provide a viable alternative to chemo-radiotherapy when radiotherapy is unavailable or radiotherapy is unavoidably delayed.[@CIT0033]--[@CIT0035] Although the safety and effectiveness of NACT in the treatment of cervical cancer were guaranteed in many reports,[@CIT0025],[@CIT0031],[@CIT0033]--[@CIT0036] the results of retrospective cohort studies, randomized controlled trials,[@CIT0037] and meta-analysis[@CIT0038] showed that NACT did not improve the survival outcome of patients with cervical cancer. More specifically, patients who received NACT had a higher recurrence rate, longer median duration of RRH, and more median estimated blood loss.[@CIT0035] Therefore, from the available studies, there is insufficient evidence to show that radical hysterectomy with or without NACT can improve the survival rate and outcomes of patients with cervical cancer. In our study, neither univariate nor multivariate analyses revealed any significant advantages of NACT in perioperative complications. In contrast, NACT was associated with postoperative complications, but not with intraoperative or severe complications. As previously reported, advanced cancer, aortic lymphadenectomy, open surgery and malnutrition were associated with a higher risk of complications.[@CIT0009],[@CIT0039]--[@CIT0041] In our study, NACT resulted in more postoperative complications, the reasons of which remained unclear. The systemic effects of NACT and adverse reactions of chemotherapy reagents may be the reason for the increased postoperative complications. However, we found that NACT was not associated with intraoperative and severe complications in this study. Intraoperative complications here included transfusion within 72 h after surgery, bladder or ureter injury and bowel injury. In previous studies, BMI \>30kg/m^2^, previous abdominal surgery, metabolic/endocrine disorders (excluding diabetes), surgical complexity and final diagnosis were significantly associated with intraoperative complications.[@CIT0042],[@CIT0043] We speculated that the occurrence of intraoperative complications would be probably induced by insufficient experiences and learning curves of the surgeon or the specific surgical situations, rather than NACT. On the other hand, NACT was not associated with severe complications in this study. In a Phase II clinical trial, the results showed that the cervical cancer tissue of patients undergoing NACT was of high sensitivity, and because of the short course chemotherapy and low degree of reactions, gastrointestinal reactions such as nausea and vomiting and bone marrow suppression reactions such as leukocyte, hemoglobin and platelet reduction could be well tolerated in most of the patients.[@CIT0044] Therefore, we speculate that NACT may not cause severe complications, which seems to reveal the feasibility and safety of RRH for cervical carcinoma after NACT in general. In order to further clarify whether there is a "special" effect between NACT and complications, we further explored the association between the times of NACT and complications. The results showed that the overall situation was very similar to the multivariate analysis, which demonstrated the risk rate of perioperative complications, especially postoperative complications, increased steadily with the increase of NACT. This result is consistent with the prospective clinical research.[@CIT0045] For the "special" effect between NACT and complications, we may speculate that, on the one hand, with the increase of times of NACT, the accumulation of chemical toxicity in patients was gradually obvious, thereby indeed increasing the incidence of perioperative complications; on the other hand, the toxic impacts of continuous NACT were not independent. The impact of previous NACT would affect the effect of the next NACT, which could lead to an increased incidence of complications. The main advantage of this study is the short-term efficacy, and use of the well-standardized CDC for standard and uniform classification of the surgical complications to supplement the short-term effect of cervical cancer after RRH. Meanwhile, the impact of NACT times on perioperative complications was analyzed to fill the gap in the short-term efficacy of NACT performed before RRH of cervical cancer. However, our study has some limitations. These data are limited to patients' condition during the period of hospitalization. Our study does not consider long-term complications, such as malnutrition recurrence and survival outcome, which may influence the patients' quality of life and mortality. Additionally, since this was a retrospective study, recall bias and selection bias are inevitable, and there is no follow-up statistics on the survival rate, such as disease-free survival and overall survival. In the future, large-scale randomized controlled prospective research is needed with multi-center and multi-sector cooperation to achieve more credible results, eliminate bias, and obtain more surgical results. Conclusions {#S0005} =========== Our study demonstrates that NACT is a special risk factor of perioperative complications for patients with cervical cancer undergoing RRH, which seems to not lead to serious disease burden due to tolerable clinical toxicity, that is, NACT was closely related to mild postoperative complications. Hence, our study demonstrates the feasibility and safety of RRH of cervical carcinoma after NACT. However, the clinical application of NACT should be selected discreetly. In general, these results provide important clues for future research and provide directions for the adjuvant therapy of cervical cancer. This study was supported by the Key Research and Development Program of Hunan Province (2017SK2011). Disclosure {#S0006} ========== The authors declare no potential conflicts of interest related to this work. [^1]: **Abbreviations:** FIGO, International Federation of Gynecology and Obstetrics; ASA, American Society of Anesthesiologists; HPV, human papillomavirus. [^2]: **Abbreviations:** FIGO, International Federation of Gynecology and Obstetrics; ASA, American Society of Anesthesiologists; HPV, Human papillomavirus. [^3]: **Abbreviations:** OR, odds ratio; CI, confidence interval. [^4]: **Notes:** ^a^Adjust I: Estimates derived from two-level logistic regression models after adjusted for age, tumor grade, histopathological type, clinical FIGO stage, lymph node metastasis; ^b^Adjust II: Estimates derived from two-level logistic regression models after adjusted for not only the above variations but also for surgical time, blood loss, number of pelvic lymph node dissection and dwelling time of drainage tube. [^5]: **Abbreviations:** OR, odds ratio; CI, confidence interval. [^6]: **Notes:** ^a^NA means not adopting neoadjuvant chemotherapy; ^b^Adjust I: Estimates derived from two-level logistic regression models after adjusted forage, tumor grade, histopathological type, clinical FIGO stage and lymph node metastasis; ^c^Adjust II: Estimates derived from two-level logistic regression models after adjusted for not only the above variations but also for surgical time, blood loss, number of pelvic lymph node dissection and dwelling time of drainage tube. [^7]: **Abbreviations:** OR, odds ratio; CI, confidence interval.

1. Introduction {#sec1} =============== Cellulases derived from microorganisms mainly fungi and bacteria have been applied to various biotechnological applications mainly related to the degradation of cellulose fibers ([@bib4]; [@bib18]; [@bib28]). The cellulase enzyme system is composed of exoglucanases, which hydrolyze the reducing and non-reducing ends of cellulose to release glucose monomers; endoglucanases, which hydrolyze internal glycosidic bonds producing cellobiose and short cellulose fragments and β-glucosidases, which mainly hydrolyze cellobiose to release glucose monomers ([@bib28]). In particular, bacteria and their cellulases have demonstrated higher growth rates, and better tolerance to changes in temperature and pH compared to fungi. Therefore their cellulases have been recognized as more suitable for industrial application ([@bib18]). Several genera of bacteria produce efficient extracellular cellulases and these include: *Pseudomonas*, *Enterobacter*, *Bacillus*, *Klebsiella*, *Paenibacillus*, *Rhodococcus*, *Cellulomonas*, *Streptomyces* and *Citrobacter* among others ([@bib21]; [@bib23]; [@bib28]; [@bib33]). Many of these bacteria produce cellulases that tolerate a wide range of pH and temperatures. These include *B. brevis*, *B*. *subtillis*, *Paenibacillus* sp. and *Cellulomonas* sp. which are active in the ranges pH 5.5--7.5 and 37--60 °C ([@bib24]). To this effect, bacteria that produce extracellular cellulases that are active across various pH and temperature ranges are important, specifically to the application of biotechnological and industrial processes. These include stone washing and improvement of fabric appearance of clothing and textiles as well as improving pulp yield and reducing environmental problems in the pulp and paper industries and the conversion of cellulose in lignocellulosic material to glucose for fermentation to second generation biofuel to provide an alternative source of liquid fuel ([@bib26]; [@bib28]). Therefore, the aim of this investigation was to examine cellulolytic ability of four bacterial species to produce glucose from pretreated sugarcane bagasse over a range of pH and temperatures. 2. Material & methods {#sec2} ===================== 2.1. General materials {#sec2.1} ---------------------- All chemicals inclusive of broths and reagents used in this investigation were obtained from Sigma Aldrich, St. Louis MO, USA except where otherwise indicated. 2.2. Substrate collection and isolation of bacterial strains {#sec2.2} ------------------------------------------------------------ Sugarcane bagasse was used as the lignocellulose substrate for bacterial hydrolysis. This was collected from stores of sugarcane bagasse from Portvale Sugar Factory located in the parish of St. James, Barbados. The total culturable bacterial population of the bagasse was determined by an enrichment culture that was prepared by incubating 2 g of bagasse in tryptic soy broth, at 37 °C and 150 rpm for 48 hours. The culture was serially diluted after 48 h and 40 uL portions of the 10^−6^ and 10^−7^ dilutions were spread plated on tryptic soy agar (TSA) plates supplemented with amphotericin B (0.25 μg/mL) and incubated at 30 °C for 48 hours ([@bib3]; [@bib8]). Colonies that varied by morphological appearance inclusive of color, size and shape were selected and purified by serial subculturing on TSA plates. 2.3. Screening culture collection for cellulose degrading bacteria {#sec2.3} ------------------------------------------------------------------ Isolated bacteria were screened for exoglucananse and endoglucanase activity using cellulose and carboxymethyl cellulose (CMC) supplemented agar plates. Agar plates contained (per L): K~2~HPO~4~ (2 g), Na~2~HPO~4~ (2 g), (NH~4~)~2~SO~4~ (1.25 g), peptone (1 g), bacto agar (7.5 g), 1% cellulose or CMC ([@bib3]; [@bib13]). Twenty-four hour cultures of bacteria were streaked on to each type of plate. A rapid detection method consisting of staining the plates with Grams iodine for 5 minutes, followed by washing with distilled water, was employed. Cleared zones of hydrolysis on the respective plate indicated cellulase enzyme production and activity ([@bib16]). 2.4. Molecular identification of bacterial strains {#sec2.4} -------------------------------------------------- Bacterial strains which demonstrated hydrolytic production of glucose from pretreated bagasse were identified by analysis of 16S rRNA gene sequences. DNA was isolated by suspending individual colonies from agar plates in 300 μL of 100 mM Tris-EDTA buffer (pH 8) in 1.5 mL microcentrifuge tubes. Lysozyme (0.1 mg) and RNase A (0.002 mg) were added to the cell suspension and incubated at 30 °C for 30 min. Cell lysates were extracted with phenol/chloroform/isoamyl alcohol (25:24:1) and the DNA precipitated from the aqueous phase with 70% ethanol after the addition of 1/10 volume of 3 M sodium acetate (pH 5.5). 16S rRNA genes were amplified using the following primers: pA (5′-AGAGTTTGATCCTGGCTCAG) and pH (5′-AAGGAGGTGATCCAGCCGCA). The PCR reaction mixture (50 μL) consisted of: 1X EconoTag Plus Green Master Mix (Lucigen, Middleton, WI, USA), 5% molecular biology grade DMSO (Sigma, Oakville, ON, USA), 1 μM each primer and 25 ng template DNA. 16S rRNA gene amplicons were sequenced using the following primers 530R 5′-GTATTACCGCGGCTGCTGG, 936R (5′-GGGGTTATGCCTGAGCAGTTTG), 1527R (5′-AAGGAGGTGATCCAGCC), 514F (5′-GTGCCAGCASCCGCGG) and 1114F (5′-GCAACGAGCGCAACCC), ([@bib11]). Sequences were assembled and analyzed using Vector NTI version 11.5 (Life Technologies, Carlsbad, CA). After assembly, nearly complete 16S rRNA gene sequences were analyzed using the EzBioCloud identification tool ([@bib14]). Phyologenetic analyses were conducted using MEGA 7 ([@bib19]). Representative 16S rRNA gene sequences of selected type strains within the *Enterobacteriaceae* were obtained from GenBank. *Aeromonas hydrophilla* ATCC7966 (NR_07484) was used as an out group. Sequences were aligned in MEGA7 using ClustalW and alignments were manually edited. A maximum likelihood test to determine the optimal evolutionary model for the dataset identified the Kimura 2 parameter with a gamma distribution with invariant sites was identified as the best model for the dataset ([@bib17]). Evolutionary histories were inferred using the Neighbor-Joining, maximum likelihood, maximum parsimony, minimum evolution and unweighted pair group method with arithmetic mean (UPGMA) methods ([@bib29]). To evaluate reproducibility of clades, bootstrap analysis was conducted using 1000 replicates ([@bib9]). 2.5. Pretreatment of bagasse lignocellulose {#sec2.5} ------------------------------------------- Bagasse lignocellulose was subjected to two pretreatment methods, which consisted of a thermal (HT) and a solvent fractionation (ST) process. Bagasse was initially ground using a Waring blender ([@bib31]). For the HT pretreatment process, bagasse was transferred to a 1 L flask, mixed with distilled water (1:5, w/v) and autoclaved at 121 °C/60 minutes. The residue was separated and washed with tap water followed by distilled water and air dried for 5 days ([@bib31]). In the ST pretreatment process, organic solvent fractionation using ethanol (50%, v/v) was performed. Ground bagasse (100 g) was mixed with 1L (50%, v/v) ethanol at room temperature for 24 hours without stirring. After this time the residue was separated, washed with tap water followed by distilled water and air dried for 5 days ([@bib27]; [@bib35]). 2.6. Glucose production from pretreated sugarcane bagasse by the bacterial strains {#sec2.6} ---------------------------------------------------------------------------------- The hydrolytic activity of the bacteria on pretreated bagasse was evaluated by the accumulation of hydrolytic product glucose. Glucose concentrations were determined by colorimetric analysis using 3,5-dinitrosalicyclic acid reagent (DNS), ([@bib25]). The DNS assay was miniaturized to fit a 96-well plate by mixing 125 μL of hydrolysate with 125 μL of DNS reagent in microcentrifuge tubes. The tubes were boiled at 100 °C for 10 minutes. The contents of each tube were then transferred to the wells of a 96 well plate and the absorbance read at 540 nm using a microplate reader (BGM labtech FLUOstar OPTIMA). The concentration of glucose was determined by interpolation of a glucose standard curve ([@bib2]). To examine the effect of pH on hydrolysis and glucose concentration, various buffering systems were utilized as follows: citric acid buffer (pH 5--6); phosphate-buffered saline (PBS) (pH 7); phosphate buffer (pH 8) and carbonate buffer (pH 9). For the effects of temperature on hydrolytic production of glucose from pretreated bagasse and glucose concentration, at constant pH the temperature was varied between 25--40 °C. The optimal conditions of pH and temperature for bacterial hydrolysis of HT and ST pretreated bagasse were determined by monitoring the concentrations of glucose at 24-hour intervals for 168 hours ([@bib6]). Three Individual replicate experiments were conducted for each bacterial strain as follows: HT and ST pretreated bagasse (2.5 g, substrate concentration) were each placed in 125 mL conical flasks, to which 45 mL of appropriate buffer was added to the flask and the mixture sterilized at 121 °C for 15 min. A 5 mL aliquot of bacteria cultured in TSB for 24 hours, was washed and resuspended in buffer by centrifugation at 10 000 rpm bringing the total volume to 50 mL ([@bib3]). The flasks were then incubated for 168 h at the appropriate pH and temperature. 2.7. Statistical analysis {#sec2.7} ------------------------- Statistical analysis was performed with Statistical Package for Social Sciences (SPSS), version 19 and GraphPad Prism version 7.03 for windows, GraphPad Software, La Jolla California USA. The data were evaluated using two-way repeated measures ANOVA. Where significant differences were found, Tukey HSD tests were performed on pairwise comparisons. All statistical analyses were performed at an alpha of 0.05 and the data were expressed as mean ± SEM. All graphs were constructed using GraphPad Prism version 7.03. Additionally, in cases where the error bars are shorter than the height of the symbol, GraphPad Prism does not draw the error bars. 3. Results {#sec3} ========== 3.1. Bacterial culture collection and screening for cellulolytic bacteria {#sec3.1} ------------------------------------------------------------------------- Twenty-six bacterial strains were isolated from sugarcane bagasse. These strains were examined for production of cellulases using cellulose and CMC plate screens. From these bacteria, 16 strains were selected to test their cellulolytic ability using pretreated sugarcane bagasse. A further 4 strains were identified by molecular analysis for ability to produce glucose over a range of pH (pH 5--9) and temperatures (25--40 °C). 3.2. Molecular identification of bacterial strains {#sec3.2} -------------------------------------------------- Four bacterial strains designated UWIBGS3, UWIBGS6, UWIBGS10 and UWIBGS12 (accession numbers: MG923793, MG923794, MG923792, MG923795 respectively) isolated from sugarcane bagasse were identified by analysis of nearly complete 16S rRNA gene sequences using the EZBioCloud identifier. The 16S rRNA gene sequence of UWIBGS3 was most similar (99.60%) to that of *Enterobacter xiangfangensis* LMG 27195^T^, while those of UWIBGS6 and UWIBGS12 were most similar to *Serratia rubidaea* JCM 1240^T^ (99.71% similarity) and *Klebsiella pneumoniae* ATCC 11296^T^ (99.93% similarity), respectively. The huge degree of 16S rRNA gene sequence similarity between these three strains and those of previously described type strains indicated these bacteria are strains of these species. UWIBGS10 was identified as belonging to the genus *Citrobacter*, however, it exhibited the closest sequence similarity to *Cedecea lepagia* GTC 346^T^ (98.42%) and *Citrobacter sedlakii* NBRC 105722^T^ (98.29%). The 16S rRNA sequence of this strain also exhibited \>98% identity to several other strains of *Citrobacter* and *Salmonella*. This level of 16S rRNA sequence similarity suggests UWIBGS10 may represent a novel species in the *Enterobacteriaceae.* As *Cedecea* sp. have been shown to belong to the *Cedecea* clade of the *Enterobacteriaceae* while *Citrobacter* and *Salmonella* belong to the *Escherichia* clade we further investigated the *Cedecea lepagei* GTC 346^T^ entry in EzBioCloud ([@bib1]). This strain is not the type strain of the species which is *[Ce. lepagei]{.ul}* JCM 1684^T^. BlastN analysis of the *Ce. lepagei* GTC 346^T^ (AB273742) sequence revealed it was most similar (98.79%) to *Citrobacter youngae* GTC 1314. This indicated UWIBGS10 was most closely related to *Citrobacter* spp. Due to similar levels of 16S rRNA gene sequence similarity to *Citrobacter* and *Salmonella* spp., we conducted a phylogenetic analysis to determine the taxonomic affiliation of this strain ([Fig. 1](#fig1){ref-type="fig"}). In the Neighbor-Joining tree shown in [Fig. 1](#fig1){ref-type="fig"}, UWIBGS10 formed a sister clade to *C. youngae* and *C. sedlakii*, albeit with low bootstrap support. The low bootstrap support is consistent with reported poor resolution of 16S rRNA gene sequence based phylogenetic analysis with in the *Enterobacteriaceae* ([@bib1]). Despite the low bootstrap support, UWIBGS10 clustered with *C. youngae* and *C. sedlakii* in phylogenetic analyses conducted using the maximum likelihood, maximum parsimony, minimum evolution and UPGMA methods (data not shown). This suggests UWIBGS10 likely represents a novel *Citrobacter* species. Further polyphasic taxonomic analyses would be required to definitively determine the taxonomic placement of this strain within the *Enterobacteriaceae*.Fig. 1Neighbor-Joining phylogenetic analysis of 16S rRNA gene sequences of selected bacteria in the *Enterobacteriaceae* ([@bib29]). The optimal tree with the sum of branch length = 0.44327546 is shown. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test (1000 replicates) are shown next to the branches ([@bib9]). The tree is drawn to scale, with branch lengths in the same units as those of the evolutionary distances used to infer the phylogenetic tree. The evolutionary distances were computed using the Kimura 2-parameter method and are in the units of the number of base substitutions per site ([@bib17]). The rate variation among sites was modeled with a gamma distribution (shape parameter = 1). The analysis involved 41 nucleotide sequences. All positions containing gaps and missing data were eliminated. There were a total of 1314 positions in the final dataset.Fig. 1 3.3. Effects of pH and temperature on hydrolysis of pretreated bagasse {#sec3.3} ---------------------------------------------------------------------- Hydrolysis of pretreated bagasse by the four strains was examined across a range of pH (5--9) and temperatures (25--40 °C). The hydrolysis was monitored by measuring net glucose concentration accumulated during hydrolysis of HT and ST pretreated bagasse every 24 h for 168 h. This was performed by subtracting the starting concentration of glucose available in the media (T~i~), from the concentration of glucose at a given time point (T~i+1~). Therefore, Δ~glucose~ = T~i+1~ --T~i~ is the glucose accumulated as a result of hydrolysis which occurred during a specific time period. The changes in glucose concentrations at different pH levels over 168 h are demonstrated in Figs. [2](#fig2){ref-type="fig"} and [3](#fig3){ref-type="fig"} for HT and ST pretreated bagasse respectively. These data indicate the accumulation of glucose which demonstrates hydrolytic ability of the strains in varying conditions of pH and temperatures.Fig. 2Net glucose concentration at 24-hour intervals during hydrolysis thermally pretreated (HT) bagasse in the range pH 5--9 for four bacterial strains.Fig. 2Fig. 3Net glucose concentration at 24-hour intervals during hydrolysis of solvent pretreated (ST) bagasse in the range pH 5--9 for four bacterial strains.Fig. 3 Statistical analysis consisting of two-way repeated measures ANOVA was conducted to determine effects of pretreatment, strain and pH on hydrolytic production of glucose from pretreated bagasse and glucose concentration. These results are displayed in [Table 1](#tbl1){ref-type="table"}, indicating a significant interaction effect (F (4, 27) = 7.48, p = 0.00) of pretreatment and pH on glucose concentration during cellulose hydrolysis. Further post hoc analysis consisting of Tukey HSD tests were conducted on pairwise comparisons of mean glucose concentrations. This was for the range pH 5--9 for each bacterial strain for HT and ST pretreated bagasse. It was determined that for *Citrobacter* sp.UWIBGS10, the concentration of glucose at pH 6 (21.46 ± 2.50 mg/mL, mean ± SEM) between 72--96 hours was significantly higher (Tukey, p \< 0.05) than glucose concentrations at pH 5, 7, 8 and 9 for HT and ST bagasse. Additionally, glucose concentration for *Citrobacter* sp.UWIBGS10 between 72--96 hours for HT and ST bagasse was significantly higher (Tukey, p \< 0.05) than glucose concentrations accumulated for *E. xiangfangensis*, *S. rubidaea*, and *K. pneumoniae* (Figs. [2](#fig2){ref-type="fig"} and [3](#fig3){ref-type="fig"}).Table 1Repeated measures ANOVA for effects of pretreatment, strain and pH on glucose concentration for four bacterial strains.Table 1SourceType III sum ofdfMean squareFpPretreatment4.5414.547.830.01Strain3.0331.011.740.18pH25.6346.4111.040.00Pretreatment\*Strain0.6230.210.360.79**Pretreatment\*pH17.3544.347.480.00**Strain\*pH11.05120.921.590.16Pretreatment\*strain\*pH4.7660.791.370.26Error15.67270.58[^1] The effects of temperature on hydrolytic ability of thermally and solvent pretreated bagasse by four bacterial strains were displayed in Figs. [4](#fig4){ref-type="fig"} and [5](#fig5){ref-type="fig"}. These data were subjected to two-way repeated measures ANOVA analysis to determine the effects of pretreatment, strain and temperature on hydrolytic production of glucose from pretreated bagasse. This analysis indicated a significant interaction effect (F (5, 19) = 3.16, p = 0.03) of pretreatment, strain and temperature on glucose concentration as highlighted in [Table 2](#tbl2){ref-type="table"}. Tukey HSD tests were performed on pairwise comparisons of glucose concentrations for HT and ST pretreated bagasse for the strains in the temperature range 25--40 °C. These comparisons indicated that glucose concentration for *Citrobacter* sp. at 25 °C (21.46 ± 2.50 mg/mL; 72--96 hours; Tukey, p \< 0.05) was also significantly higher for HT and ST bagasse compared to glucose concentrations at 30, 35 and 40 °C ([Fig. 5](#fig5){ref-type="fig"}). Also, glucose concentrations for *Citrobacter* sp.UWIBGS10 between 72--96 hours were significantly higher (Tukey, p \< 0.05) compared to glucose concentrations for *E. xiangfangensis*, *S. rubidaea* and *K. pneumoniae* (Figs. [4](#fig4){ref-type="fig"} and [5](#fig5){ref-type="fig"}). These results indicate that for *Citrobacter* sp., optimal conditions are pH 6 and a temperature of 25 °C for release of maximal concentrations of glucose from the pretreated bagasse.Fig. 4Net glucose concentration at 24-hour intervals during hydrolysis thermally pretreated (HT) bagasse in the temperature range 25--40 °C for four bacterial strains.Fig. 4Fig. 5Net glucose concentration at 24-hour intervals during hydrolysis of solvent pretreated (ST) bagasse in the temperature range 25--40 °C for four bacterial strains.Fig. 5Table 2Repeated measures ANOVA for effects of pretreatment, strain and temperature on glucose concentration for four bacterial strains.Table 2SourceType III sum ofdfMean squareFpPretreatment0.0210.020.080.78Strain1.7730.592.710.07Temperature19.6536.5530.100.00Pretreatment\*Strain2.9830.994.560.14Pretreatment\*Temperature5.5731.868.530.00Strain\*Temperature3.0580.381.750.15**Pretreatment\*strain\*Temperature3.4450.693.160.03**Error4.13190.22[^2] Further analyses consisting of paired t-tests were conducted to compare hydrolysis between the HT and ST pretreated bagasse by the bacterial strains. These results indicated that ST pretreated bagasse in the pH range 5--9 generally produced higher concentrations of glucose however, these were not statistically significantly different (t = 1.44, p \> 0.05) from glucose produced from thermally pretreated bagasse. Alternatively, higher concentrations (t = 7.23, p \< 0.05) of glucose were produced from hydrolysis of ST pretreated bagasse in the temperature range 25--40 °C by the bacterial strains examined. 4. Discussion {#sec4} ============= In this study, bacterial strains isolated from sugarcane bagasse were examined for cellulolytic ability. Four cellulolytic bacteria were identified of the *Enterobacteriaceae* family, which are Gram negative, facultatively anaerobic, commonly found in water and soil, and can cause opportunistic infection ([@bib21]; [@bib28]). *E. xiangfangensis*, originally isolated from sourdough in Heilongjiang Province, China, has not been previously investigated for hydrolysis of pretreated bagasse lignocellulose for glucose production as was observed in this examination ([@bib12]). *S. rubidaea*, which has a distinctive characteristic of production of red pigment prodigiosin, has been associated with production of cellulolytic enzymes as well as, *K. pneumoniae* ([@bib5]; [@bib7]). This work has also resulted in the identification of a novel species of *Citrobacter* which has demonstrated ability to hydrolyze pretreated sugarcane bagasse to produce glucose. Previous studies have indicated several species of *Citrobacter* as possessing cellulolytic ability ([@bib10]; [@bib30]). However, in this work we examine this ability using sugarcane bagasse as the lignocellulose substrate. In the examination of hydrolytic production of glucose from pretreated bagasse in the range of pH 5--9 the data indicate that glucose concentrations for *Citrobacter* sp. UWIBGS10 are highest between 72--96 hours at pH 6. This demonstrates that hydrolysis of pretreated sugarcane bagasse for glucose production at this pH was optimal. Similar studies have previously indicated that hydrolysis of pretreated sugarcane bagasse for glucose production is optimal in the range pH 4--7.5. This has been documented for *B. subtilis* (pH 5.5), *B. licheniformis* (pH 6.1), and *Cellulomonas sp*. ASN2 (pH 7.5) ([@bib28]). Glucose accumulated for the strains *E. xiangfangensis*, *S. rubidaea* and *K. pneumoniae* were consistent in the pH range during cellulose hydrolysis. This is useful for industrial applications, which can require cellulolytic enzymes that tolerate changes in pH. Bacteria have also been documented to produce cellulolytic enzymes that tolerate alkaline pH conditions between pH 8--11 such as *Bacillus sp*. HSH-810 (pH 10) ([@bib28]; [@bib33]). Alkaline tolerant cellulases have been documented in *Pseudomonas fluorescens* (pH 10), *Serratia marcescens* (pH 10), and other *Bacillus* species, inclusive of *Bacillus* sp N-4 (pH 9), *Bacillus* KSM-522 (pH 7--10), and *B. subtilis* (pH 10) ([@bib20]; [@bib30]). It was also observed that during hydrolytic production of glucose from pretreated bagasse for *Citrobacter* sp. at pH 6, glucose concentration decreased after 96 hours. Factors that affect hydrolytic production of glucose from pretreated bagasse causing decreases in glucose concentrations include glucose consumption for metabolism and the presence of metabolic by-products that inhibit enzyme production and activity ([@bib34]). Hydrolysis is also affected by the high concentrations of glucose, which cause end-product inhibition. Furthermore, if sufficient glucose resources are available in the media, bacteria can decrease production of cellulases during that time period, which results in lower concentrations of glucose ([@bib22]). For hydrolytic production of glucose from pretreated bagasse in the temperature range of 25--40 °C *Citrobacter* sp. UWIBGS10 demonstrated optimal hydrolysis of pretreated sugarcane bagasse for glucose production by accumulating higher concentrations of glucose at 25 °C between 72--96 hours. Glucose concentrations accumulated for *E. xiangfangensis, S. rubidaea* and *K. pneumoniae* however, were consistent during the experimental period. These results are consistent with previous studies where it was found that cellulolytic enzymes from bacterial sources exhibit optimal hydrolytic ability between 25--50 °C ([@bib15]; [@bib28]; [@bib32]). Other studies have, however, demonstrated that at temperatures approaching 50 °C, cellulolytic enzymes tend to lose 60% of their activity and lose all activity at temperatures approaching 80 °C as the enzymes become denatured ([@bib30]). Overall, between the HT and the ST pretreatment, results indicated that higher concentrations of glucose were produced from hydrolysis of ST pretreated bagasse for the bacterial strains. This was significant for the temperature range 25--40 °C and indicated that the ST pretreatment made the bagasse lignocellulose more amenable for hydrolysis. This pretreatment process favored the disruption and removal of lignin, to increase pore volume, which facilitates improved hydrolysis indicated by higher concentrations of glucose ([@bib35]). 5. Conclusion {#sec5} ============= This work has identified a novel species of *Citrobacter* that could hydrolyze sugarcane bagasse at optimal conditions of pH 6 and 25 °C. This along with *E. xiangfangensis, S. rubidaea* and *K. pneumoniae* that could act over a broad range of pH and temperature, has significant potential for commercial exploitation for the production of cellulolytic enzymes. This is important in decreasing environmental impact associated with lignocellulose conversion to fermentable sugars among other biotechnological processes, in a low energy and environmentally friendly manner. Declarations {#sec6} ============ Author contribution statement {#sec6.1} ----------------------------- Jamila A.D. Jones: Conceived and designed the experiments; Performed the experiments; Analyzed and interpreted the data; Wrote the paper. R.G. Kerr: Performed the experiments; Contributed reagents, materials, analysis tools or data. B.A. Haltli: Performed the experiments; Analyzed and interpreted the data. Winston F. Tinto: Contributed reagents, materials, analysis tools or data. Funding statement {#sec6.2} ----------------- This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors. Competing interest statement {#sec6.3} ---------------------------- The authors declare no conflict of interest. Additional information {#sec6.4} ---------------------- No additional information is available for this paper. [^1]: The bold highlights reflect the highest level of statistically significant interaction effect of the independent variables on the dependent variable. [^2]: The bold highlights reflect the highest level of statistically significant interaction effect of the independent variables on the dependent variable.

{.264}

Introduction {#s1} ============ The Gram-negative opportunistic pathogen, *Pseudomonas aeruginosa*, is a typical biofilm-forming microbe, and this characteristic allows it to thrive in a diverse range of natural and nosocomial niches (Driscoll et al., [@B9]). *P*. *aeruginosa* is a common cause of severe infections in wounds, eyes and lungs, and it is often difficult to treat, due to the high prevalence of multi-drug resistance (Breidenstein et al., [@B5]). The quorum sensing (QS) system is a population density-dependent regulatory system that enables cell-to-cell communication and coordinated control of gene expression (Fuqua et al., [@B12]). Co-ordination of gene expression through QS is an important determinant of virulence and drug-resistance in *P. aeruginosa* (Van Delden and Iglewski, [@B47]). It is proposed that inhibitors of the QS system may act as effective anti-microbial agents by deregulating these determinants of pathogenesis and thus increasing the effectiveness of host defenses and antibiotic treatment (Fothergill et al., [@B11]). *P. aeruginosa* utilizes two interconnected QS systems, termed Las and Rhl (Schuster et al., [@B39]), which are regulated by *N*-acyl-homoserine lactones (AHLs, also termed *P. aeruginosa* autoinducers; PAIs). The *las* system is the predominant of the two QS systems and consists of the LasI and LasR proteins (Gambello and Iglewski, [@B13]). LasI synthesizes the AHL molecule, *N-3-oxododecanoyl-L-homoserine* lactone (3OC12-HSL, PAI-1), which is bound by the transcription regulator, LasR (Pearson et al., [@B33]). LasR directly regulates the expression of up to 74 genes, including *lasI* (Gilbert et al., [@B17]). In the analogous Rhl system, RhlI synthesizes *N*-butyryl-L-homoserine lactone (C4-HSL, PAI-2) and RhlR acts as the sensor/transcriptional regulator (Brint and Ohman, [@B6]). The concentration of the population, and thus the two concentration of the two AHL molecules, determines the expression of multiple proteins related to virulence, drug resistance, motility and biofilm development (Williams and Camara, [@B51]). Both natural and synthetic molecules that block QS have been shown to inhibit effectively QS systems, both *in vitro* (Pejin et al., [@B34]) and *in vivo* (Wu et al., [@B52]). For example, it was previously demonstrated that a sub-MIC level of the antibiotic azithromycin (AZM), which was sufficient to inhibit QS, was also effective in treating *P. aeruginosa* infections (Imperi et al., [@B21]). *Houttuynia cordata* Thunb (Saururaceae family) is an edible plant used in traditional Chinese medicine for the treatment of a wide range of infectious diseases, including pneumonia (Gao et al., [@B14], [@B15]; Li et al., [@B25]). The major constituent of the volatile oil derived from *H. cordata*, sodium houttuyfonate \[SH, CH~3~(CH~2~)~8~COCH~2~CHOHSO~3~Na\] is a product of the addition reaction of sodium bisulfite and houttuynin \[i.e., decanoyl acetaldehyde, CH~3~(CH~2~)~8~COCH~2~CHO\] (Wang et al., [@B48]). SH is the active compound of the *Houttuynia* plant, the healing properties of which have been recorded in ancient Chinese medical writings (Gao et al., [@B15]). SH is mainly used for treating purulent skin infections, respiratory tract infections, including pneumonia in elderly patients, and chronic bronchitis (Wang et al., [@B48]). However, despite its widespread and effective use, the mode of action remains unknown. In previous studies we reported that SH can inhibit biofilm formation and motility of *P. aeruginosa* (Shao et al., [@B40], [@B41],[@B42]). We found that SH can effectively prevent biofilm formation of *P. aeruginosa, Staphylococcus epidermidis* and *Candida albicans* (Shao et al., [@B41],[@B42]) and acts synergistically with the broad-spectrum antibiotic, levofloxacin (Shao et al., [@B40]). The mode of action of SH, however, remains unknown. Therefore, in this study we focused on the effect of SH on quorum sensing. Here, we investigate the putative role of SH as a QS-inhibitor in *P. aeruginosa*. Materials and methods {#s2} ===================== Bacterial strains, media and growth conditions ---------------------------------------------- *P. aeruginosa* strain ATCC 27853, obtained from the National Institute for the Control of Pharmaceutical and Biological Products (NICPBP, Beijing, China) was inoculated in Luria--Bertani (LB) broth (Aoboxing Bio-tech, Beijing, China) and grown under standard conditions (37°C, 220 rpm). *Chromobacterium violaceum* strain CV026 (McClean et al., [@B29]) was grown in LB broth supplemented with 1% agar, fetal bovine serum (20%, w/v), L-tryptophan (0.007%, w/v) and/or kanamycin (20 μg/ml) as appropriate which is modified from original medium of McClean et al. ([@B29]). In the modified medium, fetal bovine serum and L-tryptophan are added to the original medium, because L-tryptophan is known to increase the purple pigment production of *C. violaceum* (Demoss and Evans, [@B8]) and fetal bovine serum was observed by us to accelerate the growth speed of *C. violaceum* (data not shown). For measurement by spectrophotometry, cells were harvested at 24 h after inoculation by centrifugation at 4600 × g for 10 min. The supernatant was discarded and the pellet was resuspended in sterile saline solution for optic density detection at 600 nm (OD~600~) in a UV spectrophotometer. The absorbance of cell suspensions was adjusted to 0.05 for further experiments. MIC determination ----------------- A micro-broth dilution method (Wiegand et al., [@B50]) was adopted to test the minimum inhibitory concentrations (MICs) of SH and AZM. The assay was performed using 96-well plates and consisted of a gradient of inhibitor concentrations, i.e., 2048, 1024, 512, 256, 128, 64, 32, 16, 8, 4, 2, 1, 0.5, and 0.25 μg/ml, of equal final volume (100 μl), and 100 μl of bacterial suspension (final concentration 0.75 × 10^6^ CFU/ml). After mixing, the plates were cultured for 24 h at 37°C and the OD~600~ was measured. Each assay was performed in triplicate. Growth inhibition assay ----------------------- Antibiotics were added to a *P. aeruginosa* suspension (1 × 10^6^ CFU/ml) in a constant-temperature shaker (37°C) at 220 rpm. Drugs were added into the medium at the following concentrations: 64, 128, 256, and 512 μg/ml (1/8--1 × MIC) and 64 μg/ml (1 × MIC) AZM. Growth inhibition was measured using OD~600~ relative to the control culture (no drug) over a 72 h period and it was further quantified by plating the cultures and counting the CFU at 24 h and 72 h. Extraction of signaling molecules --------------------------------- *P. aeruginosa* was grown in 100ml of medium under standard conditions for 72 h, followed by centrifugation (10,800 × g, 10 min, 4°C) and subsequent transfer of the cleared supernatant into a clean flask. An equal volume of ethyl acetate was added, with mixing, and the organic phase was separated by centrifugation (10,800 × g for 15 min at 4°C). The organic phase was transferred into another clean flask and initially concentrated by evaporation to a 1 ml volume through heating in a water bath at 37°C. The remaining solution was further concentrated and stored in a sterile EP tube at −80°C. Each concentration was repeated in triplicate. Biological detection of signaling molecules ------------------------------------------- Detection of AHLs was determined on agar plates employing the biosensor strain *C. violaceum* CV026, which produces a purple pigment only in response to exogenously added AHLs (McClean et al., [@B29]). Overnight cultures (LB broth supplemented with 20 μg/ml kanamycin, 0.007% (w/v) L-tryptophan and 20% (w/v) fetal bovine serum) were mixed (20% (v/v)) with LB broth containing 2% agar and poured into plates. Once set, a 20 μl drop of signaling molecule solution was added in the center of the plates and the plates were incubated at 48°C for 24 h for development and observation of the violet color zone. Liquid cultures containing the signaling molecule solution were harvested after 24 h growth (220 rpm, 37°C). A 300 μl sample of the culture was transferred to a centrifuge tube and 300 μl of 10% (w/v) SDS was added. The cells were vortex-mixed for 5 s and then 2.1 ml of 98% ethanol was added. The supernatant was harvested (10,800 × g, 10 min, 4°C) and the OD~580~ was measured. Each experiment was repeated in triplicate. Gene expression analysis ------------------------ Approximately 0.75 × 10^6^ CFU (in a 200 μl volume) were used to inoculate 5 ml broths containing a range of SH concentrations, alone or in combination with AZM. Cultures were grown for 72 h, the cells were harvested by centrifugation (10,800 × g, 1 min) and the supernatant was discarded. Total RNA was extracted using an RNAprep Pure Cell/Bacteria Kit (Code No. DP430, TIANGEN, China), according to the manufacturer\'s guidelines. A FastQuant RT Kit (Code No. KR106, TIANGEN, China) was used to remove genomic DNA, and the purified RNA (2 μg) was used for reverse-transcription. The oligonucleotide primers used are designed and listed in Table [1](#T1){ref-type="table"}. Reverse-transcription polymerase chain reaction (RT-PCR) was performed using LA Taq (Takara, Japan). The conditions were as follows: one step of 5 min at 95°C and 35 cycles of 95°C for 30 s, 56°C for 30 s and 72°C for 30 s. The resulting cDNA was electrophoresed on 1% agarose gel and then imaged. Quantitative RT-PCR (qRT-PCR) was performed using Realtime PCR Master Mix (SYBR Green) (Code No. QPK-201, QIAGEN, Germany) using the following conditions: one step of 60 s at 95°C and 40 cycles of 95°C for 15 s, 56°C for 15 s and 72°C for 30 s. The calculated cycle threshold (C~T~) of each gene was normalized to the C~T~ for *rpoD* amplified from the corresponding sample. The reactions for RT-PCR and qRT-PCR were performed in ABI 9700 and ABI PRISM thermal cyclers, respectively (Applied Biosystems, USA). Fold-changes in gene expression were calculated according to the 2^−ΔΔCT^ method (Livak and Schmittgen, [@B26]). ###### **Oligonucleotide primers used during RT-PCR**. **Gene** **Forward (5′--3′)** **Reverse (5′--3′)** ---------- ---------------------- ---------------------- *lasI* TTGCTCGCCGCACATC GGCACGGATCATCATCTT *rhlI* ATCCGCAAACCCGCTAC GCAGGCTGGACCAGAATAT *lasR* CATCGTCGGCAACTACCC GCGCACCACTGCAACACT *phzM* GACATGGTGCTGTTCTACGG TGGAATGCCAGGTTGCTC *lasA* CTACAGCATCAACCCGAAAG TAGCGCCGCGACAACT *pslA* TACCGGGCCTGGATGA CGGCAGCGAGTTGTAGTT *lasB* GTTCTATCCGCTGGTGTCG CGCTGCCCTTCTTGATG *rsmA* AGACCCTGATGGTAGGTG AATGGTTTGGCTCTTGAT *gacA* AACTGGCCCGCGAACT GCGCCCTTGGTCATGTAG *mexA* TCCCTGAAGCTGGAGGACG TGCTGCGGAGCGAGGAT *ropD* AGGCCGTGAGCAGGGAT GGTGGTGCGACCGATGT Pyocyanin quantification assay ------------------------------ After 24 h or 72 h cultivation, the culture of *P. aeruginosa* was centrifuged at 10,800 × g for 1 min. The resulting supernatant (5 ml) was mixed with chloroform (3 ml) and then centrifuged at 4600 × g for 10 min. The chloroform phase was transferred to another centrifuge tube, mixed with 1 ml 0.2 M HCl and then centrifuged at 4600 × g for 10 min. The upper phase was taken to detect OD~520.~ The A OD~520~ reading was normalized by dividing by the final OD~600~ reading of the culture. The quantity of pyocyanin was calculated by multiplying OD~520~ by 17.072 (Kong et al., [@B24]). LasA staphylolytic assay ------------------------ LasA protease activities of different groups were measured by measuring the ability of stationary-phase *P. aeruginosa* culture supernatant to lyse boiled *Staphylococcus aureus* (Kessler et al., [@B23]). The LasA staphylolytic assay was performed according to Kong et al. ([@B24]). Statistical analysis -------------------- All data were analyzed by SPSS 17.0 statistical software, and expressed as mean ± standard deviation (SD). Different group of data were compared by Student\'s *T*-test. All experiments were carried out at least in triplicate. Results {#s3} ======= Effect of SH on growth of *P. aeruginosa* ----------------------------------------- We first determined the MIC for SH and the broad-spectrum antibiotic AZM, which is known to inhibit QS at sub-MIC concentrations (Bala et al., [@B2]) using the Micro-broth dilution method. We found the MICs for *P. aeruginosa* strain ATCC27853 to be 512 μg/ml and 64 μg/ml, respectively. The high MIC for SH would limit the clinical use of SH as a growth inhibitor to treat *P. aeruginosa* infections. The growth curves of *P. aeruginosa* treated by SH (Figure [1A](#F1){ref-type="fig"}) showed that SH inhibits the growth of *P. aeruginosa* in the early growth stages, before 30 h, at all tested concentrations. After 30 h, the solutions containing 64 μg/ml and 128 μg/ml SH showed no inhibitory activity toward *P. aeruginosa* and only those solutions containing \>256 μg/ml showed inhibitory activity. In the declining stage of the growth curve, the concentration of *P. aeruginosa* was independent of SH, with no significant difference between SH-containing cultures and the control. Measurement of CFU at 24 and 72 h showed that only cultures containing the full SH MIC (512 μg/ml) demonstrated lower CFU at 24 h than the control group, while at 72 h no differences were detected (Figure [1B](#F1){ref-type="fig"}). As expected, the growth curve and CFU results of cultures containing the AZM showed significant growth differences when compared with the control group. These results indicate that SH alone is inefficient at inhibiting growth under the conditions tested. We therefore assessed the specific effect of SH on QS. {#F1} Effect of SH on QS-regulated systems ------------------------------------ Addition of SH was shown to affect production of the QS-regulated chromogenic toxin, pyocyanin, the presence of which is indicated by the green coloration of the growth medium (Figure [2](#F2){ref-type="fig"}). We observed that addition of SH reduced the green pigmentation of the medium to a level similar to that of the positive control, AZM (Figure [S1](#SM1){ref-type="supplementary-material"}). Purification and quantification of the pyocyanin in the culture supernatants again demonstrated that addition of SH dose-dependently inhibits pyocyanin production (Figure [2](#F2){ref-type="fig"}), which in *P. aeruginosa* is regulated by the Las system (Williams and Camara, [@B51]). {#F2} To further investigate the potential for SH to disrupt QS, we used the Gram-negative bacterium *C. violaceum*, which produces a water-soluble purple dye (violacein) under the control of an AHL-controlled QS system (McClean et al., [@B29]). Specifically, we used a mutant derivative *C. violaceum* CV026 that lacks the gene (*cviI*) required to produce AHL, thus producing violacein only in response to exogenously supplied AHLs (McClean et al., [@B29]) and providing a convenient tool with which to screen QS inhibitors (Blosser and Gray, [@B3]). SH had a clear dose-dependent inhibitory effect on the production of violacein, indicating that SH was capable of blocking QS regulation in the system (Figure [S2](#SM2){ref-type="supplementary-material"}), which we further corroborated by spectrophotometric analysis after growth in liquid culture (Figure [3](#F3){ref-type="fig"}). {#F3} Expression of the AHL biosynthesis genes in response to SH ---------------------------------------------------------- To investigate the effects of SH on the expression AHL biosynthesis genes *lasI* and *rhlI*, we performed RT-PCR and qRT-PCR experiments. The RT-PCR results (Figure [4A](#F4){ref-type="fig"}) showed that *lasI* expression was down-regulated by the presence of SH. The qRT-PCR results revealed a dose-dependent down-regulation of *lasI* in response to increasing SH concentrations (Figure [4B](#F4){ref-type="fig"}), with a fold change in *lasI* levels of 2.50, 7.40, 8.13, 9.09 in response to the presence of 64, 128, 256, and 512 μg/ml SH, respectively. AZM, which was previously shown to effectively inhibit QS, down-regulated *lasI* 4.91 fold at its MIC of 64 μg/ml. Unexpectedly, we found that the expression *rhlI* was up-regulated by SH (Figure [4C](#F4){ref-type="fig"}) indicating that SH has a specific, dose-dependent effect on expression of the main AHL biosynthesis gene, *lasI*. {#F4} Expression of LasR and related genes and in response to SH ---------------------------------------------------------- LasR is the key regulator factor of the Las system of *P. aeruginosa* (Williams and Camara, [@B51]). The qRT-PCR results showed that the expression of *lasR* gene was strongly down-regulated by SH treatments (Figure [5](#F5){ref-type="fig"}). Furthermore, we found that *lasA* and *pslA*, which are regulated by LasR, were down-regulated in response to SH in a concentration-dependent manner (Figure [5](#F5){ref-type="fig"}). And the pyocyanin biosynthesis gene, phzM was also down-regulated by Sh in the dose-dependent manner (Figure [5](#F5){ref-type="fig"}). These results indicate that SH causes inhibition of many QS-regulated genes, including the main QS regulator, *lasR*. In addition, we also detected the expression of *lasB, gacA, rsmA*s, and *mexA* related in virulence factor, virulence regulation and drug resistance under the SH treatment. However, expression of these four genes is not significantly affected by SH (Figure [S3](#SM3){ref-type="supplementary-material"}). {#F5} LasA is an important virulence factor of *P. aeruginosa* and is positively regulated by LasR. Considering our observation of significant down-regulation of *lasR* in response to SH, we decided to monitor the effect of SH on production of LasA. We found that LasA enzymatic activity was significantly repressed by SH, even at concentrations below those that inhibit growth (Figure [6](#F6){ref-type="fig"}). Considering that SH also inhibited production of the toxin and important virulence factor, pyocyanin (Figure [2](#F2){ref-type="fig"}), our data suggest SH can be used to significantly inhibit the production of key *P. aeruginosa* virulence factors, independent of a direct effect on growth rate. {#F6} Discussion {#s4} ========== *P. aeruginosa* is one of the ESKAPE pathogens (*Enterococcus faecium, Staphylococcus aureus, Klebsiella pneumoniae, Acinetobacter baumanii, P. aeruginosa, and Enterobacteriaceae*) emphasizing their strong capacity to "escape" from routine antibacterial treatments (Boucher et al., [@B4]). The quorum sensing system is a key regulatory system which is responsible for the multi-drug resistance and pathogenesis of *P. aeruginosa* (Van Delden and Iglewski, [@B47]). Thus, development of quorum sensing inhibiting agents is one of the key areas in the Pseudomonas research field (Fothergill et al., [@B11]). Here we have demonstrated that SH can effectively inhibit the production of the QS-regulated virulence factors, LasA and pyocyanin, with sub-MIC concentrations sufficient to inhibit QS-regulated systems independent of an effect on growth. We found that SH causes a specific, dose-dependent repression of two components of the Las QS system, the AHL biosynthesis gene, *lasI*, and the transcriptional regulator of QS LuxR-type receptor, *lasR*. In line with these results, we also found that SH repressed expression of the *lasR* regulated genes *lasA*, and *pslA* (Gilbert et al., [@B17]; Jimenez et al., [@B22]). Among them, *lasA* encode secreted protease LasA which is a virulence factor of *P. aeruginosa* (Kessler et al., [@B23]), and *pslA* is the first gene of *psl* operon encoding the biosynthesis enzyme of Psl polysaccharide of biofilm matrix (Colvin et al., [@B7]). *PhzM* is responsible for pyocyanin production (Huang et al., [@B20]) *LasA* and *pslA* are directly and positively regulated by LasR in *P. aeruginosa*, and pyocyanin production is positively regulated by LasR (Gilbert et al., [@B17]; Jimenez et al., [@B22]). However, *lasB, gacA, rsmA* and *mexA* were not found to be down-regulated by SH treatment. *LasB* codes for the elastase, LasB which plays a role in pathogenesis of *P. aeruginosa* respiratory infections by rupturing the respiratory epithelium (Azghani, [@B1]) and is regulated by LasR and RhlR in combination (Gilbert et al., [@B17]; Jimenez et al., [@B22]). *GacA* and *rsmA* enconde regulators of GacA and RsmA of Gac/Rsm signal transduction pathway which positively controls quorum sensing in *P. aeruginosa* (Heeb et al., [@B19]). GacA is the positive regulator of *lasR* and *rhlR*, and RsmA negatively controls the *lasI* and *rhlI* (Williams and Camara, [@B51]). *MexA* is the first gene of operon enconding MexAB-OprM efflux pump responsible for intrinsic drug resistance of *P. aeruginosa* (Breidenstein et al., [@B5]; Poole, [@B36]). Expression of *mexAB-oprM* is positively regulated by C4-HSL in Rhl system (Evans et al., [@B10]; Sawada et al., [@B38]; Sugimura et al., [@B45]), and MexAB-OprM regulates QS in *P. aeruginosa* by controlling accessibility of non-cognate acyl-HSLs to LasR (Minagawa et al., [@B30]). Thus, *lasB, gacA, rsmA*, and *mexA* are not regulated by Las system solely or directly, and not down-regulated by SH treatment. Taken together, our results imply that SH can specifically inhibits the Las system and related genes expression. Molecules modulating QS LuxR-type receptors to interfere with bacterial virulence and biofilms is the most intensively investigated in the antiquorum sensing research (Wang and Ma, [@B49]). AHL Analogs (Smith et al., [@B44]), furanones (Givskov et al., [@B18]), benzoheterocyclics (Peters et al., [@B35]), 4-Nitropyridine-N-oxide (Rasmussen et al., [@B37]), thimerosal and phenyl percuric nitrate (Taha et al., [@B46]), azithromycin (Imperi et al., [@B21]), ceftazidime and ciprofloxacin (Skindersoe et al., [@B43]), tobramycin (Garske et al., [@B16]), solenopsin A (Park et al., [@B32]), and andrographolides (Ma et al., [@B27]) were found to possess the antiquorum sensing activity by modulate QS LuxR-type receptor, i.e., LasR or RhlR. Rather than affecting both the Las and Rhl QS systems in the same manner, our data also revealed the up-regulation of the Rhl system regulator, *rhlR*, indicating a possible compensation of the Las QS system by the Rhl system (Figure [4](#F4){ref-type="fig"}). In the available antiquorum sensing drugs, the chemical structure of synthetic AHL analogs is also close to SH. Among them, the ribolactam analogs and cyclic azahemiacetals were found to significantly block Las system at all concentrations tested and to moderately stimulate rhl, which is also similar to SH (Malladi et al., [@B28]). Thus, the results suggest that SH is a natural AHL analog. Interestingly, two other AHL analog compounds with 12-carbon alkyl tails have also been identified as specific inhibitors of the Las system (Muh et al., [@B31]), while AZM and 14-alpha-lipoyl andrographolide were shown to inhibit both the Las and Rhl systems (Ma et al., [@B27]; Imperi et al., [@B21]). While AZM is believed to affect QS through a more general effect on translation (Imperi et al., [@B21]), the similarity in the chemical structures of SH and AHL suggests that SH may compete with 3-oxo-C12-HSL for binding of LasR and thus specifically inhibit the sensing of this molecule (Figure [7](#F7){ref-type="fig"}). ![**Putative mechanism for the inhibition of *P. aeruginosa* QS by SH**. The similarity in the chemical structures of SH and 3-oxo-C12-HSL suggests that SH may compete with for signal molecule binding of LasR and thus specifically inhibit the sensing of this molecule to reduce the enzyme activity of LasR. Then, LasA and LasI activity, pyocyanin production and biosynthesis of biofilm matrix component Psl, which are positively regulated by LasR, are inhibited by SH treatment. Therefore, pathogenicity of *P. aeruginosa* may be attenuated by repression of extracellular virulence factors and biofilm formation of SH. LasR also positively regulates other extracellular virulence factors (LasB, ToxA, AprA, HcnABC), biofilm matrix component Pel, regulators (PqsR, VqsR, and so on) alone or combined with RhlR (Williams and Camara, [@B51]). Las system has a positive feedback (autoinduction) that at high population densities by increasing in QS signal molecule production and consequently enhanced expression of the target genes (Williams and Camara, [@B51]). Blue arrow besides LasR stands for the positive regulatory effect of LasR. The arrow between LasI and 3-oxo-C12-HSL means that LasI is responsible for biosynthesis of 3-oxo-C12-HSL. The arrow between 3-oxo-C12-HSL and LasR means that 3-oxo-C12-HSL bind of LasR to induce the enzyme activity of LasR. The arrow between polysaccharides, Psl, Pel and biofim stands for that Psl and Pel is the main matrix components of biofilm in *P. aeruginosa*. Red line with a round at the end stands for inhibiting effects.](fmicb-05-00635-g0007){#F7} In conclusion, our results demonstrate that the natural products of plants, especially those used in traditional medicine, could be an important source of clinically-relevant quorum sensing inhibitors. Conflict of interest statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. This work was supported by the National Natural Science Foundation of China (No. 81173629), Talent Introduction Foundation (No. 2013RC003) and Youth Natural Science Foundation (No.2014qn007) of Anhui University of Chinese Medicine. Supplementary material {#s5} ====================== The Supplementary Material for this article can be found online at: <http://www.frontiersin.org/journal/10.3389/fmicb.2014.00635/abstract> ###### **Color change of *P. aeruginosa* in response to SH**. The color change of *P. aeruginosa* in liquid culture with various concentrations of SH \[Control (without any drugs) (a), 64 μg/ml (1/8 × MIC) SH (b), 128 μg/ml (1/4 × MIC) SH (c), 256 μg/ml (1/2 × MIC) SH (d), 512 μg/ml (1 × MIC) SH (e) and 64 μg/ml (1 × MIC) AZM (f)\]. ###### Click here for additional data file. ###### **Plate assay demonstrating SH affecting AHL production**. The AHLs from different groups of *P. aeruginosa* of Control (a), 64 μg/ml (1/8 × MIC) SH (b), 128 μg/ml (1/4 × MIC) SH (c), 256 μg/ml (1/2 × MIC) SH (d), 512 μg/ml (1 × MIC) SH (e) and 64 μg/ml (1 × MIC) AZM (f) were extracted and added into the center of CV026 plates. ###### Click here for additional data file. ###### **Effect of SH on the expression of *lasB, gacA, rsmA* and *mexA***. The expression of *lasB, gacA, rsmA*, and *mexA* were monitored in response to SH treatment. Expression of the house-keeping gene, *rpoD*, was used as the internal control for each sample. The drug concentration of treatments was as follows: 512 μg/ml (1 × MIC) SH, 256 μg/ml (1/2 × MIC) SH, 128 μg/ml (1/4 × MIC) SH, 64 μg/ml (1/8 × MIC) and 64 μg/ml (1 × MIC) AZM. ###### Click here for additional data file. [^1]: Edited by: Filomena Nazzaro, Consiglio Nazionale delle Ricerche (National Research Council), Istituto di Scienze dell\'Alimentazione (Institute of Food Science), Italy [^2]: Reviewed by: Dmitri Debabov, NovaBay Pharmaceuticals, USA; Yuji Morita, Aichi Gakuin University, Japan; Hidetada Hirakawa, Gunma University, Japan [^3]: This article was submitted to Antimicrobials, Resistance and Chemotherapy, a section of the journal Frontiers in Microbiology. [^4]: †These authors have contributed equally to this work.

Introduction {#s1} ============ Dystonia is known as a group of clinically and etiologically heterogeneous disorders characterized by sustained or intermittent muscle contractions causing abnormal, often repetitive, movements, postures, or both ([@B1]). Genetically defective components play an important role in the genesis of dystonia. With the advent of next-generation sequencing (NGS) technology, an expanding spectrum of dystonia associated genes have been identified ([@B2]). In 2016, two groups independently reported mutations in a newly identified gene, lysine-specific histone methyltransferase 2B (*KMT2B*) in patients with early-onset generalized dystonia ([@B3], [@B4]). Till now, more than 40 patients with different *KMT2B* variants have been reported, among which only one is from China ([@B3]--[@B13]). In this study, we screened *KMT2B* in a cohort of early-onset Chinese dystonia patients by whole-exome sequencing to expand the current knowledge on this gene. Methods {#s2} ======= Subjects -------- This study was carried out in a cohort of 52 unrelated dystonia patients (29 females, 23 males) from the Movement Disorders Clinic in the Department of Neurology, Peking Union Medical College Hospital in China. Detailed demographic and clinical characteristics of all recruited patients were summarized in the [Table S1](#SM1){ref-type="supplementary-material"}. All these patients were required to be diagnosed as dystonia with onset of dystonia before 26 years old (the cut-off age of early-onset defined by Chinese dystonia guideline). The mean age at onset was 15.7 years ranging from 1 to 26. Among these patients, 18 had focal dystonia, 10 had segmental dystonia, 12 had multifocal dystonia, 11 had generalized dystonia, and 1 had paroxysmal dystonia. Patients who were suspected of acquired etiologies were excluded. Most patients (43/52, 82.7%) were categorized as isolated dystonia, whereas nine patients manifested with additional symptoms including myoclonus (2/9), cognitive impairment (3/9), parkinsonism (3/9), chorea(1/9), epilepsy (1/9), and impaired vision (1/9). The study was approved by the ethics committee of Peking Union Medical College Hospital. Written informed consents were obtained from all patients or their legal guardians. Genetic Analysis ---------------- Genomic DNAs were extracted from peripheral blood samples of all the patients. After exclusion of the two most common genes (*TOR1A, THAP1*) for dystonia by Sanger sequencing, whole-exome sequencings (WES) were carried out in all patients. The detailed methods for WES were shown in [Supplementary Materials](#SM1){ref-type="supplementary-material"}. Briefly, variants were called, aligned, annotated and filtered. Only missense, nonsense, splice-site, in-frame insertion/deletion and frameshift variants with population minor allele frequency (MAF) \<1% in public databases of normal human variation were selected for further assessment. The referenced public databases included 1000 Genomes Project (<http://www.ncbi.nlm.nih.gov/variation/tools/1000genomes/>), Exome Aggregation Consortium (ExAC; <http://exac.broadinstitute.org/>) and genome Aggregation Database (genomAD, <http://gnomad-old.broadinstitute.org/>). Pathogenicity prediction were performed by SIFT (<http://sift.jcvi.org>), PolyPhen-2 (<http://genetics.bwh.harvard.edu/pph2>), MutationTaster (<http://www.mutationtaster.org>), and Combined Annotation Dependent Depletion (CADD; <http://cadd.gs.washington.edu>). Potential causal variants were further verified by Sanger sequencing and tested for cosegregation in their available family members. Then these variants were tested in 100 Chinese unrelated healthy individuals by Sanger sequencing. The clinical effects of identified variants were classified according to the American College of Medical Genetics and Genomics (ACMG) standards and guidelines ([@B14]). Results {#s3} ======= In this study, three novel *KMT2B* (NM_014727.2) variants were detected, including one nonsense variant c.4075C\>T (p.Q1359^\*^) in patient 1, one frameshift variant c.4458delC (p.R1487AfsTer7) in patient 2, and one missense variant c.454C\>T (p.R152W) in patient 3 ([Table 1](#T1){ref-type="table"}). The clinical features of affected patients were summarized in [Table 2](#T2){ref-type="table"}. ###### Summary of *KMT2B* novel variants identified in three probands with dystonia. **Patient no**. **cDNA** **Protein** **Variant type** **Exon** **Inheritance** **1000 Genomes** **ExAc** **gnomAD** **PolyPhen2 (score)** **SIFT (score)** **Mutation Taster** **CADD** ----------------- ------------ ---------------- ------------------ ---------- ----------------- ------------------ ----------- ------------ --------------------------- ------------------ --------------------- ---------- 1 c.4075C\>T p.Q1359\* Nonsense 15 *De novo* Not found Not found Not found NA NA Disease causing 38 2 c.4458delC p.R1487AfsTer7 Frameshift 18 *De novo* Not found Not found Not found NA NA NA 29.5 3 c.454C\>T p.R152W Missense 3 Paternal 0.000599042 0.0002531 0.0002246 Probable damaging (0.990) Deleterious (0) Disease causing 25.7 ExAC, exome aggregation consortium; PolyPhen2, Polymorphism Phenotyping v2; SIFT, Sorting intolerant from tolerant; CADD, Combined annotation dependent depletion; gnomAD, genome Aggregation Database; NA, not available. *KMT2B reference sequence used: NM_014727.2*. ###### Clinical findings of patients with *KMT2B* variants in this study. **Patient** **Patient 1** **Patient 2** **Patient 3** -------------------------- --------------------------------------------------- -------------------------- ------------------------- Genotype p.Q1359\* p.R1487AfsTer7 p.R152W Age/Sex 31/M 15/F 33/F Familial history No No No Age at onset 24 7 18 Site of onset Larynx Feet Right hand Involuntary movements Dystonia Dystonia Dystonia Distribution of dystonia Left leg, larynx Feet, right hand, larynx Neck, trunk, right hand Other features Short statue, microcephaly, and bulbous nasal tip No No Brain MRI Normal Normal Normal Patient 1 --------- Patient 1 was a 31 year-old man who was the only child born to his healthy non-consanguineous parents. Family history and delivery history of the patient were unremarkable. At the age of 24, he first developed slurred speech. The symptom progressed slowly. Mild dystonia of left leg were later noticed when he was admitted to our hospital ([Video S1](#SM2){ref-type="supplementary-material"}). Also, short statue, microcephaly and bulbous nasal tip were present in this patient. However, there was no evidence of additional neurological features such as developmental delay, intellectual disability and seizure. Brain magnetic resonance imaging (MRI) was unremarkable from the age of 24 to 31. Administration of medication, including levodopa, baclofen and benzhexol, did not show any clinical benefit. Whole-exome sequencing (WES) was performed and detected a novel heterozygous stop-gain variant c.4075C\>T (p.Q1359^\*^) in *KMT2B*. The variant was absent in dbSNP, ExAc, 1000 Genomes, and gnomAD. It was predicted to "disease causing" by MutationTaster and scored 38 by CADD. Segregation analysis revealed that the stop-gain variant was absent in his parents which indicated it was *de novo* ([Table 1](#T1){ref-type="table"}; [Figure 1](#F1){ref-type="fig"}). {#F1} Patient 2 --------- Patient 2 was a 15 year-old girl from a non-consanguineous family with negative family history. She firstly experienced abnormal gait with dystonia posturing in her feet at the age of 7 years. During the following years, she developed slurred speech, followed by abnormal posture of right hand, leading to handwriting difficulty ([Video S2](#SM3){ref-type="supplementary-material"}). The severity of these symptoms mildly progressed, especially the severe dysphonia. Neuroimaging of brain showed no obvious abnormalities. Medical interventions including levodopa were of little effect. The testings of gene *GCH1, TOR1A*, and *THAP1* did not detect any pathogenic variants. WES was then performed and uncovered a heterozygous *KMT2B* frameshift variant c.4458delC (p.R1487AfsTer7), which was never reported before. This variant was absent in dbSNP, ExAc, 1000 Genomes and gnomAD, and predicted a score of 29.5 by CADD. Segregation analysis demonstrated the *de novo* status of the variant ([Table 1](#T1){ref-type="table"}; [Figure 2](#F2){ref-type="fig"}). {#F2} Patient 3 --------- Patient 3 was a 33 year-old woman, who first developed right-handed writer\'s cramp at 18 years old. The symptom was improved after the treatment of botulinum toxin A injection. At 28 years old, a remarkable spasmodic torticollis was developed and progressively deteriorated in the following years. On current examination, in addition to the demonstrated cervical dystonia, abnormal posture of trunk was also noted ([Video S3](#SM4){ref-type="supplementary-material"}). In this patient, exome sequencing uncovered a novel heterozygous missense variant c.454C\>T (p.R152W) in *KMT2B*, which was predicted to be damaging by SIFT, PolyPhen2, and MutationTaster. The CADD score was 25.7. The variant was rare in ExAc, 1000 Genomes and gnomAD (MAF 0.02531%, 0.0599042%, and 0.02246%, respectively). Segregation analysis revealed that the variant was inherited from her father who did not show any abnormal symptoms at current examination ([Table 1](#T1){ref-type="table"}; [Figure S1](#SM1){ref-type="supplementary-material"}). In addition, several variants in other dystonia related genes were also identified, including *GCH1* c.638_641del in patient 4, *PINK1* c.1474C\>T and c.938C\>T in patient 7, *SGCE* c.304C\>T in patient 26, *VPS13A* c.7867C\>T in patient 23, and *ANO3* c.970A\>G in patient 40. Discussion {#s4} ========== *KMT2B* is a large gene (NM_014727.2: 37 exons, 8,148 bp), encoding a lysine methlytransferase specifically responsible for the methylation of histone H3 at lysine 4 (H3K4), an important epigenetic modification associated with active gene transcription ([@B4]). Its mutations have been recently reported to cause early-onset generalized dystonia ([@B3], [@B4]). Up to now, more than 40 patients with different *KMT2B* variants have been reported, including cases with interstitial microdeletion encompassing the entire gene, as well as cases with pathogenic variants (frameshift, in-frame deletion, splice-site, nonsense and missense variants) ([@B3]--[@B11]). For the majority of patients, *KMT2B* variants were confirmed as de novo, but autosomal dominant inheritance with reduced penetrance was also reported ([@B4]). In this study, we found one nonsense mutation, one frameshift mutation and one missense mutation in three dystonia patients, among which c.4075C\>T (p.Q1359^\*^) was present in patient 1, c.4458delC (p.R1487AfsTer7) in patient 2, and c.454C\>T (p.R152W) in patient 3. According to the criteria published by the ACMG ([@B14]), the variant c.4075C\>T (p.Q1359^\*^) in patient 1 was rated as "pathogenic" because this variant was a null variant in *KMT2B* \[very strong pathogenic criterion (PVS)\], proven to be *de novo* in origin \[strong pathogenic criterion 2 (PS2)\], absent from population databases \[moderate pathogenic criterion 2 (PM2)\], and predicted to be deleterious by multiple computational methods \[supporting pathogenic criterion 3 (PP3)\]. Of note, this nonsense variant occurred within the functionally important PHD domain ([@B4]). Similar to above, the frameshift variant c.4458delC in patient 2 should also be categorized to be "pathogenic," because it belonged to PVS, PS2, PM2, and PP3. In contrast, the missense variant c.454C\>T in patient 3 was of uncertain significance because it only met the criteria PP3 (*In silico* analysis supporting a deleterious effect) according to the consensus of ACMG. The variant was not located in any functional domains of KMT2B protein. Notably, co-segregations of this patient were difficult to judge because of the possible incomplete penetrance of *KMT2B*. Consequently, further functional verification of this missense variant may be helpful. Regarding the phenotypes, patients with *KMT2B* variants present a relatively similar disease course progressively evolving from childhood-onset lower-limb dystonia into generalized dystonia with prominent bulbar and craniocervical involvement ([@B3], [@B4]). In our study, between the patients with pathogenic *KMT2B* variants, patient 2 showed the similar presentations as reported whereas patient 1 seemly harbored the atypical presentations. Firstly, patient 1 had a later onset age (24 years) than the typical *KMT2B*-mutated patients (5.8 years in the original report) ([@B3], [@B4]). Secondly, the site of onset with larynx was also atypical because lower-limb dystonia was the initial symptom in the majority patients. What\'s more, as the core feature, the symptoms of dystonia in this patient were seemly much milder than previously reported patients. As reported, in addition to dystonia, other features were present to varying degrees in the majority of patients, including developmental delay, intellectual disability, oculomotor disturbances, microcephaly, and dysmorphic features (such as elongated face, bulbous nasal tip, and short stature) ([@B4]). In our study, patient 1 showed short statue, microcephaly and bulbous nasal tip in addition to dystonia, whereas patient 2 presented with isolated dystonia. Finally, symmetrical hypointensity of the globus pallidi on brain MRI were observed in some reported cases ([@B4]). However, no abnormal neuroimaging findings were showed in our patients. In addition, patient 3 also presented with atypical manifestations, including a later onset age (18 years old), atypical site of onset (right hand) and milder degree of dystonia. However, the meaning of the phenotype is unclear concerning the uncertainty of the pathogenicity of R152W. With the development of sequencing technology, more and more atypical cases with *KMT2B* mutations have been reported, such as paroxysmal cervical dystonia, or isolated oromandibular dystonia, or global development delay without any evidence of dystonia ([@B4], [@B9]). Thus, clinically heterogeneous phenotypes bring great challenge on precise diagnosis of dystonia. This study suggests that performing WES on affected individuals is an effective method for mapping genes of patients with possible genetic cause. In conclusion, we report three dystonia patients with novel variants in *KMT2B* and broaden the spectrums of genotype and phenotype of *KMT2B*. Among these variants, the novel nonsense variant c.4075C\>T (p.Q1359^\*^) and the frameshift variant c.4458delC (p.R1487AfsTer7) show high pathogenicity whereas the missense variant need to further verify. Data Availability {#s5} ================= The sequencing data in our manuscript has been uploaded to SRA (Sequence Read Archive) of NCBI. The SRA accession is [PRJNA549023](PRJNA549023). It will be accessible with the following link after the indicated release date: <https://www.ncbi.nlm.nih.gov/sra/PRJNA549023>. Ethics Statement {#s6} ================ The study was approved by the ethics committee of Peking Union Medical College Hospital. Written informed consents were obtained from all patients or their legal guardians in accordance with the Declaration of Helsinki. Author Contributions {#s7} ==================== JM: conception of the work, data acquisition, statistical analysis, and writing of the first draft. XW: design and organization of the work, manuscript review, and critique. LW, YY, and SL: data acquisition, manuscript review, and critique. Conflict of Interest Statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The authors thank all the study participants. **Funding.** This study was supported by National Key Research and Development Program of China (2018YFC1314700). Supplementary Material {#s8} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fneur.2019.00729/full#supplementary-material> ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. ###### Click here for additional data file. [^1]: Edited by: Antonio Pisani, University of Rome Tor Vergata, Italy [^2]: Reviewed by: Niccolo Mencacci, Northwestern University, United States; Yih-Ru Wu, Chang Gung Memorial Hospital, Taiwan [^3]: This article was submitted to Movement Disorders, a section of the journal Frontiers in Neurology

Li J, Wang X‐H, Hu J, Shi M, Zhang L, Chen H. Combined treatment with N‐acetylcysteine and gefitinib overcomes drug resistance to gefitinib in NSCLC cell line. Cancer Med. 2020;9:1495--1502. 10.1002/cam4.2610 31891230 Jun Li and Xiao‐Hui Wang contributed equally to this work. 1. INTRODUCTION {#cam42610-sec-0001} =============== Lung cancer, one of the most aggressive cancers, accounts for the leading cause of cancer‐related mortality worldwide,[1](#cam42610-bib-0001){ref-type="ref"}, [2](#cam42610-bib-0002){ref-type="ref"} and non‐small cell lung cancer (NSCLC) contributes to around 85% of lung cancer. Gefitinib and erlotinib, two major epithelial growth factor receptor tyrosine kinase inhibitors (EGFR‐TKIs), display promising therapeutic efficacy in patients with NSCLC carrying EGFR‐activating mutations (for example, exon 19 deletion and exon 21 L858R), and these patients regularly receive EGFR‐TKIs as a first‐line treatment.[3](#cam42610-bib-0003){ref-type="ref"}, [4](#cam42610-bib-0004){ref-type="ref"}, [5](#cam42610-bib-0005){ref-type="ref"}, [6](#cam42610-bib-0006){ref-type="ref"}, [7](#cam42610-bib-0007){ref-type="ref"} However, almost all patients become resistant to gefitinib and erlotinib within a median time period of approximately 10 months.[8](#cam42610-bib-0008){ref-type="ref"}, [9](#cam42610-bib-0009){ref-type="ref"}, [10](#cam42610-bib-0010){ref-type="ref"} However, the mechanisms of EGFR‐TKI resistance are largely unknown. Therefore, exploring the underlying mechanisms is beneficial for developing new strategies to overcome this problem, which might improve the prognosis of patients with NSCLC. Currently, multiple factors are involved in the resistance mechanisms, including the T790M secondary mutation in EGFR, HGF overexpression, MET amplification, transition into small cell lung carcinoma, obtaining cancer stem‐cell phenotypes and epithelial‐mesenchymal transition (EMT).[11](#cam42610-bib-0011){ref-type="ref"}, [12](#cam42610-bib-0012){ref-type="ref"}, [13](#cam42610-bib-0013){ref-type="ref"}, [14](#cam42610-bib-0014){ref-type="ref"} EMT is a universal phenomenon in various physiological and pathological processes. It is clear that epithelial cells display characteristics of mesenchymal cells, accompanied by upregulation of vimentin and N‐cadherin, and downregulation of E‐cadherin during EMT. EMT also contributes to tumor invasion, proliferation, metastasis, and therapy resistance to EGFR‐TKIs.[15](#cam42610-bib-0015){ref-type="ref"}, [16](#cam42610-bib-0016){ref-type="ref"} As a consequence, targeting EMT might be a potential strategy to reverse or prevent EGFR‐TKIs resistance. N‐acetylcysteine (NAC) is an effective antioxidant widely used in anticancer investigation in recent years. Our previous studies have demonstrated that NAC could overcome gefitinib resistance mediated by cigarette smoke extract (CSE). However, whether NAC plays a critical role in non‐smokers would explore the combined effect of NAC with gefitinib on gefitinib‐resistant cells and the underlying mechanisms. 2. MATERIALS AND METHODS {#cam42610-sec-0002} ======================== 2.1. Cell culture {#cam42610-sec-0003} ----------------- PC‐9 gefitinib‐sensitive cells (PC‐9)[17](#cam42610-bib-0017){ref-type="ref"}, [18](#cam42610-bib-0018){ref-type="ref"} and gefitinib‐resistant cells (PC‐9/GR) were gifts from Dr Jian Zhang at Xijing Hospital, Fourth Military Medical University, China. It is known that exon19 deletion is one of the hall‐marks in EGFR‐activating mutation, and PC‐9 cell line is characterized by exon19 deletion of lung cancer cells. The mutation profile is exon19(E746‐A750)del for PC‐9/GR.[19](#cam42610-bib-0019){ref-type="ref"} Cells were cultured in RPMI 1640 medium (Hyclone, USA) with 10% fetal bovine serum (PAN, USA) at 37°C in a cell incubator containing 5% CO~2~. In addition, PC‐9/GR cell was cultured in medium containing 10 nmol/L of gefitinib to maintain resistance. 2.2. Reagents {#cam42610-sec-0004} ------------- NAC, 4ʹ,6‐diamidino‐2‐phenylindole (DAPI), dimethyl sulfoxide (DMSO), and Triton X‐100 were purchased from Sigma‐Aldrich. PP2 and gefitinib were purchased from Tocris Bioscience and Abcam, respectively. Rabbit monoclonal antibodies against E‐cadherin (24E10), Src (32G6), Phospho‐Src (Ty416), and mouse monoclonal antibody against GAPDH were purchased from Cell Signaling Technology. Bax (WL01637) and Bcl‐2 (WL01556) were purchased from Wanleibio. Rabbit monoclonal antibody against Vimentin (EGFR3776) was purchased from Abcam. 2.3. Cell growth assay {#cam42610-sec-0005} ---------------------- Cell proliferation was evaluated with a CCK8 kit (Dojindo Laboratories). In brief, Cells (5 × 10^3^ cells) were seeded into 96‐well plates, cultured overnight and treated with various concentrations of drugs for 48 hours. Then, 10 µL of CCK8 was added to each well and cells were incubated for 2 hours. Optical density (OD) was set at 450 nm by Mircroplate Reader. 2.4. Combination studies {#cam42610-sec-0006} ------------------------ Combination studies were performed as described previously.[20](#cam42610-bib-0020){ref-type="ref"}, [21](#cam42610-bib-0021){ref-type="ref"} On the basis of the median‐effect analysis by Chou and Talalay (CalcuSyn software, Biosoft: Chou, 2010), the effects of drugs were calculated using the CI method for each experimental condition.[22](#cam42610-bib-0022){ref-type="ref"}, [23](#cam42610-bib-0023){ref-type="ref"} 2.5. Western blotting analysis {#cam42610-sec-0007} ------------------------------ The precise procedure was in accordance with our previous methods.[24](#cam42610-bib-0024){ref-type="ref"} In brief, proteins were separated by SDS‐PAGE gel and transferred to PVDF membranes. After blocking with 5% fat‐free milk or BSA, the membranes were incubated with primary antibodies overnight at 4°C, followed by incubation with secondary antibodies for 2 hours at room temperature. The protein bands were detected. 2.6. RNA isolation and qRT‐PCR {#cam42610-sec-0008} ------------------------------ The procedure of RNA extraction and cDNA synthesis was based on our previous study.[25](#cam42610-bib-0025){ref-type="ref"} The following primers were used: E‐cadherin, 5‐CGTAGCAGTGACGAATGTGG‐3(F) and 5‐CTGGGCAGTGTAGGATGTGA‐3(R); vimentin, 5‐GAGTCCACTGAGTACCGGAG‐3(F) and 5‐ACGAGCCATTTCCTCCTTCA‐3(R); GAPDH, 5‐ACCTGACCTGTCTAGAA‐3(F) and 5‐TCCACCACCTGTTGCTGTA‐3(R). The relative expression of indicated mRNAs was normalized to GAPDH. 2.7. Cell invasion assay {#cam42610-sec-0009} ------------------------ Sixty microliters of Matrigel (Becton Dickinson) was added into the center of each chamber (Millipore). The cells were seeded in the upper chamber of the insert with or without drugs, after incubation for 24 hours at 37°C with 5% CO~2~. The upper surface of the insert was scraped and the cells on the lower surface of the membrane were fixed with methanol and stained with 0.1% crystal violet solution (Becton Dickinson). Cells in the bottom of the membrane were counted using a light microscope. 2.8. Measurement of cell migration {#cam42610-sec-0010} ---------------------------------- Cells were planted into six‐well plates to create a confluent monolayer with up to 80% cell confluence, and then starved for 24 hours with serum‐free medium, followed by scratching with a sterile 200 µL tip to manually create a wound. The cells were washed with PBS and cultured in medium supplemented with NAC, gefitinib or a combination of both. Images were acquired by inverted optical microscope after creating the wound. 2.9. Measurement of apoptotic rate {#cam42610-sec-0011} ---------------------------------- Apoptotic cells were determined by Annexin V‐FITC/propidium iodide (PI) Kit according to manufacturer\'s guidelines. The specific process referred to our previous work.[24](#cam42610-bib-0024){ref-type="ref"} Briefly, PC‐9/GR cells were treated with NAC, gefitinib or a combination of both for 48 hours, respectively. Afterward, cells were harvested, and stained with Annexin V‐FITC and PI according to manufacturer\'s recommendations (Beyotime Institute of Biotechnology). 2.10. Statistical analysis {#cam42610-sec-0012} -------------------------- The data were represented as mean ± standard deviation (SD) or 95% confidence interval (CI) for three independent experiments. The GraphPad Prism software (version 5.0, GraphPad Software) was used for statistical analysis. The comparison between two independent treatment groups was analyzed by unpaired, two‐tailed Student\'s *t* test. One‐way analysis of variance (ANOVA) was utilized to analyse the variance among multi‐sample. Statistical significance was assumed at *P* \< .05. 3. RESULTS {#cam42610-sec-0013} ========== 3.1. Determination of dose‐response curves and PC‐9/GR EMT phenotype characteristics {#cam42610-sec-0014} ------------------------------------------------------------------------------------ Similar to our previous findings, the IC50 value of PC‐9/GR for gefitinib was 7.711 μmol/L (95% CI: 7.058‐9.657 μmol/L; Figure [S1](#cam42610-sup-0001){ref-type="supplementary-material"}A) which was increased 141‐fold compared with that of PC‐9 cells (IC50: 0.05471 μmol/L, 95% CI: 0.04378‐0.06835 μmol/L; Figure [S1](#cam42610-sup-0001){ref-type="supplementary-material"}B). This result showed a highly resistant effect to gefitinib in PC‐9/GR cells. As presented in Figure [S1](#cam42610-sup-0001){ref-type="supplementary-material"}C, E‐cadherin was expressed in PC‐9 cells, but downregulated in PC‐9/GR cells. In addition, vimentin was upregulated in PC‐9/GR cells, which was absent in PC‐9 cells. These results demonstrated an EMT phenotype characteristic of PC‐9/GR cells. The IC50 of NAC in PC‐9/GR cells was 15.53 mmol/L (95% CI: 14.50‐16.62 mmol/L; Figure [S1](#cam42610-sup-0001){ref-type="supplementary-material"}D). {#cam42610-fig-0001} 3.2. CalcuSyn‐based analysis of NAC and gefitinib combination treatment {#cam42610-sec-0015} ----------------------------------------------------------------------- The constant combination ratio experiments were carried out at an equipotency ratio approximating their individual IC50 (IC50~NAC~: IC50~gefitinib~ ≈ 2:1), which made sure the effect of each drug in combination was roughly equal. Figure [1](#cam42610-fig-0001){ref-type="fig"}A showed the dose‐response curves for PC‐9/GR cells exposed to NAC, gefitinib and both. CI values of the group treated with a combination of both drugs in different fractional cell growth inhibition (Fa) were shown in Figure [1](#cam42610-fig-0001){ref-type="fig"}B. CI values of less than 1 were acquired from the combination group, demonstrating that the two drugs must have a synergistic effect on growth inhibition. Then, PC‐9/GR cells were treated with 5 mmol/L of NAC adding different concentrations of gefitinib. We found that the IC50 of gefitinib was 0.3986μmol/L in the combination group, which was lower than gefitinib alone (*P* \< .001; Figure [1](#cam42610-fig-0001){ref-type="fig"}C). 3.3. Combination of NAC and gefitinib inhibited migration and invasion of PC‐9/GR cells {#cam42610-sec-0016} --------------------------------------------------------------------------------------- To investigate whether NAC (5 mmol/L) in combination with gefitinib (2 μmol/L) had an impact on biological behavior of PC‐9/GR cells, we performed migration and invasion assays. After 48 hours of treatment with NAC or gefitinib alone or in combination (NAC + gefitinib group), the number of cells passing through the Matrigel decreased in the NAC + gefitinib group compared to that in either alone group (Figure [2](#cam42610-fig-0002){ref-type="fig"}A). Cell migration assay showed that the distance of cell migration was the shortest in the NAC + gefitinib group (Figure [2](#cam42610-fig-0002){ref-type="fig"}B). These data illustrated that NAC in combination with gefitinib could inhibit the invasion and migration of PC‐9/GR cells. {#cam42610-fig-0002} 3.4. NAC in combination with gefitinib promoted apoptotic rate in PC‐9/GR cells {#cam42610-sec-0017} ------------------------------------------------------------------------------- Furthermore, we detected the apoptosis of PC‐9/GR cells under different treatments using flow cytometry analysis. NAC + gefitinib caused more apoptotic cells compared with NAC or gefitinib alone did (*P* \< .01; Figure [3](#cam42610-fig-0003){ref-type="fig"}A,B). Bax and Bcl‐2 are known as pro‐apoptotic and anti‐apoptotic molecules, respectively. The protein level of Bcl‐2 was decreased in the NAC + gefitinib group. While treatment with NAC or geftinib alone led to higher expression of Bax (Figure [3](#cam42610-fig-0003){ref-type="fig"}C). The trend of protein level was observed for Bcl‐2 and Bax in the combination group compared to the other groups. These results demonstrated that NAC in combination with gefitinib promoted apoptosis of PC‐9/GR cells. {#cam42610-fig-0003} 3.5. NAC in combination with gefitinib reversed EMT and inhibited Src activation in PC‐9/GR cells {#cam42610-sec-0018} ------------------------------------------------------------------------------------------------- We next explored the underlying mechanism that led to the superior efficacy of NAC in combination with gefitinib. Western blot analysis showed that NAC in combination with gefitnib facilitated E‐cadherin expression and inhibited Vimentin expression in PC‐9/GR cells. In addition, the protein level of p‐Src was decreased in the combination group (Figure [4](#cam42610-fig-0004){ref-type="fig"}A). In qRT‐PCR analyses, expression of the E‐cadherin increased and vimentin was decreased in the NAC + gefitinib group (Figure [4](#cam42610-fig-0004){ref-type="fig"}B). Then, we treated PC9/GR with PP2 (a potent inhibitor of Src, 10 µmol/L) in combination with gefitinib, which displayed the similar results as NAC + gefitinib did (Figure [4](#cam42610-fig-0004){ref-type="fig"}C,D). These results indicated that NAC in combination with gefitinib could inhibit Src activation and reverse EMT. {#cam42610-fig-0004} 4. DISCUSSION {#cam42610-sec-0019} ============= NSCLC patients with EGFR mutations initially get good responses to EGFR‐TKIs. However, these patients gradually acquire drug resistance inevitably. Therefore, it is of necessity to develop novel strategies to delay or overcome the acquired resistance to EGFR‐TKIs. One of the interesting approaches would be the combination treatment with alternative drug in addition to gefitinib. Drug combination is widely used and has become the primary treatment modality for cancer.[26](#cam42610-bib-0026){ref-type="ref"} A combination of gefitinib with metformin shows a synergistic effect and increases sensitivity of patients with lung cancer to gefitinib.[27](#cam42610-bib-0027){ref-type="ref"} In this study, we used NAC in combination with gefitinib, which could effectively overcome drug resistance to gefitinb. NAC is a precursor of glutathione, a powerful antioxidant used in anti‐tumor research. It has been reported that NAC inhibits proliferation and invasive behavior of human cancer cells in vitro, including colorectal cancer, bladder cancer, prostate cancer, tongue cancer, and lung carcinoma.[28](#cam42610-bib-0028){ref-type="ref"}, [29](#cam42610-bib-0029){ref-type="ref"}, [30](#cam42610-bib-0030){ref-type="ref"}, [31](#cam42610-bib-0031){ref-type="ref"} Our previous study showed that NAC could overcome gefitinib resistance induced by cigarette smoke extract. NAC has been proven to inhibit the growth of lung carcinomas by reducing cell proliferation and facilitating apoptosis in tobacco carcinogen‐treated A/J mice.[32](#cam42610-bib-0032){ref-type="ref"} NAC exerts inhibitory effect on tumor growth via modulation of EGFR/AKT signaling and HBP1 expression in EGFR‐overexpressed oral cancer.[33](#cam42610-bib-0033){ref-type="ref"} It can be inferred that NAC has promising potential to be a novel anticancer agent. In this study, we demonstrated that NAC could overcome gefitinib resistance by inhibiting Src activation and reversing EMT in PC‐9/GR cells when used in combination with gefitinib. It has been demonstrated that Src is a key regulator of EMT in cancer cells.[34](#cam42610-bib-0034){ref-type="ref"}, [35](#cam42610-bib-0035){ref-type="ref"} Our research verified that NAC modulated EMT of lung cancer cells by inhibiting the activation of Src, which might clarify the underlying mechanism theoretically. EMT is one of hallmarks in cancer and plays a crucial role in the development and progression of most solid tumors.[36](#cam42610-bib-0036){ref-type="ref"} As a candidate mechanism of drug resistance, EMT plays a crucial role in acquired EGFR‐TKIs resistance as reported by different research groups.[37](#cam42610-bib-0037){ref-type="ref"}, [38](#cam42610-bib-0038){ref-type="ref"} It has been demonstrated that EMT could be regarded as a predictor of therapy response in patients with NSCLC. Furthermore, there is a direct connection between EMT and EGFR‐TKIs sensitivity. Research has shown that E‐cadherin potentiates the sensitivity to gefitinb to improve the effect.[39](#cam42610-bib-0039){ref-type="ref"} Notably, EMT co‐occurs with EGFR T790M mutations in a study using re‐biopsies.[40](#cam42610-bib-0040){ref-type="ref"} Multiple signaling networks modulate EMT, such as transforming growth factor‐β1 (TGF‐β1), mitogen‐activated protein (MAP) kinases, RHOA, AKT, and STAT3. Multiple molecules are involved in EMT, including Src, a key oncogene. Src is the first oncogene ever identified, and its family members have also been recognized as potential targets in cancer therapy. Activation of Src promotes numerous pathological processes, including invasion, migration, proliferation, and angiogenesis in a variety of cancers.[41](#cam42610-bib-0041){ref-type="ref"} Increased Src activity boosts EMT process, while Src inhibition suppresses this process. Moreover Src‐mediated EMT is involved in the chemotherapy resistance of cancers.[42](#cam42610-bib-0042){ref-type="ref"}, [43](#cam42610-bib-0043){ref-type="ref"} Some studies concentrate on the combination of EGFR inhibitors and Src inhibitors. A previous study has revealed that the efficacy of Src inhibitors combined with EGFR inhibitors is synergistic, and Src inhibitors could improve gefitinib resistance in NSCLC with EMT.[44](#cam42610-bib-0044){ref-type="ref"} Our previous studies have shown that Src is involved in cigarette smoke‐induced EMT and EGFR‐TKI resistance, and Src inhibition sensitizes resistant cells to gefitinib. However, further investigations are needed to determine whether this combination in vivo could delay drug resistance and prolong patient progression‐free survival (PFS) and overall survival (OS). A lack of in vivo study is an obvious limitation in the current study, and we will conduct in vivo experiments in animal models to verify the cytological results in this study. We have not further investigated the up‐and‐down pathway of apoptotic proteins, which is another limitation of this study. To prove the synergistic effect of this combination therapy for EGFR mutant lung cancer, the future study should demonstrate with other EGFR mutant lung cancer cell lines. In addition to gefitinib, osimertinib has obtained positive results from the phase 3 AURA trial and phase 3 FLAURA trial.[45](#cam42610-bib-0045){ref-type="ref"}, [46](#cam42610-bib-0046){ref-type="ref"} With the approval of osimertinib in Nov 2015 (<https://www.fda.gov/>), it will be clinically meaningful if osimertinib has the similar effect when combined with NAC. In conclusion, we used the combination index value to evaluate the efficacy of NAC and gefitinib combination treatment. Our results demonstrated that NAC had synergistic effect with gefitinib. These two drugs used in combination overcame the resistance of PC‐9/GR to gefitinib by inhibiting Src activation and reversing EMT. Thus, these findings highlighted a novel insight into overcoming gefitinib resistance and provided a potential strategy for NSCLC patients. CONFLICT OF INTEREST {#cam42610-sec-0020} ==================== We declare no conflicts of interest in association with this study. Supporting information ====================== ######   ###### Click here for additional data file. This study was supported by the National Health and Family Planning Commission of the People\'s Republic of China, Grant/Award Number: W2014RQ29. The authors thank Dr Jian Zhang (Xijing Hospital, the Fourth Military Medical University, China) for kindly providing the cell lines used in the current study and Mr Bowei Li, Dr Xinxin Zhang and Dr Long‐Biao Cui (Xijing Hospital, the Fourth Military Medical University, China) for comments and suggestions on the manuscript.

{#sp1 .230} {#sp2 .231}

###### Strengths and limitations of this study - This is the most comprehensive review to date of mandatory reporting of child maltreatment, focusing on mandated reporters' (MRs) experiences with reporting. - Although a systematic search was conducted, little information about mandatory reporting from low- and middle-income countries was retrieved. - Critical appraisal of included articles followed an established checklist and reporting of synthesis findings was done according to the Enhancing Transparency in Reporting the Synthesis of Qualitative research statement. - This meta-synthesis used an established method for synthesising study findings that enabled the creation of recommendations for MRs relating to the reporting process Introduction {#s1} ============ Global estimates of child maltreatment indicate that nearly a quarter of adults (22.6%) have suffered childhood physical abuse; over a third of adults (36.3%) have suffered childhood emotional abuse; 16.3% of adults have suffered childhood neglect and 18% of women and 7.6% of men, respectively, have suffered childhood sexual abuse.[@R1] These estimates vary across countries. For example, according to 2015 US child protective services (CPS) reports, 63.4% of reported children experienced neglect.[@R4] Given the high prevalence of child maltreatment and its potentially serious, long-term health and social consequences,[@R5] many countries have taken steps to prevent child maltreatment and reduce its associated impairment, including through the introduction of mandatory reporting. Mandatory reporting law, in the context of child maltreatment, "is a specific kind of legislative enactment which imposes a duty on a specified group or groups of persons outside the family to report suspected cases of designated types of child maltreatment to child welfare agencies".[@R9] The USA enacted the first mandatory reporting laws in 1963.[@R10] These laws were more narrowly conceived, requiring certain mandated professions to report 'severe' or 'significant' physical abuse by parents or caregivers. Over time, legislation has expanded in the USA and has been replicated in other countries. Across jurisdictions, mandatory reporting can include other forms of maltreatment (notably physical, sexual and emotional abuse, neglect, children's exposure to intimate partner violence (IPV) and prenatal exposure to drug abuse), reporting by more than mandated professionals (eg, by all citizens), reporting abuse perpetrated by non-caregivers and reporting beyond 'severe' or 'significant' abuse.[@R12] Some information about the international context of mandatory reporting is available, but in general little information about this process is available from low- and middle-income countries (LMICs) (see online [supplementary file 1](#SP1){ref-type="supplementary-material"}). Furthermore, while we began this project with the intent of doing a systematic review of studies of effectiveness about mandatory reporting, we were unable to find any studies that could be used for this purpose (ie, no prospective controlled trials, cohort studies or case--control studies assessing the effectiveness of mandatory reporting in relation to child outcomes were retrieved from our systematic search). Instead, we found that while there are a handful of prospective studies assessing particular outcomes of mandatory reporting,[@R13] most of the research discussing its impact relies on retrospective analysis of CPS reports[@R15] or is related to mandated reporters' (MRs), children's and caregivers' perceptions about reporting, as discussed in surveys,[@R19] qualitative literature[@R28] or case reports[@R31] (qualitative literature is summarised in this meta-synthesis). Given the paucity of data on effectiveness of mandatory reporting, the purpose of this meta-synthesis is to summarise qualitative research about MRs' experiences with reporting. A companion paper titled, A meta-synthesis of children's and caregivers' perceptions of mandatory reporting of child maltreatment (in preparation), will address caregivers' and children's experiences with mandatory reporting. 10.1136/bmjopen-2016-013942.supp1 Methods {#s2} ======= Various methods for synthesising qualitative literature exist depending on the purpose of the review[@R34] or the philosophical[@R35] or epistemological[@R36] stance of the researcher. As there is no standard way to summarise qualitative literature, for this meta-synthesis we follow the methods of Feder and colleagues,[@R37] whose work builds on Noblit and Hare's (43) approach to meta-ethnography. Meta-ethnography does not offer suggestions for sampling or appraising articles and at times can be criticised for lack of transparency.[@R34] A benefit of Feder and colleagues'[@R37] method is that they conducted a systematic search of qualitative studies with clear inclusion and exclusion criteria, thus enhancing the transparency of their study selection process. While the benefit of appraising qualitative research is still debated,[@R38] Feder and colleagues' approach to appraising qualitative literature prioritises studies that are ranked as of higher quality, which supports increasing recommendations to consider study quality, but also does not inappropriately exclude so-called lower quality studies that make 'surface mistakes' that would not otherwise invalidate their study findings.[@R34] Finally, like Noblit and Hare's (43) work, Feder *et al*'s[@R37] approach to synthesising qualitative literature allows for the inductive creation of a set of higher order constructs (third-order constructs, discussed below) that reflect concepts identified in individual studies but also extends beyond them. While the quantification of qualitative work has been criticised, in this study, individual concepts are 'counted' to let the reader decide about the relative importance of the themes. We suggest that themes that appear at a lower frequency are not necessarily less important (eg, one account of harm to a child is significant and must be considered) but rather that this theme was less of a focus for MRs and study authors. For example, the theme of 'cultural competence' is not discussed by as many MRs as compared with all of the various factors that impact their decision to report, which is partially explained by the fact that 11 (25%) of included articles set out specifically to investigate factors that impact MRs' decision to report. The results of this meta-synthesis are reported according to the PRISMA checklist and Enhancing Transparency in Reporting the Synthesis of Qualitative (ENTREQ) research statement[@R35] (see online [supplementary file 2](#SP2){ref-type="supplementary-material"}). 10.1136/bmjopen-2016-013942.supp2 Search strategy {#s2a} --------------- The systematic search was conducted by an information professional (JRM). Index terms and keywords related to mandatory reporting (eg, 'mandatory reporting', 'mandated reporters', 'duty to report', 'failure to report') and child abuse (broadly defined, including, but not limited to terms for child welfare, physical abuse, emotional abuse, neglect, sexual abuse/exploitation and children's exposure to IPV) were used in the following databases from database inception to 3 November 2015: Medline (1947-), Embase (1947-), PsycINFO (1806-), Cumulative Index to Nursing and Allied Health Literature (1981-), Criminal Justice Abstracts (1968-), Education Resources Information Center (1966-), Sociological Abstracts (1952-) and Cochrane Libraries (see online [supplementary file 3](#SP3){ref-type="supplementary-material"} for example search strategy). Forward and backward citation chaining was conducted to complement the search. All articles identified by our database searches were screened by two independent reviewers (JRM and AA) at the title and abstract level. At the level of title and abstract screening, an article suggested for inclusion by one screener was sufficient to put it forward to full-text review. Full-text articles were screened for relevance and put forward for consideration by one author (JRM); relevance for inclusion was confirmed by a second author (MK), with discrepancies being resolved by consensus. 10.1136/bmjopen-2016-013942.supp3 Study selection criteria {#s2b} ------------------------ Our inclusion criteria were as follows: (1) primary studies that used a qualitative design; (2) published articles; (3) investigations of MRs' experiences with mandatory reporting of child maltreatment, including physical abuse, sexual abuse, emotional abuse, neglect, exposure to IPV, prenatal exposure to maternal drug abuse or child sex trafficking; (4) presence of direct quotes from the participants to facilitate the formulation of the results and (5) English-language articles only. Excluded studies include (1) all non-qualitative designs, including surveys with open-response options; (2) studies that did not examine mandatory reporting in the context of child maltreatment (eg, mandatory reporting for elder abuse or IPV only) and (3) qualitative methods that did not lend themselves to direct quotes from participants (eg, forensic interviews). Data analysis {#s2c} ------------- Data analysis followed two parallel strands: (1) first and second-order constructs ([table 1](#T1){ref-type="table"}) were identified in each article and (2) each article was appraised with a modified critical appraisal tool for qualitative literature from the Critical Appraisal Skills Programme (CASP). For data extraction, each article was analysed for the perspectives of MRs (first-order constructs) and the conclusions offered by the author(s) of the article (second-order constructs). For first-order constructs, only direct quotes from participants (and any clarifying text provided by the study author) found in the Results sections of included articles were considered for analysis. For second-order constructs, only study author recommendations (often worded as 'should' or 'ought' statements and found in the Discussion of the article) were considered for analysis. ###### First, second and third-order constructs Construct order Definition ------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- First-order constructs First-order constructs represent the experiences and understandings of mandated reporters with respect to mandatory reporting processes Second-order constructs Second-order constructs represent the conclusions or interpretations of the article author(s) who reported the study findings---some of these interpretations were inferred from the author's recommendations. Third-order constructs Views and interpretations of the meta-synthesis team Two reviewers (JRM and MK) independently placed the primary data from each study and its corresponding code into an Excel file, and these files were compared for consistency (JRM). After reviewing discrepancies across Excel files, one author (JRM) developed a master list of codes, and after discussion with a second author (MK) (where both authors reviewed all codes and corresponding data together), this list of codes was further modified. Any discrepancies identified by the two authors were resolved by a third researcher (HLM). After this point, one author (JRM) went back through and recoded all data in the excel file according to the master list of codes and a second author reviewed all recoding (MK). Readers are able to view this final Excel file, which includes all extracted data, codes (including master list of codes) and overall quality rating of included articles. Final conclusions (third-order constructs ([table 1](#T1){ref-type="table"})) were all double checked (JRM) to ensure that they were supported by articles that ranked highly on the quality appraisal forms. For critical appraisal, a modified appraisal tool from CASP was used to assess the quality of each article (see online [supplementary file 4](#SP4){ref-type="supplementary-material"}). Two independent authors (JRM and MK) appraised each article to assess if it addressed each CASP question (yes/no/unsure) and came to consensus about the final score for each article. Only the total CASP scores were considered, and studies were not excluded for poor study design, as (1) according to our inclusion criteria, we only included articles with full quotes from MRs, (2) we coded all MRs' quotes as first-order constructs and (3) we felt that the exclusion of any articles could exclude a valuable quote/perspective from an MR and that this exclusion could impact the meta-synthesis findings. 10.1136/bmjopen-2016-013942.supp4 Data coding for this meta-synthesis was primarily inductive. Data analysis focused on identifying (1) first-order and second-order constructs that appeared across studies (repeating themes); (2) first-order or second-order constructs that were conflicting across studies or within studies and (3) unfounded second-order constructs or researchers' conclusions or interpretations that were not supported by quotes from participants. First and second-order constructs that appeared across studies were re-examined to develop the third-order constructs or the conclusions of this meta-synthesis. Specifically, one author (JRM) identified third-order constructs that addressed strategies to improve MRs' experiences with the reporting process---especially when these themes were supported by strategies offered by MRs in first-order constructs---and these themes were, per Feder *et al*,[@R37] reworded as recommendations. For example, the recommendation that MRs should 'Be aware of jurisdiction-specific legislation on reportable child maltreatment' combines a second-order construct that suggests MRs need better training about jurisdiction-specific mandatory reporting legislation with the first-order construct in which MRs admitted that they lacked knowledge about mandatory reporting legislation. These third-order constructs were first discussed with the two authors (MK, HLM) involved in developing the first and second-order constructs to ensure they reflected their understanding of the data. Following this, a table that showed a 'tally' of which first and second-order constructs were combined to generate each third-order construct (and a brief rationale for combining them) was reviewed by all study authors and a discussion followed. Minor adjustments to the third-order constructs were made after this discussion. The biggest discrepancy across all authors of this meta-synthesis was whether or not we should offer recommendations specific to mandatory reporting at all, given that (1) we did not find any effectiveness data and (2) the qualitative studies suggest many negative experiences with reporting. However, the third-order constructs represent what is found in the included studies that we synthesised (ie, included studies did not recommend against mandatory reporting), and their presentation as recommendations is faithful to the approach used by Feder *et al*, which we set out to follow, and the experiences of MRs, as summarised in the included articles. Results {#s3} ======= A total of 6500 records were identified and, after deduplication, 3809 titles and abstracts were screened using the screening criteria. After full-text screening of 218 articles, 44 articles (representing 42 studies) were included in this review (see [figure 1](#F1){ref-type="fig"}). Details about participant and study characteristics are available in online [supplementary file 5](#SP5){ref-type="supplementary-material"}. 10.1136/bmjopen-2016-013942.supp5 {#F1} Study characteristics and methodological quality {#s3a} ------------------------------------------------ The methodological quality of the studies varied and the total score percentages for each article (total possible score was 20 'yeses') are reported in [table 2](#T2){ref-type="table"}. These studies represent the views of 1088 MRs, including 231 physicians, 224 nurses, 168 CPS professionals, 156 teachers, 114 psychologists and therapists, 85 social workers, 19 dentists, 16 domestic violence workers, 16 police officers. This underestimates the number of participants included because it was challenging to determine exact number of participants in some of the studies (including one study with 10 focus groups). MRs' ages were reported in 25% of studies and ranged from 20 to 60 years of age; their years of experience were reported in just over 50% of the studies and ranged from 6 months to 41 years of experience. Only six articles[@R39] discussed any training that MRs received about recognising and responding to child maltreatment; aside from one study[@R42] that was examining the impact of child maltreatment training, it is hard to determine if or how training (or lack of training) influenced MRs' responses. Over 80% of the articles had been published since the year 2000, with seven articles published between 1981 and 1999. The studies took place in nine high-income countries (USA (15), Australia (6), Sweden (5), Taiwan (5), Canada (2), Israel (2), Norway (1), Finland (1) and Cyprus (1)) and three middle-income countries (South Africa (3), Brazil (2) and El Salvador (1)). Other studies from LMICs were identified[@R45] that did not meet all of the inclusion criteria; this limitation of our study is discussed further below. ###### Methodological quality of studies \% of total score 49% and under 50%--74% 75% or above ------------------- --------------- ---------- -------------- Study reference [@R41] [@R28] [@R40] MRs' decisions to report and experiences with reporting (first-order constructs) {#s3b} -------------------------------------------------------------------------------- Seven first-order constructs (views of MRs) are detailed below; all except construct seven (experiences receiving a report) are supported by articles from the top quartile (see [table 2](#T2){ref-type="table"} above). As is shown in [table 3](#T3){ref-type="table"}, most of the articles (91%) addressed factors that influenced MRs' decision to report (construct 1). These findings suggest that MRs struggle to identify less overt forms of maltreatment, including 'mild' physical abuse, emotional abuse, children's exposure to IPV and abuse experienced by children with disabilities. MRs also were reluctant to report their suspicions of abuse and preferred to report only when they found physical evidence of abuse, such as physical injuries, bruises, broken bones, caries (and corresponding lack of treatment) or 'total' changes in behaviour. Unfortunately, most MRs did not clarify their reporting decisions in relation to specific forms of maltreatment. For example, only five articles[@R28] discussed decisions to report (including hesitance to report) in relation to sexual abuse, and four of these articles discussed maltreatment of children with disabilities (suggesting particular challenges they faced in reporting maltreatment of children with disabilities). ###### First-order constructs (views of MRs) and the number (n) and per cent (%) of articles that address each construct ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ First-order construct (n, %) Description of construct Illustrative quotes ------------------------------------------------------------------------------ ----------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ \(1\) Deciding when to report n=40, 91% Factors that influenced MRs' decision to report, including:The amount of evidence of maltreatment (eg, challenges identifying less overt forms of maltreatment);The context of the reporter (eg, institutional support; time burden);Preferred alternative responses (eg, chart and follow child progress instead of reporting);The perceived impact of the report on the child or family (eg, concern regarding stigma);Consultation (ie, MRs' decision or need to consult with a colleague or CPS before filing a report);Family context (eg, perceived parental skills). "The most obvious (signs) are easy. It's the ones that are not so obvious, the ones that you have to dig for and explore to get to...those are the hardest ones...those are the ones that just haunt you."[@R95]\ "We need more time (than 24 hours) to interact with the child, evaluate the whole thing, and make a decision."[@R99]\ "If nothing comes out of it (report to CPS is unsubstantiated)...you're scared...thinking, I just bothered this family for no reason based on my assumptions."[@R75] \(a\) Evidence n=32, 73% \(b\) Context of reporter n=28, 64% \(c\) Alternative response n=19, 43% \(d\) Perceived impact n=12, 27% \(e\) Consultation n=9, 20% \(f\) Context of family n=8, 18% \(2\) Judgements and views towards the reporting process n=34, 77% Factors related to MRs' general satisfaction with the reporting process, including:MRs' perceived level of trust or collaboration with other professionals in the reporting process (including their own colleagues or CPS);Any general burden MRs felt from the reporting process;MRs' perceptions of CPS's (in)effectiveness. "Knowing the child protection agency in our area, nothing would come of a report."[@R57]\ "It's pretty much a one way street as far as information goes. I find that really frustrating."[@R111] \(a\) Negative n=33, 75% \(b\) Positive n=11, 25% \(3\) Experiences with reporting n=33, 73% Examples of MRs' positive or negative experiences with the reporting process, including:The amount of support MRs received when reporting (eg, some MRs had little institutional support for their reporting duties);Responsiveness of the intake workers screening the report (eg, some reporters discussed rude or dismissive responses from intake workers);The scope of CPS (eg, some reporters were discouraged when their report fell outside of the scope of CPS);MRs' positive or negative feelings about filing a report;Feedback from CPS (eg, many reporters were discouraged when they received no feedback about their reported case from CPS);Perceived outcomes of the report (MRs described positive or negative outcomes of the report for themselves, the child, or the family). "You'll call and say, 'I have a such and such child who made an outcry that her uncle rubbed her breasts last night.' And they'll be like, 'Well, was it over the clothes or under the clothes?'...I know that's all part of their risk assessment and they have to get to the high-priority risk to be able to take a report, but it's really challenging to hear someone on the other line say, 'Well, you know, that's just not bad enough."[@R101]\ "She made the student describe the sexual abuse experience again after they returned from the hospital. This is so (emphasised) wrong. The student should not have to experience secondary damage by going through this again and again."[@R109] \(a\) Negative n=32, 74% \(b\) Positive n=6, 14% \(4\) MRs' values and knowledge n=19, 43% Values and knowledge that informed MRs throughout the reporting process:Conflicting values included discussions of child rights and well-being, parental rights and well-being, cultural factors and the desire to ensure family preservation;MRs' discussions about their lack of knowledge related to reporting legislation or about how to identify and respond to children in need. "Many times, we don't have adequate knowledge about child abuse and the law. It is not extensively provided to every healthcare provider or to ordinary people. Without the knowledge, it is hard for us to be sensitive about the abuse or to find evidence of child abuse."[@R39] \(5\) Strategies for responding to disclosures of maltreatment and reporting n=16, 36% Practical strategies used by MRs during the reporting process, including:\ "My sense was that this child just wanted to know that she was safe and that she could tell someone, so I used that to help, in questioning her, reassuring her that nothing would happen if she told...(When the report was made) I presented it to her as that she wouldn't get in trouble but that it was a secret that I couldn't keep, and that it was something that I could help her with...she was very aware of the decision...The child knew what was going on and she felt comfortable with my telling her I was going to make a report."[@R91] Strategies for responding to disclosures of abuse (eg, listening and consoling);Strategies for filing a report (eg, informing a child or family of the limits of confidentiality when starting a therapeutic relationship).This construct also related to MRs' struggles to engage non-judgementally with offending caregivers. \(6\) Responsibility n=15, 34% MRs' perceived responsibility in identifying and responding to child maltreatment (ie, whether MRs' felt they were responsible for engaging with children, or felt that they needed to refer the case to another colleague) "I reported my suspicions to the doctor that was looking after the child and he reported it to the consultant."[@R76] \(7\) Experiences receiving a report n=2, 5% CPS professionals' positive and negative experiences receiving a report "So part of the issue for us is because we got all of these mandated reporters and intake has to take the complaint regardless, that's the problem. It's that they're not permitted to say, well that's not enough information."[@R97] ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ CPS, child protective services; MRs, mandated reporters. Factors that influenced the decision to report were distinct from the reporters' judgements and views about mandated reporting (construct 2) and their experiences with reporting (construct 3), as expressed through specific accounts of positive or negative experiences. While six articles (14%) reported positive experiences with the reporting process, 32 articles (73%) mentioned negative experiences with the reporting process, including 13 articles (30%) that offered concerning examples regarding negative child outcomes, such as: when the child was not removed from harm and the abuse continued or intensified; when the child was removed from harm, but the foster care environment was worse than the family-of-origin environment and child death following a report or after being removed from the family of origin. First-order constructs also addressed MRs' values and knowledge related to child maltreatment and reporting (construct 4), MRs' strategies for responding to disclosures of child maltreatment or for reporting (construct 5) and whether or not MRs felt personally responsible for reporting or passed this responsibility to others, such as a supervisor (construct 6). A handful of articles included CPS professionals\' experiences with receiving a report (construct 7). Strategies for supporting MRs (second-order constructs) {#s3c} ------------------------------------------------------- All second-order constructs (views of study authors) listed in [table 4](#T4){ref-type="table"} below were supported by first-order constructs within the same study; all were also supported by articles from the top quartile of study quality score (see [table 2](#T2){ref-type="table"} above). These constructs represent study authors' suggestions for how MRs could improve their decision-making during the reporting process, including strategies for mitigating negative experiences. The majority of articles (86%) commented on the need for MRs to be trained in how to best identify, respond and report suspected child maltreatment (construct 1). Two other influential themes related to the need for increased consultation between MRs and between MRs and CPS (construct 2) and the need for increased communication among MRs, among MRs, children and families and between MRs and CPS (construct 3). Study authors also emphasised that MRs need to be better supported in their reporting process (construct 4) and that they need clear protocols related to identifying and reporting child maltreatment (construct 5). Some study authors emphasised that child rights and well-being must be prioritised throughout the reporting process (construct 6). A few study authors suggested that MRs' and CPS' responses to child maltreatment need to be culturally competent (construct 7) and emphasised that MRs must report suspicions of abuse when this is their legal obligation (construct 8). ###### Second-order constructs (views of study authors) and the number (n) and per cent (%) of articles that address each construct ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Second-order construct (n, %) Description and citations for supporting articles from the top quartile Illustrative quotes ------------------------------- ----------- ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- 1\. Training and knowledge n=38, 86% MRs must know how to identify all forms of child maltreatment, including common and less overt forms of child maltreatment (emotional maltreatment, physical neglect, emotional neglect, abuse against children with disabilities).[@R42]MRs must know how best to respond to a child and family when child maltreatment is identified or disclosed.[@R43]MRs must know common issues encountered when reporting, such as ethical conflicts, moments where MRs hesitate to report, confidentiality issues, jurisdiction-specific legislation, risks and benefits of reporting, strong feelings that arise from child maltreatment cases, consequences of failure to report.[@R76]MRs must know the purpose of mandatory reporting, that is, child safety and well-being.[@R107]MRs must know their duty to report and how this differs from their moral responsibility to respond.[@R43] "All practitioners whose patients include children should avail themselves regularly of educational opportunities to increase their knowledge of the epidemiology and evaluation of child abuse and neglect."[@R112]\ "Professionals and authorities should have increased awareness of the legislation and their duties in all forms of violence."[@R104]\ "Good guidelines are important, but missing guidelines must not be an excuse not to care."[@R107]\ "Reporting, a legal requirement, must be separated from responding, which is a moral duty."[@R99] 2\. Consultation n=23, 52% For child protection to be successful, there needs to be better collaboration between all professionals in the reporting process.[@R42]MRs should be able to discuss cases of suspected child maltreatment with others, whether that be members of their own team, a child maltreatment team at their institution or CPS personnel.[@R75] "Another important finding from the study is the urgent need to improve systematic collaboration and a trustful relationship with CPS."[@R43]\ "An important resource to develop in an effort to improve child abuse and neglect detection and reporting may be the identification and ongoing support of child abuse and neglect content experts within nonpediatric and nonacademic hospital."[@R75] 3\. Communication n=21, 47% MRs should communicate clearly with the child or family about their reporting duties and the limits of confidentiality.[@R108]MRs require feedback from CPS about reported cases.[@R75]MRs should be afforded opportunities to formally and informally talk about child maltreatment with other MRs.[@R40] "Forewarning is critical for ensuring that clients do not feel deceived into thinking that superior levels of confidentiality exist.[@R108]\ "Mandated professionals require feedback from child protection agencies."[@R76] 4\. Support n=12, 27% MRs should be supported in their reporting process by their respective institutions, both in terms of the time and costs of reporting (including support of their personal safety). Support may require additional staff experts in child maltreatment.[@R40]MRs should partake in self-care and be supported in stress and coping management.[@R76] "Employing bodies are encouraged to provide a suitable support mechanism to decrease the stress and anxiety of individuals who are emotionally traumatised by the process of mandatory reporting."[@R76] 5\. Structural concerns n=7, 16% MRs need clear protocols for identifying child maltreatment and reporting it, as well as methods for reviewing and updating protocols.[@R42] "It is recommended that a formalised national framework for reporting and feedback be established, which incorporates exemplar cases to demonstrate processes and outcomes which will positively influence future decision-making of mandated professionals."[@R76] 6\. Child rights & well-being n=6, 14% MRs should prioritise children's rights and well-being throughout the reporting process.[@R113] "If the intention is for children to have the full status of victim, the focus should not only be on reporting but also on the responses following reporting."[@R104] 7\. Cultural competence n=4, 9% MRs' and CPS' responses to child maltreatment should be culturally competent and families' preferences for alternative ways of dealing with abuse (eg, restorative justice) should not be dismissed.[@R53] "People's preference for traditional ways of dealing with problem should not be taken lightly, especially as any dismissal of it could be taken as constituting a lack of trust and understanding by the establishment of the current African ways of dealing with abuse."[@R53] 8\. Evidence n=4, 9% MRs should report suspicions of abuse rather than wait for evidence of abuse, when this is their legislative duty.[@R107] "Physicians and other healthcare workers are legally required to report cases if they have *reasonable suspicion* of child abuse."[@R92] ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- CPS, child protective services; MRs, mandated reporters. These second-order constructs show that MRs need better support at all social--ecological levels: (1) personally, in terms of better training, including skills to identify and respond to child maltreatment, as well as skills for stress and coping management; (2) interpersonally, in terms of better opportunities for dialogue among colleagues about child maltreatment generally, as well as specific cases; (3) organisationally, in terms of more support for the time it takes to report (and the potential 'costs' to other patients when taking this time), safeguards for MRs' personal safety when reporting and access to staff experts in child maltreatment; (4) in the community, especially in terms of better feedback about reported cases from CPS and in general better dialogue between different agencies involved in the reporting process and (5) nationally, in terms of national protocols about identifying, responding to and reporting child maltreatment. Apparent contradictions {#s3d} ----------------------- All of the apparent contradictions found within the studies (or constructs that conflicted within or across studies) are examples of correlates of reporting that have been discussed previously in the literature (eg, MRs' decisions to report should or should not be influenced by the context of the family, the level of evidence available, the context of the reporter or the perceived impact of reporting on the child or family; MRs should or should not report children's exposure to IPV or corporal punishment; MRs should or should not intervene with the family instead of reporting; the MR who identifies maltreatment should report it or refer it to a senior personnel). The solutions to these contradictions are more straightforward to resolve legally but less so ethically. For example, in cases where MRs suspect that harm may come to a child from the reporting process (based on their experience or their expert judgement), they are still required to report legally (when the type and severity of child maltreatment falls within their jurisdiction's legislation). Recommendations for MRs (third-order constructs) {#s3e} ------------------------------------------------ The first-order constructs draw attention to several negative experiences MRs had with the reporting process, as well as a number of factors that influenced their decision to report. The second-order constructs summarise some institutional and cross-disciplinary responses to these concerns (offered by study authors), such as the need for increased feedback from CPS about reported cases, the need for clear protocols for identifying child maltreatment and reporting it and the need for MRs to be better supported in their reporting process. Most of the second-order constructs, however, discuss how MRs' negative experiences with the reporting process can be addressed through increased training and better communication or consultation among MRs, their colleagues and CPS. The third-order constructs found in [table 5](#T5){ref-type="table"} represent study authors' interpretation, across the studies, of MRs' and study authors' strategies for mitigating negative experiences with the reporting process, which includes the level of knowledge about child maltreatment that is required by all MRs. Restriction of the analysis to studies in the top quartile of quality ratings did not change these third-order constructs. ###### Third-order constructs in terms of recommendations to MRs When What/How ----------------------------------------------------------- ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Before identification or disclosure of child maltreatment Be aware of jurisdiction-specific legislation on reportable child maltreatment. Most reporting legislation requires that you report suspicions of child maltreatment and not wait for physical evidence of maltreatment;Be aware of the level of evidence that CPS requires to substantiate a report in your jurisdiction; acquiring this knowledge will likely require discussions with your local CPS;Be aware of child maltreatment experts in your institution or jurisdiction that you can consult with about suspected cases of child maltreatment;Be aware of the roles of your colleagues and CPS in the reporting process. Try to arrange times to communicate with both groups about issues related to child maltreatment and reporting to increase opportunities for collaboration and trust;Take training related to how to identify child maltreatment, especially less overt forms of child maltreatment; how best to respond to children exposed to maltreatment; and best practices for filing a report;Be aware of the limitations of your decision-making about child maltreatment, in terms of conflicting values about parental rights, family preservation and other cultural factors. The child's rights and well-being should always be prioritised in cases of suspected child maltreatment. At the beginning of a relationship with a child or family When you start a relationship with a child or family, disclose your reporting duties and the limits of your confidentiality to whomever is in your care. Immediate response to disclosure Respond in a non-judgemental way, showing compassion, support and belief of the child's experiences;If you are unsure if the form of maltreatment is reportable, first consult with colleagues or CPS about the case, ensuring the confidentiality of your patient is maintained;If the identified form of maltreatment is reportable in your jurisdiction and it is safe to do so, take time to remind the child and parent of your role as a mandated reporter. Discuss how you will file a report and what CPS responses to your report may entail;Be sensitive to the parent's needs and well-being during the reporting process. Be professional and non-judgemental with the offending caregiver;Ensure that the child is safe during the reporting process; for example, report at the beginning of the school day or when the accused will be otherwise occupied;Remember that your moral responsibility to respond to the child or family in need is separate from your responsibility to report maltreatment. Debriefing after report In a confidential manner, take time to debrief about the reported case with a trusted colleague. Self-care is important. CPS, child protective services; MRs, mandated reporters. Discussion {#s4} ========== While our search retrieved no evidence about the effectiveness of mandatory reporting, and qualitative research cannot be mistaken for evaluation of effectiveness, findings from this review raise important questions about the effects of mandatory reporting by drawing on studies reporting the experiences of MRs across nine high-income and three middle-income countries. While some MRs have had positive experiences with reporting, the negative experiences reported in the individual studies are very concerning, especially those related to child outcomes. Some of these include accounts of children being revictimised by the reporting process, children whose abuse intensified after a report was filed, foster care environments that were perceived to be worse than family-of-origin environments and reports of child death after CPS intervention. Whether or not these negative experiences are reflective of national or international experiences must be assessed. Studies addressing MRs' attitudes towards reporting address perceptions of negative experiences but are not able to address child-specific outcomes.[@R54] For example, Flaherty and colleagues'[@R54] US national survey of paediatricians found that 56% of physicians experienced negative consequences from reporting, including 40% who lost patients after reporting and 2% who were sued for malpractice. Some of these concerns are likely to be especially salient for MRs in countries where child protection systems are not well developed or do not function properly. MRs may have real concerns that reporting cases of child maltreatment to poorly trained or poorly resourced service providers could lead to adverse outcomes for children (see, for example, the concerns raised by Devries and colleagues[@R46] about the very poor response of local services to children in Uganda). Particularly in these contexts, further research on the harms and benefits of mandatory reporting is needed. Given that negative experiences with reporting discussed in this meta-synthesis spanned decades and nine high-income and three middle-income countries, it is not surprising that some authors have suggested that the interface between MRs and CPS agencies 'requires renewed attention, in terms of both research and programming'.[@R57] We were unable to find any high-quality research studies suggesting that mandatory reporting and associated responses do more good than harm. The lack of evidence about the effectiveness of mandatory reporting has been noted by others, including the WHO.[@R58] More research addressing child-specific outcomes is needed on 1) alternative approaches to mandatory reporting as well as 2) alternative responses to investigation following mandatory reporting (such as differential response; see online [supplementary file 1](#SP1){ref-type="supplementary-material"}). Researchers citing the benefits of mandatory reporting note that mandatory reporting laws are an "essential means of asserting that a society is willing to be informed of child abuse and to take steps to respond to it"[@R11]; they also note that mandatory reporting laws have resulted in the identification of more cases of child maltreatment[@R59] and an increase in reporting from reluctant reporter groups.[@R62] It has been argued by some authors[@R64] that identification is not a sufficient justification given the problems with the mandatory reporting process; as described in this meta-synthesis, negative experiences seem to involve the reporting process itself and the associated responses (or lack of response). A key issue is the number of children identified by MRs who receive either no services or of greater concern---inappropriate, ineffective or harmful responses. MRs' discussions of ineffective responses seem to be related most closely to their reports of 'mild' physical violence, neglect, emotional abuse or children's exposure to IPV, which may lend credence to the suggestion that mandatory reporting is most appropriate for cases of severe abuse and neglect.[@R11] More research about the effectiveness of mandatory reporting across abuse types and severity, as well as associated responses and strategies for mitigating harm (including strategies for including children and family in the reporting process), is urgently needed. Implications for clinicians and policy-makers {#s4a} --------------------------------------------- Much of the research included in this meta-synthesis did not question the need for mandatory reporting (as many of the studies aimed to address MRs' decision-making process with regards to reporting); instead, it included studies that addressed MRs' negative experiences and reluctance to report with suggestions about the need for increased support, training, consultation and communication. The third-order constructs (final conclusions) of this study therefore offer recommendations for how MRs can mitigate negative experiences with the reporting process. Analysis of recommendations by study authors suggests that MRs need better support for the reporting process at many levels: personally, interpersonally, institutionally, in the community and nationally. Personal support for reporters can include training or support for secondary traumatic stress---which many healthcare professionals experience---through, for example, strategies for debriefing.[@R66] Emerging work is examining the methods by which health and social service providers can be trained to recognise and respond to child maltreatment disclosures and suspicions of child maltreatment (for example, see [@R69]). Given that the evaluation of these training programmes falls outside the scope of this review, and that mandatory reporting is but one of many components of appropriate recognition of and response to children exposed to maltreatment, further work and evaluation is needed to understand the extent to which existing training programmes are capable of improving MRs' recognition and response to children exposed to maltreatment or if further specialised training is needed. Among studies of training programmes for mandatory reporting with controlled designs, Kenny[@R72] argues that Alvarez and colleagues'[@R73] training programme shows the most promise. The components of the training programme, discussed further by Donohue *et al*,[@R69] include discussions about identifying child maltreatment, reporting requirements and procedures, strategies for involving caregivers in the reporting process and information about consultation with colleagues and CPS---all identified as important components of training in this review. Whether or not the training programmes can be successfully modified to address the training needs of different countries and multidisciplinary trainees has yet to be assessed. Interpersonal support can include increased opportunity for communication and teamwork between interdisciplinary and multidisciplinary colleagues through, for example, interdisciplinary training[@R42] or multidisciplinary conferences.[@R74] Relatedly, community support can include increased communication and collaboration between reporting professionals; the need for increased feedback from CPS about reported cases is also important.[@R54] Poor communication or collaboration between CPS and MRs has long been cited as an area for much needed improvement.[@R74] How exactly to improve collaboration, however, is complex and under-researched. As Winkworth and White[@R81] argued in relation to Australian initiatives to increase collaboration between child protection, family relationship and family support service systems, "So ubiquitous is reference to collaboration in policy documents that it is in danger of being ignored altogether by service deliverers who are not clear about its rationale, how it is built, or its real value". Finally, national support necessitates national protocols about identifying, responding to and reporting abuse, as well as increased clarity around specific reporting requirements (including increased clarity around national or jurisdictional reporting legislation). Whether or not national protocols improve the reporting process for MRs or help to improve child outcomes would need to be tested.[@R82] Strengths, limitations and future research {#s4b} ------------------------------------------ The strengths of our review include a systematic search to inform the meta-synthesis; the use of clear a priori study inclusion and exclusion criteria; use of an established study appraisal checklist and transparent and reproducible methods for analysis. This review focused on MRs' direct experiences with, or views about, the mandatory reporting process and as such does not reflect complete findings about (1) appropriate MR responses to the disclosures or identification of child maltreatment; (2) CPS workers' experiences substantiating reports; (3) children's and caregivers' experiences with mandatory reporting and (4) professionals' experiences with reporting in a non-mandated context (such as the UK). Reviews on these topics would be complementary to the findings of this review. While only English-language studies were included and only a handful of included articles discussed reporting processes in LMICs, the limited availability of research from LMICs suggests an even greater need to invest in research in these settings. Research about voluntary or policy-based reporting processes, as well as responses to mandatory reporting, may provide more information about reporting process from LMICs. Conclusion {#s5} ========== Mandatory reporting of child maltreatment has been variously implemented across jurisdictions and high-quality research on the effectiveness of this process is severely lacking. While our search retrieved no evidence about the effectiveness of mandatory reporting, through this meta-synthesis of MRs' experiences with reporting, we have summarised many accounts of harm associated with reporting. Along with focusing on approaches to improve mandatory reporting, the field needs to address whether or not mandatory reporting actually improves children's health outcomes through research that is sensitive to both severe and less overt forms of maltreatment. Our findings in no way imply that the recognition and response to children exposed to maltreatment is not a significant public health concern that requires coordinated responses. Rather, it implies that we must work to ensure that all of our methods for recognising and responding to children exposed to maltreatment demonstrate that they benefit children's safety and well-being and do no additional harm. Supplementary Material ====================== ###### Reviewer comments ###### Author\'s manuscript **Contributors:** Conceptualisation: HLM, KD, MC, JRMT, JCDM; Analysed the data: JRMT, MK, AA, HLM; Writing--original draft preparation: JRMT; Writing--review and editing: JRMT, MK, KD, MC, JCDM, CNW, HLM; ICMJE criteria for authorship read: JRMT, MK, KD, MC, JCDM, CNW, AA, HLM; Agree with manuscript results and conclusions: JRMT, MK, KD, MC, JCDM, CNW, AA, HLM. **Funding:** HLM and CNW receive funding from the Canadian Institutes of Health Research (CIHR) Institute of Gender and Health (IGH) and Institute of Neurosciences, Mental Health and Addictions (INMHA) to the PreVAiL Research Network (a CIHR Center for Research Development in Gender, Mental Health and Violence across the Lifespan---www.PreVAiLResearch.ca). HLM holds the Chedoke Health Chair in Child Psychiatry at McMaster University in Hamilton, Ontario, Canada. JRMT is supported by a postdoctoral fellowship from Violence Evidence Guidance Action (VEGA). MK is supported by Ontario Ministry of Health and Long-Term Care Women's Health Scholar Post-Doctoral Fellowship Award. The funders had no role in study design, data collection and analysis, decision to publish or preparation of the manuscript. **Competing interests:** None declared. **Provenance and peer review:** Not commissioned; externally peer reviewed. **Data sharing statement:** Additional data can be accessed via the Dryad data repository at <http://datadryad.org/> with the doi:10.5061/dryad.6d159.

1. Introduction {#sec1-materials-13-02703} =============== The development of electrocatalysts for oxygen electrochemistry has attracted considerable attention for many decades because both the oxygen evolution reaction (OER) and oxygen reduction reaction (ORR) play important roles in alternative energy sectors \[[@B1-materials-13-02703]\]. For example, during water electrolysis to produce pure hydrogen and oxygen, the OER is known to be the rate-determining step as considerable overpotential is often required. Therefore, it is of particular interest to develop effective electrocatalysts to minimize unnecessary energy loss. In fuel cells where the oxygen combines with hydrogen to produce electricity, the ORR is the most energy-consuming step and it reduces the efficiency of converting energy. In research studies, the primary electrocatalysts for OER are conductive oxides such as RuO~2~, IrO~2~, and their composites \[[@B2-materials-13-02703],[@B3-materials-13-02703]\]. It is because metallic electrocatalysts would suffer from oxidation during OER and their performances degrade steadily with time. Since both RuO~2~ and IrO~2~ are oxides of noble metals, there is a strong motivation to develop less-expensive alternatives for comparable OER activities \[[@B4-materials-13-02703],[@B5-materials-13-02703],[@B6-materials-13-02703],[@B7-materials-13-02703],[@B8-materials-13-02703]\]. For example, the cubic Co~3~O~4~ was reported to reveal impressive OER performance in alkaline electrolytes \[[@B9-materials-13-02703]\]. Another approach that has been widely explored is the fabrication of core-shell nanoparticles (core\@shell) in which the inexpensive constituent occupies the core whereas the shell is covered by the element responsible for an electrochemical action \[[@B10-materials-13-02703],[@B11-materials-13-02703]\]. This approach makes logical sense since the electrochemical reaction is an interfacial process where only the surface atoms matter. Moreover, it has been established that proper selection of a core constituent is able to enhance the electrocatalytic behavior of surface atoms via deliberate modification of the latter's lattice parameter or energy band \[[@B12-materials-13-02703]\]. For ORR electrocatalysis, the Pt and Co have been studied for their chemical stability and impressive electrocatalytic activity in both acidic and alkaline electrolytes \[[@B13-materials-13-02703],[@B14-materials-13-02703],[@B15-materials-13-02703],[@B16-materials-13-02703]\]. In the research, the Pt has been synthesized in a wide range of morphologies and nanostructures \[[@B17-materials-13-02703],[@B18-materials-13-02703],[@B19-materials-13-02703],[@B20-materials-13-02703]\]. However, due to the expensive nature of Pt, it is rather critical to explore low-cost alternatives that could still deliver similar ORR performances. A straightforward solution is the formation of alloyed nanoparticles with binary (Pt-Ni, Pt-Fe, and Pt-Co) and tertiary (Pt-Fe-Co, Pt-Cu-Ni, and Pt-Fe-Ni) constituents \[[@B21-materials-13-02703],[@B22-materials-13-02703],[@B23-materials-13-02703],[@B24-materials-13-02703],[@B25-materials-13-02703],[@B26-materials-13-02703]\]. On many occasions, these alloyed electrocatalysts not only reduce the Pt utilization but also demonstrate even stronger ORR activities \[[@B27-materials-13-02703],[@B28-materials-13-02703],[@B29-materials-13-02703],[@B30-materials-13-02703]\]. An alternative route to reduce the Pt usage is the preparation of core\@Pt nanoparticles since the heterogeneous core is able to alter both the electronic structure and the lattice parameter of surface Pt atoms, and, consequently, affects the ORR activities in distinct manners \[[@B31-materials-13-02703],[@B32-materials-13-02703]\]. So far, the preparation of core\@Pt nanoparticles is often achieved via a galvanic displacement reaction in which corrosive-prone atoms are pre-coated on selective cores, which was followed by their oxidative dissolution for reducing Pt ions from the electrolyte. For example, in our group, we adopted a Cu under-potential-deposition (UPD) to synthesize Pd~9~Ru\@Cu nanoparticles and engaged the displacement reaction to fabricate Pd~9~Ru\@Pt nanoparticles for ORR and methanol electro-oxidation activities in acidic electrolytes \[[@B33-materials-13-02703],[@B34-materials-13-02703]\]. In the literature, the UPD process is a powerful tool to fabricate a core\@shell microstructure by first depositing a Cu monolayer, which is followed by a displacement reaction in which the Cu is corrosively dissolved in conjunction with the reduction of selective ions from the electrolyte \[[@B35-materials-13-02703],[@B36-materials-13-02703]\]. Previously, a wide variety of oxides such as rutiles, perovskites, and spinels have been investigated as the ORR electrocatalysts \[[@B37-materials-13-02703],[@B38-materials-13-02703],[@B39-materials-13-02703],[@B40-materials-13-02703]\]. It is understood that the presence of cations with mixed valences in a spinel lattice could provide donor-acceptor chemisorption sites for reversible adsorption of oxygen. In addition, these mixed-valence oxides exhibit reasonable electric conductivity because of the relatively low activation energy required for electron hopping between neighboring cations. In the literature, the first documented report of oxide\@Pt nanoparticles for ORR was demonstrated by Dhavale and Kurungot using Fe~2~O~3~ as the core \[[@B41-materials-13-02703]\]. Recently, the Co~3~O~4~ was reported to reveal impressive ORR activities with a similar four electron path to that of Pt \[[@B42-materials-13-02703]\]. Similar behaviors were reported in the spinel family of cobaltites (NiCo~2~O~4~ and MnCo~2~O~4~) \[[@B43-materials-13-02703],[@B44-materials-13-02703]\]. Since the synthesis of Co~3~O~4~ nanoparticles is straightforward and its raw material cost is relatively inexpensive, it is possible to engineer the Co~3~O~4~ nanoparticles into gradient Co~3~O~4~\@CoO\@Co nanoparticles for direct OER electrocatalysis. Similar compositions have recently been demonstrated with impressive electrocatalytic activities \[[@B45-materials-13-02703],[@B46-materials-13-02703]\]. In addition, instead of conventional Cu UPD, the surface-enriched Co atoms could be leveraged to initiate the displacement reaction for forming Co~3~O~4~\@Pt nanoparticles for ORR actions. In this work, we adopted the core\@shell design to fabricate nanoparticles with radial composition of Co~3~O~4~/CoO/Co (inside/outside) and evaluated their OER activities in an alkaline electrolyte. These Co~3~O~4~\@CoO\@Co nanoparticles were produced by a controlled reduction of Co~3~O~4~ nanoparticles derived from hydrothermal synthesis. In addition, we employed the displacement reaction to synthesize Co~3~O~4~\@Pt nanoparticles and evaluated their ORR activities. Material characterization and electrochemical analysis were conducted, and the results were thoroughly discussed. 2. Materials and Methods {#sec2-materials-13-02703} ======================== 2.1. Synthesis of Co~3~O~4~/CoO/Co and Co~3~O~4~\@Pt Nanoparticles {#sec2dot1-materials-13-02703} ------------------------------------------------------------------ The synthesis of Co~3~O~4~/CoO/Co nanoparticles started with the preparation of Co~3~O~4~ nanoparticles. First, 1.2 mL of 0.2 M Co(Ac)~2~ (98%, Alfa Aesar, Ward Hill, MA, USA), 0.5 mL of NH~4~OH (28%, J.T. Baker, Phillipsburg, PA, USA), 0.7 mL of deionized water, and 12 mg of graphene (Enerage Inc., Yilan, Taiwan) were mixed in 24 mL of ethanol at 25 °C. Subsequently, the mixture was kept at 80 °C under constant stirring for 10 h. Then, the mixture was transferred to a 40 mL autoclave for a hydrothermal process for 3 h at 150 °C. Afterward, the synthesized Co~3~O~4~ nanoparticles impregnated on the graphene (Co~3~O~4~/GN) were retrieved by vacuum filtration, which was followed by repeated rinsing in ethanol. Next, the Co~3~O~4~/GN underwent a mild reduction treatment (5%H~2~-95%Ar) at 350 °C for 15 min to reduce the surface Co~3~O~4~ to CoO and Co, which formed core\@shell (Co~3~O~4~\@CoO\@Co) nanoparticles impregnated on the graphene (Co~3~O~4~\@CoO\@Co/GN). To engage the galvanic displacement reaction between the Co~3~O~4~\@CoO\@Co/GN and Pt ions in the electrolyte, 5 mg of Co~3~O~4~\@CoO\@Co/GN was suspended in 5 mL of ethanol at 10 °C for 15 min, which was followed by the slow addition of 5 mL of 5 mM K~2~PtCl~4~ (98%, UniRegion Bio-Tech, Taipei, Taiwan) aqueous solution. The galvanic displacement reaction lasted for 30 min, which allowed the formation of Pt nanoparticles on the surface of Co~3~O~4~ nanoparticles. This formed Co~3~O~4~\@Pt impregnated on the graphene (Co~3~O~4~\@Pt/GN). Afterward, the resulting powders were collected by centrifugation and washed with excessive deionized water and ethanol. 2.2. Materials Characterization {#sec2dot2-materials-13-02703} ------------------------------- X-ray diffraction (XRD, Bruker, Billerica, MA, USA) patterns were acquired by a Bruker D2 Phaser system equipped with a Cu Kα radiation source (λ = 1.54 Å) and the samples were prepared by dropping the ink slurry on a Si wafer. The resulting diffraction peaks were used to identify a relevant phase and particle size for Co~3~O~4~, Co~3~O~4~\@CoO\@Co, and Co~3~O~4~\@Pt. A transmission electron microscope (TEM, FEI Tecnai F20, Philips/FEI, Amsterdam, Netherlands) was employed to observe the size and morphology of the electrocatalysts and the TEM samples were prepared by dropping the ink slurry onto Cu grids. X-ray absorption spectroscopy (XAS, NSRRC, Hsinchu, Taiwan) was adopted to determine the oxidation state of the Co before and after galvanic displacement reaction. The XAS experiments were performed at the BL17C beam line of the National Synchrotron Radiation Research Center (NSRRC) in Hsinchu, Taiwan, China. The standard operation conditions were 1.5 GeV and 300−360 mA. The XAS profiles were obtained in a fluorescence mode at 25 °C in air with Ni and Cr foil serving as fluorescence filter screening photons entering the fluorescence detector. The resulting X-ray absorption near-edge spectra (XANES) were processed by a program (Athena, version 0.8.056) in which the Co K-edge absorption steps were normalized to unify with the removal of background signals. Commercially available Co, CoO, and Co~3~O~4~ were adopted as the reference materials. Detailed XAS processing protocols were reported earlier \[[@B47-materials-13-02703]\]. 2.3. Electrochemical Analysis {#sec2dot3-materials-13-02703} ----------------------------- To determine OER activities, 5 mg of Co~3~O~4~/GN or Co~3~O~4~\@CoO\@Co/GN, 0.016 mL of Nafion ionomers solution (5 wt %), and 1 mL of water/isopropanol (1:1 vol/vol) were mixed and sonicated for 1 h. Next, 0.005 mL of ink (25 µg electrocatalyst) was deposited on a rotating disk electrode (RDE, glassy carbon electrode with 5 mm in diameter) serving as the working electrode. A saturated calomel electrode (SCE) and a Pt foil (3 × 3 cm^2^) were used as the reference and counter electrode, respectively. The electrolyte was aqueous 0.1 M potassium hydroxide (KOH) aqueous solution and was saturated with oxygen for 30 min. The transient OER activities were determined by conducting linear *i*-V polarization between 0 and 1 V (vs. SCE) at 5 mV s^−1^. The potential for the reversible hydrogen electrode (RHE) was 0.99 V (vs. SCE). In our *i*-V profiles, the potentials were plotted against the RHE. Evaluation on the durability of Co~3~O~4~\@CoO\@Co/GN was conducted by recording the *i*-V polarization profile after 1000 cyclic voltammetric (CV) scans from 0.2 to 0.3 V (vs. RHE) at 50 mV s^−1^ in aqueous 0.1 M KOH solution. For ORR activities, 5 mg of electrocatalyst (Co~3~O~4~/GN or Co~3~O~4~\@Pt/GN), 0.016 mL of Nafion ionomers solution (5 wt %), and 1 mL of water/isopropanol (1:1 vol/vol) were mixed and sonicated for 1 h. On an RDE, 0.004 mL of ink (20 µg electrocatalyst) was deposited. The SCE was used as the reference electrode. The ORR experiments were carried out in a three-electrode cell using a Pt foil (3 × 3 cm^2^) as the counter electrode and 0.1 M KOH aqueous solution as the electrolyte. Both working and counter electrodes were in the same compartment. Before the experiments, the electrolyte was saturated with oxygen for 30 min. During the measurements, the oxygen was still flowing. The ORR activities for both Co~3~O~4~/GN and Co~3~O~4~\@Pt/GN were obtained by conducting linear *i*-V polarization between 0.11 and −0.99 V (vs SCE) at a scan rate of 5 mV s^−1^. The electrochemical active surface area (ECSA) of Co~3~O~4~\@Pt/GN was determined by imposing cyclic voltammetric (CV) scans between 0 and 1.5 V (vs. RHE) at 50 mV s^−1^ in a de-aerated 0.1 M KOH aqueous solution. The ECSA was determined by dividing the coulomb charge associated with the hydrogen UPD region between 0.05 and 0.4 V (vs. RHE) by 210 µC cm^−2^~Pt~. All the electrochemical experiments were conducted by a potentiostat (VersaSTAT4) at 25 °C. For both ORR activities and ECSA, commercially available Pt/C (20 wt % Pt on XC72R, BASF SE, Ludwigshafen, Germany) was also analyzed for a comparison purpose. 3. Results {#sec3-materials-13-02703} ========== 3.1. Characterizations of Co~3~O~4~\@CoO\@Co for OER Activities {#sec3dot1-materials-13-02703} --------------------------------------------------------------- [Figure 1](#materials-13-02703-f001){ref-type="fig"} displays the XRD patterns for Co~3~O~4~ before and after the reduction treatment (5, 10, and 15 min) as well as standard Co, CoO, and Co~3~O~4~ for a comparison purpose. Apparently, the as-synthesized Co~3~O~4~ exhibited diffraction peaks and intensities that were consistent with those of standard cubic Co~3~O~4~ (Joint Committee on Powder Diffraction Standard, JCPDS: 073--1701), suggesting its isotropic nature. For samples after hydrogen treatment, the predominant peaks were still attributed to those of Co~3~O~4~, but additional diffraction signals were observed for samples after hydrogen treatments of 10 and 15 min, and they were associated with CoO and Co, respectively. For example, for the sample of a 15-min reduction treatment, the strongest peak of CoO was recorded at 2θ of 42.5° for the (200) plane whereas the strongest peak of Co was also observed at 2θ of 44.3° for the (111) plane. It is noted that. for both CoO and Co, only their respective strongest peaks were observed, which suggested their absolute amounts were relatively insignificant as compared to that of Co~3~O~4~. This suited our purpose as the Co~3~O~4~ core was nicely maintained without experiencing excessive reduction. We were able to estimate the grain size of Co~3~O~4~, CoO, and Co using Scherrer's equation on their respective strongest diffraction peaks, and determined their size was 9.3 nm (Co~3~O~4~), 2.8 nm (CoO), and 4.1 nm (Co), respectively. The presence of surface Co atoms was further confirmed by a simple magnetic test in which the dried Co~3~O~4~\@CoO\@Co powders were easily attracted toward a magnet. Our results indicated that the reduction treatment of 15 min was successful in reducing the surface of Co~3~O~4~ nanoparticles to CoO and Co. Considering the reduction reaction was driven by the hydrogen molecules, which diffused inward, it is reasonable to expect the radial composition of our sample to be Co~3~O~4~/CoO/Co (inside-outside). Because the samples after 5-min and 10-min reduction treatments did not exhibit convincing evidence for the formation of CoO and Co. The sample of a 15-min reduction treatment was selected for subsequent characterization and electrochemical analysis. However, it is noted that further studies are warranted to conduct extended heat treatments. Therefore, the amount of CoO and Co could be varied and their effects for OER could be elucidated. [Figure 2](#materials-13-02703-f002){ref-type="fig"} displays the TEM image of Co~3~O~4~\@CoO\@Co/GN. Apparently, the Co~3~O~4~\@CoO\@Co nanoparticles exhibited a nearly spherical shape with a size of 24.3 ± 4.8 nm. In typical TEM imaging, areas with a darker appearance represent structures with higher mass density. In our sample, the metallic Co on the shell was expected to have a greater mass density, and, thus, revealed a darker image as compared to that of Co~3~O~4~ at the core, which was smaller in mass density, and, thus, appeared as a lighter image. In between, there was the formation of CoO whose image became indistinguishable with that of Co~3~O~4~ because of similar mass density. A closer look at this TEM image indicated that the average diameter for the Co~3~O~4~ core was 10.4 ± 3.8 nm and the Co shell thickness was 7.7 ± 2.5 nm. We did not perform compositional analysis on X-ray photoelectron spectroscopy because the Co~3~O~4~\@CoO\@Co revealed a strong magnetic behavior that precluded its possible diagnosis. The XAS is a powerful tool in providing qualitative information about the average oxidation state and local atomic structure of the absorbing atoms. Therefore, the XAS was utilized to investigate the absorbance of Co in Co~3~O~4~/GN, and Co~3~O~4~\@CoO\@Co/GN, and the resulting Co XANES spectra are displayed in [Figure 3](#materials-13-02703-f003){ref-type="fig"}, along with standard materials of Co, CoO, and Co~3~O~4~ for comparison purposes. In general, for K-edge XANES, the absolute magnitude of the absorption energy is proportional to the oxidation state of the absorbing atom. For example, Co with a higher oxidation state reveals a greater absorption energy. Therefore, these XANES profiles were used to compare with those of standard materials to determine the oxidation state of Co qualitatively. The oxidation state of Co in Co, CoO, and Co~3~O~4~ is 0, +2, and +2.67, respectively. The K-edge absorption energy of standard materials appeared in order of Co~3~O~4~ \> CoO \> Co. For the Co~3~O~4~/GN sample, its XANES profile was identical to that of standard Co~3~O~4~. After the reduction treatment, the XANES profile for Co~3~O~4~\@CoO\@Co/GN was shifted slightly to lower energy, becoming similar to that of the CoO standard. This result was anticipated as the XAS recorded the average oxidation state of Co. [Figure 4](#materials-13-02703-f004){ref-type="fig"}a displays the OER *i*-V profiles for RDE and Co~3~O~4~/GN as well as Co~3~O~4~\@CoO\@Co/GN before and after 1000 CV cycles. As shown, the RDE exhibited negligible OER activity because the glassy carbon is not electrocatalytically active. At 10 mA cm^−2^, the Co~3~O~4~\@CoO\@Co/GN revealed a smaller overpotential as compared to that of Co~3~O~4~. This result was consistent with what was reported by Hang et al. in which metallic Co was used to improve the electrical conductivity of Co~3~O~4~ for improved OER activities \[[@B48-materials-13-02703]\]. In addition, the Co~3~O~4~\@CoO\@Co/GN showed impressive durability with similar *i*-V profiles after 1000 cycles. It is noted that our testing methods are consistent with what was used in the literature \[[@B49-materials-13-02703]\]. Therefore, our results were reliable and reproducible. To further explore their electrocatalytic behavior, these *i*-V profiles are replotted in the Tafel curve, as shown in [Figure 4](#materials-13-02703-f004){ref-type="fig"}b. The Co~3~O~4~\@CoO\@Co/GN demonstrated a Tafel slope of 92 mV dec^−1^, which was relatively lower than that of Co~3~O~4~/GN at 147 mV dec^−1^. This Tafel slope was similar to what was reported earlier \[[@B48-materials-13-02703]\]. The determination of exact electrocatalytic contributions from CoO and Co in OER is not feasible because the Co atoms experience complicated oxidation and reduction steps during OER actions. In short, we demonstrated a simple approach to fabricate gradient nanoparticles under deliberate hydrogen reduction, and the gradient nanoparticles revealed effective improvement in OER relative to the untreated oxide nanoparticles. 3.2. Characterization of Co~3~O~4~\@Pt/GN for ORR Activities {#sec3dot2-materials-13-02703} ------------------------------------------------------------ The synthesis of Co~3~O~4~\@Pt nanoparticles was achieved by a deliberate displacement reaction between Co and Pt ions, which proceeded via Co + \[PtCl~4~\]^2−^ → Pt + Co^2+^ + 4Cl^−^. The difference between the E^0^ value of Pt/\[PtCl~4~\]^2−^ (0.758 V vs. Normal hydrogen electrode, NHE) and Co/Co^2+^ (−0.28 V vs. NHE) is sufficient for the displacement reaction to activate. The aqueous K~2~PtCl~4~ solution was in light yellow. After we finished the displacement reaction, the solution remained light yellow because excessive Pt ions were used in the electrolyte for the displacement reaction. The presence of Pt on the Co~3~O~4~\@Pt/GN was verified by XRD. The resulting diffraction pattern along with those of standard fcc Pt (JCPDS: 1--87--636) and cubic Co~3~O~4~ (JCPDS: 073--1701) are displayed in [Figure 5](#materials-13-02703-f005){ref-type="fig"}. This diffraction pattern was plotted between 2θ of 30--65° to minimize unnecessary interference from graphene at 2θ of 27°. As shown, the predominant peaks at 2θ of 40.5° and 47° were associated with Pt (111) and Pt (200), respectively, and their relative intensity agreed well with those of standard Pt. Unexpectedly, the strongest peak for (311) Co~3~O~4~, located at 2θ of 37.2°, was rather weak, and the CoO signal was clearly absent. We realized that CoO was dissolved during the displacement reaction because it was chemically unstable in an acidic solution. Despite the diffraction ability of metallic Pt being much larger than that of Co~3~O~4~, judging from the large intensity contrast between the (311) Co~3~O~4~ and (111) Pt, we surmised that the amount of Pt on the Co~3~O~4~\@Pt/GN was rather substantial. In addition, the estimation of the Pt crystallite size from Sherrer's equation using the Pt (111) plane was 6.2 nm. [Figure 6](#materials-13-02703-f006){ref-type="fig"} exhibits the TEM image for Co~3~O~4~\@Pt/GN. As expected, the Co~3~O~4~ at the core revealed a lighter image whereas the Pt occupied the outer shell in a darker image. Notably, the Pt shell appeared as aggregates of Pt nanoparticles and the primary Pt nanoparticle size was 8.3 ± 1.2 nm. This value was in reasonable agreement with that of XRD analysis in [Figure 5](#materials-13-02703-f005){ref-type="fig"}. In addition, the aggregates of Pt nanoparticles did not fully encapsulate the Co~3~O~4~ core and, therefore, partial exposure of Co~3~O~4~ to the electrolyte was possible. Successful formation of Co~3~O~4~\@Pt nanoparticles could be validated by their XAS profile. As shown in [Figure 3](#materials-13-02703-f003){ref-type="fig"}, after the completion of the displacement reaction, the XANES profile for Co~3~O~4~\@Pt/GN demonstrated a higher oxidation state than that of Co~3~O~4~\@CoO\@Co/GN. This phenomenon was anticipated because the metallic cobalt and CoO in Co~3~O~4~\@CoO\@Co was replaced by Pt, which led to the increase of the average oxidation state of the remaining Co atoms. The Pt had its characteristic K-edge absorption energy at 78,393 eV and L~III~-edge absorption energy at 11,564 eV. These peaks were not shown in this Co K-edge XANES profile. The ORR activities in apparent current density from Co~3~O~4~/GN and Co~3~O~4~\@Pt/GN are shown in [Figure 7](#materials-13-02703-f007){ref-type="fig"} along with commercially available Pt/C for a comparison purpose. The apparent current density is based on the actual geometric area of RDE. According to the research, at a potential below 0.6 V, the ORR curve is under mass transport control limited by the diffusion of dissolved oxygen in the electrolyte, whereas, at a potential between 0.8 and 1 V, the ORR response is controlled by the kinetics of electrocatalysts \[[@B50-materials-13-02703]\]. Therefore, the most common method to compare the ORR activities of electrocatalysts is based on the half-way potential, which is defined by the potential at the half magnitude of limiting current. Generally, a larger half-way potential indicates a greater ORR activity. For Pt/C, at a rotation speed of 1600 rpm, the diffusion-limiting current was approaching 5.6 mA cm^−2^, which is a value that was close to what was reported for Pt previously \[[@B51-materials-13-02703],[@B52-materials-13-02703]\]. For Co~3~O~4~/GN, the current response appeared in two distinct stages, which indicated that two separate steps were involved in the ORR because of a weak electrocatalytic activity \[[@B42-materials-13-02703]\]. However, the Co~3~O~4~\@Pt/GN revealed a similar current response like that of Pt, albeit with a smaller limiting current. Nevertheless, the Co~3~O~4~\@Pt/GN behaved much better than that of Co~3~O~4~/GN, which showed the advantage of depositing Pt as the shell layer. From [Figure 7](#materials-13-02703-f007){ref-type="fig"}, the half-way potential for Co~3~O~4~/GN, Co~3~O~4~\@Pt/GN, and Pt/C was 0.53, 0.79, and 0.83 V, respectively. The minor difference between Co~3~O~4~\@Pt/GN and Pt/C was due to the incomplete coverage of Pt on the Co~3~O~4~. It is noted that, in our study, the GN was employed as the support for Co~3~O~4~\@CoO\@Co nanoparticles to reside on. In electrocatalysis, electrocatalysts in nanoparticulate forms are often impregnated on carbonaceous materials for better distribution and utilization. In the literature, the GN has been used as the substrate to support electrocatalysts for many electrochemical reactions \[[@B53-materials-13-02703],[@B54-materials-13-02703]\]. To further investigate the kinetics of Co~3~O~4~\@Pt/GN, we employed the Koutecky-Levich equation, which is listed below, to calculate the electron transfer number. where the *i* is the experimentally-measured apparent current, the *i~diffusion\ limit~* is the diffusion limiting current due to the limitation of mass transport of dissolved oxygen in the 0.1 M aqueous KOH solution, the *i~kinetic~* is the kinetic current associated with the ORR activity, *n* is the number of electrons transferred in the ORR process, *F* is the Faraday constant, *A* is the reaction area of the RDE (0.196 cm^2^), *D~O2~* is the diffusivity of dissolved oxygen in the 0.1 M aqueous KOH solution (1.9 × 10^−5^ cm^2^ s^−1^), *ω* is the rotation speed of the RDE, *v* is the kinematic viscosity of the 0.1 M aqueous KOH solution (1.09 × 10^−2^ cm^2^ s^−1^), and *C~O2~* is the concentration of dissolved oxygen in the 0.1 M aqueous KOH solution (1.2 × 10^−6^ mol cm^−3^). [Figure 8](#materials-13-02703-f008){ref-type="fig"}a exhibits the ORR curves of Co~3~O~4~\@Pt/GN at various rotation speeds of RDE. Among these curves, the ORR curves at a voltage below 0.4 V were almost stabilized at the limiting currents whose magnitudes were proportional to the rotation speed. [Figure 8](#materials-13-02703-f008){ref-type="fig"}b demonstrates the Koutecky-Levich plots at different potentials. These curves displayed a consistent pattern with an average slope of 12.2 (1/(mA × s^1/2^)). Accordingly, the electron transfer number, *n*, was close 3.8, which is a value rather close to 4 of typical Pt \[[@B52-materials-13-02703]\]. Our methodology was consistent with what was reported in the literature \[[@B42-materials-13-02703],[@B55-materials-13-02703]\]. According to a prior study, the Co~3~O~4~ catalyzed the ORR in a 2.9 to 3.4-electron route \[[@B56-materials-13-02703]\]. Therefore, our results agreed well with the microstructural information from XRD and TEM that the surface of Co~3~O~4~\@Pt/GN was predominantly Pt with a minor presence of Co~3~O~4~. [Figure 9](#materials-13-02703-f009){ref-type="fig"} demonstrates the CV curves for ECSA measurements in which the highlighted area was used to estimate the ECSA of Co~3~O~4~\@Pt/GN and Pt/C, respectively. Between them, the Pt/C exhibited a significantly larger CV profile in both capacitive response of the carbon substrate and the redox behavior of Pt. The resulting ECSA for Co~3~O~4~\@Pt/GN and Pt/C was 1.11 and 8.96 cm^2^, respectively. This smaller ECSA for our Co~3~O~4~\@Pt/GN was expected as the Pt nanoparticles were aggregated on the Co~3~O~4~ surface whereas, in commercial Pt/C, the Pt nanoparticles were homogeneously distributed on a carbonaceous substrate. Since the electrochemical reaction is an interfacial phenomenon, the magnitude of ECSA becomes an important factor as a greater ECSA typically leads to a larger electrocatalytic current. Therefore, to compare different electrocatalysts fairly, it is necessary to obtain the specific activity, which includes a parameter defined as *I~k~/ECSA*, where the *I~k~* is the kinetic current. In our case, the specific activity of Co~3~O~4~\@Pt/GN and Pt/C was 0.549 and 0.248 mA cm^−2^~Pt~. In the literature, the rationale behind the enhanced electrocatalytic activity of core\@Pt nanoparticles involves both electronic and structural effects. The electronic effect is concerned with the difference in the electronegativity between the core and Pt that shifts the d-band center and, thus, alters the electrocatalytic activity. The structural effect entails the changeup in the lattice constant of Pt because the core constituent adopts a different crystal structure. In the case of alloyed Pt nanoparticles, there is a mechanism known as a ligand effect in which the Pt and its alloying constituent cooperate in completing the electrocatalytic action. In our case, from the TEM and XRD analysis, we surmised that both electronic and structural effects were negligible. In alkaline electrolytes, the magnitude of ORR activities for Pt, Co~3~O~4~, and GN is in the order of Pt \> Co~3~O~4~ \> GN. In our case, we prepared the core\@shell structure in Co~3~O~4~\@Pt so the Pt nanoparticles were anticipated to be used more effectively. The Co~3~O~4~ was adopted as a chemically stable core that functioned as a catalyst support but was more electrocatalytically active than that of GN. In practical fuel cell devices, during long-term operation, the detachment/poisoning of Pt nanoparticles become inevitable. Therefore, the Co~3~O~4~ is expected to provide moderate ORR activities at a relatively low cost to relieve the undesirable degradation in fuel cell performances. In short, we explored a simple mild reduction treatment instead of conventional Cu UPD to produce Co~3~O~4~\@Pt nanoparticles for ORR action. 4. Conclusions {#sec4-materials-13-02703} ============== We successfully fabricated Co~3~O~4~\@CoO\@Co nanoparticles and evaluated their OER activities in an alkaline electrolyte. The Co~3~O~4~ nanoparticles were synthesized by a hydrothermal process, which was followed by a mild hydrogen reduction to produce Co~3~O~4~\@CoO\@Co nanoparticles. The XRD patterns confirmed the presence of crystalline Co~3~O~4~, CoO, and Co, and the TEM image confirmed a core-shell structure. In OER *i*-V profiles, the Co~3~O~4~\@CoO\@Co/GN revealed an impressive transient behavior and durability with a Tafel slope near 92 mV dec^−1^. The Co~3~O~4~\@CoO\@Co nanoparticles were successfully converted to Co~3~O~4~\@Pt nanoparticles via a displacement reaction. The TEM image showed that the Pt existed as nanoparticles aggregated on the surface of Co~3~O~4~ with a coverage ratio of 90%. The ORR activity followed the sequence of Pt/C \> Co~3~O~4~\@Pt/GN \> Co~3~O~4~/GN. Measurements from the RDE at various rotation speeds indicated a 4-electron transfer path for Co~3~O~4~\@Pt/GN. The specific activity of Co~3~O~4~\@Pt/GN was more than two times greater than that of Pt/C. Conceptualization, S.-C.C. and P.-W.W. Methodology, S.-C.C., K.-C.T., Y.-C.H. B.-Y.S., and J.-F.L. Writing---original draft preparation, S.-C.C. Writing---review and editing, P.-W.W. All authors have read and agreed to the published version of the manuscript. The Ministry of Science and Technology of Taiwan (107-2221-E-009-006-MY3, 107-2633-B-009-003, 107-2218-E-009-011, 108-2321-B-010-008-MY2) and the Ministry of Education provided funding for the "Center for Neuromodulation Medical Electronics Systems" from the featured areas research center program within the framework of the higher education sprout project. The authors declare no conflict of interest. {#materials-13-02703-f001} {#materials-13-02703-f002} {#materials-13-02703-f003} {#materials-13-02703-f004} {#materials-13-02703-f005} {#materials-13-02703-f006} {#materials-13-02703-f007} {#materials-13-02703-f008} {#materials-13-02703-f009}

Introduction {#sec1-1} ============ Bone marrow stem cells stimulate endogenous neural cell regeneration and inhibit apoptosis (Isele et al., 2007), repair and replace nerve cells after cerebral ischemia, promote or participate in angiogenesis in the ischemic region (Qiao and Cao, 2014). In conjunction with a mobilizing agent, they improve the survival rate of rats with cerebral ischemia by increasing the number of CD34-positive cells and the expression of Bcl-2 protein at the edges of the infarct, reducing the area of cerebral infarction (Yanqing et al., 2006). Studies have shown that bone marrow stem cell transplantation can improve the expression of Bcl-2 after cerebral ischemia reperfusion in rats and decrease the expression of Bax. Thus it reduces the cerebral ischemia/reperfusion injury-induced apoptosis and improves the neurological status in rats with focal cerebral ischemia (Liu et al., 2005; Huang et al., 2006). Bone marrow stem cell mobilization has a protective effect on cerebral ischemia (Chen et al., 2012). Granulocyte colony stimulating factor is a strong bone marrow stem cell mobilizing agent and has been widely applied clinically. Granulocyte colony stimulating factor can mobilize bone marrow stem cell release into the peripheral blood and help them reach the ischemic region. Here they protect injured neurons, improve the blood supply and promote the recovery of neurological function (Zhang and Lei, 2014). Promotion of circulation and removing blood stasis method can improve microcirculation, encourage bone marrow stem cells to move from the bone marrow into the blood circulation and so exert the mobilization effect. These stem cells can easily traverse blood vessels and reach the ischemic region (Zhang and Zhang, 2014). Radix *Ilicis Pubescentis*, a commonly used Chinese herbal medicine that can remove blood stasis, alter cerebral blood flow, and assist in the prevention and treatment of cerebral vascular disease (Cai et al., 2001). Radix *Ilicis Pubescentis* total flavonoids (RIPTFs) is one of the effective ingredients extracted from the dried root of the holly, *Ilicis Pubescentis* (Liu et al., 2014). Our previous studies found that RIPTFs had an anti-ischemic effect (Zhang et al., 2012a, b). In this study, we used different concentrations of RIPTFs to treat rat models of cerebral ischemia/reperfusion injury after they received transplants of bone marrow stem cells that were mobilized by recombinant human granulocyte colony stimulating factor. We investigated the mechanism of RIPTFs on enhancing bone marrow stem cell mobilization. Materials and Methods {#sec1-2} ===================== Preparation of rat models of cerebral ischemia/reperfusion {#sec2-1} ---------------------------------------------------------- A total of 96 healthy, clean, male Sprague-Dawley rats aged 6--7 weeks and weighing 280--300 g were provided by the Hebei Provincial Experimental Animal Center of China (license SCXK (Ji) 2003-1-003). The rats were housed at 25 ± 3°C and humidity of 55 ± 10%, and were allowed free access to food and water. The experiment was approved by the Animal Ethics Committee of Henan University of Traditional Chinese Medicine in China. The 96 rats were equally and randomly divided into sham operation, model, mobilization, high-, moderate-, low-dose RIPTFs groups. In all groups, except the sham operation group, rats were intraperitoneally anesthetized with 10% chloral hydrate 0.3 mL/100 g (Tianjin Municipality Kemi'ou Chemical Reagents Development Center, Tianjin, China) after fasting for 12 hours. The rats were fixed on the operation table in a supine position. A median incision was made on the neck. According to the method of Kitagawa et al. (1991), a nylon thread (approximately 0.30 mm in diameter) coated with 1 mm of silicone rubber on the head end was inserted into proximal end of the middle cerebral artery through the left common carotid artery to a depth of 18--20 mm to block the blood supply of the middle cerebral artery for 2 hours. The nylon thread was then taken out, and the left middle cerebral artery was occluded. Intraoperative temperature was maintained at 23--24°C. In the sham operation group, middle cerebral artery was exposed under anesthesia but had no embolism. Rats were scored at 24 hours after ischemia and reperfusion in accordance with the Longa\'s method (Longa et al., 1989): 0 points, no nerve damage symptoms; 1 point, cannot fully extend right forepaw; 2 points, circling to the right; 3 points, dumping to the opposite side; 4 points, cannot walk by itself, loss of consciousness; 5 points, death. The rats scoring 1 point or above were considered successful ischemic models. Drug intervention and bone marrow stem cell mobilization {#sec2-2} -------------------------------------------------------- Radix *Ilicis Pubescentis* extract was provided by the Laboratory of Analytical Chemistry, Henan University of Traditional Chinese Medicine, China. The RIPTF content was 52%. Radix *Ilicis Pubescentis* is the root of the plant *Ilex pubescens* Hook. et Arn, belongs to Ilex L. (Aquifoliaceae), and was found in Anhui Province of China, and identified by Professor Cheng-ming Dong from Henan University of Traditional Chinese Medicine, China. Radix *Ilicis Pubescentis* coarse powder was taken and soaked with 10 times the amount of 70% ethanol for 0.5 hour, followed by 2 reflux extractions, the first time for 1.5 hours and the second time for 1 hour. After filtration, the two filtrates were combined and subjected to vacuum ethanol recovery until no alcohol smell was detected. The Radix *Ilicis Pubescentis* extract obtained was put on an AB-8 macroporous resin column (Tianjin Haiguang Chemical Co., Ltd., Tianjin, China). The concentration of RIPTF in the sample liquid was 0.4 g/mL; the sample liquid pH was 4.5; the ratio of the crude drug to macroporous resin was 1:8; the ratio of the diameter to the height of the macroporous resin column was 1:15. Total saponins were eluted by 12 bed volume water and 10 bed volume 30% ethanol. The 30% ethanol eluent was collected, followed by vacuum ethanol recovery, and then dried at 50°C, to obtain the RIPTFs (Feng et al., 2012b). Ultraviolet spectrophotometry was used and rutin was used as the control. The content of RIPTFs was determined (Xu et al., 2011). RIPTFs were dissolved in 0.5% sodium carboxymethyl cellulose (Tianjin Fuchen Chemical Reagent Factory, Tianjin, China) to make 1.5, 0.75, and 0.375 mg/mL suspensions. In accordance with Miao\'s method (Miao, 2003), rats were intragastrically administered 0.2 mL/10 g drug solutions, once a day, from 3 days before the modeling to the 2 days after the modeling. The rats in the sham operation, model and mobilization groups were given 0.5% sodium carboxymethyl cellulose. The rats in the high-, moderate- and low-dose RIPTFs groups received 0.3, 0.15 and 0.075 g/kg body weight of RIPTFs solutions, respectively. From 3 days before the ischemic modeling, rats in the mobilization group, high-, moderate- and low-dose RIPTFs groups were also given normal saline-diluted recombinant human granulocyte colony stimulating factor (Approval No. GYZZ S19990049; Shandong Qilu Pharmaceutical Co., Ltd., Jinan, Shandong Province, China) by subcutaneous injection (10 μg/kg daily), until the second day after injury for a total of 5 days. Rats in the model and sham operation groups were injected with an equal volume of normal saline. At 1 day after the ischemic injury, rats were intraperitoneally injected with 50 mg/kg BrdU (Shanghai Hao Ran Biological Co., Ltd., Shanghai, China) twice, with an interval of 2 hours, so as to mark the dividing bone marrow stem cells. A total of 35 rats died from poor drug tolerance and severe reaction to ischemic modeling during the experiment. The data from the 61 remaining rats were used in the analysis, including 16 in the sham operation group, 7 in the model group, 9 in the mobilization group, 12 in the high-dose RIPTFs group, 8 in the moderate-dose RIPTFs group and 9 in the low-dose RIPTFs group. Sample collection {#sec2-3} ----------------- In each group, the rats were sacrificed at 1 hour after administration on day 5. The brain was immediately separated on an ice plate. In accordance with a previous method (Paxinos and Watson, 2005), coronary sections were obtained at the optic chiasm and 4 mm posterior to the optic chiasm, fixed with 10% formaldehyde solution, and embedded with paraffin. Subsequently, samples were sliced into 5 μm thick serial sections, pasted on the polylysine-treated glass slides, and stored at room temperature. Pathological observation of brain tissue {#sec2-4} ---------------------------------------- Sections of the left hemisphere underwent hematoxylin-eosin staining (Sun et al., 2014) and Nissl staining (Wu et al., 2015). Under a light microscope at 400× magnification (Olympus, Japan), nerve cells and Nissl bodies were observed in the brain. According to the results of hematoxylin-eosin staining, pathological change of nerve cells could be divided into four levels: "--" nerve cells, cytoplasm and nucleus were normal; "+" most of nerve cells, cytoplasm and nucleus were normal, but a few nerve cells were atrophied; "++" a small number of nerve cells were atrophied, and the majority of nerve cells, cytoplasm and nucleus were normal; "+++" most of nerve cells were atrophied, cytoplasm was apparently reduced, their nuclei and nucleoli were blurred and indistinct. According to the results of Nissl staining, pathological change of nerve cells can be divided into four levels: "--" abundant Nissl bodies in the cytoplasm; "+" less than 2/10 Nissl bodies had decreased or disappeared; "++" 3/10--6/10 Nissl bodies had decreased or disappeared; "+++" 7/10--10/10 Nissl bodies had decreased or disappeared. Bcl-2, Bax, CD34 and BrdU immunoreactivities in the brain tissue as detected by immunohistochemical method {#sec2-5} ---------------------------------------------------------------------------------------------------------- Sections of ischemic brain tissue in each group were attached to the glass slide coated with polylysine, dewaxed, hydrated, incubated with 3% H~2~O~2~ for 10 minutes, and washed twice with PBS each for 5 minutes. Antigen was retrieved by exposure to microwave radiation in an oven. Following two washes with PBS each for 3 minutes, sections were blocked with normal goat serum at 37°C in a wet box for 20 minutes. Redundant liquid was drawn off. Sections were incubated with rabbit anti-rat BrdU monoclonal antibody (Boster, Wuhan, China), rabbit anti-CD34 polyclonal antibody (1:200) (Boster), rabbit anti-Bax polyclonal antibody (1:200) (Boster), and rabbit anti-Bcl-2 polyclonal antibody (1:200) (Boster) at 4°C overnight. After two washes with PBS each for 3 minutes, sections were incubated with biotinylated goat anti-rabbit IgG (1:100) (Boster) at 37°C for 20 minutes in a wet box, washed with PBS (3 minutes × 2), incubated with streptavidin biotin peroxidase complex at 37°C for 20 minutes in a wet box, washed with PBS (5 minutes × 4), and visualized with 3,3′-diaminobenzidine. Visualization was terminated with PBS. Subsequently, sections were counterstained with hematoxylin and mounted with a neutral resin. Bcl-2 and Bax expression and the number of CD34- and BrdU-positive cells were observed under 400× light microscope (Olympus). According to results of staining, the expression of different proteins in nerve cell membrane and cytoplasm was graded: "--" negative expression of Bax; "+" weakly positive expression of Bax; "++" positive expression of Bax; "+++" strongly positive expression of Bax. Statistical analysis {#sec2-6} -------------------- Data were analyzed using SPSS 13.0 for windows medical statistical package (SPSS, Chicago, IL, USA). Ranked data were measured using a Ridit test. The difference between groups was compared using *t*-test. A value of *P* \< 0.05 was considered statistically significant. Results {#sec1-3} ======= RIPTFs improved the effects of bone marrow stem cell mobilization on pathological changes in rat brain tissue after cerebral ischemia/reperfusion injury {#sec2-7} -------------------------------------------------------------------------------------------------------------------------------------------------------- Hematoxylin-eosin staining showed that, compared with the sham operation group, nerve cell atrophy and cytoplasmic volume shrinkage were observed and nuclei and nucleoli were indistinct. Compared with the model group, nerve cells, cytoplasm, nucleus were normal in the mobilization group and high-dose RIPTFs group. In most nerve cells, cytoplasm and nucleus were normal in the moderate-dose RIPTFs group, but some nerve cells were atrophic, cytoplasm volume was clearly reduced, and nuclei and nucleoli were indistinct. In the low-dose RIPTFs group, most nerve cells were atrophic, cytoplasm had shrunk and the nuclei and nucleoli were indistinct, and some nerve cells, cytoplasm, nucleus were normal (**[Figure 1](#F1){ref-type="fig"}, [Table 1](#T1){ref-type="table"}**). {#F1} ###### RIPTFs improved the effect of bone marrow stem cell mobilization on pathological changes in rat brain tissue with cerebral ischemia/reperfusion injury  RIPTFs improved the effects of bone marrow stem cell mobilization on Nissl body in rat brain tissue with cerebral ischemia/reperfusion injury {#sec2-8} --------------------------------------------------------------------------------------------------------------------------------------------- Nissl staining results show that compared with the sham operation group, neuronal cell edema and vacuolar degeneration were observed in the experimental groups. Nissl bodies disappeared in the model group. Compared with the model group, a large number of Nissl bodies were found in the cytoplasm of most neuronal cells, and Nissl bodies disappeared in a few cells in the mobilization, high- and moderate-dose RIPTFs groups. Nissl bodies disappeared in most neurons, and Nissl bodies could be seen in the cytoplasm of some cells (**[Figure 1](#F1){ref-type="fig"}, [Table 2](#T2){ref-type="table"}**). ###### RIPTFs improved the effects of bone marrow stem cell mobilization on Nissl bodies in rat brain tissue after cerebral ischemia/reperfusion injury  RIPTFs elevated the effect of bone marrow stem cell mobilization on Bax immunoreactivity in rat brain tissue after cerebral ischemia/reperfusion injury {#sec2-9} ------------------------------------------------------------------------------------------------------------------------------------------------------- Immunohistochemical staining showed that no Bax immunoreactivity was detected in the sham operation group. Compared with the sham operation group, Bax immunoreactivity did not obviously alter in nerve cells of rats from the mobilization group (*P* \> 0.05), but Bax expression was weakly positive in the model group (*P* \< 0.05). Compared with the model group, Bax immunoreactivity became weaker in the high-dose RIPTFs group (*P* \< 0.01), but no significant changes in Bax immunoreactivity were detected in the moderate- and low-dose RIPTFs groups (*P* \> 0.05; **[Figure 2](#F2){ref-type="fig"}, [Table 3](#T3){ref-type="table"}**). {#F2} ###### RIPTFs elevated the effect of bone marrow stem cell mobilization on Bax immunoreactivity in rat brain tissue after cerebral ischemia/reperfusion injury  RIPTFs elevated the effect of bone marrow stem cell mobilization on Bcl-2 immunoreactivity in rat brain tissue after cerebral ischemia/reperfusion injury {#sec2-10} --------------------------------------------------------------------------------------------------------------------------------------------------------- Immunohistochemical staining showed that Bcl-2 was not detected in the sham operation and model groups. Compared with the sham operation group, Bcl-2 immunoreactivity was significantly greater in the mobilization, high- and moderate-dose RIPTFs groups (*P* \< 0.01). No significant changes were detectable in Bcl-2 immunoreactivity in the low-dose RIPTFs group (*P* \> 0.05). Compared with the model group, Bcl-2 immunoreactivity was significantly higher in the mobilization and high-dose RIPTFs groups (*P* \< 0.05). No significant changes in Bcl-2 immunoreactivity were found in the moderate- and low-dose RIPTFs groups (*P* \> 0.05; **[Figure 2](#F2){ref-type="fig"}, [Table 4](#T4){ref-type="table"}**). ###### RIPTFs elevated the effect of bone marrow stem cell mobilization on Bcl-2 immunoreactivity in rat brain tissue after cerebral ischemia/reperfusion injury  RIPTFs elevated the effect of bone marrow stem cell mobilization on CD34 immunoreactivity in rat brain tissue after cerebral ischemia/reperfusion injury {#sec2-11} -------------------------------------------------------------------------------------------------------------------------------------------------------- Immunohistochemical staining showed that, compared with the sham operation group, CD34 immunoreactivity was significantly greater in the mobilization, high- and moderate-dose RIPTFs groups (*P* \< 0.01). No significant changes in CD34 immunoreactivity were detected in the ischemic region in the model and low-dose RIPTFs groups (*P* \> 0.05). Compared with the model group, CD34 immunoreactivity in the ischemic region was significantly higher in the high-dose RIPTFs group (*P* \< 0.05). No significant changes in CD34 immunoreactivity were found in the ischemic region of the mobilization, moderate- and low-dose RIPTFs groups (*P* \> 0.05; **[Figure 2](#F2){ref-type="fig"}, [Table 5](#T5){ref-type="table"}**). ###### RIPTFs elevated the effect of bone marrow stem cell mobilization on CD34 immunoreactivity in rat brain tissue following cerebral ischemia/reperfusion injury  RIPTFs elevated the effect of bone marrow stem cell mobilization on BrdU immunoreactivity in rat brain tissue after cerebral ischemia/reperfusion injury {#sec2-12} -------------------------------------------------------------------------------------------------------------------------------------------------------- Immunohistochemical staining showed that there was no BrdU immunoreactivity in the model group. Compared with the sham operation group, BrdU immunoreactivity was significantly greater in the high-dose RIPTFs group (*P* \< 0.05). No significant change in BrdU immunoreactivity was detected in the mobilization, moderate- and low-dose RIPTFs groups (*P* \> 0.05). BrdU immunoreactivity was significantly higher in the high-, moderate- and low-dose RIPTFs groups than in the model group (*P* \< 0.01). BrdU immunoreactivity was significantly higher in the mobilization group than in the model group (*P* \< 0.05; **[Figure 2](#F2){ref-type="fig"}, [Table 6](#T6){ref-type="table"}**). ###### RIPTFs elevated the effect of bone marrow stem cell mobilization on BrdU immunoreactivity in rat brain tissue after cerebral ischemia/reperfusion injury  Discussion {#sec1-4} ========== Ischemic stroke is the most common cerebrovascular disease, and cerebral ischemia is frequently detected in the middle cerebral artery (Zhu et al., 2015). The methods of inducing animal models of focal cerebral ischemia include the thread embolism method, micro-embolism method, and chemical stimulus-induced thrombotic occlusion method (Chen et al., 2015). The thread embolism method is recognized as the ideal method for modeling of focal cerebral ischemia as there is no craniotomy, little trauma and good repeatability (Fan et al., 2014). We chose this model as it is similar to the clinical conditions. Nissl body is a morphological index of nerve cell viability (Zeng et al., 2015) as its function was strongly related to neuronal function. Therefore, it is often used to judge the pathological changes of nerve cells. Moreover, the changes in Nissl bodies can reflect nerve cell changes in the hippocampus. Nissl bodies are very sensitive to ischemia and hypoxia in nerve cells. When axons are damaged, Nissl bodies of nerve cells tend to dissolve or disappear (Zhong et al., 2011). This study demonstrated that the number of Nissl bodies in the cytoplasm of nerve cells in the hippocampus of the rats in the model group clearly diminished after cerebral ischemia/reperfusion injury. The decrease in the number of Nissl bodies was lessened by drug administration following ischemia. The use of the RIPTFs drug increased the number of Nissl bodies, promoted axonal regeneration and lengthened the nerve growth cone, indicating that RIPTFs protected nerve cells of rats against ischemia/reperfusion injury. Nerve cell apoptosis is strongly correlated with cerebral ischemia and is regulated by apoptosis-related genes (Luo et al., 2005). Bcl-2 is an apoptosis-inhibiting gene, and Bax protein is a pro-apoptotic gene; both are considered the most important genes to suppress and promote apoptosis (Banjerdpongchai et al., 2011; Song et al., 2011). Our experimental results showed that RIPTFs suppressed neuronal apoptosis caused by cerebral ischemia by decreasing the expression of Bax and enhancing the expression of Bcl-2. CD34 cells can migrate to the ischemic region or injury site and participate in local angiogenesis (Mu et al., 2013). Immunohistochemical staining can be used to measure the microvessel density and area of the ischemic tissue using a CD34 antibody so as to determine neovascularization in the early stage (Zhu et al., 2011). This study confirmed that RIPTFs effectively improved capillary blood flow and oxygen supply and reduced the degree of damage to capillary endothelial cells in the hypoxic and ischemic brain. BrdU as an analogue of thymidine can compete with endogenous thymidine and integrate into the nucleus during the DNA synthesis of S phase of the cell cycle. BrdU as a marker of cell proliferation has been widely used in the field of neural stem cells (Seaberg and van der Kooy, 2002). This study verified that, after cerebral ischemia, BrdU expression was apparently stronger in the RIPTFs groups than that of the model group, and BrdU expression was higher in the mobilization group than in the model group, which suggested that RIPTFs and bone marrow mobilizing agent (recombinant human granulocyte colony stimulating factor) may have a synergistic effect, to help improve blood circulation in the ischemic region, and promote the reconstruction of neural circuits. The present study verified that RIPTFs improved the mobilization of bone marrow stem cells, reduced the apoptosis of nerve cells in brain tissue, could prevent and repair damage from cerebral ischemia, and had synergistic effect on bone marrow stem cell mobilizing agent in rat models of cerebral ischemia/reperfusion injury. However, a more comprehensive dose-effect relationship of RIPTFs should be investigated. Despite the simplicity of the mobilization of bone marrow stem cells and that immunological rejection of allogeneic transplantation can be avoided so it is an effective treatment for cerebral ischemia (Rüster et al., 2006; Dietz et al., 2007; Wang et al., 2007), its efficacy and safety need further evaluation. The honing of concordant remedies for the mobilization of bone marrow stem cells towards ischemic tissue, and the precise mechanism of the differentiation into functional cells should be clarified. New drugs to enhance the targeting and effectiveness of bone marrow stem cell mobilization should be developed. It would be beneficial to establish a complete technology platform for animal models, and explore the possibility of clinical application of the various technologies. Thus, bone marrow stem cell mobilization co-medication to improve cerebral ischemia/reperfusion strategy will become a new means of cell transplantation therapy incorporating traditional medicine, hyperacute thrombolytic therapy and stem cell therapy. *We are very grateful to Professor Jian-guo Li from Henan University of Traditional Chinese Medicine, China for assistance in the implementation of pathology*. ***Funding:*** *This study was supported by a grant from the State "Major New Drug Creation" Science and Technology Major Project in China, No. 2009ZX09103-324; a grant from the Henan Provincial Science and Technology Innovation Team in University in China, No. 2012IRTSTHN011.* **Conflicts of interest:** *None declared.* **Plagiarism check:** *This paper was screened twice using Cross-Check to verify originality before publication.* **Peer review:** *This paper was double-blinded and stringently reviewed by international expert reviewers.* Copyedited by Dawes EA, Pack M, Yu J, Qiu Y, Li CH, Song LP, Zhao M [^1]: **Author contributions:** *MSM designed this experiment and was responsible for the paper authorization. LG, RQL and XM performed experiments and collected the data. MSM and LG estimated experiments. LG wrote the manuscript. All authors approved the final version of the paper.*

Introduction {#s1} ============ Biological knowledge, accumulated over centuries of observation and experimentation, is contained within the legacy format of printed text, which may or may not be digitized [@pone.0089550-Miller1]. Because many observations are expensive or impossible to duplicate, Biology as a discipline, needs to leverage the scalability of computing to optimize use of all existing data. Because the volume of information is far more than any human can read in a lifetime, a key challenge for biology is migrating this vast amount of knowledge into modern, machine-readable formats so a computer can do the work of data discovery. Tools designed to aid in the manual annotation of text have improved the rate of digitization, but this process needs to be streamlined further if it is to catch up to the current rates of species discovery (approximately 19,000 new species every year, [@pone.0089550-IISE1]). Approaches that involve generating marked-up manuscripts from databases are emerging [@pone.0089550-Penev1] but so far only in taxonomy and not in ecology, and mark-up is not yet as detailed as needed. Decades of research into Natural Language Processing and Machine Learning have resulted in a suite of freely available algorithms capable of extracting information from general text, such as newspaper articles [@pone.0089550-Jackson1]. The specialized nature of biological text requires the development of new algorithms or modification of existing algorithms [@pone.0089550-Thessen1]. Many specialized algorithms exist and continue to be refined for the purpose of identifying species names in text and extracting morphological information and molecule interaction information (See review of biodiversity related NLP tools in [@pone.0089550-Thessen1]). Machine learning is not yet advanced enough to allow large-scale extraction of knowledge from biological text without human input (at least not life-wide, see [@pone.0089550-Cui1] for a tool that works well for plant descriptions). Numerous annotator tools have been developed to aid the process of human text markup to guide machines (see [@pone.0089550-Agosti1] for a biodiversity example). In addition to using algorithms to extract information from text, placing data in machine readable formats and making it more interconnected increases the usefulness of data by making it easier to find and combine in novel ways by people other than the original collector [@pone.0089550-Heath1]. The linked open data cloud (LOD) is one manifestation of this type of data mobilization [@pone.0089550-Bizer1]. There are over 31 billion triples in LOD and 9.6% of them are from life sciences data sets [@pone.0089550-Bizer1]. Many of these life sciences data sets have a molecular focus, but some, like TaxonConcept (<http://datahub.io/dataset/taxonconcept>) and GeoSpecies (<http://datahub.io/dataset/geospecies>) are biodiversity-related. Some members of the taxonomy community have embraced Semantic Web technology as a tool that can streamline the process of describing species, make optimal use of the data collected by taxonomists and efficiently manage species information [@pone.0089550-Deans1]. RDF (Resource Description Framework), one of the languages of the Semantic Web, has been identified as a machine-readable way to express taxonomic information [@pone.0089550-Page1]. Organizations such as TDWG (Taxonomic Databases Working Group) and others have been working to develop a semantic framework for describing biodiversity information [@pone.0089550-Webb1], [@pone.0089550-Page1], [@pone.0089550-Page2]. A major roadblock to developing large quantities of machine-readable data is widespread adoption of community standards, such as unique identifiers for species, specimens and observations and how to model such data. Proposed biodiversity semantic structures focus on describing nomenclature (TDWG), phenotype (PATO) and observations (TDWG and OBOE; [@pone.0089550-Madin1]). The field of evolutionary informatics has recently been defined and strives to organize information about species around the structure of a tree of life [@pone.0089550-Parr1]. The related field of ecoinformatics has developed a strong network of tools and support for ecological data throughout the data lifecycle [@pone.0089550-Michener1]. Standard vocabularies for habitats exist and can be used to relate species to their habitats (for example, EnvO (<http://environmentontology.org>). Ontologies and schema treating ecological species associations have been attempted several times ([@pone.0089550-Williams1]; (<http://wiki.tdwg.org/twiki/bin/view/DarwinCore/InteractionExtension>; SWEET (sweet.jpl.nasa.gov), SEEK (seek.ecoinformatics.org), ETHAN ([@pone.0089550-Parr2]; bioportal.bioontology.org/ontologies/1530), SWISST (foodwebs.org), and TRIN (<http://bit.ly/1cXHoHs>), but a leading standard has not yet emerged. A complete semantic representation of biodiversity data will likely use components from all of these ontologies. The biomedical community has fully embraced machine learning, informatics and semantics as a way to improve information sharing and discovery. Numerous tools exist for extracting molecular entities, diseases and interactions from text [@pone.0089550-Ananiadou1], [@pone.0089550-Krallinger1]. Large standardized repositories exist for storing biomedical and molecular data (GenBank, uniprot, KEGG, ArrayExpress, OMIM, PubMed and BTRIS) that offer archiving, citation and visualization tools. The U.S. National Library of Medicine has developed the Unified Medical Language System, which functions as a Biomedical WordNet, providing standardized definitions for biomedical terms and linking related terms in a Semantic Network [@pone.0089550-Bodenreider1]. Ecoinformatics has not received the same level of investment as biomedical informatics, but both disciplines suffer from ambiguous language, use of abbreviations and the constant creation of new terms [@pone.0089550-Chen1]--[@pone.0089550-Chang1]. Biodiversity sciences are particularly crippled by data heterogeneity because of the distributed nature of data collection and the need to consider the entire body of knowledge [@pone.0089550-Page2]. Thus, a semantic infrastructure that improves data discovery and integration, has the potential to make a disproportionate, positive effect on advances in the discipline. There are thousands of databases and journals holding biodiversity information (too many to name all here) all with a different taxonomic and geographic scope. The advantages of using RDF, or some other semantic markup, to manage biodiversity data stem from the ability to assign unique identifiers (URIs) to names, taxa, concepts, etc. and then link information using those identifiers [@pone.0089550-Page1]. Much of the conversation has been dominated by linking data phylogenetically and taxonomically, but semantic structures also have the potential to link information based on interactions between taxa, between a taxon and its environment, and ultimately to look at ecological or environmental interactions in another context (phylogeny, for example). Building this semantic structure is non-trivial because many data sources use their own terms and identifiers; however, within biodiversity science, standards are being developed and followed by some major biodiversity data sources [@pone.0089550-Clark1], [@pone.0089550-Wieczorek1]. The Encyclopedia of Life ([www.eol.org](http://www.eol.org)) is a web site with the ambitious goal of creating a web page for every species and enabling global access to knowledge about life on earth. They are accomplishing this goal by being a content aggregator rather than authoring species pages [@pone.0089550-Patterson1]--[@pone.0089550-Rotman1]. EOL develops relationships with existing data sources to have their content featured on an EOL species page with full attribution and creative commons licenses (content partners include more than 250 museums, government agencies and research consortia). EOL has also developed a network of credentialed curators who label aggregated content as trustworthy or untrustworthy. All content featured in EOL can be accessed through an API that gives data in XML or JSON formats. Anyone with internet access can gain EOL content easily through this API. The goals of this project were to develop and evaluate workflows to mine information held in EOL text objects in order to 1) annotate them with DBpedia URIs for biological entities relevant to EOL user interest, and 2) generate a global species association network. These two tasks, even at a very basic level of success, provide the foundational structure needed to explore EOL information from an ecological perspective rather than its current taxonomic perspective. Outside of EOL, these goals are a step toward large-scale analyses of biodiversity data based on ecological relationships. Since 1) there are approximately 2 million described species, 2) each species can have up to 100 text objects or more, and 3) EOL content is updated weekly from hundreds of content partners, any tool for interacting with EOL text must be dynamic, scalable and capable of handling heterogeneous structure. To demonstrate proof-of-concept, we have chosen 21 test species. In this paper we will describe and evaluate our workflows and the structure of the extracted data. We also provide pointers to data files ([Appendix S1](#pone.0089550.s002){ref-type="supplementary-material"}) and recommendations for further analysis. Materials and Methods {#s2} ===================== Selection of test species {#s2a} ------------------------- To explore approaches to achieving our two goals we chose 21 test species across the tree of life ([Table 1](#pone-0089550-t001){ref-type="table"}). To limit overfitting, we chose species that are diverse in their taxonomy, life history, habitat, ecological niche and EOL content. Any process that operates over the entire tree of life will require attention to scalability. We have taken some basic measures to improve scalability within this project and discuss strategies to further improve scalability in future work (see [discussion](#s4){ref-type="sec"}). 10.1371/journal.pone.0089550.t001 ###### Test Species and URI Annotation Results. {#pone-0089550-t001-1} Common Name Scientific Name EOL Taxon ID Number of Text Objects in English Number of Correct URIs Number of Correct and Unique URIs Precision Recall F1 Score ------------------- ------------------------------ -------------- ----------------------------------- ------------------------ ----------------------------------- ----------- -------- ---------- Great White Shark *Carcharodon carcharias* 213726 56 169 45 0.918 1 0.956 Lion *Panthera leo* 328672 52 173 45 0.961 1 0.980 Moss *Hypnum fauriei* 861365 2 0 0 0 0 0 Diatom *Pseudo-nitzschia australis* 904440 19 23 14 0.852 1 0.92 Mosquito *Culex pipiens* 740671 8 8 6 1 1 1 Bacteria *Escherichia coli* 972688 7 36 17 0.783 1 0.878 Cactus *Harrisia simpsonii* 594934 20 13 9 1 1 1 Oak *Quercus robur* 1151323 36 37 20 0.949 1 0.974 Worm *Lumbricus terrestris* 3126801 17 28 15 0.903 1 0.949 Crab *Callinectes sapidus* 312939 45 94 36 0.95 1 0.97 Spider *Loxosceles reclusa* 1182604 17 27 16 0.871 1 0.931 Ciliate *Paramecium aurelia* 484359 2 3 3 1 1 1 Mushroom *Psilocybe cubensis* 190054 9 9 7 0.9 1 0.947 Beetle *Oryctes nasicornis* 982103 2 5 5 1 1 1 Walrus *Odobenus rosmarus* 328627 57 100 35 0.909 1 0.952 Water Bear *Hypsibius dujardini* 1053826 4 5 5 0.833 1 0.909 Virus Tobacco mosaic virus 8615186 2 5 4 1 1 1 Copepod *Eurytemora affinis* 1020941 6 6 4 1 1 1 Kelp *Nereocystis luetkeana* 902899 9 7 5 0.875 1 0.933 Tube Worm *Riftia pachyptila* 393274 55 68 17 0.523 1 0.687 Honey Bee *Apis mellifera* 1045608 61 222 54 0.978 1 0.989 GRAND TOTAL 487 1038 362 0.889 1 0.941 Description of text {#s2b} ------------------- The text data objects used in this study were accessed through the EOL API. EOL accesses data objects from its 250 content partners that deliver structured data so that information from many sources can be integrated on an EOL page based on taxon and subject. The text data objects used in this study were written for a general audience and vary in length from one to 12,075 words. All text data objects used in this study were in English and labeled as "Trusted" by EOL curators. Content partners include, but are not limited to, Wikipedia (<http://wikipedia.org>), Animal Diversity Web (<http://animaldiversity.org>) and ARKive (<http://www.arkive.org>). Data retrieval {#s2c} -------------- EOL exposes their content via an API that serves data in XML or JSON. Every taxon and data object in EOL has its own identifier which can be used to call it and its associated metadata through the API. EOL updates content from hundreds of providers every week, so interacting with EOL content through the API ensures relevance. API documentation can be found at <http://eol.org/info/api_overview>. Building the dictionary {#s2d} ----------------------- Text data objects connected to each of the test species were read and manually annotated for ecologically relevant words and phrases. These words and phrases were placed in a dictionary as keys and the appropriate DBpedia URIs as the corresponding values. For example, 'uterine cannibalsm': '<http://dbpedia.org/resource/Oophagy>' is included in the dictionary. DBpedia URIs were collected using the DBpedia Keyword API and manually placed in the dictionary. The documentation for this API can be found at <http://wiki.dbpedia.org/lookup/>. Any URIs can be placed in the dictionary (or removed) to suite user need. An important issue for dictionary creation is how much of and which word or phrase to include. For example 'predator', 'predation', 'predatory' are all words used to describe a predator prey interaction that we would want tagged with the DBpedia URI for Predation (<http://dbpedia.org/resource/Predation>). This can be handled by having three separate keys in the dictionary with identical values or by having one key, 'predat', that will find all of the words. We used the latter strategy to handle plurals and any other form of a word/phrase. This reduces the size of the dictionary and thus the time it takes to build it and iterate over it in the workflow. This strategy would generate many false positives if applied to general text, but in the biological literature chances of encountering "predat" outside the context of predation are smaller. URI annotation {#s2e} -------------- One code module was written to annotate EOL text objects with relevant DBpedia URIs (<https://github.com/EOL/pseudonitzchia>). The module used a list of Data Object IDs to call the API and get a JSON response. Each text data object was converted to lowercase. This helped control the size of the dictionary by not requiring separate entries for 'prey' and 'Prey'. The module then went through each key in the dictionary and searched for it in the data object text returned in the JSON response. If the key was found, the module returned the corresponding value (the DBpedia URI) and moved on to the next key. If the key was not found, the module returned nothing and moved on to the next key. Information extraction and network building {#s2f} ------------------------------------------- The important information extraction task for this project was to generate a species association network using information in EOL text objects. Another code module was written to call the EOL API using EOL taxon IDs and retrieve the text objects under the "Associations", "Trophic Strategy" "General Ecology" and "Habitat" subchapters (<https://github.com/EOL/pseudonitzchia>). We focused on these subchapters so we could extract information from text specifically describing ecological interactions (see [discussion](#s4){ref-type="sec"}). The JSON response was parsed and cleaned using Beautiful Soup (<http://www.crummy.com/software/BeautifulSoup/>) and then passed to the GNRD API. GNRD (Global Names Recognition and Discovery) is a tool for finding scientific names in web pages, pdfs, Microsoft Office documents, images or freeform text using the TaxonFinder and NetiNeti name finding engines [@pone.0089550-Leary1], [@pone.0089550-Akella1]. Documentation for the GNRD API can be found at <http://gnrd.globalnames.org/api>. GNRD has a "resolver" function that is capable of reconciling multiple names for one species to a single, current name based on a user-chosen authority. For this project we chose EOL as the authoritative source of names and synonyms. More information about the resolver tool can be found at <http://resolver.globalnames.org/>. If a name was found on a page that, according to EOL, was an old name, the API returned the EOL ID of the new name. The extracted associations were returned in the format of Taxon A (Subject of Taxon page): Taxon B (Mentioned on Taxon page). Thus, if a taxon was mentioned on a page it was considered associated with the topic of the page. The data, in tabular form, was imported into Cytoscape for visualization and analysis [@pone.0089550-Smoot1]. In the network visualization, the nodes represent taxa and the edges represent an interaction between the taxa. Metrics testing and analysis {#s2g} ---------------------------- ### URI annotations {#s2g1} A human annotator read all of the text objects for the test species and annotated each of them with appropriate DBpedia URIs. Precision, recall and F1 Score were calculated for each test species and overall ([Table 1](#pone-0089550-t001){ref-type="table"}). Errors were noted and categorized ([Table 2](#pone-0089550-t002){ref-type="table"}). Precision was calculated as the URIs correctly assigned divided by the total number of URIs assigned to a text object. Recall was calculated as the URIs correctly assigned divided by the total number of URIs that were supposed to be assigned. The F1 Score is an overall measure of accuracy and is the harmonic mean of precision and recall. It is calculated by dividing precision by recall, then dividing that quantity by the sum of precision and recall, then multiplying that quantity by 2. 10.1371/journal.pone.0089550.t002 ###### URI Annotation Errors. {#pone-0089550-t002-2} Species Negation Describing Related Taxa Word Part Generalities Homonym Total Errors ------------- ---------- ------------------------- ----------- -------------- --------- -------------- Shark 2 7 1 2 0 12 Lion 0 4 2 0 0 6 Moss N/A N/A N/A N/A N/A N/A Diatom 0 1 1 2 0 4 Mosquito 0 0 0 0 0 0 Bacteria 0 1 1 3 1 6 Cactus 0 0 0 0 0 0 Oak 0 2 0 0 0 2 Worm 0 1 0 1 1 3 Crab 0 3 0 0 1 4 Spider 0 1 0 0 2 3 Ciliate 0 0 0 0 0 0 Mushroom 0 0 0 1 0 1 Beetle 0 0 0 0 0 0 Walrus 1 0 2 1 3 7 Water Bear 0 0 0 0 1 1 Copepod 0 0 0 0 0 0 Virus 0 0 0 0 0 0 Kelp 0 0 1 0 0 1 Tube Worm 1 1 1 0 1 4 Honey Bee 0 0 2 0 1 3 GRAND TOTAL 4 21 11 10 11 57 ### Associations network {#s2g2} Three human annotators with biodiversity expertise read all of the text objects for the test species and developed a list of association statements representing real ecological relationships between taxa. Taxonomic relationships were not included. If the text said that a species was, for example, pollinated by members of a family, the association was recorded as one between a species and a family, not as several between that species and every member of that family. Some annotators worked before the automatic process and others worked after, but in all cases the annotators had no prior knowledge of the algorithm results. Annotator agreement was calculated using Fleiss\' Kappa [@pone.0089550-Fleiss1]. The workflow result was evaluated against performance of the GNRD API working directly on the EOL taxon page without the intervention of our workflow (See [Appendix S1](#pone.0089550.s002){ref-type="supplementary-material"} baseline.txt). Precision, recall and F1 Score were calculated for the baseline result (GNRD alone) and the workflow result compared to a human-created "gold standard" (See [Appendix S1](#pone.0089550.s002){ref-type="supplementary-material"} gold_standard_int.txt). Precision was calculated as the interactions correctly extracted by the workflow divided by the total number of interactions the workflow extracted. Recall was calculated as the interactions correctly extracted by the workflow divided by the total number of interactions that were supposed to be extracted. The F1 Score was calculated in the same way as for the URI annotations. When constructing the network by hand, the annotator must not only look for terms that refer to taxa, but make a judgment call as to whether or not the author is describing an interaction. We did not want to include taxa that were mentioned solely for comparative purposes, for example. The annotator-created and automated networks were visualized using Cytoscape software [@pone.0089550-Smoot1]. Results {#s3} ======= URI annotations {#s3a} --------------- The workflow successfully annotated 487 text data objects associated with 21 species in EOL using a biologically-focused dictionary with 239 ecologically relevant entries (See biodict.txt in [Appendix S1](#pone.0089550.s002){ref-type="supplementary-material"}). Some of those text objects did not receive any annotations (14%). The most annotations that any one text object received was 33 and the average number of URIs assigned to a text object was 2.1. The performance metrics for this particular method depended almost entirely on the construction of the dictionary. Overall, 80% of the URI annotations were correct. Precision = 0.889, recall = 1, and the F1 Score = 0.941 ([Table 1](#pone-0089550-t001){ref-type="table"}). The lowest F1 Score (0.687) was for the tube worm, *Riftia pachyptila*. Recall is perfect, indicating the ability of the computer to find strings in text, a simple task. False positives decrease the precision and, in our data set, can be caused by several factors ([Table 2](#pone-0089550-t002){ref-type="table"}). We have divided all of the errors into five categories: 1. Negation - This workflow cannot distinguish between "This species migrates." and "This species does not migrate." Both texts would be annotated with the URI for migration. Negation is a common NLP problem with rules-based solutions that rely heavily on context [@pone.0089550-SanchezGraillet1]. Coping with negation in biodiversity and ecology text is possible, but would require additional analysis. 2. Describing related taxa - Some text objects will use terms describing a related species that will cause a text object to be annotated with a URI that does not apply to that taxon. One example found on the Great White Shark page is the text "white sharks often attack dolphins and porpoises from above, behind or below to avoid being detected by their echolocation". The text object containing that line would be annotated with the URI for echolocation, even though sharks do not have echolocation. However, since sharks have to cope with echolocation in their prey species, one could argue that this annotation would be relevant. For the metrics calculations in this publication, we took the conservative approach and did not consider them relevant. 3. Word or phrase part - Some terms will occur within other terms that mean something totally different. For example, the term "forest" in "kelp forest". This will result in an oceanic species being annotated with a URI for a terrestrial habitat. This error also occurs if a content provider\'s name appears in the text object and contains a relevant term like "garden" or "marine". 4. Generalities - Some terms are used in a general sense, not specifically referring to the taxon of interest and may cause a text object to be annotated with an irrelevant URI. This often occurs in headings. For example, a text object could give examples of why a species would migrate as an aside "e.g., to breeding or wintering grounds, to hibernation sites". This would result in the text object being annotated with the URI for hibernation even if the species does not hibernate. 5. Homonym - Sometimes the same term can have a subtly different meaning in different contexts. The available URIs or the dictionary may not have the appropriate level of specificity. For example, the DBpedia URI for migration refers specifically to bird migration and thus cannot be used for any other group. The DBpedia URI for molt refers specifically to arthropod molting and cannot be used for bird molting. The term "attack" can describe an act of predation, self defense or disease, so it is not clear which URI to use. The largest problem in our data set is the "describing related taxa" problem (37% of errors) followed by the "word part" and "homonym" problem (each are 19% of errors). Negation is the least of our problems (7% of errors). Association network {#s3b} ------------------- We used the EOL and GNRD APIs to locate ecologically relevant text on a species page and find scientific names in that text. The output from this process was imported into Cytoscape [@pone.0089550-Smoot1] for visualization ([Fig. 1](#pone-0089550-g001){ref-type="fig"}). Our automated methods found 585 relationships between 581 taxa just by looking at the content under the "Ecology" Chapter of the 21 test species pages. A manual construction of the associations network using human annotation of the test species found 675 relationships between 657 taxa ([Fig. 2](#pone-0089550-g002){ref-type="fig"}; includes references made using scientific and common names). Agreement between annotators was 0.840 (Fleiss\' Kappa; See [Appendix S1](#pone.0089550.s002){ref-type="supplementary-material"}, annotator_agreement.xlsx). Our automated methods had an overall precision of 0.844, a recall of 0.930 and an F1 Score of 0.885 ([Table 3](#pone-0089550-t003){ref-type="table"}). {#pone-0089550-g001} {#pone-0089550-g002} 10.1371/journal.pone.0089550.t003 ###### Performance Metrics for Associations Workflow. {#pone-0089550-t003-3} True Positives False Positives False Negatives True Negatives Precision Recall F1 Score ------------------------------ ---------------- ----------------- ----------------- ---------------- ----------- -------- ---------- *Carcharodon carcharias* 9 2 2 23 0.818 0.818 0.818 *Panthera leo* 30 2 4 16 0.938 0.882 0.909 *Hypnum fauriei* 0 0 0 1 *Pseudo-nitzschia australis* 0 0 0 3 *Culex pipiens* 0 0 0 6 *Escherichia coli* 0 0 1 119 *Harrisia simpsonii* 1 0 0 5 1.000 1.000 1.000 *Quercus robur* 81 2 3 15 0.976 0.976 0.976 *Lumbricus terrestris* 0 0 2 3 *Callinectes sapidus* 49 9 11 13 0.845 0.875 0.860 *Loxosceles reclusa* 0 0 0 11 *Paramecium aurelia* 0 0 0 1 *Psilocybe cubensis* 0 0 0 51 *Oryctes nasicornis* 0 0 1 24 *Odobenus rosmarus* 1 1 1 17 0.500 0.500 0.500 *Hypsibius dujardini* 0 0 0 7 Tobacco mosaic virus 5 0 1 0 1.000 0.833 0.909 *Eurytemora affinis* 0 0 0 4 *Nereocystis luetkeana* 0 0 0 2 *Riftia pachyptila* 0 0 0 7 *Apis mellifera* 318 75 17 19 0.809 0.952 0.875 TOTAL 494 91 43 347 0.844 0.930 0.885 The workflow (using the GNRD API) found 585 taxon names and 494 of those represented ecological species interactions (84%). Our results had a total of 91 false positives and 43 false negatives. The largest source of false positives was the inclusion of higher taxon names for species that were mentioned (89%). For example, if a species and the Family it belongs too are both mentioned in a text object the algorithm will include them both, even though the interaction is with the species, not the entire Family. The largest source of false negatives was interactions discussed in text objects that were not under the Ecology Chapter and thus not analyzed (46%). Other important sources of error include formatting confounding GNRD (especially virus nomenclature, 14% of false negatives) and self referencing (7% of false positives). Out of the 583 unique names GNRD found, 36 (6%) were identified as an "old" name and needed to be reconciled to the new name. For the task of identifying ecological interactions in text, GNRD alone had a precision of 0.477, a recall of 0.957 and an F1 Score of 0.636 ([Table 4](#pone-0089550-t004){ref-type="table"}). The workflow, which makes use of GNRD, was much more effective at extracting ecological interactions, with a relative improvement in the F1 Score of 39% over the baseline. 10.1371/journal.pone.0089550.t004 ###### Performance of GNRD without Associations Workflow. {#pone-0089550-t004-4} True Positives False Positives False Negatives True Negatives Precision Recall F1 Score ------------------------------ ---------------- ----------------- ----------------- ---------------- ----------- -------- ---------- *Carcharodon carcharias* 10 37 1 0 0.213 0.909 0.345 *Panthera leo* 33 45 2 0 0.423 0.917 0.579 *Hypnum fauriei* 0 2 0 0 *Pseudo-nitzschia australis* 0 7 0 0 *Culex pipiens* 0 7 0 0 *Escherichia coli* 0 135 1 0 *Harrisia simpsonii* 1 7 0 0 0.125 1.000 0.222 *Quercus robur* 82 26 2 3 0.759 0.976 0.854 *Lumbricus terrestris* 2 4 0 0 0.333 1.000 0.500 *Callinectes sapidus* 53 28 2 1 0.654 0.964 0.779 *Loxosceles reclusa* 0 16 0 0 *Paramecium aurelia* 0 3 0 0 *Psilocybe cubensis* 0 55 0 1 *Oryctes nasicornis* 1 25 0 0 0.038 1.000 0.074 *Odobenus rosmarus* 2 29 0 1 0.065 1.000 0.121 *Hypsibius dujardini* 0 8 0 0 Tobacco mosaic virus 5 3 1 0 0.625 0.833 0.714 *Eurytemora affinis* 0 6 0 0 *Nereocystis luetkeana* 0 4 0 0 *Riftia pachyptila* 0 12 0 0 *Apis mellifera* 321 101 15 0 0.761 0.961 0.849 TOTAL 510 560 24 6 0.477 0.957 0.636 Discussion {#s4} ========== URI annotations {#s4a} --------------- The performance of the URI annotation workflow depended entirely on the contents of the dictionary; building this dictionary was a manual process. This type of dictionary matching method has limited utility when the dictionary is not complete relative to user need. This problem can be partially addressed through methods currently employed by community ontology projects. For example, terms and their corresponding URIs in an ontology can be automatically formatted into a dictionary. Many ontologies are curated by experts who add new terms, reconcile synonyms and add connections to other knowledge bases. One could leverage Uberon, a multi-species anatomy ontology containing over 10,000 terms [@pone.0089550-Mungall1] that can be easily transformed into a dictionary and used to annotate text based on morphology. A user can run the program with whatever dictionary he/she chooses to target an area of interest. This strategy can be further improved by focusing annotations on taxa rather than text objects and thereby requiring less specificity in the dictionary. For example, if the goal were to annotate *Panthera leo* (EOL ID 328672) with the URI for predation, rather than every text object that discusses predation, we can be much more general with our dictionary key values. We only need "predat" instead of also needing "attack" "feed on" and "hunt" which can describe interactions not related to predation. Automating dictionary creation may also speed the process, but this would entirely depend on user need. For example, if a user were only interested in one key term, automation would not be helpful. There are many existing tools that can add semantic annotations to text. Many of these have been tested and developed using news articles and other non-specialized text [@pone.0089550-Rizzo1], [@pone.0089550-Milne1]. We have explored two annotators within the context of ecology text: Terminizer [@pone.0089550-Hancock1] and DBpedia Spotlight ([@pone.0089550-Mendes1]; [Fig. 3](#pone-0089550-g003){ref-type="fig"}, [Table S1](#pone.0089550.s001){ref-type="supplementary-material"}). Terminizer annotates text when it finds strings that match ontological terms from 40+ biological ontologies in the OBO Foundry. DBpedia Spotlight annotates text with DBpedia URIs when it finds a relevant string. To explore these systems, we submitted a single text object to Terminizer and DBpedia Spotlight ([Fig. 3](#pone-0089550-g003){ref-type="fig"}). Terminizer made 65 annotations from 21 different ontologies. Most of the annotations Terminizer made were correct (40%), but there were still large numbers of incorrect annotations (37%) and we were not able to verify some annotations (25%) because the terms were not defined in their respective ontologies. DBpedia Spotlight made 36 annotations, all with DBpedia URIs. Most of these annotations were correct (70%) and only one annotation could not be verified as correct or incorrect (\<1%). The errors made by Terminizer were mostly due to specificity and domain. For example, some annotations were made that described unrelated model organisms or molecular interactions. Another common error included annotating only one word in a two-word term, like "litter size" and "food chain". DBpedia Spotlight was able to find two-word terms like "food chain" and "marine mammal", but had disambiguation problems with terms that also had a pop culture reference. For example, "carcass" was annotated with a URI for a music band and "seal" was annotated with the URI for the Unites States Navy SEALs. The performance of these annotators may be improved by restricting the ontologies they are using. Terminizer has an option to choose specific ontologies in OBO or insert your own (in OBO format). DBpedia Spotlight has settings that can allow for refinement or it can take your own dictionary, if you can set up your own instance. Our method annotated the same text with six URIs from our dictionary, including "social animal", "electroreception", "gestation", "ovoviviparous", "predation" and "marine". These terms represent our ecological knowledge annotation goal. As expected, the tool we designed ourselves best addressed our needs. Only half of these desired terms were found by Terminizer and/or DBpedia Spotlight: "gestation" was found by both and "predator" and "marine" was found by DBpedia Spotlight. Our tool avoided some disambiguation problems by only working with biological text (so we don\'t have to worry about pop culture references like DBpedia Spotlight) and carefully selecting the terms in the dictionary. For example, "ovoviviparous" has a single meaning. A possible next step would be to fashion either a dictionary or an ontology file containing our terms so that Terminizer or DBpedia Spotlight could be used as an attractive user interface. We would like to suggest that annotators include pathways for users to add terms of interest within the functionality of the interface. {ref-type="supplementary-material"}.](pone.0089550.g003){#pone-0089550-g003} Association network {#s4b} ------------------- The ability to build an automated taxon associations network from narrative depends on three factors 1) the ability to recognize taxon names in text, 2) the ability to reconcile multiple terms used for the same species and 3) the ability to determine if a taxon mention refers to an ecological interaction. Recognizing taxon names in text is a major area of research. Several tools have been developed that can find scientific names [@pone.0089550-Thessen1] and we use one of them, GNRD, in this project. None of these tools can find common names in text, which results in missing approximately 13% of the total associations and taxa. Some work has been done in the field of Earth Science to recognize common names for geomaterials using contextual clues with some success (Jenkins pers. comm.). EOL has a compilation of over 780,000 common names across the tree of life that are linked to an EOL ID. The EOL API can return an EOL taxon for a common name submitted, but recognizing a common name in narrative is still a major challenge. One possible strategy is to search for the common names corresponding to the found scientific names (Mozzherin pers. comm.), but that would not find taxa only mentioned by common name and would not improve the performance of the workflow. One cannot assume that all of the taxa mentioned on an EOL taxon page have an ecological relationship with the topic of that page. Taxa are mentioned for several reasons including comparison, taxonomic relationships and in discussion of a common phenomenon, like poisonous mushrooms. Users must balance their need for precision and recall against the amount and type of text they process. As mentioned above, the largest cause of false negatives was names found in text objects that were not under the Ecology chapter. Most of those text objects were from Wikipedia. Including Wikipedia content in our analysis is likely to also dramatically increase false positives because Wikipedia text often includes a list of taxonomically related species and subspecies. This annotation and information extraction exercise represents the initial explorations for making EOL data semantically available and creating a semantic infrastructure describing species interactions. Most available information about species is organized taxonomically or phylogenetically. EOL is an example of this, allowing users to browse information using a taxonomic classification. Using a taxonomic infrastructure, a user is not able to navigate from a lion directly to its prey, its parasites or its competitors. Long term goals include enabling an ecologically-focused browse capability in EOL and making EOL content semantically available. Our short term vision is that users who are learning about Oophagy on the Wikipedia page can easily find EOL text objects and taxon pages that discuss the phenomenon through the linking of DBpedia and EOL URIs. These workflows are not perfect and, considering the complexity of biology and language, are not likely to ever be perfect. A text object might say "oak trees provide habitat for many animals". Animals refers to Animalia, but the authors were not intending to say that all Animalia live in oak trees. A text object may use a high level taxon name that in practice only applies to a single species in a particular area. For example, referring to deer (Cervidae) on the US East Coast really only refers to White Tailed Deer (*Odocoileus virginianus*). Some terms are used as though they correspond to a taxonomic group, but in reality do not, such as seabirds and worms. From an EOL perspective, it is far more important to have high precision than high recall (Wilson pers. comm.). Fortunately, EOL has curators and staff to cope with the errors that may result. The algorithm may be accurate enough to reduce the manual work to a level that is reasonable for the curators and staff. There are approximately 1,200 EOL curators of which 182 are considered "active" and make approximately 1,500 actions per month. This has resulted in approximately 22,000 EOL pages that have been subject to one or more curatorial actions since the curatorial functions were enabled in 2008. Still, other methods of assisting users in avoiding bad text-mining results are recommended to any data consumer, including EOL, such as not importing until an average F1 Score threshold is met, assigning numerical ratings to datasets based on rate and type of errors, and flagging datasets that are the result of text-mining. Future Work {#s5} =========== Using machines to extract knowledge from centuries of biological descriptive data is a promising strategy that still needs significant investment. Below we discuss future work specifically relating to knowledge extraction from sources like EOL. 1\. *Classify associations using a well-developed interaction vocabulary*. An association network like the one developed in this project could be made even more informative if the association was defined more specifically. At the moment, the network only states that a relationship exists. No information is given about the nature of the relationship. The most efficient way to do this is to develop a standard vocabulary for ecological relationships that can be applied using clues from the surrounding text. A new generation of researchers is working on normalizing terms and relationships that can be applied to found associations [@pone.0089550-Poelen1]. 2\. *Inferring meaning with contextual clues*. Humans use context to cope with ambiguous terms. Computers can do this through distributional semantics. This is where meaning is inferred by quantitative analysis of word-associates using statistical methods such as Cosine Similarity, Jaccard Index and Maximum Entropy [@pone.0089550-Tan1]. Algorithms exist that analyze text using distributional semantics and have been used successfully in the geosciences (Jenkins pers. comm.). We plan to apply these methods to EOL content. The URI annotations discussed in this paper can also provide contextual clues. For example, if a text object is annotated with the URI for predation and an association is also detected, a predatory relationship can be inferred. However, if multiple associations and multiple URI annotations are available for the same text object, it is unclear which goes with which. Development of these types of intelligence systems for ecological text is a step above the dictionary methods used here and represents our anticipated research trajectory. 3\. *Scaling up*. There are two important ways to increase the speed of this process: improving data retrieval and removing as much manual intervention as possible. Considering the frequency with which EOL updates, interacting with content through the API is ideal; however, these types of requests are relatively slow. There may be small ways to improve efficiency of these methods in the future, such as not rerunning the algorithms over EOL content that has not changed, but increases in the speed of queries using an API will be limited. SPARQL queries, because of their bulk processing capabilities, have the potential to be a much faster way to interact with heterogeneous EOL data, but cannot be immediately implemented. Any manual step is a bottleneck, including the manual dictionary creation and the correction of errors. Scale is and will continue to be an issue for any project trying to work over the entirety of life and all of its data. Conclusions {#s6} =========== Semantic strategies for representing knowledge are finding a practical foothold in biological research [@pone.0089550-Washington1], [@pone.0089550-Deans2]. The potential of semantics for revolutionizing the way biodiversity knowledge is recorded and data are shared and used is vast. The work described in this paper is a modest step toward large-scale knowledge extraction from an aggregator of biological text, EOL. As we continue to develop our workflows and data structures we will be increasingly able to utilize the advantages of semantics, including inferencing and reasoning. Supporting Information {#s7} ====================== ###### **Terminizer and DBpedia Spotlight Annotation Details.** Results from the annotation of an EOL text object with Terminizer and DBpedia Spotlight. (TXT) ###### Click here for additional data file. ###### **Data Sets.** List of data files available from this study and their description. (PDF) ###### Click here for additional data file. The authors would like to acknowledge Peter DeVries for early discussion, David Shorthouse, Dima Mozzherin and David Patterson for the creation and support of GNRD and Nathan Wilson, PyLadies Boston and the Boston Python Group for coding advice. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: AET. Performed the experiments: AET. Analyzed the data: AET. Contributed reagents/materials/analysis tools: AET CSP. Wrote the paper: AET CSP.

Recent developments in super resolution microscopy have allowed unprecedented insight into the supramolecular organization of cellular machineries in their native environments[@R1]--[@R7]. Despite those developments, high-throughput innovations in optical nanoscopy[@R8]--[@R11] have, thus far, been unmatched with the demand for rapid and robust analytical tools. The segmentation, analysis and classification of super-resolved subcellular assemblies is heavily prone to human cognitive biases and is still reliant on the use of semi-automated expensive image analysis software packages or non-standard custom-written protocols. Determining the architecture of nanoscopic structures in the cell using available methods can be expensive, time consuming and irreproducible, thereby hindering statistical inference across scales and the understanding of biological function at the molecular level. To address this limitation, we developed ASAP (Automated Structures Analysis Program), a generic, simple and freely available tool that enables rapid and objective detection, classification and analysis of the macromolecular architecture of cellular assemblies in nanoscopy images. ASAP is provided with a user-friendly graphical interface ([Figure 1A](#F1){ref-type="fig"}). A scripted version that allows pipelining of the analysis workflow for full automation and faster analysis is also included. ASAP follows a generic approach to the analysis of nanoscopic structures ([Figure 1B](#F1){ref-type="fig"}). It identifies structures using connectivity-based clustering for small and connected structures, or density-based clustering for large and fragmented structures. Accelerated by CPU- and GPU-based parallel computing, ASAP analyzes geometrically the segmented structures to extract relevant shape descriptors. The varied topology of the potential structures of interest and strong behavioral biases involved in describing the extracted topologies has inspired us to deploy a generic machine-learning approach for the automated classification of the analyzed structures. The user starts by training a classifier through annotation of a subset of the structures. The classifier learns to recognize the shape of a structure based on its descriptors. The accuracy of the training process can be improved by annotating more structures, changing the descriptors used for training or selecting a different type of classifier. Once the user is satisfied with the accuracy of the classifier, a classification model can be exported and subsequently used to automatically classify the rest of structures. The classified structures can be revised manually or using semi-automated re-assignment based on values imposed to shape descriptors. Using cluster analysis, ASAP can then group the analyzed structures into distinct populations based on statistical differences in their shape descriptors to help users answer important biological questions related to quantifying assembly states, detecting uneven spatial distributions or analyzing co-recruitment. All generated data can be plotted, modeled, montaged and exported for further analysis, dissemination and reproduction. ASAP was first tested on ground truth data of exemplary ring structures ([Figure 2A](#F2){ref-type="fig"}). We simulated super-resolved images of 30 nm to 60 nm rings constituted of 8 subunits or epitopes labeled at 50% and 75% efficiencies and sampled with 10 nm and 20 nm precisions. A number of other conditions were also assigned to mimic localization-based imaging experiments ([Figure 2A](#F2){ref-type="fig"}). The simulated structures were input to ASAP, segmented by density, analyzed geometrically and a small subset (146 structures) was used to train the classifier on recognizing rings from fragments ([Supplementary note 1](#SD1){ref-type="supplementary-material"}). More than 500,000 structures were subsequently segmented, analyzed and classified automatically with ASAP. Rings were selected for further analysis and their radial profiles were fitted to Gaussian functions to extract their radii ([Supplementary note 1](#SD1){ref-type="supplementary-material"}). We obtained median differential radii ranging from 0.7 nm to 3.3 nm depending on actual radius, labeling efficiency and spatial precision ([Figures 2B and C](#F2){ref-type="fig"}). These remarkable values were obtained for low labeling efficiencies (50 or 75%), large pixel size (10 or 20 nm) and small training volume (146 structures) considering current experimental standards in super-resolution imaging. To test for ASAP robustness against user input, we made systematic changes to the segmentation parameters and measured the radii of the extracted rings ([Supplementary note 1](#SD1){ref-type="supplementary-material"}). Despite the magnitude of the induced variations (\> 20% of the average size of the structures and \> 30% of their average intensity), differences between the median absolute differential radii measured for each set of parameters were remarkably low (\< 5 Å) ([Figure 2D](#F2){ref-type="fig"}). We subsequently evaluated the performance of ASAP using super-resolution images of different cellular structures and compared the results with published analysis. [Figure 3](#F3){ref-type="fig"} shows the results of re-analyzing the architecture of the nuclear pore scaffold published by Szymborska and coworkers[@R12] ([Supplementary note 2](#SD1){ref-type="supplementary-material"}). Nucleoporins (or Nups) are the main constituting proteins of the nuclear pore complex; a large machinery in the nuclear envelope responsible for the transport of macromolecules between the cell nucleus and cytoplasm[@R13],[@R14]. In the published work[@R12], nucleoporin datasets were segmented and analyzed with a combined approach employing strict human-based quality control and electron microscopy image analysis packages. In our analysis, we segmented the structures by density, analyzed them geometrically and used a subset of them to train a classifier ([Supplementary note 2](#SD1){ref-type="supplementary-material"}). All analyzed structures were automatically classified by ASAP using the classification model previously generated ([Figure 3A](#F3){ref-type="fig"}). We fit the radial profiles of the rings to Gaussian functions to extract the center of their peaks ([Supplementary note 2](#SD1){ref-type="supplementary-material"}). When comparing the results of the published analysis[@R12] with those generated with ASAP ([Figure 3B and C](#F3){ref-type="fig"}), we obtained a median differential radius of just 4.5 Å corresponding to a root mean squared error (RMSE) of \~9 Å (\~1.8% relative to the average radii of the analyzed structures). We have additionally used ASAP to analyze the nanoscale organization of the protein kinase TORC1[@R1] ([Supplementary note 3](#SD1){ref-type="supplementary-material"}), proteins involved in clathrin-mediated endocytosis[@R15] ([Supplementary note 4](#SD1){ref-type="supplementary-material"}) and the proapoptotic protein BAX[@R2] ([Supplementary note 5](#SD1){ref-type="supplementary-material"}) from super resolution microscopy images from previous studies ([Supplementary Information](#SD1){ref-type="supplementary-material"}). In all cases, we obtained remarkably small deviations (RMSE \< 5 nm) between our and published results. This similarity demonstrates the wide applicability of ASAP to a broad range of biological systems at a fraction of the cost and time. In addition, ASAP is entirely image-based and as such, it can process localization-based, coordinate-targeted and computationally-assisted super-resolved images. There is strong demand for tools that turn super resolution microscopy into a smarter structural biology methodology capable of quantitatively describing single structures in a dynamic biological context. We expect ASAP to fill this important niche and to become a unique and powerful tool enabling full exploitation of the recent developments in high-throughput super resolution imaging[@R9],[@R16],[@R17] for nanoscopy-based structural quantification. ASAP standalones for Mac OS and Windows, 5 guidance documents and extensive examples are attached as [supplementary material](#SD1){ref-type="supplementary-material"}. Supplementary Material {#SM} ====================== We would like to thank Christian Sieben and Suliana Manley (EPFL) for providing super-resolved images of TORC1, Markus Mund and Jonas Ries (EMBL) for datasets of proteins involved in clathrin-mediated endocytosis and useful discussions, Raquel Salvador-Gallego (University of Colorado Boulder) for helpful discussions on the datasets of the apoptotic protein Bax and Stephanie Alexander and Jan Ellenberg (EMBL) for super-resolved images of nucleoporins. We acknowledge a Max Planck Society (Max-Planck-Gesellschaft) postdoctoral fellowship awarded to J.S.H.D. This work was supported by Deutsche Forschungsgemeinschaft (DFG) grant GA164/3-1 and the European Research Council (ERC StG -- 309966) awarded to A.J.G.S. **Author contributions** J.S.H.D and A.J.G.S conceived and designed the study, J.S.H.D wrote the software and performed analysis. J.S.H.D and A.J.G.S assessed performance and wrote the manuscript. **Competing financial interests** The authors declare no competing financial interests. **Data Availability Statement** The data that support the findings of this study are available from the corresponding authors upon request. **Code Availability Statement** Source code for ASAP can be obtained from <https://github.com/jdanial/ASAP>. Compliations of ASAP for Windows and MacOS are available as **supplementary software**. {#F1} {ref-type="supplementary-material"}. Exact radii can be found in [supplementary note 1](#SD1){ref-type="supplementary-material"}.](EMS83286-f002){#F2} {ref-type="supplementary-material"}.](EMS83286-f003){#F3} [^1]: Present address: Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge CB2 1EW, United Kingdom.

For the abstract or full text in other languages, please see [Supplementary files](http://www.eurojnlofpsychotraumatol.net/index.php/ejpt/rt/suppFiles/17188/0) under Reading Tools online Over the last number of years the obesity rate has reached epidemic proportions in most countries (WHO, [@CIT0035]), and new studies show that 52% of the Danish population is overweight (Larsen, Ankersen, Poulsen & Christensen, [@CIT0018]). The parallel rise in the prevalence of eating disorders such as Anorexia Nervosa (Hay, Mond, Buttner & Darby, [@CIT0012]; Hudson, Hiripi, Pope & Kessler, [@CIT0013]) has been mentioned less frequently. Posttraumatic stress disorder (PTSD) and exposure to potentially traumatic events have been linked to adverse health outcomes, the onset of specific diseases, and premature death (Boscarino, [@CIT0003]; Scott et al., [@CIT0030]). One of these adverse health outcomes may be unhealthy weight. Some studies have focused on possible explanations for the connection between adverse life events and weight. A recent study (Akkermann et al., [@CIT0001]) showed that adverse experiences in childhood may heighten susceptibility to serotonergic dysregulation, which was associated with both binge eating and drive for thinness in a cohort of the European Youth Heart Study. Several studies of specific groups such as military veterans (Vieweg et al., [@CIT0034]) have found that the prevalence of overweight and obesity among male veterans with PTSD significantly exceeds national findings. Also studies with samples from the general population have found an association between PTSD and obesity (Mamun et al., [@CIT0017]; Scott, McGee, Wells & Oakley Browne, [@CIT0029]). A nationally representative survey in New Zealand (Scott et al., [@CIT0031]) found that obesity was significantly associated with both mood disorders and anxiety disorders. The specific disorder that was most strongly associated with obesity was PTSD. Furthermore, studies (Swinbourne & Touyz, [@CIT0032]) investigating the other end of the weight spectrum have shown that stressful life events are associated with anorexia nervosa. One study (Tagay, Schlegl & Senf, [@CIT0033]) found that 63% of the anorectic patients in an outpatient treatment sample had experienced at least one trauma in their life, and 10% fulfilled the criteria for a current diagnosis of PTSD. A Danish study found that women with a history of eating disorders had experienced significantly more childhood abuse than women with no prior history of eating disorder (Blaase & Elklit, [@CIT0002]). However, most studies on the relationship between weight and PTSD have focused on obesity or have focused solely on eating disorder patients (Perkonigg, Owashi, Stein, Kirschbaum, & Wittchen, [@CIT0022]; Scott et al., [@CIT0031]). This article is an attempt to extend the knowledge of factors involved in a potential unhealthy weight at both ends of the weight spectrum. Additionally, a number of studies have found that childhood trauma is associated with both being underweight (Reyes-Rodriguez et al., [@CIT0025]) and overweight (Mamun et al., [@CIT0017]; Midei, Matthews & Bromberger, [@CIT0019]; Pederson & Wilson, [@CIT0021]). Despite these consistent findings, the mechanism for the added risk of obesity following childhood trauma is unknown. A recent review (Gustafson & Sarwer, [@CIT0010]) suggested that disordered eating behavior, psychological, and/or physiological reactions to a traumatic experience may account for the added risk of adult obesity. For example, binge-eating behavior has been associated with childhood sexual abuse and an increased incidence of obesity (Gunstad, Schofield, et al., [@CIT0009]). Therefore, the objective of the present study was to compare some core PTSD symptoms in a group of normal weight individuals with a group of persons with underweight and overweight respectively. To our knowledge, there have been no previous attempts to include both underweight and overweight, respectively in a general population sample. Also, this study explored the association between different categories of childhood abuse and weight status. Methods {#S0001} ======= Procedure {#S20002} --------- A stratified random probability survey was conducted in Denmark by the Danish National Centre for Social Research between 2008 and 2009 in order to retrieve mental health related data from young Danish people. Statistics Denmark randomly selected 4,718 participants, aged 24 in 2008, from the total birth cohort of Denmark in 1984. To increase the number of participants who had experienced childhood abuse and neglect, children who had been in child protection were over-sampled by stratifying the number of "child protection cases" versus "non-child protection cases" (1/3:2/3) (see [Table 1](#T0001){ref-type="table"}). A child protection case was defined as a case where the council (according to the files of local social workers) had provided support for the child and the family or placement with a foster family due to concerns about the well-being and development of the child. Individuals who had previously refused to participate in national research, who were incarcerated, deemed to have learning difficulties and who had moved out of the country were excluded. Individuals who were deemed to be child-protection cases (as evidenced through the files of local social work departments) were oversampled. Structured interviews were conducted by trained interviewers either in the home or via the telephone. Due to the sensitive nature of a number of the questions, those respondents who were interviewed in their homes could respond by entering their answers into a laptop computer. The average duration of the interviews equated to 43 min. Participation was entirely voluntary and post-interview psychological help was offered to all respondents via a telephone help-line. The study was approved by the Danish Data Protection Agency. Further details pertaining to the procedure will be published elsewhere. ###### Distribution of child non-protection cases and child protection cases Gender Total sample weighted *N*=2,980 (%) Child non-protection sample *N*=2,128 (%) Child protection case *N*=852 (%) -------- ------------------------------------- ------------------------------------------- ----------------------------------- Male 1,555 (52.2) 1,106 (52.0) 473 (55.5) Female 1,425 (47.8) 1,022 (48.0) 379 (44.5) Participants {#S20003} ------------ A total of 2,981 interviews were successfully conducted, equating to a response rate of 67%. Eight hundred and fifty-two interviews were conducted with individuals who had been previously identified by the Danish authorities as child protection cases. The most common reasons for non-participation were: refusal to take part in the study (21%), being uncontactable (13%), and illness or disability (2%). Unfortunately, a non-response analysis was not possible. Fifty-two percent of the sample was men. All demographics were analyzed employing a weight variable to account for the oversampling of child protection cases. Child protection status (weighted) was given to 6% of the total sample. The data was also weighted for non-response and the proportions in socio- demographic factors found in the general population. Measures {#S20004} -------- The interview administered a series of questions pertaining to several psychological and physical domains. The response format was pre-coded; however, respondents were given the opportunity to elaborate if they wanted. PTSD symptomatology was measured through four questions reflecting some core symptoms of PTSD: avoidance, re-experience, numbing, and hypervigility. The four questions were:Have you had unpleasant experiences that you sometimes re-experience?Have you experienced an event that was so terrifying that you have tried not to think of the event and avoid things that remind you of the event during the last month?Have you experienced an event that was so terrifying that you have felt alert during the last month?Have you experienced an event that was so terrifying that you have felt numb and detached from other people during the last month? Because of the small numbers of individuals in some categories participants were categorized based on number of PTSD symptoms; (1) no symptoms, (2) 1--2 symptoms, or (3) 3--4 symptoms. Furthermore, because the number of traumatic experiences have been linked to adverse negative health outcomes (Kolassa, Kolassa, Ertl, Papassotiropoulos & De Quervain, [@CIT0015]), this was also included in the analyses. The participants were asked to point out which of 17 selected, potentially traumatizing events they had been exposed to (see [Table 2](#T0002){ref-type="table"}). The included events were selected from literature (Rhiger, Elklit & Lasgaard, [@CIT0026]) and clinical experience, covering a broad spectrum of possible life-threatening experiences and distressing conditions. Data from previously similar studies supports that the included events are frequently experienced by youths across nations and cultures, and that these events may be potentially traumatizing (Elklit & Petersen, [@CIT0007]). There exists conceptual overlap between a few of the potentially traumatizing events and the questions pertaining to child abuse experiences in this study. ###### Potentially traumatizing events ---- ---------------------------------- 1 Traffic accident 2 Fire 3 Accident 4 Physical assault 5 Threats of being beaten 6 Drowning 7 Robbery 8 Maltreatment 9 Humiliation 10 Rape 11 Suicide (self-attempted) 12 Suicide/suicide attempt (family) 13 Suicide/suicide attempt (friend) 14 Death (family) 15 Serious illness 16 Sexual abuse 17 Punishment ---- ---------------------------------- Weight status was measured as Body Mass Index (kg/m^2^) calculated from self-reported height and weight. BMI is regarded as an accurate method for comparing weight status (Deitel & Greenstein, [@CIT0006]). Participants were divided into four sub-groups according to their BMI: Underweight was defined as a BMI \< 18.5, normal weight as a BMI ≥18.5 and \<25, overweight as a BMI ≥25 and \<30, and obesity as a BMI ≥30. Childhood abuse was measured using a similar approach as May-Chacal and Cawson ([@CIT0018]). The present study asked participants questions regarding specific experiences of four types of maltreatment; physical abuse, sexual abuse, emotional maltreatment and physical neglect, a method not dependent of subjective interpretations of whether maltreatment has occurred or not. Childhood abuse categories were based on latent class analyses (further information will published elsewhere) and the respondents were divided into four distinct childhood abuse categories:Non-abused.Emotional abused.Sexual abused.Abused overall. The category of "abused overall" covers a distinct group of individuals who were exposed to all three types of abuse (i.e., physical abuse, emotional abuse, and neglect). Statistical analyses {#S20005} -------------------- Chi-square analyses were conducted to examine the relationship between (1) PTSD symptomatology and BMI categories, (2) PTSD symptomatology and child abuse categories, and (3) child abuse categories and BMI categories. A one way ANOVA analysis was used to examine the relationship between BMI categories and number of potential traumatic events. All analyses were conducted using IBMs SPSS statistics 19. Finally, logistic regression was performed to assess the impact of a number of factors on the likelihood that respondents would be in either the normal weight group or in the underweight, overweight or obese group, respectively. Results {#S0006} ======= Descriptives {#S20007} ------------ ### BMI The mean BMI was 23.84 (SD 3.81). Of the included participants 83 (3%) of the participants were underweight, 2,018 (68%) of the participants were normal weight, 650 (22%) of the participants were overweight, and the remaining 197 (7%) participants were obese. Information on BMI was lacking on 33 individuals (1%). ### PTSD The majority, 2,176 (73%) of the included participants, did not have any symptoms of PTSD. A total of 637 (21%) individuals reported having one or two symptoms of PTSD, and 168 individuals (6%) had three or four symptoms of PTSD. ### Childhood abuse Two hundred and sixty-three (9%) of the participants reported being emotionally abused during childhood, 59 individuals (2%) reported being sexually abused, and 64 (2%) of the participants had experienced overall abuse (e.g., physical abuse, emotional abuse, and neglect). Relationships between BMI categories and PTSD symptom categories {#S20011} ---------------------------------------------------------------- The chi-square for independence indicated a significant association between BMI categories and PTSD status (*χ* ^2^ *=*20.23, *p \<* .001; see [Table 3](#T0003){ref-type="table"}). As illustrated in [Fig. 1](#F0001){ref-type="fig"}, people who were underweight or overweight/obese reported more PTSD symptoms than respondents with a normal BMI. Furthermore, the percentage of persons with 3--4 symptoms of PTSD was higher in the population of underweight individuals compared to the group of overweight and obese individuals. {#F0001} ###### BMI and PTSD symptom categories Number of PTSD symptoms --------------- ------------------ ------------------------- ------ ------ Underweight *n* 50 21 12 \% of PTSD class 2.3 3.3 7.3 Normal weight *n* 1,503 417 98 \% of PTSD class 69.7 66.5 59.4 Overweight *n* 478 131 42 \% of PTSD class 22.2 20.9 25.5 Obese *n* 126 58 13 \% of PTSD class 5.8 9.3 7.9 Post hoc tests were performed comparing respondents in the normal weight group with the respondents in the underweight group, the overweight group, and the obese group independently. These analyses revealed that there was no statistically significant differences between the normal weight group and the overweight group in the number of reported PTSD symptoms (*χ* ^2^=.29, *p*=.287). The largest difference found in the analysis was between the underweight group and the normal weight group (*χ* ^2^ *=*17.41, *p \<* .002) followed by the difference between the obese group and the normal weight group (*χ* ^2^ *=*10.28, *p*\<.006). Relationships between BMI and total number of potential traumatic events {#S20012} ------------------------------------------------------------------------ A one-way ANOVA was conducted to compare the number of reported potentially traumatic events between the different BMI categories. The means and standard deviations are presented in [Table 4](#T0004){ref-type="table"}. As illustrated in [Fig. 2](#F0002){ref-type="fig"} the number of potential traumatic events were highest in the groups of overweight and obese individuals (*F* (3.2944) = 8.02, *p*\<.001). {#F0002} ###### BMI categories and number of events *N* Mean SD --------------- ------- ------ ------ Underweight 83 1.89 1.96 Normal weight 2,018 1.85 1.67 Overweight 650 2.12 1.85 Obese 197 2.40 1.88 Relationships between BMI, PTSD symptoms, and childhood trauma categories {#S20013} ------------------------------------------------------------------------- The chi-square for independence indicated a significant association between childhood trauma categories and PTSD symptoms (*χ* ^2^=309.20, *p*\<.001). Specifically, the categories of sexual abuse and overall abuse were associated with a higher number of PTSD symptoms. The chi-square for independence also indicated a significant association between childhood trauma categories and BMI categories (*χ* ^2^=20.36, *p*\<.016). Cross tabulations revealed that different categories of childhood abuse were related to underweight and overweight/obesity, respectively (see [Table 5](#T0005){ref-type="table"}). Childhood emotional abuse was especially associated with underweight, whereas sexual abuse and overall abuse were associated with overweight/obesity in specific. ###### BMI and childhood abuse categories Childhood abuse categories --------------- ----------------------- ---------------------------- ------ ------ ------- Underweight *n* 13 0 2 68 \% within abuse class 5.0 .0 3.2 2.6 Normal weight *n* 166 44 35 1,773 \% within abuse class 64.3 74.6 55.6 69.0 Overweight *n* 62 8 17 563 \% within abuse class 24.0 13.6 27.0 21.9 Obese *n* 17 7 9 164 \% within abuse class 6.6 11.9 14.3 6.4 Post hoc test was performed comparing respondents in the normal weight group with the respondents in the underweight group, the overweight group, and the obese group independently. These analyses revealed that the only statistically significant difference was between the obese group and the normal weight group in the type of childhood trauma reported (*χ* ^2^ *=*9.20, *p*\<.027). The result of the analysis of the underweight group versus normal weight group approached statistical significance (*p =* .057). Impact of a number of factors on weight status {#S20014} ---------------------------------------------- Logistic regression was performed to assess the impact of a number of factors on the likelihood that respondents would be in either the normal weight group or in the underweight, overweight or obese group, respectively. The model contained four independent variables (gender, number of PTSD symptoms, type of childhood trauma, and total number of potentially traumatizing events). ### Underweight versus normal weight The model reached statistical significance (*χ* ^2^(4, 2026) = 36.91, *p*\<.001) indicating that the model was able to distinguish between respondents who were underweight or normal weight. The model as a whole explained between 1.7% (Cox and Snell *R* ^2^) and 6.2% (Nagelkerke *R* ^2^) of the variance in BMI. As shown in [Table 6](#T0006){ref-type="table"}, only two of the independent variables made a unique statistically significant contribution to the model (gender \[being female\] and number of PTSD symptoms). ###### Logistic regression predicting likelihood of being underweight *p* Odds ratio -------------------------- -------- ------------ No. of PTSD symptoms .001 1.45 Total no. of events .212 .91 Gender \<.001 1.31 Type of childhood trauma .319 .89 Constant \<.001 .02 ### Overweight versus normal weight The model reached statistical significance (*χ* ^2^(4, 2604) = 44.25, *p*\<.001) indicating that the model was able to distinguish between respondents who were underweight or normal weight. The model as a whole explained between 1.6% (Cox and Snell *R* ^2^) and 2.5% (Nagelkerke *R* ^2^) of the variance in BMI. As shown in [Table 7](#T0007){ref-type="table"}, only two of the independent variables made a unique statistically significant contribution to the model (gender \[being male\] and total number of potentially traumatizing events). ###### Logistic regression predicting likelihood of being overweight *p* Odds ratio -------------------------- -------- ------------ No. of PTSD symptoms .416 1.05 Total no. of events .010 1.08 Gender \<.001 .88 Type of childhood trauma .951 1.0 Constant \<.001 .40 ### Obesity versus normal weight The model reached statistical significance (*χ* ^2^(4, 2173) = 17.81, *p \<* .001) indicating that the model was able to distinguish between respondents who were obese or normal weight. The model as a whole explained between 0.8% (Cox and Snell *R2*) and 1.8% (Nagelkerke *R* ^2^) of the variance in BMI. As shown in [Table 8](#T0008){ref-type="table"}, only the total number of potentially traumatizing events made a unique statistically significant contribution to the model. ###### Logistic regression predicting likelihood of being obese *p* Odds ratio -------------------------- -------- ------------ No. of PTSD symptoms .192 1.12 Total no. of events .001 1.15 Gender .831 .99 Type of childhood trauma .729 1.03 Constant \<.001 .06 Discussion {#S0018} ========== The specific aims of the current study were two-fold. First, we wanted to determine if there was an association between BMI and PTSD symptomatology in a general population based survey of Danish young adults. Secondly we wanted to explore the association between different categories of childhood abuse and BMI status. In relation to the first objective of this study, results supported an association between PTSD symptomatology and BMI status. Both underweight and obese participants reported significantly more PTSD symptoms than participants within a normal weight range. Participants with PTSD symptoms were more than twice as likely to be underweight than participants in the general sample. Likewise, respondents with PTSD symptoms were more prevalent among obese individuals than among participants in the normal weight range. Post hoc analyses revealed that the association between PTSD symptoms and obesity was less pronounced as the association between PTSD symptoms and underweight. Furthermore, PTSD symptoms were not significantly associated with overweight but only with obesity. In relation to the second objective of this study, results indicated a significant association between childhood abuse and BMI. Post hoc analyses revealed that the only statistical significant difference was between the obese group and the normal weight group in the type of childhood trauma reported, whereas the result of the analysis of the underweight group versus normal weight group approached statistical significance. Recent research suggests that obesity and eating disorders such as Anorexia Nervosa may have both shared and distinct susceptibility factors (Day, Ternouth & Collier, [@CIT0005]). Although Anorexia Nervosa is not necessarily equal to the concept of underweight, this study indicates that PTSD symptomatology and exposure to childhood abuse may be shared susceptibility factors for developing both underweight and overweight. It has been suggested that childhood abuse may interfere with mechanisms of emotional regulation (Wonderlich et al., [@CIT0036]). Emotional dysregulation has been associated with both traditional eating disorders and obesity. Wonderlich et al. ([@CIT0036]) suggest that restrictive eating or binge-purge behavior may reflect instrumental behaviors aimed at regulating negative affect. Similarly, emotional eating has been associated with obesity, and it is described as the tendency to eat in an effort of regulating negative emotions. Research shows that obese individuals engage in significantly more emotional eating compared to non-obese people (Pinaquy, Chabrol, Simon, Louvet & Barbe, [@CIT0023]). Other studies have focused on other risk factors involved in both under- and overweight. These factors include various physiological factors such as similar genetic dispositions (Gunstad, Paul et al., [@CIT0008]) and changes in the serotonin system, which have been linked to PTSD (Bremner, Southwick & Charney, [@CIT0004]) and to both overeating and a drive for thinness (Akkermann et al., [@CIT0001]). Finally, changes in the HPA axis and thereby the secretion of cortisol have been linked to both PTSD (Nemeroff et al., [@CIT0020]) and the metabolic syndrome which is closely associated to overweight (Rosmond, [@CIT0028]). However, research (Day et al., [@CIT0005]) also indicates that other risk factors affect whether a person becomes underweight or overweight after experiencing a shared risk factor. The present study indicates that type of childhood abuse may act as a separating susceptibility factor. The reporting of a history of childhood emotional abuse was associated with underweight, whereas reporting a history of childhood sexual abuse and overall abuse were associated with overweight. These results are consistent with other research supporting that different forms of child maltreatment may bear specific relationships to different health behaviors (Wonderlich et al., [@CIT0036]). Although research on eating disorders are not directly transferable to underweight Wonderlich et al. ([@CIT0036]) have found that emotional abuse was particularly associated with bulimia nervosa and anorexia nervosa, and suggests that emotional abuse may influence risk factors that are especially relevant for those disorders. Furthermore, our study indicates that sexual abuse and overall abuse may influence risk factors of special relevance for the development of obesity. Authors have suggested that obesity might be a defense mechanism for victims of sexual abuse, seeking to minimize the risk of future abuse by the opposite sex (Ray, Nickels, Sayeed & Sax, [@CIT0024]). More research is needed in order to fully understand the effect of distinct childhood abuse categories on the development of underweight and overweight. Furthermore, the performing of logistic regression analyses in this study indicated that PTSD symptoms were a significant predictor of underweight whereas the exposure of several traumatic events was a significant predictor of obesity. These findings may indicate that PTSD and the amount of traumatic experiences may be important factors in the understanding of the association between trauma and BMI. Prospective studies are needed in the future to understand the causal connections between childhood abuse, PTSD and BMI. The findings of the present study must be interpreted in the context of several limitations. A limitation pertaining to this study is inherent in the research concerned with childhood maltreatment. Different definitions of what constitutes childhood abuse exist, and there appears to be no standardized measure of childhood maltreatment. Therefore, comparisons of results between studies are fraught with difficulty. Furthermore, the current study did not assess the frequency, or duration of abuse; rather participants were asked to endorse an experience by responding yes or no. Other studies have shown that severity of childhood abuse affect the magnitude of psychological disorders (Wonderlich et al., [@CIT0036]). In addition, the current study relies on the self-reporting of abuse. Another limitation in the present study is the measure of PTSD symptomatology. The study did not include a complete PTSD measurement but consisted of only four PTSD symptoms which limit the possibility of comparing the results of our study with findings from other studies. Furthermore, using a measure of only four PTSD symptoms is a very brief measure and provides only an estimate of PTSD symptomatology. However, the four symptoms included in this study are representative of the three PTSD clusters---re-experiencing, avoidance, and hypervigility---and are considered central to the PTSD diagnosis. Earlier studies with brief measures of PTSD like SPAN (4 items; Sijbrandij, Olff, Opmeer, Carlier & Gersons, [@CIT0031]) and PTSD-8 (8 items; Hansen et al., [@CIT0011]) have shown that using core symptoms as a screening measure accurately predict trauma survivors at risk for developing PTSD. A third limitation concerns the limited age-span. The participants in this study were 24 years of age, and therefore the results of the present study might not be representative of other age groups. Additionally, BMI was calculated from self-reported height and weight. Studies have shown that overweight individuals have a tendency to underreport weight (Roberts, [@CIT0027]). Finally, the cross-sectional design limits the ability to examine causal relationships between the investigated variables. In the context of these limitations, this study provides valuable knowledge of possible associations between PTSD and BMI, and childhood abuse and BMI. To our knowledge, this is the first study that has included individuals with both an underweight and overweight BMI in relation to PTSD. Research in this area seems only present in relation to clinical populations, i.e., anorexia nervosa and bulimia nervosa or in studies on the association between psychological distress and weight not including a PTSD measure (Hwang, Lee, Kim, Chung & Kim, [@CIT0014]; Zhao et al., [@CIT0037]). The results of this study seem to support findings from this research and extend earlier findings to a specific birth cohort in the general population. Conclusion {#S0019} ========== The serious health risks associated with an unhealthy body weight demand attention to the potential risk factors for an unhealthy weight status. Therefore, this study set out to explore potential relationships between PTSD, childhood abuse and weight status in a national sample. This study indicates that PTSD symptomatology and exposure to childhood abuse may be shared susceptibility factors for developing both underweight and obesity. Furthermore, different types of childhood abuse categories may be particularly relevant for underweight or obesity, respectively. These findings indicate that health care professionals may profit from assessing PTSD in the treatment of both obese and underweight individuals. Prospective studies are needed in order to fully understand the causal relationship between PTSD, childhood abuse and unhealthy body weight. Conflict of interest and funding {#S0020} ================================ There is no conflict of interest in the present study for any of the authors.

Introduction ============ Giant cell tumour of bone (GCTOB) is the prototype of giant cell rich neoplasms of the skeleton. The term giant cell tumour was coined by Bloodgood in 1912 \[[@B1]\] and it was not until 1940 that Jaffe distinguished giant cell tumour of bone from other bone tumours containing many osteoclast-like giant cells \[[@B2]\]. This lesion represents 4% to 5% of all primary bone tumors and mainly occurs in skeletally mature patients (peak incidence between ages 20 and 45 years) with a slight female predominance \[[@B3]-[@B5]\]. It most commonly arises at the epiphyses of long bones like the distal femur, proximal tibia, distal radius and proximal humerus \[[@B6]\]. This tumor can be locally aggressive with a tendency for recurrence. Lung metastases occur infrequently; more rarely, this tumor behaves as a sarcoma \[[@B4],[@B7]\]. Because of its different evolution and prognosis, GCTOB must be distinguished from other multinucleated giant cell-containing tumors and pseudotumors. Differential diagnosis can be challenging, particularly in instances of limited sampling such as with needle-core biopsies. It is based not only on histology, but also on clinical and radiological data. There is currently no well-accepted diagnosis marker available for GCTOB, but recent studies using immunohistochemistry and molecular methods have demonstrated overexpression of p63 in the stromal cells of most giant cell tumors of bone and advocate its use as a diagnostic marker \[[@B3],[@B4],[@B6]\]. P63 was identified in 1998 \[[@B8]\]. It belongs to the family of transcription factors that also includes p53 and p73 \[[@B9]\]. It is mostly used as a diagnostic aid in breast, prostate, and salivary gland cancer because of its high sensitivity and specificity for mammary and salivary myoepithelial cells and prostatic basal cells \[[@B3],[@B10]-[@B12]\]. It can be a useful tool in distinguishing urothélial carcinoma from prostatic carcinoma \[[@B13]\] and it can also be used as a prognosis factor as in adenoid cystic carcinoma \[[@B14]\]. The purpose of this study is to determine whether GCTOB expresses p63, and whether p63 can be used as a biomarker to discriminate GCTOB from other giant cell-rich tumors. Methods ======= This study concerns 48 giant cell-containing tumors and pseudotumors of bone that were retrieved from department of pathology of Hassan II University Hospital in Fez, from January 2009 to February 2012. They include 12 osteosarcomas, 8 osteoblastomas, 5 GCTOB (Figure [1](#F1){ref-type="fig"}), 5 aneurysmal bone cysts (ABCs) (Figure [2](#F2){ref-type="fig"}), 4 osteoid osteomas (OO), 4 central giant cell granulomas (CGCGs) (Figure [3](#F3){ref-type="fig"}), 4 non ossifiant fibromas (NOFs), 3 chondromyxoid fibromas (CMFs), 1 fibrous dysplasia (FD), 1 chondroblastoma and 1 Langerhans cell histiocytosis (LCH). The data were collected prospectively from pathology reports, from forms filled by trauma surgeons, pediatric surgeons and otorhinolaryngologists, and from radiographs. A form was filled for each patient, including the following informations: patient's name, age, sex, tumor location, histological type and P63 expression. The demographic data and location of these cases are shown in Table [1](#T1){ref-type="table"}. {#F1} {#F2} {#F3} ###### Demographic data and location of tumours **Case number** **Diagnosis** **Age** **Sex** **Location** ----------------- ----------------- --------- --------- ----------------------- 1 *osteosarcoma* 34 F femur 2 *osteosarcoma* 40 F femur 3 *osteosarcoma* 19 M femur 4 *osteosarcoma* 19 M femur 5 *osteosarcoma* 12 F femur 6 *osteosarcoma* 14 F femur 7 *osteosarcoma* 23 M humerus 8 *osteosarcoma* 21 F humerus 9 *osteosarcoma* 20 M humerus 10 *osteosarcoma* 25 M Tibia 11 *osteosarcoma* 18 F Tibia 12 *osteosarcoma* 27 M mandible 13 osteoblastoma 14 M femur 14 osteoblastoma 19 F femur 15 osteoblastoma 20 F radius 16 osteoblastoma 25 F Cuneiform bone 17 osteoblastoma 23 M 5th metatarsal bone 18 osteoblastoma 23 M astragal 19 osteoblastoma 21 M mandible 20 osteoblastoma 57 M vertebrae 21 GCTOB 29 F femur 22 GCTOB 30 M femur 23 GCTOB 21 F 5th metacarpal bone 24 GCTOB 21 F Fibula 25 GCTOB 40 F Tibia 26 ABC 19 F Fibula 27 ABC 14 F Fibula 28 ABC 16 M Femur 29 ABC 13 F Tibia 30 ABC 15 M 1^st^ metatarsal bone 31 OO 30 F femur 32 OO 28 M femur 33 OO 24 M astragal 34 OO 24 M Fibula 35 CGCG 9 M mandible 36 CGCG 15 M mandible 37 CGCG 48 F Maxilla 38 CGCG 53 F Maxilla 39 NOF 16 F Tibia 40 NOF 19 M Tibia 41 NOF 16 F Tibia 42 NOF 8 M femur 43 CMF 23 M Toe 44 CMF 33 M Tibia 45 CMF 59 F Sphenoid bone 46 FD 19 M Femur 47 chondroblastoma 23 M calcaneus 48 LCH 7 M Ilium M: male; F: female. All specimens were fixed in 10% buffered formalin, embedded in paraffin and 4 micron-thick sections were stained with hematoxylin and eosin for routine histological examination. Immunohistochemical staining ---------------------------- P63 expression was evaluated by immunohistochemistry. All immunohistochemical stains were performed on a Ventana Benchmark LT automated immunostainer, on 3 micron-thick sections that were incubated with a mouse monoclonal antibody against p63 (clones 463M-17, prediluated, ready to use, Cell Marque Datasheet). The stained slides were examined without knowing the original histologic diagnosis. As there is no consensual scoring, we evaluated intensity of staining as weak (1+), moderate (2+), and strong (3+), and percentage of staining cells. A case was considered positive when nuclear staining of a single lesional cell or more was found. Statistical analysis -------------------- The calculation of average age, median age, sex ratio and rate of P63 expression was done using Epi-info software. Sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated in the GCTOB vs. not GCTOB and P63 positive vs. P63 negative groups using the SPSS software package (version 17). Results ======= The patients' age ranged between 7 and 59 years with an average of 23.8 years and a median of 21 years. A discrete male predominance was noted (sex ratio = 1.2). Immunohistochemical analysis showed a P63 nuclear expression in all GCTOB (Figure [4](#F4){ref-type="fig"}), 2 of 4 osteoid osteomas (50%), 2 of 5 ABCs (40%) (Figure [5](#F5){ref-type="fig"}), 3 of 8 osteoblastomas (37.5%), 1 of 3 CMFs (33.3%), 1 of 4 NOFs (25%), 1 of 12 osteosarcomas (8.3%) and in the single case of chondroblastoma included in this series. The staining was observed only in the nucleus of the mononuclear cells and no staining was present in the multinucleated giant cells. No P63 immunoreactivity was detected in any of the cases of CGCG (Figure [6](#F6){ref-type="fig"}), LCH, and FD. Strong staining was seen in 40% of GCTOB (2 cases) and in one case of osteoblastomas (33.3% of P63 positive osteoblastomas). Moderate staining was seen in 2 cases of GCTOBs (40%) and in one case of ABCs. In other tumors expressing P63, staining intensity was weak. Staining was seen in 30%-60% of tumor cells in GCTOB and in 20% and 50% of tumor cells in ABCs. In other tumors, percentage of reactive cells was lower (5%-30% in osteoblastomas, 10% in osteoid osteomas, osteosarcomas and CMFs, and 5% in chondroblastoma and NOFs). {#F4} {#F5} {#F6} The sensitivity and negative predictive value (NPV) of P63 immunohistochemistry for the diagnosis of GCTOB were 100%. The specificity and positive predictive value (PPV) were 74.42% and 59.26% respectively. Discussion ========== In this study, we showed that all GCTOBs express P63. Dickson \[[@B4]\] and Linden \[[@B15]\] found similar results by immunohistochemistry. They reported P63 overexpression in all GCTOB. In De la Rosa's study \[[@B3]\], P63 immunoreactivity was seen in 20 of 23 GCTOBs (86.9%). Similar results were reported by Lee \[[@B6]\] who showed a P63 overexpression by immunohistochemistry in 81% of cases (n=26) with a strong staining in 69% (Table [2](#T2){ref-type="table"}). The immunostaining was mostly confined to the mononuclear component \[[@B3],[@B4],[@B6]\]. This strong expression of P63 suggests that this protein may be implicated in the pathogenesis of GCTOB but determining its exact role requires further investigation. ###### P63 expression of in current series and in other published series **Diagnosis** **Our series** **Dickson's series \[**\[[@B4]\]**\]** **De la Rosa's series \[**\[[@B3]\]**\]** **Lee's series \[**\[[@B6]\]**\]** ----------------- ---------------- ---------------------------------------- ------------------------------------------- ------------------------------------ ---- ------- ---- ------- GCTOB 5 100% 17 100% 23 86.9% 6 81% Osteosarcoma 12 8.3% 0 \- 4 50% 13 15.3% Osteoblastoma 8 37.5% 0 \- 0 \- 0 \- ABC 5 40% 7 28.6% 8 62.5% 25 20% CGCG 4 0% 12 0% 4 100% 12 0% NOF 4 25% 0 \- 6 16.6% 0 \- OO 4 50% 0 \- 0 \- 0 \- CMF 3 33.4% 0 \- 0 \- 12 0% Chondroblastoma 1 100% 10 30% 12 83.3% 15 40% FD 1 0% 0 \- 2 0% 4 0% LCH 1 0% 0 \- 0 \- 0 \- chondrosarcoma 0 \- 0 \- 0 \- 0 \- Brown tumor 0 \- 0 \- 0 \- 4 0% The relationship of GCTOB and central giant cell granuloma has long been controversial. The absence of p63 expression in CGCG suggests that these tumors may have a pathogenesis that differs from that of GCTOB. P63 negativity found in all cases of CGCG in our study is consistent with results obtained by Dickson \[[@B4]\] and Lee \[[@B6]\] who found negativity in all cases (n=12 in each series). De la Rosa \[[@B3]\] showed different results with p63 positivity in all cases (n = 4) (Table [2](#T2){ref-type="table"}). Only one case (8.3%) of osteosarcomas included in our study showed overexpression of P63. Proportion of immunoreactive cells was less than 10% and staining was 1+ in intensity. The rate of P63 expression in other series remains low (2 cases/13 in Lee's study, with low intensity \[[@B6]\], and 2 cases/4 in De la Rosa's study \[[@B3]\]) (Table [2](#T2){ref-type="table"}). In this work, we recorded a single case of chondroblastoma. The immunohistochemical study showed P63 expression by less than 10% of tumor cells with low intensity. The rate of expression in other studies is variable. In Dickson's study, 3 of 10 chondroblastomas expressed p63 (30.0%); this ranged from 7--75% of cells, and staining was predominantly mild-moderate in intensity \[[@B4]\]. De Larosa found a higher expression (83.3%, 10 of 12 chondroblastomas) with moderate staining in 6 cases, weak staining in 3 cases and strong staining in only one case \[[@B3]\]. Lee showed P63 staining in 40% of cases (6 of 15). To differentiate between chondroblastoma expressing P63 and GCTOB, he used PS100: chondroblastoma shows positive S-100 immunostaining whereas only occasional weak S-100 immunostaining is seen in GCTOB \[[@B6]\]. In the same study, no P63 staining was seen in chondromyxoïd fibromas (n=12) (Table [2](#T2){ref-type="table"}). The rate of P63 expression in ABC in Dickson's \[[@B4]\] and Lee's \[[@B6]\] studies is lower than that obtained in our study: 28.6% (2 cases/7) and 20% (5cases/25) respectively. De la Rosa \[[@B3]\] and Linden \[[@B15]\] found higher results: 62.5% and 100% respectively (Table [2](#T2){ref-type="table"}). If some cases of ABC are P63 +, they could be a component of a GCTOB. In fibrous dysplasia, our results are concordant with those found by De La Rosa \[[@B3]\] (two cases all negative) and Lee (4 cases all negative) \[[@B6]\]. Non ossifiant fibroma showed P63 expression in one case with weak and focal staining. De la Rosa found similar results with P63 expression in 1 of 6 cases (16.6%) (Table [2](#T2){ref-type="table"}). Proportion of positive cells was less than 10% and staining intensity was weak \[[@B3]\]. In current study, 50% of osteoid osteomas and 37.5% of osteoblastomas expressed P63. LCH showed no P63 immunostaining. These tumors were not included in the other studies. P63 contribution in the differential diagnosis between GCTOB and other multinucleated giant cell-containing lesions of bone is variable. Dickson \[[@B4]\] considers that P63 can be useful as a biomarker for the differential diagnosis between GCTOB and other lesions particularly central giant cell granuloma, since these do not express P63. De La Rosa \[[@B3]\] found a high P63 negative predictive value (91.17%) but a low specificity (53.36%) which limits the use of this protein as an immunohistochemical marker for differential diagnosis. Lee \[[@B6]\] considers that the use of P63 can help in histological diagnosis of GCTOB. In current study, the P63 negative predictive value is 100%, this means that in difficult cases, P63 negativity can eliminate a GCTOB. The positive predictive value is low (59.26%). However, except a case of osteoblastoma, a strong staining was found only in GCTOB. Therefore, it is strongly suggestive of this tumor. Conclusion ========== This study shows that P63 may serve as a biomarker for the differential diagnosis between GCTOB and other morphologically similar lesions, particularly CGCG since the latter does not express P63. Other giant cell-containing lesions express P63, decreasing its specificity as a diagnostic marker, but a strong staining was seen, except a case of chondroblastoma, only in GCTOB. Abbreviations ============= ABC: Aneurysmal bone cyst; CGCG: Central giant cell granuloma; CMF: Chondromyxoid fibroma; FD: Fibrous dysplasia; GCTOB: Giant cell tumor of bone; LCH: Langerhans cell histiocytosis; NOF: Non ossifiant fibroma; OO: Osteoid osteomas. Competing interests =================== The authors declare that they have no competing interests. Authors' contributions ====================== NH, LC, and AA performed the histological examination of bone lesions and were major contributors to writing the manuscript. HE and TH assisted in histological interpretation. YA and ST performed the radiological examination. All authors read and approved the final manuscript. Acknowledgements ================ The authors thank Batoul Ech-chahdi which participated in the English translation of the manuscript.

Background ========== Introduction ------------ Plant breeding is making increasing use of molecular markers to speed up selection cycles and thereby more rapidly introgress interesting zones of the genome in varieties of agronomic interest. In addition to QTL analyses, which only focus on a single progeny, association genetics can be used to focus on a set of populations to identify zones of interest of the genome in breeding. However, for association studies to be successful it is necessary to know the degree of linkage disequilibrium (LD) beforehand, along with its extent and its distribution within the species studied. This disequilibrium, which measures the intensity of linkage between markers, is used to determine the molecular marker density of the genome, which is necessary for association studies to be effective \[[@B1]-[@B3]\]. Association studies are particularly relevant in perennial species, as they can be carried out on pre-existing populations, in collections or in selection trials. They do not require the creation of specific populations by controlled crossing, as is the case with conventional genetic mapping approaches \[[@B4]\]. Association studies have proven effective in a large number of plant species, including maize \[[@B5]\] and grapevine \[[@B6]\]. A prior study of the LD in a species guides the choice of one of the two association study possibilities: genome-wide scan or candidate gene approach. No large-scale LD study has been undertaken yet on the coffee tree. *Coffea canephora* Pierre ex A. Froehner (2n = 2X = 22) is a strictly outcrossing diploid species with a genetic self-incompatibility system \[[@B7],[@B8]\]. It provides 41.3% of the world coffee production \[[@B9]\]. One of the most ambitious genetic improvement programs for this species has been conducted in the Ivory Coast and is based on a reciprocal recurrent selection scheme (RRS) that was launched in the 1990s \[[@B10],[@B11]\]. This selection program uses the genetic diversity of the species by creating hybrids between the genotypes of two genetic groups: Congolese from central Africa and Guinean from western Africa. The combined use of the RRS and association genetics could help to more effectively guide crosses and speed up the introgression of specific alleles identified as being of interest through the early selection of genotypes derived from crosses. However, natural populations of *C. canephora* are relatively small in size. The strictly outcrossing reproduction system of this species and the different levels of existing kinship create complex genetic structures at the population scale. This structure in populations is superimposed on the larger diversity group (DG) scale \[[@B12]\]. Our work, therefore, consisted in assessing the LD within the Guinean and Congolese DGs of *C. canephora* as identified using molecular markers \[[@B12]-[@B14]\]. The expected results were as follows: i) knowledge of the LD dynamics at the genome level for a certain number of coffee diversity groups or populations, ii) enhanced knowledge of the genetic structure of the Guinean diversity group, and iii) identification of populations that can be used in association genetics. Methods ======= Plant material -------------- We studied 356 genotypes of *C. canephora* divided into five DGs based on the diversity analyses carried out in previous studies \[[@B12],[@B13]\] (see Table [1](#T1){ref-type="table"}): DG G, Guinean diversity group: a mixture of different wild or cultivated genotypes collected in the Ivory Coast and Guinea. It includes the populations Mouniandougou, Ira 1, Ira 2, Fourougbankoro, Piné, and cultivated Guinean clones DG GP, Pélézi diversity group: a natural Guinean population related to the Guinean diversity group, isolated from a forest of the Ivory Coast DG C, Nana diversity group: a population collected in the Central African Republic (CAR) DG SG2: genotypes cultivated in the Ivory Coast and Uganda or in collections in Brazil, most likely originating from the Congo basin DG SG1: genotypes cultivated in Togo and Benin (Niaouli population) or from surveys in the Luki collection (Democratic Republic of Congo, Luki population), originating from the Atlantic seaboard, from Gabon to the Democratic Republic of Congo ###### **Characteristics and origins of the*C****.****canephora*diversity groups** **Name** **Origin** **Diversity group** **Size** **Type** ---------- ----------------------------------------------------------------- --------------------- ---------- -------------------------- Nana Central African Republic C 92 Spontaneous Pélézi Ivory Coast GP 35 Spontaneous Guinean Ivory Coast -- Guinea G 128 Cultivated & spontaneous SG1 Bulk Atlantic Coast (Gabon -- Congo -- Democratic Republic of Congo) SG1 16 Cultivated & spontaneous SG2 Bulk Congo basin SG2 85 Cultivated Choice of microsatellite markers and genotyping ----------------------------------------------- All individuals were genotyped using 108 microsatellite markers mapped to eleven linkage groups (LG, called A to K) on a *C. canephora*\[[@B15]\] linkage map that spanned a length of 1320 cM (Figure [1](#F1){ref-type="fig"}). The markers are described in Additional file [1](#S1){ref-type="supplementary-material"}: Table S1. {#F1} All LGs were studied with a larger number of markers on LGs A, B, D, F, G and H. The average distance between markers was 13 cM, ranging from 0 cM for the closest marker pairs to 243.3 cM for the most distant marker pairs. The genotyping was performed according to the protocol described in \[[@B16]\]. Size controls were replicated on the different gels to ensure the uniformity of genotyping data. Data were imported into the Microsoft EXCEL® spreadsheet from SAGA GT® (LI-COR Biosciences, Lincoln, Nebraska, USA) and were formatted for the different data analysis software used. Genetic structure validation of the sample ------------------------------------------ A model-based Bayesian analysis implemented in STRUCTURE \[[@B17],[@B18]\] was performed to validate the DG structure of our sample. We ran a correlated-allele model with 10,000 iterations and 10,000 burn-ins; ten runs were made for each assumed K (putative number of populations), with K varying from one to ten. The ad-hoc statistics Δ(K) proposed by \[[@B19]\] were used to assess the number of populations. The Hardy-Weinberg Equilibrium (HWE) and summary statistics, including the number of alleles and the expected and observed heterozygosity for each locus, were computed for the entire sample and for each DG using ARLEQUIN software v. 3.5.1.2 \[[@B20]\]. We also computed the AMOVA and F-statistics for diverse levels of population structure using ARLEQUIN. Genetic structure of diversity group G -------------------------------------- The Guinean diversity group is composed of a large number of natural populations from the forests of the Ivory Coast, including the Pélézi population (DG GP, see above), which exhibits some original characteristics. The wide variety of these populations, as well as the existence of Guinean genotypes taken from smallholder plantations in the living collections in the Ivory Coast, further justify an in-depth study of this diversity group. Earlier diversity studies did not offer sufficient resolution to analyze the genetic structure of this group in a satisfactory manner (lack of markers). The genetic structure of this group was analyzed using DARwin software based on the calculation of genetic dissimilarities between individuals followed by a factorial analysis from a dissimilarity matrix (FADM) and the construction of a neighbor-joining tree (NJ) \[[@B21],[@B22]\]. For comparison purposes, we ran a STRUCTURE analysis within the DG G using the same parameters and procedure as above. Statistical data analysis and LD analysis ----------------------------------------- ### Reconstruction of haplotypes The species *C. canephora* is a highly heterozygous species. Therefore, it is difficult to distinguish the allele phase of the double heterozygotes Aa/Bb, i.e., whether A is associated with B or with b at the haplotype level \[[@B23]\]. However, the most common and powerful measurements of the LD (D, D', r^2^) rely on an estimation of the haplotypic or gametic linkage disequilibrium \[[@B24]\], i.e., using the allele phase information at the gamete level. Consequently, to estimate these measurements, it is necessary to have access to the haplotypes. Therefore, we used PHASE software \[[@B25]-[@B27]\] to reconstruct the haplotypes. This software estimates the most probable haplotypes for each genotype based on an EM algorithm (Expectation-Maximization) that incorporates a coalescence hypothesis in a maximum likelihood model. The haplotypes are reconstructed following a certain number of strong hypotheses using parameters such as allele frequencies, the possibilities of recombination between markers and simulated allele pedigrees. We used the entire set of 356 genotypes, as the algorithm functions better with genetically structured data \[[@B27]\]. Nevertheless, the dataset was partitioned by the LG of the genetic map (11 matrices in all), as the markers situated on different LG could not be in phase. This approach has been shown to be effective, and the gain in power has been proven, as in the case of grapevine \[[@B28],[@B29]\]. To enable the use of PHASE while maintaining the Stepwise Mutation hypothesis for the microsatellite markers, our data that were expressed in allele sizes were converted into repeat numbers using CREATE software \[[@B30]\]. Five algorithm repeats based on 1000 iterations, 100 thinning intervals and 1000 burn-ins were performed, and the repeat displaying the greatest maximum likelihood was used for the remaining analyses. The resulting tables adopted for each LG were then merged and formatted for incorporation into PowerMarker to analyze the LD, declaring genotype data of the known phase as the data type. ### LD analysis The LD was analyzed for the five DGs, together and separately. We calculated the *D'* and r^2^ values for the set of possible combinations of markers, two-by-two, using PowerMarker software \[[@B31]\]. These measurements were initially developed for bi-allelic loci. Nevertheless, an estimation of these measurements for multi-allelic loci can be performed by establishing a weighted mean for the set of disequilibria between allele pairs \[[@B32],[@B33]\]. Exact Fisher tests were carried out for all the possible combinations to test whether the haplotype frequencies between two loci were the product of the allele frequency corresponding to the two loci. Allele counting was organized in a contingency table, and permutations, following an algorithm using a Markov-Monte-Carlo chain, were used to calculate the unbiased p-values associated with the test \[[@B33]\]. We corrected the significance limit of the p-values associated with the exact test using Bonferroni's correction to more effectively overcome the effect of the very large number of tests performed. We graphically represented the r^2^ values as a function of distance for each of the groups studied. Results ======= Genetic structure of the entire sample -------------------------------------- The ad-hoc statistics Δ(K) indicate an uppermost level of structure in five populations. The Δ(K) and bar-plot from the STRUCTURE run showing the highest lnP(D) for K=5 are presented in Figure [2](#F2){ref-type="fig"}. The expected and observed heterozygosity and Hardy-Weinberg exact test associated p-values for each locus are provided in Additional file [2](#S2){ref-type="supplementary-material"}: Table S2. At the level of the entire population, all HWE departure tests are significant, indicating a high level of structure. Even at the population level (GP), we still detect a significant departure from HWE for a high proportion (56%) of the polymorphic loci. The AMOVA and F-statistics also indicate a high structure in our sample, with more than 38% of the variance due to the DG and with all the *Fst* between DG significant at a 5% level. {#F2} ### Genetic structure of diversity group G The ad-hoc statistics Δ(K) based on the STRUCTURE analysis gave an uppermost level of structure within DG G in three populations. The Δ(K) and bar-plot from the STRUCTURE run showing the highest lnP(D) for K=3 are presented in Figure [3](#F3){ref-type="fig"}. The factorial analysis based on dissimilarity index and the NJ tree are shown in Figure [4](#F4){ref-type="fig"}. Both analyses revealed a marginal structure in three clusters. On the tree, these three clusters corresponded to the Mouniandougou population plus a few individuals from the Piné and cultivated Guinean populations (Gsub1); to the majority of the Ira2 population (Gsub2); and to the Fourougbankoro and Ira1 populations plus a few cultivated Guinean and Piné individuals (Gsub3). For the rest of our work, we considered DG G to be a composite group of different populations, and we also tried to analyze separately the three identified subgroups. The expected and observed heterozygosity and Hardy-Weinberg exact test associated p-values for each locus are provided in Additional file [3](#S3){ref-type="supplementary-material"}: Table S3. The AMOVA analysis and F-statistics of the entire sample, including Gsub1, Gsub2 and Gsub3, are presented in Additional file [4](#S4){ref-type="supplementary-material"}: Table S4. {#F3} {#F4} ### LD analysis at the pan-genomic level #### LD estimates and their within- and between-linkage group significance by exact Fisher tests and Bonferroni's correction Table [2](#T2){ref-type="table"} presents a summary of the exact tests carried out on the different clusters considered. These results demonstrate the importance of the significant associations generated by the genetic structure. In fact, for the set of genotypes, it can be seen that 98% of the marker pairs displayed a significant disequilibrium, regardless of whether they were linked. In addition, DG C and DG GP and the three subgroups of DG G were the only ones to show a within-group: between-group ratio over 1. When trying to consider a finer structure, a certain number of significant between-group associations were no longer detected. For DG G, we found a considerable correction of genomic LD for the set of subgroups (Gsub1, Gsub2 and Gsub3) in relation to DG composite G. ###### Analyses of associations between markers per exact Fisher test for the different genetic groups considered **Diversity group** **Number of individuals** **Number of polymorphic markers** **Number of 2-by-2 tests** **Limit(5%)after Bonferroni's correction** **Number of significant associations(exact test)** **Number of significant within-linkage group associations and percentage compared to the total significant associations** **Number of significant between**-**linkage group associations and percentage compared to the total significant associations** **Within**-**linkage group:Between-linkage group ratio** --------------------- --------------------------- ----------------------------------- ---------------------------- -------------------------------------------- ---------------------------------------------------- --------------------------------------------------------------------------------------------------------------------------- -------------------------------------------------------------------------------------------------------------------------------- ---------------------------------------------------------- ------ ------------- ------ All genotypes 356 108 5778 8.65351E-06 5684 98% 680 ***12****%* 5004 ***88****%* 0.14 C 92 98 4753 1.05197E-05 72 2% 62 ***86****%* 10 ***14****%* 6.20 SG1 16 93 4278 1.16877E-05 177 4% 29 ***16****%* 148 ***84****%* 0.20 SG2 85 107 5671 8.81679E-06 389 7% 144 ***37****%* 245 ***63****%* 0.59 Pélézi 35 74 2701 1.85117E-05 116 4% 85 ***73****%* 31 ***27****%* 2.74 G 128 97 4656 1.07388E-05 483 10% 99 ***20****%* 384 ***80****%* 0.26 G sub1 54 85 3570 1.40056E-05 73 2% 49 ***67****%* 24 ***33****%* 2.04 G sub2 17 73 2628 1.90259E-05 38 1% 32 ***84****%* 6 ***16****%* 5.33 G sub3 57 96 4560 1.09649E-05 36 1% 20 ***56****%* 16 ***44***% 1.25 ### Decrease in LD with genetic distance for the set of genotypes Figure [5](#F5){ref-type="fig"} shows the r^2^ and *D'* values (after Bonferroni's correction) as a function of distance for the set of genotypes. In general, the *D'* values were much higher than the r^2^ values. In addition, it seems that these values were much more stochastic with large proportions of very high LD that corresponded to non-significant associations. When the different DGs were considered (data not shown), *D'* seemed to be very sensitive to the structure effects and was also sensitive to the variations in allele frequencies between the different markers. Consequently, although the values of the two measurements appeared to decrease with distance, it seems that *D'* was less sensitive than r^2^ to that decrease, with values remaining high at very long distances. {#F5} As these two parameters have different properties, the information provided is not redundant. Indeed, *D'* measures only the recombination history, whereas r^2^ measures both the recombination and mutation \[[@B34]\]. These different properties signify that *D'* is globally higher than r^2^ and can potentially reveal more associations. The calculated values of *D'* are lower when population sizes are small, which is why this parameter is preferred when studying the evolutionary history of large populations. However, with association studies, the most indicative value of potential power is the r^2^ measurement, as it provides an indication of the way the markers and phenotypic traits being studied will be correlated. We cite the *D'* for a general comparison on the set of individuals before discussing in detail some differences in r^2^ between the different populations. ### Decrease in LD with distance for the different diversity groups The results of this analysis revealed a general decrease in the LD with distance. Nevertheless, we found as many different cases as there were DGs studied. ### Diversity groups SG1, SG2 and G For DG SG1 (Figure [6](#F6){ref-type="fig"}), we found moderate to high r^2^ values compared to the other populations, with a large proportion of between-LG LD. The major genetic structure in two populations (Niaouli and Luki) helps to explain such results. In light of these results, it is difficult to choose an r^2^ limit that can be used in association genetics for this DG and that will avoid the risks of false detections. {#F6} In DG SG2 (Figure [6](#F6){ref-type="fig"}), a large share of the significant r^2^ values was found between unlinked markers. The r^2^ values were extremely low, even for some very close markers. This observation may be explained by the origin of this DG, which must have undergone greater genetic mixing than the other origins. Indeed, this group originated from a major center of *Coffea canephora* diversity and is constituted mainly by genotypes resulting from the cultivation and selection processes. It can also be linked to the high number of alleles found in this DG, resulting in a more effective recombination process and diluting the LD signal. For this DG, it is possible to hypothesize that obtaining comparable values between linked and unlinked markers is not due to a structural phenomenon, but rather to very low or non-existent LD in this population. Indeed, structure analyses indicated that this diversity group is not structured in sub-populations and that most of the genotypes within this group are of a cultivated origin. Consequently, more than one microsatellite per cM would be needed in this DG to have any hope of satisfactorily covering the entire genome. For DG G (Figure [7](#F7){ref-type="fig"}), higher r^2^ values between linked markers than between unlinked markers were observed, with a clear decrease in those values with distance. Although some values over 0.1 could be seen for the DG as a whole, this was not the case for subgroups Gsub1 and Gsub2. Indeed, in these two subgroups, we found significant and high r^2^ values between close markers. By choosing an empirical r^2^ limit of 0.2, we arrived at an LD covering approximately 5 cM for subgroup Gsub1 and 20 cM for subgroup Gsub2 (Figure [7](#F7){ref-type="fig"}); these values suggest that it would be possible to cover the entire genome with a reasonable number of markers, i.e., approximately 290 and 70 markers, respectively. {#F7} A very low LD was found for subgroup Gsub3, as well as for DG SG2. The LD remained very low for this group, even at low distances. Therefore, more than 1500 markers would most likely be required to cover the entire genome in this population. Lastly, DG C and DG GP (Figure [8](#F8){ref-type="fig"}) exhibited high r^2^ values between close markers, displaying a strong decrease with distance, with a faster decrease in C than in GP. When an r^2^ value limit of 0.2 was chosen for GP and 0.1 was chosen for C, the LD was significant over a distance of approximately 23 cM for GP and 5 cM for C. These distances suggest that association studies can be conducted on these two populations with approximately sixty-five and 290 markers for GP and C, respectively. {#F8} Knowledge of the global LD thus provides a general idea of the variations in the LD over the entire genome. However, as the LD is highly variable depending on the region, a separate analysis of the various LGs for DGs GP, C, SG2 and G was carried out to compare the LD depending on the genome region. ### Comparison of LD patterns between linkage groups The graphs for r^2^ as a function of distance for LGs A, B, D, F, G and H are given in Figure [9](#F9){ref-type="fig"} for DGs GP, C, SG2 and G. {#F9} The analysis of these results shows that the LD varied depending on the LGs and the DGs that were considered. We confirmed the virtual absence of any usable r^2^ values for DGs SG2 and G. In contrast, for DG GP, we found moderate to high values for the set of LGs, although there were some large differences between the LGs. For DG C, only LGs A, F and G seemed to display r^2^ values over 0.1. However, these results need to be examined further, particularly with regards to LG G, which had a much greater marker density than the other LGs. When observing the LD matrices per DG (data not shown), significant p-values were preferentially located close to the diagonal, and hence, between linked markers. This preferential localization was particularly apparent for DGs GP and C. In contrast, DGs SG2 and G displayed a large share of significant p-values outside the LGs. DG C only had a few r^2^ values over 0.1, reflecting the short distance at which the LD can be detected. However, DG GP had a larger number of values over 0.1 within the different LGs with a LD that seemed to be organized in blocks. Discussion ========== This LD study is the first of its kind in the genus *Coffea*. We identified numerous different cases of LD evolution along the genome within the different DGs. We were also able to identify a cryptic structure within DG G. This very fine genetic structure had not been discovered in earlier studies either due to a lack of resolution or because too few markers were used. What is the point of a genome-wide LD study? -------------------------------------------- As in all species, we found a decrease in the LD with distance. For a population in equilibrium between mutation and genetic drift, the LD (measured by r^2^) is expected to depend on both the effective size of the population and the recombination rate between the loci considered \[[@B23],[@B24]\]. The closer any two markers are, the longer it will take for the LD to dissipate. Therefore, one expects to find some LD values greater between close markers than between distant markers at a given moment in the evolution of a population. The approach we adopted, with one marker every 13 cM on average but with highly densified portions of the genome, seemed to be the best strategy for constructing an initial pan-genomic view of the LD properties in the species studied. Moreover, our approach enabled a comparison of the LD behavior over different LGs. This comparison should lead to a clearer understanding of the possibilities for association studies within the studied populations. Is LD in *Coffea canephora* an insurmountable problem for association studies? ------------------------------------------------------------------------------ Our results demonstrate that it is possible to perform association studies by working specifically on each population, but not on the global diversity level, in the species. Depending on the populations, the needed marker density varies, but the prospects of using association studies to support breeding programs for our species are quite interesting. Indeed, the graph showing the set of r^2^ and *D'* values on the scale of the 356 genotypes, combined with the importance of the genetic structure found in the entire sample, clearly illustrates the importance of the structure effect on the detection of associations between unlinked markers. The Hardy-Weinberg disequilibrium could not be reduced from the whole sample scale to the GD level. However, disequilibrium still exists within natural populations of *C. canephora,* as shown by \[[@B12],[@B35]\]. This disequilibrium thus prevents the implementation of association studies for an entire set of genotypes using simple correlation models, which do not take into account structure and kinship effects. This result was confirmed by the large number of significant correlations between markers located on the different LGs for the 356 genotypes compared to those found on the DGs. Therefore, it seems necessary to work at the population level to more effectively study the LD dynamics in *C. canephora*, as the analyses showed that the most valuable results were obtained on DG GP, DG C and two Guinean subgroups (Gsub1 et Gsub2) corresponding more or less to natural populations. We were thus able to reveal a high variability in LD within the different DGs, with a large share of residual between-linkage group disequilibrium in DG SG2 and DG SG1. These results may potentially lead to the detection of false positives in association studies, even at low levels of genetic structure. The importance of this "genomic" LD (as opposed to local LD) was variable depending on the groups, and by taking into account structure and kinship in association studies, it will be possible to overcome this variability. The residual genomic LD values for the less-structured DGs may be explained by different kinship levels within the natural populations of our species. Nevertheless, for DG SG2, the low r^2^ values obtained suggest that the LD is significant at very short distances. Indeed, natural populations of coffee trees are usually small, isolated populations with a small number of mother trees and a few juveniles, involving major relations of kinship, despite the strict outcrossing of the studied species. We used Bonferroni's correction to consider only truly significant values. Nevertheless, this correction is very conservative and may lead to a substantial loss of power in association studies. Many other corrections have been proposed in the literature in recent years, but none seems to be satisfactory. Moreover, we have shown that, in our case, this correction mainly made it possible to eliminate a certain number of disequilibrium values between unlinked or very weak markers. Normally, in association studies, such a correction will not be necessary because the main source of error (genetic structure) will be controlled. These questions should be given due consideration along with updates in the proposed models. Models that take into account structure and kinship in association studies appear to be a major advance in these approaches, helping to increase both the power and the resolution of such studies \[[@B32]\]. Genetic structure of DG G ------------------------- Our study enabled us to more effectively determine a fine genetic structure for DG G. Structure seems to exist in these populations, but there are indications of major gene exchanges between them. We found a structure in three subgroups (Gsub1, Gsub2 and Gsub3) with both model-based and distance-based analyses. This very fine genetic structure can only be studied with a large number of markers. DG G was initially described by Berthaud \[[@B36]\] using isozyme markers. Berthaud concluded at the time that there was an absence of genetic structure within this group. However, Cubry et al. \[[@B12],[@B13]\] showed the existence of a Guinean population that was different from the others (GD GP). It will also be important to study kinships within natural populations, along with the gene flow existing between them, to understand the dynamics of those populations on the forest scale in Guinea and the Ivory Coast. What models can be used for association studies in *Coffea canephora*? ---------------------------------------------------------------------- Our results show, particularly for the Guineans, that a large number of "control" markers (i.e., control markers that can be used to estimate structure and kinship independently from the association study) are needed to separate the fine structure into populations. Therefore, our case seems to be quite similar to the case of maize, where a set of eighty-nine microsatellite markers was used by Flint-Garcia et al. \[[@B37]\] to study structure and kinship on 302 lines. After correction of the p-values by the Bonferroni method, some large and significant values of the two LD measurements (*D'* and r^2^) were found both between unlinked and linked markers, preventing any distinction between associations based on a physical link between markers and those created by the structure. Therefore, the genomic control approach (adaptation of the significance limit to the number of associations detected between unlinked markers) appears to be less efficient and may lead to a large number of false negatives. This observation is one of the greatest criticisms of this model advanced by Yu et al. \[[@B38]\]. Moreover, this approach estimates that structure has the same effect at any point of the genome \[[@B5]\]. The structured association approach proposed by Pritchard et al. \[[@B17]\] seems to be more efficient than the genomic control. Nevertheless, the degree of kinship in the populations studied, as shown by the diversity trees obtained (particularly for DG GP), indicates that a share of the confounding effect of genetic structuring is not taken into account in this model. Consequently, it seems that the model best adapted to the species and populations in our study is the mixed model proposed by Yu et al. \[[@B38]\]. This approach has shown its power and its superior control of false positives when compared to other methods using simulated data. These association study models are becoming increasingly efficient, and we seem to be arriving at a critical point in the development of these approaches. Even so, particular attention must be given to the choice of traits studied and their distribution within the sample on which association studies are performed. Indeed, by correcting the structure effect, there is a risk of not being able to detect traits that would have a distribution superimposed on the population structure \[[@B39]\]. Which target populations should be used for association studies? ---------------------------------------------------------------- The purpose of our work was to make an initial assessment of the LD at the pan-genomic level in *C. canephora*. We thus discovered considerable variation in the LD between populations. The DGs comprising natural populations, such as GP or C, appear to have a moderate to high LD, at approximately 5 to 25 cM. In these DGs, it seems feasible to carry out genome-wide scan type studies. Nevertheless, given the stochasticity of the LD between LGs and its sensitivity to low allele frequencies, we have certain reservations regarding this type of approach, notably when using highly polymorphic multi-allelic microsatellite markers. The DGs comprising improved populations, such as SG2, seemed to have undergone substantial genetic mixing with greater diversity and a virtually undetectable LD on the scale at which we worked. Consequently, this type of population seems more suited to regional or candidate gene type approaches. To conclude, the current association study models can be used to consider structure and kinship effects, enabling this type of approach for use even with composite and structured samples. Which approach for association studies in *Coffea canephora*? ------------------------------------------------------------- For most of the considered DGs, a high density of markers is required to perform association studies. Using SNPs in additions to the SSR should be of great value. Considering ongoing work on SNP discovery and genotyping by sequencing in coffee, a high number of markers will likely be obtained in the short term. Then, genome-wide association studies (GWAS) can be easily applied to populations used in breeding processes, such as populations involved in the RRS in the Ivory Coast. In countries such as Uganda, GWAS should be applied to the entire germplasm used for breeding for tolerance to Coffee wilt disease (CWD). Therefore, the candidate gene approach should only be used in very low LD populations or for specific purposes. Conclusions =========== We were able to demonstrate a cryptic structure within the Guinean diversity group (DG G). This very fine genetic structure was not detected in earlier studies either due to a lack of resolution or because too few markers were used. Nevertheless, complex dynamics seem to exist within the coffee populations of the Guinean region that are substantially different from those found in the forests of Uganda, as described by Musoli et al. \[[@B35]\]. A more in-depth study of these populations will be undertaken to establish the relationships and gene flow existing between them. The LD study, the first of its kind in the genus *Coffea*, has enabled us to consider association studies in the species *C. canephora*. In the different DGs, we identified numerous different cases of LD evolution along the genome. This study provided us with a basis for carrying out association studies and thereby optimizing the reciprocal recurrent selection scheme for the genetic improvement of *C. canephora*. This selection scheme is based on the complementary traits of diversity groups and the agronomic value of intergroup hybrids. Association studies within groups will allow us to significantly improve each of the complementary populations for their traits of interest. An initial application of this approach can be employed in Guinean populations (exhibiting high LD) with a limited number of markers throughout the genome. A selection process should be efficient and quick using association studies for the early selection of heritable traits (such as vigor or bean size) in intergroup hybrids. The major genetic structure of our species, which may have limited the feasibility of association studies a few years ago, can now be taken into account in the latest models. Nevertheless, it will be important to carefully consider all the models and, in particular, to avoid over-correcting the structure effect, as this action may lead to numerous false negatives and a major loss of detection power. Lastly, the results of our study on *C. canephora* could be extrapolated to *C. arabica.* Although *C. arabica* has a very restricted genetic base, as shown by Lashermes et al. \[[@B40]\] and Anthony et al. \[[@B41]\], one of the species participating in its creation is very close to *C. canephora*. Abbreviations ============= LD: Linkage Disequilibrium; DG: Diversity Group; LG: Linkage Group; cM: Centimorgan. Competing interests =================== The authors declare that they have no competing interests. Authors' contributions ====================== PC, MD, DP, KA, SB and TL designed the study. HL provided the plant material. PC, FDB and KA carried out the genotyping experiments. PC and DP analyzed the data. PC, FDB and TL wrote the manuscript. All authors read and approved the final manuscript. Supplementary Material ====================== ###### Additional file 1: Table S1 Detailed information on the 108 SSR markers used. ###### Click here for file ###### Additional file 2: Table S2 Number of alleles, expected heterozigosity(He), observed heterozigosity and Hardy-Weinberg equilibrium exact test associated p-values (P-value )and standard deviation (s.d.) for each locus and for the whole population and the Diversity Groups SG2, SG1, C, G and GP. NA stands for not available due to monomorphic locus. ###### Click here for file ###### Additional file 3: Table S3 Number of alelles and Hardy-Weinberg equilibrium statistics for the Diversity Groups G, GP, Gsub1, Gsub2 and Gsub3. NA stands for not available due to monomorphic locus. ###### Click here for file ###### Additional file 4: Table S4 Amova and F-statistics computation over the 7 DG (SG1, SG2, C, GP, Gsub1, Gsub2 and Gsub3). \* : significant value at 5%, ns: non significant value. ###### Click here for file Acknowledgements ================ We thank the Grand Plateau Technique Régional de Génotypage (GPTR-GENOTYPAGE) and Ronan Rivallan for help in carrying out the genotyping experiments. We also thank Dr. Colin Kelleher from the DBN Plant Molecular Laboratory, National Botanic Gardens of Ireland, and Dr. Jean-Pierre Labouisse, CIRAD Montpellier, for correction of the manuscript. The authors are grateful to Nestlé Tours and Dr. Dominique Crouzillat for providing some genomic and EST-derived microsatellites markers. PC was supported by a French Grant from the research ministry for his PhD.

Introduction ============ Salivary duct carcinoma (SDC) arises from the ductal epithelium of the salivary gland and comprises rare tumors that account for approximately 1--3% of all salivary gland malignancies ([@b1-ol-0-0-8525]). SDC was first described by Kleinsasser *et al* in 1968 owing to its histologic similarity to invasive ductal carcinoma (IDC) ([@b2-ol-0-0-8525]). SDC constitutes one of the most aggressive salivary gland malignancies and is resistant to radiation therapy and chemotherapy ([@b1-ol-0-0-8525],[@b3-ol-0-0-8525]). Although extended resection and postoperative irradiation are performed as standard treatments, the therapeutic outcome is not generally improved ([@b1-ol-0-0-8525],[@b4-ol-0-0-8525]). Considering the similarities with ductal carcinoma of the breast and prostate cancer, overexpression of androgen receptor (AR), epidermal growth factor receptor (EGFR), and human epidermal growth factor receptor 2 (HER2) has also been investigated in SDC ([@b1-ol-0-0-8525],[@b5-ol-0-0-8525]--[@b8-ol-0-0-8525]). HER2 expression serves as a predictive factor in IDC as well ([@b9-ol-0-0-8525]); moreover, HER2 protein in IDC constitutes the most important target for molecular targeted therapy. Previously, rates of amplification of the *HER2* gene and HER2 protein overexpression in SDC were reported to range widely from 15 to 100% ([@b3-ol-0-0-8525],[@b10-ol-0-0-8525],[@b11-ol-0-0-8525]). Recently, androgen and/or estrogen deprivation therapy ([@b12-ol-0-0-8525],[@b13-ol-0-0-8525]) and molecular targeted therapy for HER2 have been attempted as adjuvant therapies ([@b14-ol-0-0-8525]--[@b17-ol-0-0-8525]) with anti-HER2 therapy in particular expected to become a useful tool for adjuvant therapy ([@b15-ol-0-0-8525],[@b17-ol-0-0-8525]); however, satisfactory results have not been obtained ([@b16-ol-0-0-8525]). Thus, additional novel therapeutic strategies are required for SDC. DNA methylation, i.e., the modification of cytosine to form 5-methylcytosine, is essential for normal development but is also associated with carcinogenesis. In many cases, suppression of tumor suppressor genes by DNA hypermethylation of the promoter region can induce carcinogenesis. Thus, elucidation of the DNA methylation profile in SDC might facilitate the development of novel therapeutic strategies for SDC. Our previous studies demonstrated that DNA methylation of several G-protein coupled receptors (GPCRs) was associated with the survival rate of patients with head and neck squamous cell (HNSCC) ([@b18-ol-0-0-8525]). The galanin receptors, GALR1 and GALR2, are members of the GPCR superfamily, and serve as important tumor suppressor genes for HNSCC ([@b19-ol-0-0-8525]--[@b21-ol-0-0-8525]). Specifically, GALR1 mediates cell cycle arrest ([@b19-ol-0-0-8525]) whereas GALR2 mediates both cell cycle arrest and apoptosis ([@b20-ol-0-0-8525]) via common pathways including p27^kip1^, p57^kip2^, and cyclin D1 ([@b22-ol-0-0-8525]). DNA methylation of *GALR1* and *GALR2* promoters was significantly associated with the survival and recurrence rates of patients with HNSCC and is considered as a potential therapeutic target and prognostic factor for HNSCC ([@b23-ol-0-0-8525]--[@b25-ol-0-0-8525]). GALR promoter methylation is observed in other squamous cell carcinomas as well as adenocarcinomas such as breast, colon, and hepatocellular carcinoma ([@b26-ol-0-0-8525],[@b27-ol-0-0-8525]), and thus appears to constitute a carcinoma type-independent prognostic factor. The aim of the present study was therefore to first define the *GALR1, GALR2*, and galanin methylation status in SDCs at the time of diagnosis and then to evaluate its significance as a biomarker for prognosis. Patients and methods ==================== ### Patient characteristics Tumor specimens were obtained from 34 patients diagnosed with SDC based on histological findings at the Department of Otolaryngology-Head and Neck Surgery, Jichi Medical University, School of Medicine, the Department of Otolaryngology-Head and Neck Surgery, Hamamatsu University, School of Medicine, and the Division of Head and Neck, Cancer Institute Hospital, Japanese Foundation of Cancer Research, from March 1995 to March 2012. The present study was approved by the Institutional Ethics Review Board of the ethics committee of each of the three institutions that participated in this study. The need to obtain informed consent was waived owing to the retrospective nature of the analysis. In this study, we analyzed only cases of de novo SDC; SDC ex pleomorphic adenomas were excluded. Patient characteristics were also reviewed with regard to sex, age, TNM classification, clinical stage, surgical procedures, and additional adjuvant therapy. ### Immunochemical analysis The tissues were fixed in 10% formalin and embedded in paraffin in a routine manner, and stained with hematoxylin and eosin. All cases were histologically reviewed according to the definition of SDC. Briefly, SDC showed a cribriform growth pattern, Roman bridge formation, and comedonecrosis of tumor cells having abundantly eosinophilic cytoplasm and a large pleomorphic nucleus with prominent nucleoli and coarse chromatin. Immunohistochemistry was performed on 4-µm sections from paraffin blocks using antibodies directed against androgen receptor (AR) (mouse monoclonal antibody clone AR441, Dako Corporation, Glostrup, Denmark), estrogen receptor (ER) (clone 6F11, Leica Biosystems, Nussloch, Germany), HER2 (rabbit polyclonal, HercepTest, Dako), EGFR (clone 31G7, Nichirei Biosciences Inc., Tokyo, Japan), p27^Kip1^ (clone Y236, GeneTex, Irvine, CA, USA), p57^Kip2^ (clone: DO-7, Dako), and cyclin D1 (clone: SP4, Thermo Scientific, Waltham, MA, USA). The results of immunohistochemical staining were independently scored by two of the authors (TK and YS). AR positively was evaluated in a manner similar to ER according to the American Society of Clinical Oncology/College of American Pathologist guideline ([@b28-ol-0-0-8525]) for evaluation of breast cancer predictive factors: if ≥1% of tumor cell nuclei are immunoreactive, the tumor was considered to be positive for AR. The evaluation of HER2 expression was in accordance to the criteria for evaluating responsiveness of breast carcinoma to anti-HER2 treatment, with a score of 0--2 being considered as HER2 negative and a score of 3 was considered as HER2 positive. For EGFR, according to the criteria for evaluating responsiveness of colorectal carcinoma to anti-EGFR treatment, a score of 0--2 was considered as EGFR negative and a score of 3 was considered as EGFR positive. p27 scoring was determined by the criteria of ovarian carcinoma: 1+ \<5%, 2+ 5--50%, 3+ \>50%; p57 was in accordance to vulva carcinoma criteria: 1+ \<10%, 2+ 10--50%, 3+ \>50%; and cyclin D1 was scored according to breast carcinoma criteria (−\<10%, + ≥10%). ### DNA promoter methylation analysis Genomic DNA was extracted from 8-µm sections of paraffin blocks using the QIAamp DNA FFPE Tissue Kit (Qiagen, Venlo, The Netherlands). Extracted DNA was bisulfite-modified using the MethylEasy™ *Xceed* Rapid DNA Bisulphite Modification Kit (TaKaRa Bio., Tokyo, Japan). Methylation in the region near the transcription start site was assessed using bisulfite-treated DNA polymerase chain reaction (PCR) amplified with methylation-specific PCR primers (MSP) and unmethylation-specific PCR primers (UMSP) using FastStart Taq DNA polymerase (Roche Lifescience Inc., Basel, Switzerland). The primers are shown in [Table I](#tI-ol-0-0-8525){ref-type="table"}. The PCR conditions were 94°C for 5 min; optimal cycle numbers between 35 and 45 at 94°C for 30 sec, 60°C for 30 sec, and 72°C for 40 sec; and a final extension at 94°C for 5 min. The PCR products were separated by 3% agarose gel electrophoresis and stained with ethidium bromide. The PCR products amplified by MSP or UMSP were visualized and quantified using Image J software (<http://imagej.nih.gov/ij/>), and the ratio of MSP/UMSP was defined as the methylation rate. Receiver operating characteristic (ROC) curve analysis was performed using the methylation rate for 34 SDC and 19 adjacent normal parotid gland tissues. The cutoff value determined from this ROC curve was applied to determine the frequently of *GALR1, GALR2*, and galanin methylation in this study. ### Statistical analysis For frequency analysis in contingency tables, statistical analyses of association between variables were performed using Fisher\'s exact test. To evaluate the galanin and GALR pathway in SDC, the Pearson\'s correlation coefficients between the methylation rate and expression score of p27, p57, and cyclin D1 were calculated. Furthermore, the survival interval was estimated as the length of time from the start of treatment to the final date of confirmed survival. Overall survival (OS) probabilities were estimated using the Kaplan-Meier method and the log-rank test was applied to assess the significance of differences among actuarial survival curves. Results ======= ### Patient characteristics [Table II](#tII-ol-0-0-8525){ref-type="table"} summarizes the characteristics of the 34 patients with SDC evaluated in this study. Men were predominant (20 cases, 58.8%) compared to women (14 cases, 41.2%). Median age was 63.4 years old (range, 45--79 years), and median follow-up time was 32.3 months (range, 5--59 months). Regarding tumor and nodal stage, T2, T4a, N0, and N2 were predominant. Over half of cases (55.9%) were classified as Stage IV. Surgery was performed for all cases with partial parotidectomy in 7 cases (20.6%), total parotidectomy in 16 cases (47.1%), and extended parotidectomy in 11 cases (32.4%). Postoperative irradiation was applied for 28 cases (82.4%), whereas no cases received preoperative irradiation. ### Clinicopathological factors associated with OS The median OS was 37.2 months. The results of univariate Kaplan-Meier survival analyses are summarized in [Table III](#tIII-ol-0-0-8525){ref-type="table"}. Increasing T stage, N stage, tumor stage, tumor size, preoperative facial paralysis, and resection margin status were negative prognostic factors for OS. Tumors in T3-T4 stage were associated significantly worse OS than those in T1-T2 stage. N2-N3 stage tumors had significantly worse OS than N0-N1 stage tumors. Stage IV tumors had significantly worse OS compared to Stage I--III tumors. Tumors over 30-mm diameter had significantly worse OS than those less than 30-mm. Tumors with preoperative facial paralysis had significantly worse OS than those without paralysis. Tumors with a positive surgical margin had significantly worse OS than negative tumors. Other factors such as lymphovascular invasion and extra-nodal spread did not affect the length of OS. Contrary to prior findings ([@b15-ol-0-0-8525],[@b16-ol-0-0-8525]), there was no association between HER2 positively and survival. Other immunochemical factors such as EFGR, AR, and ER were also not associated with survival. p27^kip1^, p57^kip2^, and cyclin D1 are encoded by cell cycle associated genes, the expression of which is controlled by GALR signaling in HNSCC ([@b19-ol-0-0-8525],[@b20-ol-0-0-8525]). Although cyclin D1 overexpression was associated with the length of OS, p27^kip1^ and p57^kip2^ expression did not affect OS. ### Promoter methylation of GALR1, GALR2, and galanin To investigate whether *GALR1, GALR2*, and galanin were methylated in SDC, the methylation level of these genes in tumor and normal tissue were compared. *GALR1, GALR2*, and galanin promoter hypermethylation exhibited highly discriminative ROC curve profiles, which clearly distinguished HNSCC from normal mucosal tissues ([@b23-ol-0-0-8525],[@b24-ol-0-0-8525],[@b29-ol-0-0-8525]). The ROC curve with corresponding area under the curve for *GALR1, GALR2*, and galanin of SDC vs. normal mucosal tissues is presented in [Fig. 1](#f1-ol-0-0-8525){ref-type="fig"}. The methylation rates of *GALR1* in tumor tissues were significantly higher (9.85-fold) than those in normal tissues ([Fig. 1A](#f1-ol-0-0-8525){ref-type="fig"}). The cutoff methylation rate (0.2) for *GALR1* was chosen from the ROC curve to maximize sensitivity (70.6%) and specificity (78.9%) ([Fig. 1D](#f1-ol-0-0-8525){ref-type="fig"}). The cutoff methylation rate (0.34) for *GALR2* in tumor tissues was also significantly higher (4.49-fold) than that in normal tissues ([Fig. 1B](#f1-ol-0-0-8525){ref-type="fig"}). *GALR2* methylation rates yielded sensitivity (67.6%) and specificity (78.9%) ([Fig. 1E](#f1-ol-0-0-8525){ref-type="fig"}). However, the cutoff methylation rate of galanin was not determined because no significant difference of methylation rate was observed between SDC and normal tissue ([Fig. 1C and F](#f1-ol-0-0-8525){ref-type="fig"}). According to the cutoff values for *GALR1* and *GALR2*, the tumors were divided into methylated and unmethylated tumors. ### Correlation between GALR methylation and expression of downstream proteins Both GALR1 and GALR2 induced cell cycle arrest though up-regulation of p27^kip1^ and p57^kip2^, and down-regulation of cyclin D1 in HNSCC ([@b19-ol-0-0-8525],[@b20-ol-0-0-8525]). To confirm whether this pathway exists in SDC, the correlation between *GALR* methylation and expression of these proteins was evaluated. As shown in [Fig. 2](#f2-ol-0-0-8525){ref-type="fig"}, *GALR1* methylation showed a significant inverse association with p27^kip1^ and p57^kip2^. The p27^kip1^ or p57^kip2^ lower expressing tumors were more often observed among *GALR1* methylated tumors than unmethylated tumors ([Fig. 2A and B](#f2-ol-0-0-8525){ref-type="fig"}). Similarly, *GALR2* methylation was also significantly inversely associated with p27^kip1^ and p57^kip2^. p27^kip1^ or p57^kip2^ higher expressing tumors were more often observed among *GALR2* unmethylated tumors than methylated tumors ([Fig. 2D and E](#f2-ol-0-0-8525){ref-type="fig"}). However, a significant correlation between cyclin D1 expression and *GALR* methylation was not observed ([Fig. 2C and F](#f2-ol-0-0-8525){ref-type="fig"}). These results indicate that GALR1 and GALR2 signaling pathways likely act as tumor suppressors in SDC. ### Prognostic value of GALR1 and/or GALR2 promoter methylation status To examine the prognostic value of *GALR1* and/or *GALR2* promoter methylation status, the OS of methylated and unmethylated tumors were compared. *GALR1* methylation was associated with a statistically significant decrease in OS (log-rank test, P=0.02609) ([Fig. 3A](#f3-ol-0-0-8525){ref-type="fig"}). The OS of *GALR1* methylated tumors was 27.5% and of unmethylated tumors was 67.5% at 4 years. Methylation of *GALR2* was also significantly associated with OS: The OS of *GALR2* methylated tumors at 4 years was 21.2% and that of unmethylated tumors was 96.2%. *GALR2* methylation was thus significantly associated with OS decrease (log-rank test, P=0.03028) ([Fig. 3B](#f3-ol-0-0-8525){ref-type="fig"}). Methylation in both *GALR1* and *GALR2* was associated with an OS rate of 22.2%, as compared with an OS rate of 42.1% for any methylation and 100% for unmethylation of both *GALR1* and *GALR2* (log-rank test, P=0.0229) ([Fig. 3C](#f3-ol-0-0-8525){ref-type="fig"}). These results indicate that *GALR1* and *GALR2* methylation status would be sufficient to determine the prognosis for SDC. Discussion ========== Limited knowledge is available regarding SDC, a rare tumor arising mainly from the salivary gland. A large study by Jayaprakash *et al* described that negative factors for SDC comprised age 50 years or older, tumor size, and lymph node involvement, with no apparent survival benefit of radiation therapy ([@b30-ol-0-0-8525]). In the present study, age and gender did not affect the survival rate and were not prognostic factors. Conversely, clinocopathological factors were important prognostic factors in SDC, similar to other carcinomas. T stage, N stage, disease stage, tumor size, preoperative facial paralysis, and positive resection margin significantly decreased the survival rate. These results provide important information for therapeutic selection, suggesting that extended surgery should be chosen for locoregional advanced cases. As facial nerve paralysis was observed in 7 of 34 cases, the local invasive potential of SDC appears very aggressive. However, the surgical margin is limited by anatomical necessity, as the site is close to the skull base, cervical vertebra, and carotid artery. Thus, effective adjuvant therapies are required. Alternatively, genetic alterations in SDCs have been reported, leading to the investigation of HER2, EGFR, ER, and AR as therapeutic targets and prognostic factors ([@b1-ol-0-0-8525],[@b4-ol-0-0-8525]--[@b8-ol-0-0-8525]). In the present study, 61.7% of cases expressed a high level EGFR (3+), 52.9% expressed a high level HER2 (3+), 5.9% expressed ER, and 64.7% of cases expressed AR. However, although the expression of these proteins was also observed in this study, significant correlations to survival rates were not observed. In comparison, HER2 positively is considered to be a predictor of poor prognosis in breast cancer, wherein the determination of HER2 status is reported to be crucial to select patients who may benefit from HER2-targeted therapy. Based on previous results, HER2-targeted therapy may therefore not have received sufficient evaluation as a standard therapy in SDC ([@b16-ol-0-0-8525]). In SDCs, however, although Jaehne *et al* ([@b3-ol-0-0-8525]) reported that HER2 overexpression was linked to poor survival in their analysis of 50 cases, it remains unclear whether HER2 gene amplification and/or protein overexpression are predictors of poor prognosis in carcinomas other than breast cancer. In particular, a recent report indicates that HER2 is not a prognostic factor in SDC ([@b1-ol-0-0-8525]). Thus, molecular targeted therapies based on the reported genetic alterations require further investigation. To develop novel therapeutic strategies for HNSCC, we have previously investigated the epigenetic silencing of tumor suppressor genes, with the most promising tumor suppressor genes being *GALR1* and *GALR2* ([@b19-ol-0-0-8525],[@b20-ol-0-0-8525]). The effects of GALR1 and GALR2 are clearly reflected in clinical outcome ([@b23-ol-0-0-8525],[@b24-ol-0-0-8525],[@b29-ol-0-0-8525]). Our previous experiments using HNSCC cell lines demonstrated that *GALR1* and *GALR2* promoter methylation is significantly correlated with a decrease of the respective mRNA expression ([@b23-ol-0-0-8525]). *GALR1* promoter methylation was significantly correlated with reduced survival rates, tumor stage, lymph-node status, increased tumor size, cyclin D1 expression, and p16 methylation ([@b23-ol-0-0-8525]). However, in multivariate analysis, only *GALR1* methylation and tumor stage were significant predictors of poor survival ([@b23-ol-0-0-8525],[@b31-ol-0-0-8525]). *GALR2* promoter methylation was significantly related to methylation of *COL1A*, H-cadherin, *DAPK, GALR1*, and galanin ([@b24-ol-0-0-8525]). *GALR2* promoter methylation was also related to a significant decrease in disease free survival. Specifically, in a multivariate logistic regression analysis, *GALR2* promoter methylation in the primary tumor was related to an adjusted odds ratio for recurrence of 3.12 ([@b24-ol-0-0-8525],[@b31-ol-0-0-8525]). Based on these results, we investigated the promoter methylation status of galanin, *GALR1*, and *GALR2* in SDC to confirm the value as prognostic biomarkers in this disorder. The methylation rates of *GALR1* in SDC tumor tissues were significantly higher (10.31-fold) than those in normal tissues. *GALR2* promoter methylation in tumor tissues was also significantly higher (4.51-fold) than that in normal tissues. *GALR1* methylation further showed a significant inverse association with p27^kip1^ and p57^kip2^. p27^kip1^ or p57^kip2^ lower expressing tumors were more often observed among *GALR1* methylated tumors than unmethylated tumors. Similarly, *GALR2* methylation was significantly inversely associated with p27^kip1^ and p57^kip2^. p27^kip1^ or p57^kip2^ higher expression tumors were more often observed among GALR2 unmethylated tumors than in methylated tumors. These results suggested that GALR1 and GALR2 pathways likely exist in SDC and that their methylation states may constitute potential prognostic biomarkers. Furthermore, *GALR1* methylation was associated with a statistically significant decrease in OS: 38.8% for *GALR1* methylated tumors vs. 68.2% for unmethylated tumors. Methylation of *GALR2* was also associated with OS, with the OS of *GALR2* methylated tumors being 21.2% and compared to 96.2% for unmethylated tumors. *GALR2* methylation thus was associated with significantly decreased OS. Methylation in both *GALR1* and *GALR2* was associated with an OS rate of 22.2%, as compared with that of 42.1% for any methylation and of 100% for both promoters being unmethylated. Thus, *GALR1* and *GALR2* resemble other major tumor suppressor genes in terms of frequency of aberrant promoter methylation *in vivo*. The survival curves clearly show the correlation between methylation status of GALRs and OS, however, the downstream proteins expressions such as p27 and p57, and OS are not related. Cyclin D1 overexpression was related to the length of OS, but not associated with GALR methylation status. Although this discrepancy was not fully understood, other signaling pathways and many kinds of molecules controlled by GALR would be related to survival of SDC. Further investigation about GALRs signaling pathway in SDC are required. In summary, in this study, we showed for the first time, to our knowledge, that silencing of the *GALR1* and *GALR2* genes by methylation may constitute a critical event in SDC. The current data further suggest that *GALR1* and *GALR2* are potentially significant therapeutic targets and prognostic factors in SDC. This paper was orally presented at the 41st annual meeting of Japan Society for Head and Neck Cancer on June 8, 2017 in Kyoto, Japan. Funding ======= A Grant-in-Aid for Scientific Research (grant nos. 26462620, 17K11403 and 17K11402) from the Ministry of Education, Culture, Sports, Science, and Technology of Japan. Availability of data and materials ================================== The datasets during and/or analysed during the current study available from the corresponding author on reasonable request. Authors\' contributions ======================= TK, SI and HN saw the patients and drafted the clinical detail at Jichi Medical University. KM, YM and HM saw the patients and drafted the clinical details at Hamamatsu University School of Medicine. HF and KK saw the patients and collected the clinical data at Cancer Institute, Japanese Foundation of Cancer Research. YS made pathological diagnosis with GK and TK. MM, TK and GK performed the DNA methylation experiments. TEC supervised this study. TK primarily compiled the data into this report. All authors read and approved the final manuscript. Ethics approval and consent to participate ========================================== The present study was approved by the Institutional Ethics Review Board of the ethics committee of each of the three institutions that participated in this study. Jichi Medical University, Hamamatsu University, School of Medicine, and Cancer Institute Hospital, Japanese Foundation of Cancer Research. The need to obtain informed consent was waived owing to the retrospective nature of the analysis, however, consent was obtained from patients at the time of data collection. Consent for publication ======================= Not applicable. Competing interests =================== The authors declare that they have no competing interests. {#f1-ol-0-0-8525} {#f2-ol-0-0-8525} {#f3-ol-0-0-8525} ###### Sequences of primers used in this study. Gene Methylation-specific primer sequence (5′-3′) Unmethylation-specific primer sequence (5′-3′) --------- ---------------------------------------------- ------------------------------------------------ Galanin Forward: TGACGCGATTTCGGGCGGTT Forward: TGATGTGATTTTGGGTGGTT Reverse: TATCCGCCGCCCGATATAAC Reverse: TATCCACCACCCAATATAAC *GALR1* Forward: GGTTCGCGGTATTCGGTAGT Forward: GGTTTGTGGTATTTGGTAGT Reverse: TCGCCGCCCACCTCCCGACTA A Reverse: TCACCACCCACCTCCCAACTAA *GALR2* Forward: CGATTGCGGGGGTTGGAGTTCGGA Forward: CCAACAACGACCGACGACGCTA Reverse: TGATTGTGGGGGTTGGAGTTTGGA Reverse: TTATCCCCAACAACAACCAACAACACTA ###### Characteristics of patients with salivary duct carcinoma of the parotid gland. Characteristics No. (%) ------------------------------- ------------ Sex   Male 20 (58.8)   Female 14 (41.2) Age   Mean 63.4   Range 45--79 Pathological T classification   T1   3 (0.09)   T2 10 (24.9)   T3   7 (20.6)   T4a 14 (41.2) Pathological N classification   N0 11 (32.3)   N1   7 (20.6)   N2 16 (47.1) Tumor stage   Stage I 3 (8.8)   Stage II   6 (17.6)   Stage III   6 (17.6)   Stage IV 19 (55.9) Surgical procedure   Partial parotidectomy   7 (20.6)   Total parotidectomy 16 (47.1)   Extended parotidectomy 11 (32.4) Postoperative irradiation   Negative   6 (17.6)   Positive 28 (82.5) ###### Univariate analysis of clinicopathological factors associated with overall survival. Variable 4-year OS (%) P-value ------------------------------- --------------- ------------------------------------------------------ T stage 0.00803^[a](#tfn1-ol-0-0-8525){ref-type="table-fn"}^   T1-2 (n=20) 65.7   T3-4 (n=14) 20.6 N stage 0.00098^[a](#tfn1-ol-0-0-8525){ref-type="table-fn"}^   N0-1 (n=18) 74.2   T3-4 (n=16) 11.8 Disease stage 6.1E-0.5   Stage I--III (n=14) 90.9   Stage IV (n=19)   9.4 Tumor size 0.00089^[a](#tfn1-ol-0-0-8525){ref-type="table-fn"}^   \<30 mm (n=20) 68.4   \>30 mm (n=14) 19.8 Preoperative facial paralysis 0.00635^[a](#tfn1-ol-0-0-8525){ref-type="table-fn"}^   Negative (n=21) 57.9   Positive (n=7) 14.3 Resection margin 0.00550^[a](#tfn1-ol-0-0-8525){ref-type="table-fn"}^   Negative (n=22) 67.5   Positive (n=9) 0 Lymphovascular invasion 0.06100   Negative (n=8) 77.3   Positive (n=20) 54.6 Extra-nodal spread 0.23000   Negative (n=18) 54.3   Positive (n=13)   41.04 EGFR 0.40320   0-2 (n=17) 52.2   3 (n=17) 44.9 HER2 0.05100   0-2 (n=16) 58.3   3 (n=18) 29.6 Androgen receptor 0.15900   Negative (n=12) 62.3   Positive (n=22) 39.0 Estrogen receptor 0.05640   Negative (n=28) 56.5   Positive (n=6) 33.3 p27 0.18465   1--2 (n=16) 24.4   3 (n=18) 60.0 p57 0.28940   1--2 (n=25)   40.99   3 (n=9) 63.5 Cyclin D1 0.03410^[b](#tfn2-ol-0-0-8525){ref-type="table-fn"}^   0 (n=25) 57.4   1 (n=9) 17.7 P\<0.01. P\<0.05. OS, overall survival; EGFR, epidermal growth factor receptor; HER2, human epidermal growth factor receptor 2.

Objectives ========== Nerve growth factor (NGF) plays a crucial role in the differentiation and survival of sympathetic neurones, including the ones represented in the cardiovascular system. NGF has also been reported to be an intermediary product in various inflammatory reactions. The aim of this study was to investigate whether cardiopulmonary bypass (CPB) has any effects on NGF release and to determine the time course of any changes in its release. Methods ======= Twelve consecutive children undergoing elective cardiac surgery were recruited for the purposes of this study. The plasma levels of NGF, interleukin 6 (IL-6) and interleukin-8 (IL-8) were measured by enzyme-linked immunosorbent assay (ELISA). Samples were obtained from the patients at the following time points: induction of anaesthesia, end of CPB, and at 1, 2 and 24h following CPB. Results ======= There was a highly significant increase in the release of NGF following CPB, reaching its highest concentration at 2 h following CPB. The pattern of release was very similar to that observed for IL-6 and IL-8. The results are summarised in the [Table](#T1){ref-type="table"}. Pre-op End-CPB 1 h post-CPB 2 h post-CPB 24 h post-CPB -------------- -------------- ---------------- ----------------- ----------------- ---------------- NGF (pg/ml) 199.5 1023.3 3198.9 3382.2 1398.6 Cl (64.7-615.2) (429.6-2437.2) (1692.8-6045.0) (1838.6-6221.6) (504.5-3827.4) *P* value ≤ 0.05 ≤ 0.05 ≤ 0.05 ≤ 0.05 IL-6 (pg/ml) 4.1 40.4 167.5 252.4 66.1 Cl (1.4-12.2) (17.0-96.3) (101.6-276.0) (152.1-418.8) (39.9-109.6) *P* value ≤ 0.05 ≤ 0.05 ≤ 0.05 ≤ 0.05 IL-8 (pg/ml) 5.5 36.0 111.1 97.2 20.3 Cl (2.0-15.2) (20.6-62.7) (74.9-165.0) (57.0-165.6) (8.1-50.9) *P* value ≤ 0.05 ≤ 0.05 ≤ 0.05 ≤ 0.05 Data are expressed as geometric means of log-transformed values with 95% confidence interval (CI). Single factor analysis of variance (ANOVA) and Student *t* test were used, as appropriate, to determine differences in NGF and cytokine levels at different time points. A *P* value ≤ 0.05 was accepted as significant. Conclusions =========== NGF is released following CPB in paediatric patients, with a similar pattern to the one observed for known mediators of the systemic inflammatory response to CPB. Further studies, both in paediatric and adult patients, are needed to elucidate the role of NGF in the pathophysiology of the body response to CPB and to evaluate its clinical impact.

**To the Editor:** In the past 30 years, war in the Balkans, the fall of Communist regimes, and economic recession in Europe have undermined the economic stability of countries in eastern Europe and eventually favored occurrence of so-called neglected infections of poverty ([@R1]). Parasitic infections causing eye disease in persons living in areas with low socioeconomic standards might be caused by parasites not well known by healthcare providers. A good example is *Thelazia callipaeda* (Spirurida, Thelaziidae) nematode infections in children and elderly persons living in rural and poor communities in countries in Europe and Asia ([@R2]). In Europe, vectors for this nematode are male *Phortica variegata* drosophilids, which feed on ocular secretions of hosts and transmit infective stage larvae to domestic and wild carnivores, lagomorphs, and humans ([@R3]). Possible outcomes of this infection include conjunctivitis, lacrimation, corneal ulcers, perforation, and blindness ([@R3]), but differentiating *T. callipaeda* infection from other ocular conditions, such as conjunctivitis-causing pathogens and allergies, can be difficult because signs and symptoms might be similar. *T. callipaeda* was previously known as the oriental eyeworm because of its original description in countries in eastern Asia (e.g., China, Japan, and Thailand), where it has caused \>1,000 cases of human infections in the past 2 decades ([@R2]). Since 1989, this nematode has also been detected in many countries in Europe, including Italy, France, Spain, Portugal, Switzerland, Germany, and Greece, as an agent of animal and human ocular infection ([@R3]). However, data on the occurrence of this parasite in countries in eastern Europe were not available until 2014. Over the past 2 years, several autochthonous cases of ocular thelaziosis in dogs and cats (Romania, Croatia, Serbia, Bosnia and Herzegovina, Bulgaria) and foxes (Bosnia and Herzegovina) were reported ([@R4]--[@R7]) ([Table](#T1){ref-type="table"}). In 2016, the zoonotic potential of this parasite in those regions was further confirmed by 2 human cases of thelaziosis, one in a 36-year-old man living in Serbia ([@R7]) and one in an 82-year-old man living in Croatia ([@R8]) ([Table](#T1){ref-type="table"}). ###### Cases of thelaziosis reported in animals and humans in eastern Europe Country Host No. infected hosts Reference ------------------------ ------- -------------------- ----------------- Bosnia and Herzegovina Fox 51 ([@R5]) Bosnia and Herzegovina Dog 4 ([@R5]) Bosnia and Herzegovina Cat 1 ([@R5]) Croatia Dog 2 ([@R5]) Croatia Human 1 ([@R8]) Romania Dog 1 ([@R6]) Serbia Dog 6 ([@R4]*,*[@R7]) Serbia Cat 2 ([@R4]) Serbia Human 1 ([@R7]) Hungary Dog 1 This study Bulgaria Dog 9 This study We report 10 new cases of ocular infection by *T. callipaeda* in dogs living in Bulgaria (n = 9) and Hungary (n = 1). All animals had no history of travel outside their native countries and were brought to the Department of Parasitology (Stara Zagora, Bulgaria) and to a veterinary practitioner (Pécs, Hungary) with various ocular disorders (i.e., epiphora, conjunctivitis). Nematodes detected in the conjunctival sac were collected by flushing the sac with saline solution. These nematodes were then stored in 70% ethanol and morphologically identified according to the procedure of Otranto et al. ([@R9]). Molecular characterization by using PCR amplification and sequencing of a partial region of the cytochrome oxidase subunit 1 gene were performed as described ([@R10]). Nucleotide sequences were identical to those of *T. callipaeda* nematode haplotype-1 (GenBank accession no. AM042549), which is the only haplotype circulating in animals and humans in Europe. Our confirmed autochthonous cases of thelaziosis in Hungary and Bulgaria have extended the geographic distribution of *T. callipaeda* nematodes from neighboring countries (e.g., Bosnia and Herzegovina, Croatia, Romania and, Greece), where occurrence of the parasite in humans and animals was already documented. Cases of human thelaziosis are reported in areas where the infection is highly prevalent in animals ([@R3]). Although no large-scale prevalence study has been conducted in countries in eastern Europe, 51 (27.7%) of 184 foxes in Bosnia and Herzegovina were infected with *T. callipaeda* nematodes ([@R5]). Isolation of *T. callipaeda* eyeworms from dogs in Bulgaria and Hungary should increases awareness of medical and veterinary communities in countries in eastern Europe for this zoonotic parasitosis. Use of a One Health approach is imperative for preventing additional eyeworm infections in persons living in eastern Europe. *Suggested citation for this article*: Colella V, Kirkova Z, Fok É, Mihalca AD, Tasić-Otašević S, Hodžić A, et al. Increase in eyeworm infections in eastern Europe \[letter\]. Emerg Infect Dis. 2106 Aug \[*date cited*\]. <http://dx.doi.org/10.3201/eid2208.160792>

INTRODUCTION ============ Transitional cell carcinoma (TCC) of the upper urinary tract (UUT), including ureteral and renal pelvic TCC, is relatively uncommon. Renal pelvic TCC accounts for 5% of all urothelial tumors, 10% of all renal tumors, and is 3 to 4 times more common than ureteral TCC ([@B1]-[@B3]). Nephroureterectomy with bladder cuff excision is the standard treatment for UUT-TCC. In general, UUT-TCC shows high local or systemic failure, even after radical surgery ([@B4], [@B5]). T stage, tumor grade, and lymphovascular invasion have been suggested as prognostic factors of UUT-TCC. Of these factors, T stage is the most closely associated risk factor ([@B4]-[@B9]). The renal pelvis is surrounded by renal parenchyma and perirenal fat posterolaterally, but is covered with peripelvic fat anteromedially. Despite this structural complexity, all renal pelvic tumors that invade beyond muscularis into peripelvic fat or renal parenchyma are diagnosed as stage pT3 according to the TNM system by the American Joint Committee on Cancer (AJCC) ([@B10]). Some urologists and pathologists have been interested in this problematic issue, but few authors have completed studies on substaging pT3 renal pelvic TCC because of their rarity ([@B11], [@B12]). To elucidate the prognostic impact of peripelvic fat invasion in pT3 renal pelvic TCC, we retrospectively reviewed our single center experience with patients who had been treated for renal pelvic TCC by open surgery. MATERIALS AND METHODS ===================== The study population -------------------- We retrospectively reviewed patients who had been treated surgically for renal pelvic TCC at our institution between 1986 and 2004. Exclusion criteria included the presence of a distant metastasis at diagnosis, unresectable disease, and concomitant invasive bladder cancer. Patients who had concomitant ureteral cancer with a higher T stage than the renal pelvic lesion were also excluded. A study population of 128 consecutive patients was identified with 99 male patients (77.3%) and 29 female patients (22.7%). The median age was 63 yr (30-81) and the median follow-up duration was 56 months (range, 6 to 240). Treatments and follow-ups ------------------------- All patients had undergone radical nephroureterectomy, and lymphadenectomy was performed on selected patients when enlarged lymph nodes were encountered intraoperatively or noted on preoperative computed tomography. In this procedure, the hilar and regional nodes adjacent to the ipsilateral great vessel were resected. As postsurgical adjuvant therapy, cisplatin-based chemotherapy was administered to 22 patients. Patient follow-up was relatively uniform including surveillance cystoscopy, urine cytology, abdomen-pelvis computed tomography, and a chest radiography. These tests were performed at 3-month intervals for the initial 2 yr, every 6-months for the subsequent 3 yr, and annually after the initial 5 yr. In addition, whole-body bone scan was checked annually during follow-up. Study methods ------------- Clinical information was obtained by a retrospective review of medical records in all patients. Survival information was obtained from medical records and our cancer center database. At the last follow-up, 80 patients (62.5%) were alive, 39 patients (30.5%) had died of cancer, and 9 (7.0%) had died of other causes. Pathology slides were reassessed by one urologic pathologist (NHC). Nineteen tumors (14.9%) were stage Ta, 25 tumors (19.5%) were T1, 14 tumors (10.9%) were T2, 60 tumors (46.9%) were T3, and 10 tumors (7.8%) were T4, based on the 2002 American Joint Committee on Cancer (AJCC) TNM staging system ([@B10]). Ninety-three tumors (72.6%) were high-grade, and 35 tumors were low-grade (27.4%) according to the 1998 World Health Organization/ International Society of Urologic Pathologists (WHO/ ISUP) classification of papillary urothelial neoplasm ([@B13]). Therefore, 60 patients with pT3 disease were eligible for the analyses. The clinical and pathological characteristics of these patients are specified on [Table 1](#T1){ref-type="table"}. We investigated the impact of various clinicopathological factors on disease-specific survival. We assessed the following prognostic factors: sex, age, concomitant bladder tumors, concomitant ureter tumors, lymphadenectomy, adjuvant chemotherapy, tumor grade, multiplicity, parenchymal invasion, peripelvic fat invasion, lymph node invasion, carcinoma in situ, and lymphovascular invasion. Kaplan-Meier curves were generated and compared using the log-rank test for the univariate survival analyses. To assess the independent impact of clinicopathological factors on the disease-free survival, Cox proportional hazards regression was used for the multivariate survival analyses. The final result for the multivariate models was obtained using a stepwise forward selection strategy. SPSS for Windows version 12.0 was used for statistical analyses, and a two-sided *p* value of less than 0.05 was considered to be significant. RESULTS ======= Of the 60 patients with pT3 renal pelvic TCC, 30 patients (50.0%) were alive at the final follow-up, and 4 patients (6.7 %) had died of other causes. Twenty-six patients (43.3%) had died of renal pelvic TCC, and the median time to cancer- related death was 28 months (6-102 months). The 5-yr disease-specific survival rate was 59.0%, and the 10-yr disease- specific survival rate was 38.8%. According to the univariate analysis, sex, age, concomitant bladder tumors, concomitant ureter tumors, lymphadenectomy, adjuvant chemotherapy, tumor grade, multiplicity, parenchymal invasion and carcinoma in situ did not influence disease-specific survival. On the other hand, peripelvic fat invasion (*p*=0.038), lymph node invasion (*p*\<0.001), and lymphovascular invasion (*p*=0.017) were each significantly associated with disease-specific survival ([Fig. 1](#F1){ref-type="fig"}) ([Table 2](#T2){ref-type="table"}). Statistically proven prognostic factors including peripelvic fat invasion, lymph node invasion, and lymphovascular invasion were subjected to multivariate Cox proportional hazards regression analysis to assess the independent impact of these factors. Using this multivariate analysis, significant differences in disease-specific survival rates were found in patients with peripelvic fat invasion (*p*=0.012) and lymph node invasion (*p*=0.004). The hazard ratios of disease-specific survival were 2.90 for cases with peripelvic fat invasion and 3.70 in cases with lymph node invasion. However, cases with lymphovascular invasion did not show a significant difference through multivariate analysis ([Table 3](#T3){ref-type="table"}). Based on these results, we subdivided the conventional pT3 stage patients into those with peripelvic fat invasion and those without. The prognosis of those with peripelvic fat invasion was more similar to that of pT4 tumors than conventional pT3 tumors, but a statistical difference still existed between them (*p*=0.043). The prognosis of those without peripelvic fat invasion was similar to that of pT2 disease, and we could not find any difference in the prognosis of patients with pT2 tumors and pT3 disease without peripelvic fat invasion (*p*=0.453) ([Fig. 1](#F1){ref-type="fig"}). DISCUSSION ========== The current staging system of UUT-TCC has been challenged for many reasons. The pathologic findings, pathogenesis, clinical manifestation, and natural history of renal pelvic and ureteral TCC are similar, and thus they have been considered identical diseases. However, questions remain as to whether renal pelvic and ureteral TCC are similar in terms of prognosis because of their apparent anatomical differences. Moreover, several studies have shown that ureteral TCC is associated with a higher local or distant failure rate than renal pelvic TCC and that ureteral TCC is associated with a poorer prognosis ([@B7], [@B14], [@B15]). The ureter is surrounded by an extensive plexus of ureteral blood vessels and lymphatics, while on the other hand, the renal pelvis is more firmly covered by adjacent tissue, such as the kidney and perihilar adipose tissue. Park et al. suggested that this anatomical weakness of the ureter was responsible for their study results ([@B7]). Another problematic issue has been raised regarding renal pelvic TCC. The staging of renal pelvic TCC has frequently been discussed in the literature ([@B11], [@B12], [@B16], [@B17]). In 1971, Grabstald et al. classified 70 cases of renal pelvic tumors into 4 staging groups based on a previous staging of transitional cell carcinoma of the bladder ([@B16], [@B18]). In this study, group IV tumors were defined as those with peripelvic or perirenal fat extensions and adjacent structural invasion. In other words, those patients with peripelvic fat invasion (group IV) were regarded as having higher stage tumors than those with renal parenchymal invasion (group III) ([@B16]). Rubenstein et al. modified the Grabstald classification and also separated tumors with peripelvic fat invasion (stage C) from tumors with renal parenchymal extension (stage B) ([@B17]). Unfortunately, these series of studies on renal pelvic cancers were small, and although survival percentages for each stage were given, the differences between these stages could not be analyzed statistically. Nonetheless, the AJCC TNM system for UUT-TCC was formulated to standardize staging in 1988, but renal pelvic TCC with renal parenchymal or peripelvic fat invasion were both considered to be identical pT3 stages of renal pelvic TCC ([@B19]). Two previous reports have statistically evaluated the prognostic impact of peripelvic fat invasion. Yoshimura et al. demonstrated that the 3-yr cause-specific survival rates were 55.6% in patients with pT3 renal pelvic tumors containing peripelvic fat invasion and 64.9% in patients without this invasion ([@B12]). However, no statistical difference was found between the two groups. A study by Olgac et al. also reported that cases with hilar fat invasion and renal parenchymal invasion had similar outcomes ([@B11]). By contrast, in our study, the disease-specific survival rate of pT3 renal pelvic TCC showed a significant difference according to peripelvic fat invasion and lymph node invasion. The prognosis of pT3 tumors with peripelvic fat invasion was more similar to that of pT4 tumors than conventional pT3 tumors. Because of this adjustment, the survival curve of those without peripelvic fat invasion was more similar to pT2 disease. In other words, disease risk has been overestimated in a subset of pT3 renal pelvic TCC patients. Our results could be explained by a number of mechanisms. The anteriomedial portion of the renal pelvis is surrounded by an extensive plexus of ureteral blood vessels and lymphatics similar to the ureter. This led to the hypothesis that it is easier for those with peripelvic fat invasion than those without to invade lymph vessels. Another explanation may be a problem related to surgical techniques. Because some extent of hilar dissection is required for vascular control, a part of the peripelvic fat layer might be destroyed during this procedure. For this reason, a more meticulous hilar dissection should be required in renal pelvic TCC, especially in the tumors that are located in the anteriomedial part of the renal pelvis. Although adjuvant cisplatin-based chemotherapy did not improve oncological outcomes in the pT3 patients of our study ([Table 2](#T2){ref-type="table"}), more aggressive adjuvant chemotherapy should be considered for those with peripelvic fat invasion. Outcomes of this strategy, such as whether adjuvant chemotherapy has survival benefits for these patients, should be evaluated based on a larger study population. There are some limitations in our current study. To clarify risk factors in pT3 patients, we excluded tumors of other stages (pTa, pT1, pT2 and pT4) for the main analysis. In addition to their rarity, this exclusion criterion also caused a relatively small study population. Another limitation is that we did not perform routine lymphadenectomies. Of the 60 patients with pT3 renal pelvic cancers, the rate of lymph node positive pT3 tumors was 15%, though there is a possibility that this rate may be underestimated. Meanwhile, we classified patients with pN0 or pNx as the same entity during our univariate analysis because we could not identify any prognostic differences between them (data not shown). However, this factor might play a role as a confounding factor. Thus, a more refined study population through routine lymphadenectomy would be helpful for controlling these confounding variables, and a larger number of cases would be required to elucidate these observations. In conclusion, peripelvic fat invasion is a strong prognostic factor in pT3 renal pelvic TCC. More meticulous hilar dissections would be helpful in the surgical treatment of renal pelvic TCC, especially for the tumors located in the anteriomedial region of the renal pelvis. In addition, aggressive systemic chemotherapy should be considered in the presence of peripelvic fat invasion, even if the lymph nodes are not involved. {#F1} ###### Clinical and pathological characteristics of the patients with pT3 renal pelvic cancer  ###### Univariate analysis for the disease-specific survival rates  ###### Multivariate analysis for the disease-specific survival rates 

Itch (pruritus) is an unpleasant somatic sensation that elicits a desire to scratch[@b1]. Acute itch serves as a self-protective mechanism against potential harmful environmental irritants or parasites[@b2]. However, chronic itch is a debilitating symptom that arises from many systemic disorders, such as dermatologic diseases (e.g. atopic dermatitis and psoriasis), chronic kidney failure, chronic liver diseases (e.g. cholestasis), infections, and hematologic diseases[@b3][@b4]. Scratching transiently relieves acute itch[@b5], but has limited effects on chronic itch and paradoxically evokes itch-scratch-itch cycle. Chronic itch disrupts sleep and substantially reduces the quality of life of patients[@b1]. Antihistamines are often clinically prescribed for treating allergy itch; however, they are inefficient for many aforementioned chronic itch conditions[@b4]. Although the recent discovery for itch-specific neural pathway[@b6][@b7][@b8], novel itch mediators and receptors[@b9][@b10][@b11][@b12], greatly improves our understanding on acute itch[@b2], the pathogenesis of chronic itch associated with systemic disorders remains enigmatic. Cholestasis is defined as diminished delivery of bile into the intestine resulting either from a functional defect at hepatocyte level or obstruction at the bile duct level of any cause, such as primary biliary cirrhosis (PBC), primary sclerosing cholangitis (PSC), intrahepatic cholestasis of pregnancy (ICP), and benign recurrent intrahepatic cholestasis[@b13]. Infections, autoimmune, metabolic diseases and drug side effects often results in cholestasis[@b14]. Patients with cholestasis experience severe and intractable pruritus[@b15][@b16], and also exhibit altered pain perception[@b17][@b18]. Analgesia has been observed in cholestatic patients, for example, patients undergoing orthotopic liver transplantation for cholestatic diseases reported lower postoperative pain scores and required less morphine for analgesia than patients undergoing liver resection[@b19]. It was demonstrated peripheral endogenous opioid-mediated and naloxone-reversible analgesia in mice or rats with bile duct ligation, which induced obstructive cholestatic disease[@b20][@b21]. Recently, there were several important findings to elucidate the pathogenesis of cholestatic pruritus. For example, serum concentrations of lysophosphatidic acid (LPA) and autotoxin (the enzyme that forms LPA) were significantly increased in cholestatic patients with pruritus and could directly activate dorsal root ganglia (DRG) neurons reflected by increased intracellular calcium[@b22], suggesting LPA is a potential mediator of cholestatic pruritus. Bile acids, such as deoxycholic acid (DCA) and taurolithocholic acid (TLCA), bind to TGR5 (also known as GPR131 or GPBAR1) and activate transient receptor potential cation channel A1 (TRPA1) in primary sensory neurons and subsequently stimulates the release of itch-related neuropeptides, such as gastrin-releasing peptide (GRP) and natriuritic precursor peptide B (NPPB), in the spinal cord to transmit itch in mice[@b17][@b23]. Previous report also demonstrated that protease-activated receptor 2 (PAR2)-induced sensitization of transient receptor potential cation channel V1 (TRPV1) contributed to the pathogenesis of cholestatic pruritus in rats[@b24]. Although several novel itch mediators (LPA, bile acids and PAR2 agonists) and receptors (TGR5, TRPV1/A1, and PAR2) were demonstrated to be involved in cholestatic itch, the peripheral and central mechanisms underlying cholestatic pruritus and antinociception are still not completely understood. Among monoamine neurotransmitters, serotonin, or 5-hydroxytryptamine (5-HT), is considered to play key and complex role in pain sensation[@b25], and recently it was also demonstrated its crucial role in itch sensation[@b26]. To date, seven classes of 5-HT receptors (5-HT~1~-5-HT~7~) have been identified and comprise at least 15 subtypes[@b25]. Except 5-HT~3~ receptor as a ligand-gated cation channel, all other 5-HT receptors are G protein-coupled receptors (GPCRs)[@b25]. As 5-HT is not able to penetrate the blood-brain-barrier (BBB), peripheral and central 5-HT systems are considered to be separated compartments and employ distinct rate-limiting enzymes for 5-HT synthesis, such as tryptophan hydroxylase 1 (*Tph1*) in the skin and tryptophan hydroxylase 1 (*Tph2*) in the brain[@b26]. 5-HT in the skin released from mast cells, as a critical component of "inflammatory soup", was identified as a potent inducer of pain or itch via activation distinct 5-HT receptors subtypes, such as 5-HT~2A~, 5-HT~2B~, 5-HT~2C~, 5-HT~3~, 5-HT~4~, 5-HT~6~, and 5-HT~7~[@b27][@b28][@b29]. Descending 5-HT pathways, originating from the nucleus raphe magus (NRM) within the vicinity of the rostral ventromedial medulla (RVM), may exert either inhibitory or facilitatory influence on pain or itch transmission in spinal cord dorsal horn, possible dependent on the physiological or pathophysiological status[@b26][@b30][@b31]. The discovery of multiple 5-HT receptors subtypes may explain their complex influence on the processing of pain or itch signal[@b25][@b28][@b30][@b32][@b33]. Interestingly, several clinical trials found that administration of 5-HT~3~ receptor antagonist ondansetron[@b34][@b35][@b36][@b37][@b38] or selective serotonin reuptake inhibitor (SSRI) sertraline[@b39][@b40] were able to alleviate cholestatic pruritus, providing important clinical evidence to support critical role of 5-HT system in cholestatic pruritus. Although the importance of 5-HT system is emphasized in the modulation of pain and itch[@b25][@b26][@b41], the possible roles of peripheral and spinal 5-HT receptors in itch and antinociception induced by cholestasis are still unclear. The aim of the present study was to reveal the possible roles of peripheral and spinal 5-HT receptors in itch and antinociception in common bile duct ligation (BDL) rats, a severe model of obstructive cholestasis. We firstly demonstrated behavioral phenotypes for itch and antinociception in BDL rats and then examined the changes of the 5-HT level in the skin and spinal cord. Furthermore, we used quantitative real-time polymerase chain reaction (qPCR) to screen the expression change of difference 5-HT receptors subtypes at the peripheral and central nervous systems in BDL rats. Finally, we employed several pharmacological 5-HT receptors agonists and antagonists to elucidate the distinct roles of 5-HT receptors subtypes, especially 5-HT~1A~, 5-HT~2~, 5-HT~3~ and 5-HT~7~, in the modulation of itch and antinociception in BDL rats. Materials and Methods ===================== Animals ------- Adult male Sprague-Dawley rats (weighing 200--250 g) were obtained from the Shanghai SLAC Laboratory Animal CO. LTD. Rats were housed in groups of 3 rats per cage with food and water available *ad libitum* and kept in controlled room temperature (22 ± 2 °C) and humidity (60--80%) under a 12 h/12 h light/dark cycle. All experimental procedures and animal handing were performed in accordance with the guidelines of the International Association for the Study of Pain and the animal protocols were approved by Soochow University Animal Committee. The authors tried all efforts to minimize the number of animals used. Surgical laparotomy and common bile duct ligation (BDL) ------------------------------------------------------- The surgical laparotomy and common bile duct ligation was performed as previously described[@b42][@b43]. Briefly, laparotomy was performed under anesthesia with isoflurane and common bile duct was ligated using two ligatures. Rats with laparotomy, bile duct identification without ligation were used as sham rats. The body temperature of rat was maintained at 37 ± 1 °C during the entire surgical procedure. Cholestasis was confirmed by increased serum level of bilirubin as well as the intact ligature and proximal dilation of the common bile duct at the sacrifice time. Drugs and administration ------------------------ Serotonin hydrochloride (5-HT), deoxycholic acid (DCA), 5-HT~2~ receptor agonist α-methylserotonin maleate salt (α-methyl-5-HT), 5-HT~3A~ receptor agonist 2-Methyl-5-hydroxytryptamine hydrochloride (2-Methyl-5-HT), and the 5-HT~2A~ receptor antagonist ketanserin tartrate salt (ketanserin) was obtained from Sigma-Aldrich (St. Louis, MO, USA). The 5-HT~1A~ receptor agonist 8-hydroxy-2-(dipropylamino) tetralin hydrobromide (DPAT), the 5-HT~1A~ receptor antagonist WAY-100635 maleate salt, the 5-HT~3A~ receptor agonist 2-Methyl-5-hydroxytryptamine hydrochloride (2-Methyl-5-HT), 5-HT~7~ receptor antagonist (2R)-1-\[(3-Hydroxyphenyl)sulfonyl\]-2-\[2-(4-methyl-1-piperidinyl)ethyl\]pyrrolidine hydrochloride (SB269970) were obtained from Tocris Bioscience (Bristol, UK). 5-HT~3A~ receptor antagonist Ondansetron Hydrochloride Injection was obtained from China Jiangsu yabang Pharmaceytical factory CO. Ltd. (Jiangsu, China). The 5-HT re-uptake inhibitor Fluoxetine Hydrochloride Dispersible Tablets (PROZAC) was obtained from Patheon France. Ketanserin, WAY-100635 and Fluoxetine were dissolved in 10% DMSO and other reagents were dissolved in sterile saline if not specified. The information of 5-HT receptor agonists and antagonists used in this study was list in [Table 1](#t1){ref-type="table"}. After BDL surgery 5, 10, 15, and 30 days, intradermal injection of 5-HT (200 μg) into the rat cheek (total volume 20 μl) or neck (total volume 50 μl) was perform to assess itch response in the sham and BDL rats. After BDL surgery 15 days, intradermal injection 5-HT receptors agonists into the rat cheek (total volume 20 μl), including 5-HT~1A~ receptor agonist 8-hydroxy-2-(dipropylamino) tetralin hydrobromide (DPAT; 100 μg), 5-HT~2A~ receptor agonist α-methylserotonin maleate salt (α-methyl-5-HT; 100 μg), 5-HT~3~ receptor agonist 2-Methyl-5-hydroxytryptamine hydrochloride (2-Methyl-5-HT; 100 μg), 5-HT~7~ receptor agonist 4-\[2-(Methylthio)phenyl\]-N-(1,2,3,4-tetrahydro-1-naphthalenyl)-1-piperazinehexanamide hydrochloride (LP44; 100 μg) was perform to assess itch response in the sham and BDL rats. After BDL surgery 15 days, we also examined the effects of intrathecal injection of 5-HT receptors agonists on itch and antinociception in the sham and BDL rats. The doses used were as follow: DPAT (10 μg), α-methyl-5-HT (10 μg), 2-Methyl-5-HT (10 μg), and LP44 (10 μg). After BDL surgery 15 days, the effects of 5-HT receptors antagonists on itch and antinociception in the sham and BDL rats were assessed by intrathecal (i.t.) or intraperitoneal (i.p.) injection of them, including 5-HT~1A~ receptor antagonist WAY-100635 maleate salt (i.t. 10 μg), 5-HT~2A~ receptor antagonist ketanserin tartrate salt (i.p. 1 mg/kg), 5-HT~3~ receptor antagonist ondansetron hydrochloride (i.p. 3 mg/kg), and 5-HT~7~ receptor antagonist (2R)-1-\[(3-Hydroxyphenyl)sulfonyl\]-2-\[2-(4-methyl-1-piperidinyl)ethyl\]pyrrolidine hydrochloride (SB269970; i.t. 10 μg). The injection of 5-HT receptors agonists or antagonists was performed before 30 min of i.d. injection of 5-HT (200 μg) and the doses used in this study were based on previous reports[@b29][@b59][@b60][@b61][@b62] and our pilot experiments. Neck model of itch ------------------ As described previously[@b12][@b59][@b63], rats were shaved at the nape of the neck at least 2 day before experiments. On the day of behavioral testing, rats were individually placed in small plastic chambers (15 × 20 × 25 cm) on an elevated metal mesh floor and allowed at least 60 min for habituation. Under brief anesthesia of isoflurane, rats were given an intradermal injection of 50 μl of 5-HT (200 μg) via a 26 G needle into the nape of the neck. Immediately after the injection, rats were returned to their chambers and were video recorded for 60 min. The video was subsequently played back offline and itch behavior was quantified by counting the number of scratches in a blinded manner. A scratch was counted when a rat lifted its hind paw to scratch the shaved region and returned the paw to the floor or to the mouth for licking. The neck model of itch is usually used to test the possible effects of intrathecal or systemic injection of drugs. Cheek model of itch ------------------- As described previously[@b64][@b65][@b66], we intradermally injected chemicals into the cheek of rats. Rats were shaved on cheeks (approx. 5 × 8 mm area) at least 2 day before the experiment. On the day of experiment, rats were intradermally injected of 20 μl of 5-HT (200 μg) via a 26G needle into the cheek under brief anesthesia with isoflurane. After the injection, rats were immediately returned to their chambers and were video recorded for 60 min. The video was played back and the wipes and scratches were quantified by counting their number in a blinded manner. Scratching is counted when rats scratch the injection site on the cheek by hind paw and then returning to the floor or to the mouth. Wiping is counted when rats unilaterally wipe the injected site using the forelimb, which was not part of grooming behavior. As intrathecal injection of drugs by lumbar puncture may be difficult to reach the central nerve system that innervates the cheek area, the cheek model of itch is usually used to test the possible effects of systemic injection of drugs. Hargreaves test --------------- According to previously report[@b67], we put the rats in plastic boxes and measured the hindpaw withdrawal latency to radiant heat apparatus (IITC Life Science). Rats were placed on a glass floor maintained at 25 °C in a clear plastic chamber and a radiant heat source, which was located under the glass floor, was focused onto the plantar surface of the hindpaw. We measured paw withdrawal latency 2--3 times for each hindpaw, and we used the mean of the results for analysis. We set a cut off of 20 s to prevent potential tissue injury. Tail immersion test ------------------- As previously described[@b43][@b66], tail immersion test was employed to determine heat pain sensitivity in rats. Briefly, the terminal 3 cm of a rat's tail was immersed in hot water bath at 52 °C and the latency of tail flick was recorded with a cutoff time of 15 seconds to avoid tissue damage. von Frey filament test ---------------------- To test mechanical hypersensitivity, we determined the mechanical sensitivity of hind paws by using series of von Frey filaments (4 g, 15 g, and 26 g) as previous report[@b66]. Rats were placed on a metal mesh floor and von Frey filaments were applied from underneath the floor. The plantar surface of hind paw was received consecutively by 10 stimulations. The measurement was repeated twice at an interval of 10 minutes. We used response frequency to estimate the mechanical sensitivity of rats. Rota-rod test ------------- We used Rota-rod test to evaluate the motor function of rats at day 15 after BDL surgery[@b43][@b68]. Before testing 2 days, all of the rats were trained at various speeds of rotation. Each rat was placed on the rungs and allowed to remain stationary at 0 rpm for 10 sec, after which the speed was increased to 5 rpm for a maximum of 5 min. Rats were tested at 5, 10 until 25 rpm, during this time, rats that fell off the cylinder were placed back on the rotarod for three times. All rats were tested at 25 rpm for the 5-min period for each animal. Rats were tested for three trials with an interval of 20 min and the latency for falling was averaged. RNA isolation and quantitative real-time polymerase chain reaction (qPCR) ------------------------------------------------------------------------- The sham and BDL rats were sacrificed at 15 days after surgery. The rat brainstem, trigeminal ganglia, cervical DRGs and spinal cord were dissected out and collected. Total RNA was extracted by homogenizing tissues using Trizol Reagent (Invitrogen, Carlsbad, California) according to the protocol supplied by manufacturer. Chloroform (Sigma-Aldrich) was added after homogenization and the tubes were vortexed, followed by incubation at room temperature for 5 min and centrifugation at 14,000 rpm for 20 min at 4 °C. The supernatant was transferred to a new tube and isopropanol was added. The aqueous phase was centrifuged at 14,000 RPM for 20 min at 4 °C. Pellets were washed using 70% ethanol and resuspended in diethylpyrocarbonate (DEPC)-treated water. The purity and concentration of RNA were determined with NanoDrop ND-1000 Spectrophotometer (NanoDrop Technologies, Wilmington, DE, USA) with absorbance at 260 and 280 nm. One microgram of total RNA was reverse transcribed for synthesizing cDNA using RevertAid First Strand cDNA Synthesis Kit, according to the protocol supplied by the manufacturer (Thermo Fisher Scientific, Waltham, USA). qPCR experiment was performed by SYBR Green PCR Master Mix (Roche, Basle, Switzerland) using Opticon real-time PCR Detection System (ABI 7500, Life technology, USA). Data were normalized to the housekeeping gene β-actin. The 2(-Delta Delta C(t)) method was used to analyze the relative level of gene expression. Primers used are listed in [Table 2](#t2){ref-type="table"}. Immunofluorescence ------------------ Rats were deeply anesthetized with chloral hydrate and intracardially perfused first with saline, then with 4% of the fixative solution paraformaldehyde in 0.1 M phosphate buffer (pH = 7.4). The L4-6 lumbar spinal cords were collected and postfixed in the same solution overnight. The spinal cords were transferred into a phosphate-buffered 15% sucrose solution overnight. After that, they were transferred into a phosphate-buffered 30% sucrose solution overnight. The spinal cord sections were cut at the thickness of 14-μm in a cryostat. The sections were mounted on siliconized slides for immunostaining. Sections were rinsed three times for 10 min with PBS, and then all incubated in PBS containing 10% normal goat serum (GIBCO) and 0.3% Triton X-100 (Sigma) for 1 h. Then, the sections were incubated overnight at 4 °C in rabbit polyclonal anti-5-HT antibody (diluted 1:500; Sigma, S5545, USA). We used the same anti-5-HT antibody as Prof. Yaobo Liu's laboratory and the specificity of the antibody had been demonstrated by his lab[@b69]. After this, the sections were rinsed three times for 10 min with PBS and then incubated in a solution containing a goat anti-rabbit secondary antibody conjugated to Alexa Fluor 488 (1:500; Molecular Probes) for 1 h. They were then rinsed three times for 10 min with PBS, quickly rinsed with ddH~2~O, and left to dry on a plate at 37 °C for 15 min. Finally, the slice sections were mounted on glass slides and coverslipped with a drop of mounting medium (Dako North America Inc., Carpinteria, CA, USA). The coverslip was sealed with nail polish for preventing drying and movement. All slice sections were stored in dark at 4 °C. The sections were observed and photographed were examined under a Zeiss fluorescence microscope AXIO SCOPE A1 (Oberkochen, Germany), and images were analyzed with AxioVision software. Histology --------- Rats were terminally anesthetized with isoflurane and transcardially perfused with phosphate buffersaline (pH 7.4, 0.1 M) followed by fixation with 4% (w/v) paraformaldehyde. Liver samples were post-fixed in 4% (w/v) formalin prepared in PBS for 24 h. This was followed by the dehydration of fixed tissue in various grades of alcohol (70%, 90%, 100%, v/v) and then cleared in benzene. Samples were removed; transverse sections were embedded in molten paraffin wax and 5 μm thick sections were cut using a microtome. Liver sections were stained with standard H&E staining. The images were analyzed using Adobe PhotoShop and the average number of bile duct-like structures per high-power field (HPF) was quantified. High performance liquid chromatography (HPLC) analysis for 5-HT --------------------------------------------------------------- Rats were terminally anesthetized with isoflurane on day 15 after BDL surgery. The skin and lumbar spinal cords were dissected quickly, weighed and then homogenized with an ultrasonic homogenizer (Microsonic, Dortmund, Germany) in 400 μl 0.4 mol/L perchloric acid. Samples were then centrifuged at 15000 rpm at 4 °C for 10 min, filtered through the 0.22 μm syringe filter and stored at −80 °C until HPLC analysis. The concentration of 5-HT were measured by applying reverse-phase HPLC with electrochemical detection (Waters, USA). The reversed phase column was YWG-C~18,~ which was perfused for analysis with amobile phase composed of 0.1 mol/L NaAc (including 0.1 mol/L EDTA-Na~2~) and 10% methanol at pH 5.1. The flow rate was 1 mL/min. The data was quantified using the area under the peaks and external standards. The obtained results were presented in ng per gram of wet tissue (ng/g). Statistical analysis -------------------- All data were analyzed using GraphPad Prism 6 software (GraphPad, San Diego, CA, USA). All data were expressed as the mean ± S.E.M. The statistical significance between two groups was analyzed by unpaired two-tailed Student's *t*-test. One-way ANOVA followed by post-hoc Bonferroni test was used for multiple comparisons. Two-way repeated-measured ANOVA was also used to analyze the data with multiple time points. Differences with *P* \< 0.05 were considered as statistical significance. Results ======= Obstructive cholestasis model was established by common bile duct ligation (BDL) in rats ---------------------------------------------------------------------------------------- We firstly established an obstructive cholestatic rat model by common bile duct ligation (BDL). The biochemical and morphological evidence of liver injury and cholestasis were confirmed in BDL rats by multiple approaches ([Fig. 1](#f1){ref-type="fig"}). H&E staining of liver section in BDL rats showed typical features of cholestasis, including expansion of the liver capillary bile duct, fibrosis around the small bile duct, and connection of the portal area ([Fig. 1A](#f1){ref-type="fig"}). The bile duct-like structures in liver sections from BDL rats were increased compared to the sham rats ([Fig. 1B](#f1){ref-type="fig"}). In addition, qPCR analysis showed that the mRNA expression of CK-7, which is a specific cholangiocytes marker, and proliferating cell nuclear antigen (PCNA), which is a cell proliferation-related marker, were significantly increased in BDL rats compared to the sham rats ([Fig. 1C](#f1){ref-type="fig"}). Thus, the results showed that the bile duct had obvious hyperplasia in the BDL rats at the 4^th^ week after surgery. The increased serum total bilirubin (BR) was also demonstrated in BDL rats ([Fig. 1D](#f1){ref-type="fig"}; t~8~ = 2.508; *P* = 0.0365). In accordance with previous study[@b70][@b71], There was no significant difference for the body weight between the sham and BDL rats within 1 to 15 days after BDL surgery ([Fig. 1E](#f1){ref-type="fig"}; F~(1,52)~ = 0.1752; *P* = 0.6772). 5-HT-evoked itch behavior was significantly enhanced in BDL rats ---------------------------------------------------------------- Previous report by Belghiti M *et al*. demonstrated spontaneous scratching behavior was significantly increased in BDL rats[@b24], suggesting BDL rat may be a suitable model for exploring the mechanisms underlying cholestatic itch. Aimed to replicate this cholestatic itch model in rats, we carefully observed and quantified the spontaneous behaviors after BDL surgery, such as scratching, grooming, and biting behaviors, which may reflect chronic itch in rats[@b24]. Surprisingly, none of aforementioned itch-related behaviors showed significant difference between the sham and BDL rats ([Fig. 2A](#f2){ref-type="fig"}). This discrepancy may be attributed to the different rat stains employed: we used SD rats while Belghiti M *et al*. used Wistar rats. Our results argued lack of BDL-associated spontaneous itch in SD rats. To further determine whether chemical-evoked itch sensation was changed in BDL rats, we next investigated whether 5-HT-evoked itch was affected in BDL rats. After BDL surgery 15 to 30 days, scratching induced by intradermal (i.d.) injection of 5-HT into the nape of the neck was significant enhanced ([Fig. 2B](#f2){ref-type="fig"}; F~(1,22)~ = 45.76; *P* \< 0.0001) in BDL rats. In order to distinguish itch and pain behaviors in rats, we used the cheek model by i.d. injection of chemicals into cheek of rats, which demonstrated that painful stimuli elicit forelimb wiping, while itchy stimuli elicit hindlimb scratching[@b72]. After BDL surgery 10 to 30 days, scratching induced by i.d. injection of 5-HT into the cheek was significant enhanced ([Fig. 2C](#f2){ref-type="fig"}; F~(1,43)~ = 34.88; *P* \< 0.0001) in BDL rats. In contrast, there was no significant difference for the wiping behavior induced by 5-HT in cheek model between the sham and BDL rats ([Fig. 2C](#f2){ref-type="fig"}; F~(1,43)~ = 1.775, *P* = 0.1898). Antinociception responding to mechanical and heat stimuli was induced in BDL rats --------------------------------------------------------------------------------- We next asked whether pain sensation was affected in BDL rats. After BDL surgery 6 to 30 days, the mechanical sensitivity, evaluated by von Frey test, was significantly lower in BDL rats than that in sham rats ([Fig. 3](#f3){ref-type="fig"}A; [4](#f4){ref-type="fig"}g: F~(1,125)~ = 39.40, *P* \<  0.0001; 15 g: F~(1,125)~ = 44.21, *P* \< 0.0001; 26 g: F~(1,126)~ = 19.52, *P* \< 0.0001). The heat sensitivity was determined by Hargreaves test and tail-flick test. The results showed the paw thermal withdrawal latency to radiated heat was significantly increased from 2 to 30 days following BDL surgery in rats ([Fig. 3B](#f3){ref-type="fig"}; F ~(1,104)~ = 60.80, *P* \< 0.0001). Tail-flick latency to 52 °C hot water was also significantly prolonged in BDL rats ([Fig. 3C](#f3){ref-type="fig"}; F ~(1,200)~ = 350.2, P \< 0.0001). After BDL 30 days, there was no significant difference for the falling latency evaluated by Rota-rod test between sham and BDL rats ([Fig. 3D](#f3){ref-type="fig"}), suggesting that pain or itch phenotypes in BDL rats may not be attributed to the motor coordination dysfunction. The levels of 5-HT in the skin and spinal cord were increased in BDL rats ------------------------------------------------------------------------- We subsequently evaluated the changes of 5-HT level in the skin and the lumbar spinal cord in sham and BDL rats by using immunofluorescence and high-performance liquid chromatography (HPLC) analysis. The immunostaning of 5-HT in the spinal cord dorsal horn was increased in BDL rats compared to sham rats ([Fig. 4A,B](#f4){ref-type="fig"}; t~14~ = 5.707, *P* \< 0.0001). As shown in [Fig. 4](#f4){ref-type="fig"}, HPLC analysis also confirmed that 5-HT level in the skin (t~7~ = 2.435, *P* = 0.0451) and the spinal cord (t~6~ = 2.594, *P* = 0.0410) significantly increased in BDL rats ([Fig. 4C](#f4){ref-type="fig"}). To determine the origin of increased 5-HT levels in the skin and spinal cord, we further used qPCR to examine the mRNA expression of tryptophan hydroxylase 1 (*Tph1*) in the skin and tryptophan hydroxylase 1 (*Tph2*) in the brain stem, the rate-limiting enzymes for 5-HT synthesis[@b26]. The results showed that the expression of both *Tph1* in the skin and *Tph2* in the brain stem were up-regulated in BDL rats ([Fig. 4D](#f4){ref-type="fig"}). Thus, the results suggested that increased 5-HT level in the skin and the spinal cord, possible due to the up-regulated 5-HT synthesis, may actively participate in itch and antinociception under cholestasis condition. The expression profile of 5-HT receptor subtypes changed in peripheral and central nervous system (CNS) ------------------------------------------------------------------------------------------------------- We next used qPCR to examine the expression profile changes of 5-HT receptor subtypes in peripheral and central nervous system (especially spinal cord and brain stem). The results demonstrated that mRNA expression of multiple 5-HT receptors, including 5-HT~1B~, 5-HT~1D~, 5-HT~2A~, 5-HT~2C~, 5-HT~3A~, 5-HT~5A~, 5-HT~5B~, 5-HT~6~, and 5-HT~7~, were up-regulated in dorsal root ganglia (DRG) from BDL rats compared to sham rats ([Fig. 5A](#f5){ref-type="fig"}). The results also demonstrated that mRNA expression of multiple 5-HT receptors, including 5-HT~1B~, 5-HT~1D~, 5-HT~1F~, 5-HT~2A~, 5-HT~2B~, 5-HT~3A~, 5-HT~5B~, 5-HT~6~, and 5-HT~7~, were up-regulated in trigeminal ganglia (TG) from BDL rats compared to sham rats ([Fig. 5B](#f5){ref-type="fig"}). In sharp contrast, the mRNA expression of multiple 5-HT receptors, including 5-HT~1A~, 5-HT~1F~, 5-HT~2B~, and 5-HT~3A~, were down-regulated in the spinal cord from BDL rats compared to sham rats ([Fig. 5C](#f5){ref-type="fig"}). Similarly, the mRNA expression of multiple 5-HT receptors, including 5-HT~1A~, 5-HT~1F~, 5-HT~2B~, 5-HT~2C~, and 5-HT~3A~, were down-regulated in the brain stem (especially cervicomedullary junction) from BDL rats compared to sham rats ([Fig. 5D](#f5){ref-type="fig"}). Thus, these results suggested up-regulation of 5-HT receptors in primary sensory neurons in DRG or TG and down-regulation of 5-HT receptors in the spinal cord or brainstem may contribute to the enhanced itch behavior and antinociception in BDL rats. We subsequently performed pharmacological experiments to investigate the possible roles of four 5-HT receptors subtypes, including 5-HT~1A~, 5-HT~2~, 5-HT~3~ and 5-HT~7~, in itch and antinociception in BDL rats. Peripheral 5-HT receptors, especially 5-HT~2~, 5-HT~3~ and 5-HT~7,~ contributed to enhanced itch response in BDL rats --------------------------------------------------------------------------------------------------------------------- We tried to identify which subtypes of 5-HT receptors in the periphery are involved in the enhanced itch behavior in BDL rats through i.d. injection of 5-HT receptors agonists into rat cheek. We found that i.d. injection of 5-HT~1A~ receptor agonist 8-hydroxy-2-(dipropylamino) tetralin hydrobromide (DPAT; 100 μg) failed to induce scratching behavior in both sham and BDL rats ([Fig. 6A](#f6){ref-type="fig"}). In contrast, i.d. injection of 5-HT~2~ receptor agonist α-methylserotonin maleate salt (α-methyl-5-HT; 100 μg) was able to induce scratching behavior in both sham and BDL rats ([Fig. 6B](#f6){ref-type="fig"}; for sham: t~11~ = 12.43; *P* \< 0.0001; for BDL: t~12~ = 46.09; *P* \< 0.0001). Interestingly, i.d. injection of 5-HT~3~ receptor agonist 2-Methyl-5-hydroxytryptamine hydrochloride (2-Methyl-5-HT; 100 μg) could only induce scratching response in BDL rats, but not in sham rats ([Fig. 6C](#f6){ref-type="fig"}; for sham: t~12~ = 1.156; *P* = 0.2702; for BDL: t~11~ = 5.216; *P* = 0.0003). Finally, i.d. injection of 5-HT~7~ receptor agonist 4-\[2-(Methylthio)phenyl\]-N-(1,2,3,4-tetrahydro-1-naphthalenyl)-1-piperazinehexanamide hydrochloride (LP44; 100 μg) was also able to induce scratching behavior in both sham and BDL rats ([Fig. 6D](#f6){ref-type="fig"}; for sham: t~9~ = 11.93; *P* \< 0.0001; for BDL: t~8~ = 11.52; *P* \< 0.0001). As recent work showed that bile acid, such as deoxycholic acid (DCA) was able to induce itch behavior in mice via activation of TGR5 and TRPA1 in primary sensory neurons in DRG, suggesting bile acid may contribute to cholestatic itch[@b17][@b23]. We also tried to investigate the possible itch-inducing effect of DCA in rats and the interaction of peripheral bile acid and 5-HT system. Unexpected, i.d. injection of DCA (1 to 200 μg) into cheek failed to induce scratching behavior in rats ([Fig. 6E](#f6){ref-type="fig"}). Thus, although bile acid can induce itch in mice[@b17][@b23], our results clearly showed that bile acid was not able to induce scratching in rats, suggesting species difference for bile acid-induced itch. Interestingly, co-administration of DCA was able to enhance 5-HT-induced scratching in rats ([Fig. 6E](#f6){ref-type="fig"}; t~9~ = 3.999; *P* = 0.0031), suggesting DCA may potentiate 5-HT-induced itch in rats. Increasing level of 5-HT in the central nervous system (CNS) inhibited itch and induced antinociception in sham and BDL rats ---------------------------------------------------------------------------------------------------------------------------- After we demonstrated the role of peripheral 5-HT and 5-HT receptors in itch, we subsequently investigate the role of 5-HT in the CNS for modulating itch and antinociception in sham and BDL rats. Firstly, we found that intrathecal (i.t.) injection of 5-HT (1 μg) could significantly inhibit i.d. injection of 5-HT (200 μg)-induced scratching in both sham and BDL rats ([Fig. 7A](#f7){ref-type="fig"}; for sham: t~12~ = 6.854; *P* \< 0.0001; for BDL: t~9~ = 4.003; *P* = 0.0031). Additionally, i.t. injection of 5-HT could significantly increase the tail-flick latency response to 52 °C hot water in naïve rats ([Fig. 7B](#f7){ref-type="fig"}; F~(1,48)~ = 41.96, *P* \< 0.0001). We also found that intraperitoneal (i.p.) injection of 5-HT re-uptake inhibitor fluoxetine (10 mg/kg) could significantly suppressed 5-HT-induced scratching in sham and BDL rats ([Fig. 7C](#f7){ref-type="fig"}; for sham: t~8~ = 6.442; *P* = 0.0002; for BDL: t~10~ = 14.62; *P* \< 0.0001). Finally, i.p. injection of fluoxetine also significantly increased the tail-flick latency response to 52 °C hot water in sham and BDL rats ([Fig. 7D](#f7){ref-type="fig"}; for sham: F~(1,80)~ = 163.6, *P* \< 0.0001; for BDL: F~(1,100)~ = 167.2, *P* \< 0.0001). Thus, increasing 5-HT level in CNS suppressed itch and induced antinociception in sham and BDL rats. Itch could be modulated by 5-HT receptors agonists or antagonists in sham and BDL rats -------------------------------------------------------------------------------------- In order to reveal the distinct roles of 5-HT receptor subtypes on itch response in sham and BDL rats, 5-HT receptors agonists or antagonists were administrated after BDL surgery 15 days. It was found that i.t. injection of 5-HT~1A~ receptor agonist DPAT ([Fig. 8A](#f8){ref-type="fig"}; for sham: t~10~ = 5.799; *P* = 0.0002; for BDL: t~14~ = 5.102; *P* = 0.0002), 5-HT~2~ receptor agonist α-methyl-5-HT ([Fig. 8B](#f8){ref-type="fig"}; for sham: t~10~ = 8.398; *P* \< 0.0001; for BDL: t~11~ = 11.18; *P* \< 0.0001), 5-HT~3~ receptor agonist 2-Methyl-5-HT ([Fig. 8C](#f8){ref-type="fig"}; for sham: t~8~ = 9.602; *P* \< 0.0001; for BDL: t~9~ = 2.433; *P* = 0.00378), and 5-HT~7~ receptor agonist LP44 ([Fig. 8D](#f8){ref-type="fig"}; for sham: t~9~ = 4.746; *P* = 0.0010; for BDL: t~8~ = 6.597; *P* = 0.0002), suppressed 5-HT-induced scratching in both sham and BDL rats. Thus, these results suggested over-activation of 5-HT receptors subtypes 5-HT~1A~, 5-HT~2~, 5-HT~3~, and 5-HT~7~ might suppress itch in sham and BDL rats. We next assessed the effects of intrathecal (i.t.) or intraperitoneal (i.p.) injection of 5-HT receptors antagonists on itch in sham and BDL rats after BDL surgery 15 days. The results showed that i.t. injection of 5-HT~1A~ receptor antagonist WAY-100635 increased 5-HT-induced scratching in sham rats ([Fig. 8E](#f8){ref-type="fig"}; t~12~ = 3.328; *P* = 0.0060), but suppressed that in BDL rats ([Fig. 8E](#f8){ref-type="fig"}; t~10~ = 5.101; P = 0.0005). It was found that i.p. application of 5-HT~2~ receptor antagonist ketanserin significantly inhibited 5-HT-induced scratching in both sham and BDL rats ([Fig. 8F](#f8){ref-type="fig"}; for sham: t~8~ = 9.300; *P* \< 0.0001; for BDL: t~10~ = 9.926; *P* \< 0.0001). Interestingly, i.p. application of 5-HT~3~ receptor antagonist ondansetron hydrochloride failed to inhibit 5-HT-induced itch in sham rats ([Fig. 8G](#f8){ref-type="fig"}; t~7~ = 1.011; *P* = 0.3456, but significantly inhibited that in BDL rats ([Fig. 8G](#f8){ref-type="fig"}; t~9~ = 9.498; *P* \< 0.0001). Finally, i.t. injection of 5-HT~7~ receptor antagonist SB269970 significantly inhibited 5-HT-induced scratching in both sham and BDL rats ([Fig. 8H](#f8){ref-type="fig"}; for sham: t~9~ = 15.33; *P* \< 0.0001; for BDL: t~8~ =9.374; *P* \< 0.0001). Thus, these data suggested antagonism of 5-HT receptor subtypes 5-HT~1A~, 5-HT~2~, 5-HT~3~, and 5-HT~7~, might suppress itch response under cholestasis condition. It also suggested antagonism of 5-HT~1A~ might increase 5-HT-induced itch, however, antagonism of 5-HT~2~ and 5-HT~7~ might suppress itch under physiological condition. Antinociception could be modulated by 5-HT receptors agonists or antagonists in sham and BDL rats ------------------------------------------------------------------------------------------------- In order to reveal the possible roles of 5-HT receptor subtypes on antinociception in sham and BDL rats, 5-HT receptors agonists or antagonists were administrated after BDL surgery 15 days. It was found that i.t. injection of 5-HT~1A~ receptor agonist DPAT ([Fig. 9A](#f9){ref-type="fig"}; F~(1,72)~ = 82.89, *P* \< 0.0001), 5-HT~2~ receptor agonist α-methyl-5-HT ([Fig. 9B](#f9){ref-type="fig"}; F~(1,60)~ = 12.15, *P* = 0.0009), 5-HT~3~ receptor agonist 2-Methyl-5-HT ([Fig. 9C](#f9){ref-type="fig"}; F ~(1,60)~ = 41.07, *P* \< 0.0001), but not 5-HT~7~ receptor agonist LP44 ([Fig. 9D](#f9){ref-type="fig"}; F~(1,48)~ = 3.308, *P* = 0.0752), reduced the latency of tail-flick in response to 52 °C hot water in BDL rats. In contrast, as shown in [Fig. 9A--D](#f9){ref-type="fig"}, only 5-HT~2~ receptor agonist α-methyl-5-HT significantly reduced the latency of tail-flick in response to 52 °C hot water (F~(1,\ 55)~ = 8.913, *P* = 0.0042) in sham rats,, but 5-HT~1A~ receptor agonist DPAT (F~(1,96)~ = 0.6788, *P* = 0.4120), 5-HT~3~ receptor agonist 2-Methyl-5-HT (F~(1,55)~ = 0.8980, *P* = 0.3475), and 5-HT~7~ receptor agonist LP44 (F~(1,54)~ = 0.3522, *P* = 0.5553) had no effects. Thus, it was suggested that over-activation of 5-HT receptors 5-HT~1A~, 5-HT~2~, and 5-HT~3~ might attenuate antinociception in BDL rats. We next assessed the effects of intrathecal (i.t.) or intraperitoneal (i.p.) injection of 5-HT receptors antagonists on antinociception in sham and BDL rats after BDL surgery 15 days. The results showed that i.t. injection of 5-HT~1A~ receptor antagonist WAY-100635 significantly increased the latency of tail-flick in response to 52 °C hot water in sham and BDL rats ([Fig. 9E](#f9){ref-type="fig"}; for sham: F~(1,72)~ = 95.79, *P* \< 0.0001; for BDL: F(1, 60) = 7.657, *P* = 0.0075). Interestingly, i.p. application of 5-HT~2~ receptor antagonist ketanserin ([Fig. 9F](#f9){ref-type="fig"}) and 5-HT~3A~ receptor antagonist ondansetron ([Fig. 9G](#f9){ref-type="fig"};) significantly reduced the latency of tail-flick in response to 52 °C hot water in BDL rats (for ketanserin: F~(1,60)~ = 42.43, *P* \< 0.0001; for ondansetron: F~(1,\ 50)~ = 24.14, *P* \< 0.0001), but not in sham rats (for ketanserin: F~(1,\ 48)~ = 3.409, *P* = 0.0710; for ondansetron: F~(1,\ 40)~ = 0.8852, *P* = 0.3524). Finally, i.t. injection of 5-HT~7~ receptor antagonist SB269970 failed to change the latency of tail-flick in response to 52 °C hot water in sham and BDL rats ([Fig. 9H](#f9){ref-type="fig"}; for sham: F~(1,48)~ = 0.01346, *P* = 0.9081; for BDL: F~(1,60)~ = 1.261, P = 0.2660). Thus, it was suggested that antagonism of 5-HT~1A~ might produce antinociception in sham and BDL rats, while antagonism of 5-HT~2~ and 5-HT~3~ might attenuate antinociception in BDL rats. Additionally, i.t. injection of 5-HT~7~ receptor antagonist SB269970 had little effects on antinociception in sham and BDL rats. Discussion ========== Several clinical observation showed that administration of 5-HT~3~ receptor antagonist ondansetron[@b34][@b35][@b36][@b37][@b38] or selective serotonin reuptake inhibitor (SSRI) sertraline[@b39][@b40] were able to alleviate cholestatic pruritus, providing important clues to support key role of 5-HT system in cholestatic pruritus. Unfortunately, little is known of the modulatory effects of peripheral and central 5-HT system, especially 5-HT receptor subtypes, on cholestatic pruritus and antinociception so far[@b73][@b74]. Based on the important and complex roles of 5-HT system in the regulation of pain and itch[@b75], the present study investigated the roles of peripheral and central 5-HT system in itch and antinociception in BDL rats, which is a severe model of obstructive cholestasis[@b42]. Our results revealed that peripheral and central 5-HT system played important roles in modulation of itch and antinociception in BDL rats. Although no obvious spontaneous scratching was observed in BDL rats, we found an enhanced scratching response was induced by i.d. injection of 5-HT in BDL rats, suggesting itch hypersensitivity in BDL rats. We further showed that the 5-HT level in the skin and spinal cord significantly increased, possible due to the increased 5-HT synthesis, in BDL rats compared to sham rats. In BDL rats, several 5-HT receptor subtypes were up-regulated in the DRG or TG, including 5-HT~1B~, 5-HT~1D~, 5-HT~2A~, 5-HT~3~, 5-HT~5B~, 5-HT~6~, and 5-HT~7~. In sharp contrast, multiple 5-HT receptor subtypes were down-regulated in the spinal cord or brainstem in BDL rats, including 5-HT~1A~, 5-HT~1F~, 5-HT~2B~, and 5-HT~3~. Pharmacological activation of 5-HT~2A~, 5-HT~3~, and 5-HT~7~ induced itch in BDL rats. Administration of 5-HT~1A~, 5-HT~2A~, 5-HT~3~, and 5-HT~7~ agonists or antagonists differentially influenced the itch and antinociception in sham and BDL rats. Thus, these results suggested 5-HT system dynamically participated in itch and antinociception under cholestasis condition and distinct 5-HT receptor subtypes might be involved in these processes. There are several animal models to mimic cholestasis-induced itch and antinociception, including BDL in rats[@b20][@b24] or mice[@b21], administration of α-naphthylisothiocyanate (ANIT) or 17α-ethynylestradiol in rats[@b76] or mice[@b77]. Interestingly, we noticed that spontaneous scratching behavior was observed in BDL rats[@b24] and in 17α-ethynylestradiol-treated rats[@b76]; However, impaired itch perception was also demonstrated in ANIT or 17α-ethynylestradiol-treated mice[@b77]. In the current study, we employed BDL rat model to investigate itch and antinociception associated with severe obstructive cholestasis. We carefully quantified the spontaneous itch-related behaviors, such as scratching, grooming, and biting behaviors after BDL surgery in SD rats[@b24]. Surprisingly, we found itch-related behaviors were not significantly increased in BDL rats compared to sham rats. We postulated the discrepancy for spontaneous itch in BDL rats may be attributed to the different rat stains employed: SD rats versus Wistar rats. Although lack of spontaneous scratching behavior in BDL rats, we observed an enhanced scratching induced by intradermal injection of 5-HT in BDL rats, suggesting evoked itch was potentiated in cholestatic rats. Thus, species difference among mice, rats and human and the etiology of cholestasis may explain the distinct itch phenotype under cholestasis condition. After BDL surgery, rats developed antinociception responding to mechanical or heat stimuli, which were consistent with others' reports[@b20][@b42][@b78]. In addition, the motor performance of BDL rat was intact evaluated by Rota-rod test, suggesting that itch and antinociception phenotypes of BDL rats may not be attributed to motor dysfunction. Patients with cholestatic pruritus also showed increased number of dermal mast cells, which release histamine, 5-HT and proteases[@b24]. Our immunostaining data and HPLC analysis provided evidence that 5-HT level in the skin and the spinal cord significantly increased in BDL rats compared to sham rats. We further found the mRNA expression of *Tph1* and *Tph2*, the rate-limiting enzymes for 5-HT synthesis in the skin and in the brain stem, respectively[@b26], were up-regulated in BDL rats. Thus, it suggested that increased 5-HT level may be attributed to enhanced 5-HT synthesis. Because histamine seems to play little role in cholestasis pruritus[@b79], our data suggested 5-HT, possible from mast cells, may serve as potential itch mediator under cholestatais condition. We subsequently demonstrated that intradermal injection 5-HT~2A~, 5-HT~3~ or 5-HT~7~ receptor agonists induced scratching in BDL rats, whereas only 5-HT~2A~, or 5-HT~7~ receptor agonists induced scratching in the sham rats, suggesting 5-HT~3~ receptor is involved in pathological cholestatic itch but not physiological itch. Our results also demonstrated application of 5-HT~3~ receptor antagonist ondansetron attenuated itch in BDL rats but not in sham rats. It was consistent with the clinical observation that 5-HT~3~ receptor antagonist ondansetron alleviated cholestatic pruritus[@b34]. We further showed that DCA was not able to induce scratching behavior in rats, although it induced itch in mice[@b17][@b23], suggesting the species difference between rat and mouse existed for bile acid-induced itch. Interestingly, our results showed that BDL rats did not scratch spontaneously but exerted enhanced scratch induced by 5-HT. These data was consistent with the observation that DCA did not evoke scratch in rats, but potentiated 5-HT-induced scratch in rats. Together, these data suggested 5-HT, possible not bile acids, may serve as a potential pruritogen in rats under cholestasis condition. Both increased 5-HT level in skin and increased expression of 5-HT~2A~, 5-HT~3~ or 5-HT~7~ receptors in peripheral nervous system contributed to itch hypersensitivity under cholestasis condition. However, the roles of other 5-HT receptors subtypes, such as 5-HT~1D~, 5-HT~5B~, and 5-HT~6~, on cholestatic itch remain unclear and warrant further investigation. As we demonstrated 5-HT-induced itch was enhanced in BDL rats, we subsequently investigated the dynamic expression changes of 5-HT receptors subtypes in peripheral and central nervous systems by using qPCR analysis under cholestasis condition. The results demonstrated that mRNA expression of multiple 5-HT receptors, including 5-HT~1B~, 5-HT~1D~, 5-HT~2A~, 5-HT~3A~, 5-HT~5B~, 5-HT~6~, and 5-HT~7~, were up-regulated in DRG or TG from BDL rats compared to sham rats. In sharp contrast, the mRNA expression of multiple 5-HT receptors, including 5-HT~1A~, 5-HT~1F~, 5-HT~2B~, and 5-HT~3A~, were down-regulated in spinal cord or brainstem (especially cervicomedullary junction) from BDL rats compared to that from sham rats. As we showed increased 5-HT level in CNS could suppress 5-HT-induced scratch in sham and BDL rats, the down-regulation of 5-HT receptors may produce dis-inhibition to contribute to the enhanced itch in BDL rats. Thus, these results suggested up-regulation of certain 5-HT receptors in primary sensory neurons and down-regulation of certain 5-HT receptors in central nervous system may contribute to the enhanced itch behavior in BDL rats. It is unclear for the precise causes to drive the expression changes of 5-HT receptors in BDL rats. As 5-HT is not able to penetrate blood-brain-barrier (BBB), peripheral and central 5-HT systems are considered as two separated compartments[@b25] and the synthesis of 5-HT also employs distinct TPH, a rate-limiting enzymes for 5-HT synthesis. To investigate the possible role of central 5-HT in cholestatic itch, we performed i.t. injection 5-HT or i.p. injection of selective serotonin reuptake inhibitor (SSRI) fluoxetine in the sham and BDL rats. It was found that 5-HT (i.t.) or fluoxetine (i.p.) significantly suppressed 5-HT-induced scratching in sham and BDL rats, suggesting increased 5-HT level in CNS is sufficient to suppress itch in sham and BDL rats. We further used pharmacological 5-HT receptor agonists to reveal the distinct role of certain 5-HT receptor subtypes that involved in. Our results showed that intrathecal injection of 5-HT~1A~, 5-HT~2A~, 5-HT~3~ or 5-HT~7~ receptors agonists suppressed 5-HT-induced scratching in sham and BDL rats, which was consistent with previous report[@b80]. Thus, the results suggested application 5-HT or 5-HT receptor agonists, including 5-HT~1A~, 5-HT~2A~, 5-HT~3~ or 5-HT~7~, inhibited itch sensation possible through over-activation of these 5-HT receptors subtypes in the spinal cord. We next examined whether increased endogenous 5-HT level in spinal cord contributed to itch hypersensitivity in BDL rats by using selective 5-HT receptor antagonists. Our results showed that antagonism of 5-HT~1A~, 5-HT~2A~, 5-HT~3~ or 5-HT~7~, inhibited itch sensation in BDL rats, suggested increased endogenous 5-HT in spinal cord of BDL rats might play an itch-facilitating effect through 5-HT~1A~, 5-HT~2A~, 5-HT~3~ or 5-HT~7~. In sham control rats, antagonism of 5-HT~1A~ increased, while antagonism of 5-HT~2A~, or 5-HT~7~ inhibited itch. Meanwhile, antagonism of 5-HT~3~ had little effect on itch in sham rats. Thus, it was suggested tonic activation of spinal 5-HT~1A~ might play a suppressive effect, while activation of 5-HT~2A~ and 5-HT~7~ might play a facilitating effect on itch transmission in physiological condition. What is the role of spinal 5-HT~1A~ in regulating cholestatic pruritus? Our results showed 5-HT~1A~ antagonist exerted itch-enhancing effect in sham rats but alleviated itch in BDL rats. Previous study demonstrated that descending serotonergic system from brainstem facilitated gastrin-releasing peptide (GRP)-GRP receptor (GRPR) signaling via spinal 5-HT~1A~ for mediating itch transmision. In sharp contrast, activation of 5-HT~1A~ hyperpolarizes spinal neurons without GRPR to dampen neuronal excitability, suggesting opposing modulation of descending serotonergic system on itch and pain. The dual actions of 5-HT mediated by 5-HT~1A~ suggested neuronal phenotypes (excitatory versus inhibitory) that expresses this receptor might play a key role in determining which actions 5-HT would finally exert on itch neurotransmission. Whether the down-regulation of 5-HT~1A~, possible other subtypes 5-HT~1F,~ 5-HT~2B~ and 5-HT~3A~, in spinal cord contributes to itch hypersensitivity in BDL rats warrants further study. To investigate the possible role of spinal 5-HT in cholestasis-associated antinociception, we performed i.t. injection of 5-HT or i.p. injection of selective serotonin reuptake inhibitor (SSRI) fluoxetine in the sham and BDL rats. 5-HT (i.t.) or fluoxetine (i.p.) significantly increased the heat pain threshold in sham and BDL rats, suggesting increased 5-HT level in spinal cord might be sufficient to induced antinociception in sham and BDL rats, which was consistent with previous report[@b81]. We next used selective 5-HT receptors agonist to investigate the distinct role of certain 5-HT receptor subtypes that involved in. Our results showed that i.t. injection of 5-HT~1A~, 5-HT~2A~, and 5-HT~3~, but not 5-HT~7~ receptor agonists suppressed antinociception in BDL rats. Thus, the results suggested over-activation of 5-HT~1A~, 5-HT~2A~, and 5-HT~3~ attenuated antinociception under cholestasis condition. We next examined whether increased endogenous 5-HT level in spinal cord contributed to antinociception in BDL rats by using selective 5-HT receptors antagonists. Our results showed that antagonism of 5-HT~2A~ and 5-HT~3~ inhibited antinociception in BDL rats, suggesting increased endogenous 5-HT in spinal cord of BDL rats exert antinociception possible through 5-HT~2A~ and 5-HT~3~. Antagonism of 5-HT~1A~ increased heat pain threshold in both sham and BDL rats, suggesting activation spinal 5-HT~1A~ might play a pronociceptive role in sham and BDL rats. Additionally, agonism or antagonism of 5-HT~7~ had little effect on antinociception in sham and BDL rats. The limitation of our pharmacological study is the lack of dose-response curves for different 5-HT receptor agonists or antagonists. Thus, our results suggested activation of spinal 5-HT~1A~ might play a pronociceptive role, while activation of 5-HT~2A~ and 5-HT~3~ might play an antinociceptive role in BDL rats. Previous reports have proposed the endogenous opioidergic system involved in itch and antinociception under cholestasis condition[@b75]. Our results suggested serotonergic system was also important for modulating cholestasis-induced itch and antinociception. Endogenous opioids-induced analgesic effects partially mediated by activation of descending serotonergic inhibitory pathways terminating on spinal cord dorsal horn[@b31]. 5-HT~1A~ receptor acts as a regulator for 5-HT release, its down-regulation could increase 5-HT release[@b82]. Previous report demonstrated the expression of 5-HT~1A~ was down-regulated in hippocampus of BDL mice[@b82]. Our results also showed the expression 5-HT~1A~ was down-regulated in spinal cord and brainstem in BDL rats, which may also contribute to the increased release of 5-HT. In summary, our findings showed that itch hypersensitivity and antinociception developed in BDL rats. We found that 5-HT level increased in the skin and spinal cord in BDL rats. 5-HT receptors subtypes, including 5-HT~1B~, 5-HT~1D~, 5-HT~2A~, 5-HT~3A~, 5-HT~5B~, 5-HT~6~, and 5-HT~7,~ were up-regulated in peripheral nervous system and 5-HT~1A~, 5-HT~1F~, 5-HT~2B~, and 5-HT~3A~ were down-regulated in central nervous system. Peripheral activation of 5-HT~2~, 5-HT~3~, and 5-HT~7~ might contribute to cholestatic itch in rats. Agonism or antagonism of 5-HT receptors 5-HT~1A~, 5-HT~2A~, 5-HT~3~, and 5-HT~7~ might suppress cholestatic itch. Agonism or antagonism of 5-HT receptors 5-HT~1A~, 5-HT~2A~, and 5-HT~3~ differently modulate antinociception in BDL rats. Thus, targeting 5-HT system may provide an effective treatment for cholestatic pruritus and possible other comorbidities. Additional Information ====================== **How to cite this article**: Tian, B. *et al*. Peripheral and spinal 5-HT receptors participate in cholestatic itch and antinociception induced by bile duct ligation in rats. *Sci. Rep.* **6**, 36286; doi: 10.1038/srep36286 (2016). **Publisher's note:** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. This work was supported by grants from the NSFC (National Natural Science Foundation of China) to Tong Liu (31371179, 81300968), and by grant form the Nature Science Foundation of Jiangsu province to Li-Hua Chen (BK20140372). Tong Liu was also supported by funding of Jiangsu Province (2015-JY-029). Xue-Long Wang was supported by grants from Jiangsu Province (SJLX15_0582). This project is subject to the second affiliated hospital of Soochow university preponderant clinic discipline group project funding (XKQ2015007). This project is subject to A Project Funded by the Priority Academic Program Development of Jiangsu Higher Education Institutions. **Author Contributions** B.T., J.C.L and T.L. designed and supervised this study. B.T., X.L.W., Y.H., L.H.C., R.X.C., F.M.Z. and R.G. carried out the experiments, collected and analyzed the data. X.L.W., B.T., and T.L. prepared the manuscript. {#f1} {#f2} {#f3} {#f4} {#f5} {#f6} {#f7} {#f8} {#f9} ###### 5-HT receptor agonists and antagonists used in this study. Chemicals Target Formula M.Wt Biological Activity Ref. --------------- ------------------------------ ------------------------------------ -------- --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------ DPAT 5-HT~1A~ receptor agonist C~16~H~25~NO.HBr 328.29 Full 5-HT~1A~ serotonin receptor agonist; more active enantiomer. Reduces hippocampal 5-HT levels following systemic administration in rats *in vivo*. [@b44] WAY-100635 5-HT~1A~ receptor antagonist C~25~H~34~N~4~O~2~.C~4~H~4~O~4~ 538.64 Potent, silent antagonist of 5-HT~1A~ receptors (IC50 = 2.2 nM; Ki = 0.84 nM for rat 5-HT~1A~ receptors). Displays 100-fold selectivity for 5-HT~1A~ over other 5-HT subtypes. [@b45], [@b46] α-methyl-5-HT 5-HT~2~ receptor agonist C~11~H~14~N~2~O · C~4~H~4~O~4~ 306.31 a-methyl-5-HT displays a high affinity (Ki = 3 nM with \[^3^H\]DOB) for 5-HT~2~ site and little selectivity for 5-HT~1A~, 5-HT~1B~, 5-HT~1C~, and 5-HT~1D~ sites (Ki = 42, 85, 150, and 150 nM, respectively) and a very low affinity for 5-HT~1E~ (Ki greater than 10,000 nM) sites. [@b47], [@b48] ketanserin 5-HT~2A~ receptor antagonist C~22~H~22~FN~3~O~3~ · C~4~H~6~O~6~ 545.51 Selective 5-HT~2~ serotonin receptor antagonist. Ketanserin significantly reduces nicotine self-administration in rats, supporting an unexpected involvement of serotonin in nicotine addiction. [@b49], [@b50] 2-Methyl-5-HT 5-HT~3A~ receptor agonist C~11~H~14~N~2~O.HCl 226.71 5-HT~3~ agonist (K~i~ = 1200 nM) and potent 5-HT~6~ ligand (K~i~ = 46 nM). [@b51], [@b52] Ondansetron 5-HT~3A~ receptor antagonist C~18~H~19~N~3~O.HCl 329.83 Selective 5-HT~3~ receptor antagonist (K~i~ = 6.16 nM). Antiemetic; prevents emesis induced by cytotoxic drugs and radiation. [@b53], [@b54], [@b55] LP44 5-HT~7~ receptor agonist C~27~H~37~N~3~OS.HCl 488.13 High affinity 5-HT~7~ receptor agonist (K~i~ = 0.22 nM) that displays selectivity over 5-HT~1A~ and 5-HT~2A~ receptors (200- and \>1000-fold respectively). Induces relaxation of substance P-stimulated guinea pig ileum (EC~50~ = 2.56 μM). [@b29], [@b56] SB 269970 5-HT~7~ receptor antagonist C ~18~H~28~N~2~O~3~S.HCl 388.95 Potent and selective 5-HT~7~ receptor antagonist (pK~i~ values are 8.9, 7.2 and 6.0 for 5-HT~7A~, 5-HT~5A~ and 5-HT~1B~ and \< 6.0 for 5-HT~1A~, 5-HT~1D~, 5-HT~1E~, 5-HT~1F~, 5-HT~2A~, 5-HT~2B~, 5-HT~2C~, 5-HT~4~ and 5-HT~6~ receptors respectively). [@b29], [@b57], [@b58] ###### Sequences of primers used for real-time quantitative PCR. Target gene Forward (5′-3′) Reverse (5′-3′) Product size (bp) Accession number ------------- ------------------------ -------------------------- ------------------- ------------------ 5-HT~1A~ CCATCAGCAAGGACCACGGCTA CCCGTAGAGAACCAGCATGAGCAA 86 NM_012585.1 5-HT~1B~ GTCAAGCCAAAGCGGAGGA GCAGGGTGGGTAAATAGAAAGC 105 NM_022225.1 5-HT~1D~ CCCGAGAAAGGAAAGCCACT GAGGACCAAGGATACCACAAAGAA 92 NM_012852.1 5-HT~1F~ CTGTGACCTTTGGCTGAGTGTT CGACTGCGTCTGTGATTGCTC 104 NM_021857.3 5-HT~2A~ CTTCCAACGGTCCATCCACA GGGCACCACATTACAACAAACAG 132 NM_017254.1 5-HT~2B~ CGCCATCCCAGTCCCTATT CAGCCAGTGACCCAAAGAGC 116 NM_017250.1 5-HT~2C~ GACTGAGGGACGAAAGCAAAG GAAGGACCCGATGAGAACGA 83 NM_012765.3 5-HT~3A~ GTGACCGCCTGTAGCCTTGA GATGCTCTTGTCCGACCTCA 147 NM_024394.2 5-HT~4~ TGCCTTCCTTATCATCCTCTGC CACCACATTCCACTGTATCCCT 134 NM_012853.1 5-HT~5A~ CGCTGTGCTCCTGGGATAT CCTGTTGAACGCCGTGTAGAT 104 NM_013148.1 5-HT~5B~ CGTGGTGCTCTTCGTCTACTG TCCTGAGGTGCTTCCTTTGC 117 NM_024395.1 5-HT~6~ GCACGAACTGGGCAAAGCT GGACGCCACGAGGACAAAA 82 NM_024365.2 5-HT~7~ TTCTGTCGGTCTGGCTGCTCTC ACCGCAGTGGAGTAGATCGTGTAG 130 NM_022938.2 CK-7 CGAGGAGATGGCCAACCATA GAGCGGTTCATCTCCGCAAT 138 NM_001047870.1 PCNA CTTACTCTGCGCTCCGAAGG TGATGTCTTCATTACCAGCACA 115 NM_022381.3 TPH1 ACGAACTCTTAGGCCACGTCC TTGCACAGTCCAAACTCCACA 151 NM_001100634.2 TPH2 CAGCCCGCAATGATGATGTTT CGCTTCTCTTGTCCTCGCTTT 145 NM_173839.2 Actin ACTATCGGCAATGAGCGGTTCC AGCACTGTGTTGGCA TAGAGGTC 152 NM_031144.3 [^1]: Present address: Capital Medical University Electric Power Teaching Hospital, Beijing, 100073, China. [^2]: These authors contributed equally to this work.

Background {#Sec1} ========== There is a growing agenda for the promotion of a healthy lifestyle in young people for the reduction in the risk of cardiovascular and other non-communicable diseases \[[@CR1]--[@CR4]\]. Non-communicable diseases (NCDs) place a heavy burden on society and on hospital and community-health services; their prevention not only benefits the individuals at risk, but reduces the pressure on limited health resources. Our research in Southampton has focused on the processes by which the developmental environment affects later risk of ill health such as obesity or NCDs. The insights from this research are highly relevant to today's society, because they raise issues about personal choice, responsibility for health and the need for better informed decisions about diet, lifestyle and the ethical dilemmas faced by technological societies. The Shanghai Declaration of the Worldwide Universities Network, led by the University of Southampton and prepared at the request of the World Health Organisation ahead of the 2011 United Nations high-level meeting of the General Assembly on the prevention and control of non-communicable diseases (NCDs), concluded that 'Particular attention should be paid to both population- and individual-based approaches to increase access to education, to promote health literacy in children, adolescents and parents, ... to both reduce the burden of NCDs and provide other benefits' \[[@CR2]\] and The Lancet Series on Adolescent Health \[[@CR5]\] states that the 'Strongest determinants of adolescent health are national wealth, income inequality and access to education'. There are great inequalities in dietary quality and the adoption of healthy lifestyles in the UK. Southampton is a relatively deprived city in the South of England, and it falls in the 25 % most deprived local authorities in England and Wales. The Southampton Women's Survey (SWS) is an ongoing, prospective cohort study which started in 1998 to assess preconception characteristics of a general population sample of young women and to investigate a wide range of maternal influences on pregnancy outcomes and child health. A detailed description of the rationale, study design and protocol has been published elsewhere \[[@CR6]\]. Dietary data from the SWS \[[@CR6]\] were analysed using principal components analysis. The first component of the analysis of the women's data can be interpreted as a 'prudent' diet score: this score summarises the degree to which each woman complies with a prudent/healthful dietary pattern, characterised by high intakes of fruit, vegetables, whole grain, fish and poultry \[[@CR7]\]. This pattern has been found to be positively associated with micronutrient intake \[[@CR7]\] and negatively associated with saturated fat intake \[[@CR8]\]. A striking finding from the SWS is the strong and graded relationship between women's educational attainment and their prudent diet score, showing for example that among women who left school with no educational qualifications, 55 % had a prudent diet score in the lowest quarter of the distribution in contrast to only 3 % of those women with degree level qualifications \[[@CR9]\]. Analysis of other lifestyle data from the SWS has also shown great differences across the population, with higher levels of obesity and smoking and lower levels of physical activity in the most disadvantaged areas of the City \[[@CR10]\]. In Southampton, we have shown that women change their diets and lifestyles only marginally when planning a pregnancy and during pregnancy itself \[[@CR11], [@CR12]\], and thus lifestyles adopted earlier in life have long-term consequences both for women themselves and for their children. Importantly, dietary patterns established by women before they become pregnant \[[@CR13], [@CR14]\] are strongly associated with the way in which they feed their children in infancy and childhood, and in turn with the health and development of the child \[[@CR15]--[@CR17]\]. Women's diets, however, are also influenced by their partners, and fathers play an important role in determining the diet and lifestyle of the family \[[@CR18]\], which reflects on their own cardiovascular health too. The effects of a woman's diet and general health as she embarks on pregnancy can have profound and lasting effects on both early development and the lifelong health of her children \[[@CR19]--[@CR23]\]. Specifically, unhealthy lifestyles in mothers, including unbalanced diets, unhealthy body composition, excessive stress and low levels of physical activity, have been associated with increased risks in their children in later life of heart disease, diabetes, poor respiratory health and osteoporosis \[[@CR24], [@CR25]\]. A recent review of six European studies concluded that interventions are needed before, as well as during, pregnancy to improve the diets of families with young children \[[@CR26]\]. Health literacy is considered as 'a means to enabling individuals to exert greater control over their health and the range of personal, social and environmental determinants of health' \[[@CR27]\]. This view of health literacy takes into account the important role of educating individuals about their health and enabling them to develop the critical and evaluative thinking skills required to make informed health-related decisions, and develop healthy behaviours which are sustained across their lifetime \[[@CR28]\]. The development of such critical thinking skills and the ability to consider and weigh evidence are part of core scientific practices and one of the advantages that science education has to offer to the general education of young individuals \[[@CR29]\]. We aim to promote health literacy through science education. Science education is important not only because it provides the basis for the development of future scientists but also because it can sustain and improve the scientific literacy of the wider population \[[@CR30]\]. Scientific literacy describes both the ability to understand scientific content and the ability to engage in critical evaluation and discussions about scientific issues as these present themselves in everyday life \[[@CR31]\]. Knowledge of scientific principles, especially those related to human biology, critical thinking dispositions and the ability to make informed decisions about science related issues, are attributes closely linked to health literacy. The evidence points to a need to improve health and nutrition literacy to promote healthy lifestyles in young people in order to prevent cardiovascular disease in them and in their children and to reduce health inequalities. Part of the solution to these problems lies in education. We have undertaken a systematic review of approaches to behaviour change interventions, and it is clear that successful interventions include educational components \[[@CR32]\]; education can change attitudes, alter health-related behaviours and increase health literacy in young people \[[@CR33], [@CR34]\]. Interventions targeting teenagers have a double advantage. During adolescence health behaviours are being developed and established, and improving them at this age should lower the risks of later chronic disease. Additionally, however, such improvements in health behaviours would enable them to be better prepared when they embark on having their own children. Interventions with groups of school classes have been shown to play a key role in helping to prevent the adoption of high-risk health-related behaviours \[[@CR35]\]. These considerations have led us to collaborate with schools in Southampton and Hampshire to develop a classroom project, LifeLab, embedded within the new Southampton Centre for Biomedical Research (SCBR) at UHS. As many young people have never been inside a hospital or visited a research laboratory, such an experience can make a great impression. Indeed, learning outside the classroom has been recognised as often providing the most memorable learning, with experiences that are remembered into adulthood and affect behaviour, lifestyle and work \[[@CR36]--[@CR40]\]. The theme of LifeLab is ***Me, My Health & My Children's Health***. The teaching programme is tailored to reach students of all abilities and is primarily aimed at secondary school students aged 13--14 years. It includes professional development for all teachers involved in the project, a series of before and after lessons delivered in school and a hands-on practical activity day in the LifeLab facility. The school lessons are explicitly linked to the UK National Curriculum and embed the messages of the LifeLab visit. During the LifeLab activity day, students have opportunities to carry out hands on practical activities designed to build on the lessons they have been taught in school. The programme also has scope for training teenagers in cardio-pulmonary resuscitation (CPR) techniques, enabling them to play a part in reducing cardiovascular mortality and also providing strong messages about the consequences of failure to prevent cardiovascular disease. Pilot work showed that participation in a science programme focusing on health, and experiencing learning within a hospital-based classroom had a positive influence on teenagers' awareness of the importance of making healthy lifestyle choices \[[@CR37]\]. Currently, a small scale evaluation of the LifeLab intervention is being undertaken in six schools. However, as LifeLab is an intervention to improve knowledge, attitudes and behaviour in young people from across society, a large-scale cluster randomised control trial is now being conducted. While our particular focus is on the prevention of cardiovascular disease and other NCDs this type of educational approach needs to be fully assessed to see if does have an impact on the health behaviours of young people. Aim {#Sec2} --- The aim of this cluster randomised controlled trial is to evaluate whether an educational intervention in the form of LifeLab targeting teenagers improves the following in the teenagers:nutrition, health and lifestyle literacy.ability to use CPR techniques.health behaviours with respect to diet and lifestyle.understanding of the long-term influences of their health behaviours on their subsequent health and that of their future children.self-efficacy in relation to diet and lifestyle.engagement with science (intention to study science post GCSE/ pursue a career in a science-related job). Methods/Design {#Sec3} ============== Trial design {#Sec4} ------------ The target population for the evaluation is the state secondary schools in Southampton and surrounding areas. We propose to evaluate LifeLab through a cluster randomised control trial, recruiting Year 9 students (aged 13 to 14 years) from 32 state secondary schools/academies (approximately 2,500 students) to take part in the project. Each school will be randomly allocated to either 'control' or 'intervention' status. During year 1 (school year 2014--15) we aim to recruit 16 schools; eight will therefore be intervention schools and their Year 9 students will take part in the LifeLab programme, whereas the other eight schools will form the control group. During year 2 (school year 2015--2016), this process will be repeated. Schools that are randomised to the control arm of this intervention will be invited to attend LifeLab in the academic year *following* the year they contribute to the control group. Questionnaires will be administered to control and intervention students after recruitment (baseline questionnaires) and again approximately 12 months after the baseline questionnaires were completed. There may be some small loss to follow-up due to children moving schools during the school year, but we do not anticipate this having major impact; in our current small-scale trial, at the 12-month follow-up, 2 % of the students were unavailable to participate due to moving schools. We emphasise in our discussions with schools during the recruitment phase the need to track the students over the length of the trial and ask that schools provide us with the new school details for those students who have left. We will then contact the new schools to request permission to work with those students who have moved. Good working relationships with the schools have been developed over many years and should assist in minimising any dropout at the school level. The Heads of Science of all the Southampton and many Hampshire secondary schools are enthused by LifeLab and are committed to it. They have been involved from a very early stage and have always been keen for their students to take part in our taster and pilot activities. We envisage good take-up within the schools and do not foresee recruitment difficulties. We appreciate that the control schools may wish to receive the intervention straight away, but all control schools are offered participation in the following year. The flow of the trial is shown in the CONSORT diagram in Fig. [1](#Fig1){ref-type="fig"}.Fig. 1Consort diagram to show the flow of participants through the trial. This CONSORT diagram shows the order and timing of the stages which the participants in the trial take part in Recruitment {#Sec5} ----------- Schools will be recruited through presentations at the Secondary Heads of Science forum meetings, building on previous excellent working relationships. All secondary state schools/academies are eligible for participation. We exclude participation of independent, grammar and special schools, as the population of students would be more selective and could potentially bias the results of the trial. The main approach will be through recruitment letters and phone calls to both the Head of Science and Head Teacher at all schools throughout our target region. The recruitment pack includes a letter of agreement which the Head Teacher is required to sign, and upon receipt of this letter a further meeting will be arranged with the Head of Science and a member of the Senior Leadership Team. At this meeting, the requirements of signing up to the randomised controlled trial will be clarified and the expectations of the school as either an intervention or a control school will be fully explained. It will also be an opportunity for the school to ask questions around implementation of the intervention for forward planning on their behalf. Schools are asked to allocate a minimum of three classes to participate in the trial (approximately 90 pupils), and we ask that these are their 'middle ability' students. As the teaching programme is planned for this age range and is differentiated for ability, there are no exclusion criteria for pupils. For pupils who may require more input (for example, English as a second language), we provide support for schools in planning the provision. This is discussed at the teachers' professional development day. Schools will already have provisions in place for these students, and so it is a matter of ensuring that the LifeLab materials are accessible by all students who will be likely to participate in each school. Following this meeting, if the school is still happy to participate it will be put through to the randomisation phase. Randomisation procedure {#Sec6} ----------------------- Groups of recruited schools will be randomised in blocks of even numbers. Recruiting schools is an on-going process and the smallest block size will be two, such that once a pair of schools is recruited, and if others are unlikely to be recruited soon, the pair will be randomised. Where possible, the block size will be larger. Once at least two schools have been recruited and the Head Teachers' letters returned, the schools are numbered and documented and dated. The randomisation is then conducted off-site by a statistician at the MRC Lifecourse Epidemiology Unit using computer-generated sequences, with no knowledge of the schools in question. Prior to randomisation, no matching of schools takes place. Schools require the information about their status in a timely fashion in order to plan their curriculum/allocate staff and resources in advance. Schools are recruited throughout the year, and it would be difficult to match schools by exact criteria. We are recruiting from a large area and a major consideration when calculating the numbers of schools necessary to recruit was to ensure a wide representation of schools/pupils. Ethics approval and research governance {#Sec7} --------------------------------------- The study has been approved by the research ethics committee in the Southampton Education School, Faculty of Social and Human Sciences, University of Southampton (ERG reference: 12328). The study is funded by the British Heart Foundation (PG/14/33/30827) and has been registered on the ISRCTN database (ISRCTN71951436, registered 25 March 2015). The research sponsor is the University of Southampton. The LifeLab Directors provide oversight of the trial, acting as a trial steering committee, with reference to the LifeLab External Liaison Group for scrutiny. Consent {#Sec8} ------- Participant information sheets are sent with the initial recruitment pack to potential schools. Once the school has agreed to participate and signed the agreement letter and following the randomisation process, parent and pupil information sheets are provided for the schools to disseminate to pupils and parents directly. Two versions of the information sheets have been produced - depending on whether the school is part of the 'intervention' or 'control' arm of the trial. Parents are provided with contact details of the research team in case they have questions about the LifeLab programme and associated research trial. For the intervention schools, consent will be opt-in and collected prior to any research data being collected. For the control schools, consent will be opt-out. While not ideal, discussions with schools around this have been carried out, and the request from schools was that an 'opt-out' pathway was followed as this would place less of a burden on the control schools. Parent/pupil information sheets will be sent home, and parents who do not want their children to take part in the research questionnaires will be required to return the form indicating this. In all cases, pupil assent is built into the questionnaire and, at any point, a pupil or parent can request that their information be withdrawn. The intervention {#Sec9} ---------------- The proposed intervention comprises the following:Professional development for the teachers.A 2 to 3 week module of work \[[@CR41]\] for use with Year 9 school students (13 to 14 year olds), linked to the UK National Curriculum encompassing both before and after lessons to be delivered in school.A hands-on practical day visit to LifeLab in the Southampton Centre for Biomedical Research at University Hospital Southampton, held part way through the module. The materials for the work in school have been produced \[[@CR41]\] and continue to be developed, and the activities for the LifeLab visit have been tested in our pilot work. Students will have opportunities to experience a variety of ways to measure health: assessing carotid artery blood flow and structure using ultrasound, measuring body composition, performing lung function tests, training in CPR and testing grip strength and flexibility. They are also able to extract their own DNA and carry out gel electrophoresis experiments that illustrate how a healthy diet can induce epigenetic changes that alter DNA structure and are passed from parents to offspring, with implications for cardiovascular and lifelong health for themselves and their children. Health messages are linked to the hands-on practical activities at LifeLab and to the school-based activities, to ensure that the teenagers understand the long-term implications of their current diet and lifestyle on their cardiovascular health. Outcome measurements {#Sec10} -------------------- Our primary outcome is a measure of nutrition, health and lifestyle literacy based on the critical nutrition literacy scale developed by Guttersrud et al. \[[@CR42]\], adapted for use by teenagers and supplemented with broader lifestyle questions. Secondary outcomes will be the students' understanding of influences on their cardiovascular health and that of their subsequent children, their health behaviours such as dietary patterns \[[@CR43]\], physical activity, their self-efficacy scores in relation to diet and lifestyle and their intent to continue studying science post-GCSE/seek a career in a science-related job. Summary statistics at the school level will be obtained. Information on the resources required to provide LifeLab will be obtained. Outcome measures are collected in schools using web-based questionnaires. The LifeLab team provides support to the teachers to administer the process, but the students complete the questionnaires independently on their computers. A written script is used in all schools to explain the process to the students to ensure consistency across all schools, both control and intervention (please see Additional file [1](#MOESM1){ref-type="media"}). Data management {#Sec11} --------------- Questionnaire data will be collected via a website-based questionnaire (hosted by the University of Southampton). These data will be downloaded via a dedicated computer and stored on a University server. All data will be kept in accordance with the Data Protection Act, University of Southampton Data protection policy and in accordance with the protocols of the MRC Lifecourse Epidemiology Unit. It will be stored in password-protected areas on computer by the research team and only accessible by them. Data will be stored in Access databases and analyses conducted in Stata and SPSS. The data will be managed with support from the data management staff of the MRC Lifecourse Epidemiology Unit, which has extensive data management expertise and houses data from more than 200 studies. Identifying information will be collected about participants, purely for the purposes of matching pre- and post-questionnaires. All identifying data will be removed from the rest of the data after linkage is complete and will be stored separately. It will only be kept in case there is future research in which a follow-up is planned. Process evaluation {#Sec12} ------------------ The intervention will be accompanied by a full and detailed process evaluation to include assessment of the context, fidelity, implementation and reach of the intervention, and barriers to implementation \[[@CR44], [@CR45]\]. It will include an assessment of contamination of control schools where science teachers have moved from intervention schools. Oversight of the process evaluation will be led by a health psychologist (MB) who has extensive experience in the field of process evaluation and was involved in the recent MRC publication of guidelines for process evaluation \[[@CR46]\]. She is independent from the implementation of the intervention and so can provide an unbiased oversight. An education researcher (AC) will be responsible for conducting observations in schools to ensure fidelity to the teaching programme and whilst she is involved in the professional development for the school teachers, in order to establish a credible relationship with the education community, she is independent of the implementation of the intervention from that point onwards. Statistical analyses {#Sec13} -------------------- We will perform an intention-to-treat analysis using multi-level models for the quantitative data. Specifically, random effects models will be used to take account of the clustered nature of our data. Multi-level logistic (for rare outcomes) or Poisson regression models with robust variance (for more common outcomes) will be used for binary outcome variables and multi-level linear regression for continuous measures. The number of clusters is relatively small, and summary analyses will be conducted at school level, not least as this will allow us to feedback school-level data to the schools that have been involved in the trial. Our endpoint will be the results of a 12-month follow-up with adjustment for baseline responses included in the modelling. Planned subgroup analyses would focus on whether there are different effects for boys and girls. We will analyse the uptake of LifeLab in different ethnic groups and in those who are more disadvantaged. Compliance will also be assessed in relation to various confounding factors to assess the bias associated with the different consent procedures imposed on us by the schools for the intervention and control groups. Sensitivity analyses will be conducted as appropriate using sampling from the intervention arm to try to account for this bias. In practice, during our pilot work, as schools do not want children remaining in school when the rest of the class is out on a trip, very few children are not included in the intervention arm, and thus we are confident that there is likely to be little difference in response rates between the two arms of the trial. Financial evaluation {#Sec14} -------------------- These data will be combined with appropriate unit cost data to estimate the costs of LifeLab. We will use the health and nutrition literacy score as the outcome measure in a cost-effectiveness analysis and will estimate the cost per unit change in this score. These data will be important in determining the level of support needed to sustain LifeLab on a longer term basis. Power calculations {#Sec15} ------------------ Using data on behaviour (z-scores), validated in our previous research in young adults in and around Southampton \[[@CR43]\], we have estimated an intra-class correlation of 0.035, but to be conservative we have rounded this up to 0.04. We propose to sample three Year 9 (aged 13 to 14 years) classes from each school, averaging 90 students from each school. Using 90 as our cluster size we estimate that with 14 clusters in each group, we will have 80 % power at the 5 % level of significance to detect a difference in change in our primary outcome score of 0.25 SDs from the beginning to end of the school year between the intervention and control schools. Comparable effect sizes have been considered in other health interventions as being meaningful in terms of the impact on health behaviours, and our level of 0.25 SDs falls in the mid-range of effect sizes reported in a meta-synthesis of meta-analyses of behaviour change interventions in the general population \[[@CR47]\]. As described above, we will study 32 year groups, thus allowing some leeway for dropouts. To minimise loss of pupils to follow-up, we will request class lists for each participating class (both from the intervention and control schools), so that missing participants at each stage of the follow-up can be identified and asked to complete the questionnaires. Discussion {#Sec16} ========== A formal assessment of LifeLab would show the value of the programme in educating teenagers about their cardiovascular health and about the value of science. It would point to ways to improve and develop the LifeLab concept and identify the activities that have the most profound effect on the students. If shown to be successful, this would provide the impetus to sustain LifeLab in Southampton and encourage the development of similar initiatives in other cities in the UK and around the world. There are a few projects similar to LifeLab which have been established in other cities worldwide. The Liggins Institute was at the forefront of this work with LENScience \[[@CR48]\], and Lord Winston established the Reach Out Lab at Imperial College, which, although a similar concept offering school students opportunities to experience science outside of the school environment, has different aims, focusing on motivating high attaining children to consider careers in science-related subjects. Plans are now being developed for laboratories elsewhere, similar to LifeLab, including in other countries such as Australia, South Africa and Ireland. Improving the diets and lifestyles of young people is an important route both to reducing NCDs in their later lives but also improving the health of the next generation. By targeting children from all backgrounds this provides a form of societal intervention, such as has recently been advocated as likely to have a greater impact on cardiovascular disease inequalities than targeting those at high risk of the disease \[[@CR49]\]. Trial status {#Sec17} ============ Recruitment for the RCT has been initiated, and 18 schools have been recruited and have already progressed through the randomisation phase. Additional file {#Sec18} =============== Additional file 1:**Pre-questionnaire script.** (PDF 66 kb) NCD : non-communicable disease RCT : randomised control trial CPR : cardio-pulmonary resuscitation SWS : Southampton Women's Survey **Competing interests** KMG has received reimbursement for speaking at conferences sponsored by nutrition companies and is part of an academic consortium that has received research funding from Abbott Nutrition, Nestec and Danone. The University of Southampton has received an unrestricted donation from Danone Nutricia to support LifeLab's work with schools. **Authors' contributions** KWT carried out the pilot work; was involved in the conception, design, development and delivery of the programme of work; and drafted the manuscript with help from HI and the other authors. LB and HD were involved in the design, development and delivery of the teaching programme of work, including the professional development programme for teachers. MEB was involved in the design of the cluster randomised trial and concurrent process evaluation. MEB, AC, MG and HI were involved in the design of the research questionnaires. AC, MG, JG, HI, KG and MH were involved in the initial design of the teaching programme and in the design and delivery of the professional development programme for participating teachers. MG, KG, JG, MH and HI have driven the development of LifeLab from its early origins in 2007, including finding funding and have designed the cluster randomised trial. All authors have contributed to the manuscript and approved the final submitted version. We would like to thank the schools, teachers, students and parents who participate so enthusiastically in the LifeLab teaching programme. In conducting pilot activities for LifeLab we have collaborated closely with colleagues from the highly acclaimed Liggins Education Network for Science (LENS) project at the University of Auckland ([www.auckland.ac.nz/uoa/liggins/lens/lens.cfm](http://www.auckland.ac.nz/uoa/liggins/lens/lens.cfm)), and would like to extend thanks for their on-going support and continued useful discussions, in particular Jacquie Bay, director of LENScience. LifeLab, has been developed as a joint initiative, involving the University of Southampton and the University Hospital Southampton NHS Foundation Trust, the MRC Lifecourse Epidemiology Unit, the University of Southampton (UoS) Faculty of Medicine and Southampton Education School, the Mathematics and Science Learning Centre (MSLC), the NIHR Southampton Biomedical Research Centre in nutrition (BRC), and the secondary schools themselves, Southampton City Council and Hampshire County Council. The cluster randomised trial is funded by the British Heart Foundation. LifeLab has also received research funding from the Wellcome Trust, Cancer Research UK, Research Councils UK, the BUPA Foundation, the Primary Science Teaching Trust (formerly the Astra Zeneca Science Teaching Trust), the EPSRC (via the UoS Pathways to Impact funding scheme). KWT & MAH are supported by the National Institute for Health Research through the NIHR Southampton Biomedical Research Centre. KMG is supported by the National Institute for Health Research through the NIHR Southampton Biomedical Research Centre and by the European Union's Seventh Framework Programme (FP7/2007-2013), project EarlyNutrition under grant agreement n° 289346. This report is independent research by the National Institute for Health Research Biomedical Research Centre Funding Scheme. The views expressed in this publication are those of the author(s) and not necessarily those of the NHS, the National Institute for Health Research or the Department of Health. The Garfield Weston Foundation gave a generous donation to the building of LifeLab.

Introduction {#s1} ============ Invasive fungal infections, particularly those caused by *Candida* and *Cryptococcus*, afflict millions of patients annually resulting in more than 1,350,000 deaths despite the introduction of new antifungal agent (Brown et al., [@B10]; Pfaller, [@B47]; Perlin, [@B45]; Perlin et al., [@B46]; Sanguinetti et al., [@B53]). Unfortunately, current antifungal therapies have limited effectiveness in treating invasive fungal infections and suffer from restrictions in route of administration, spectrum of activity, and bioavailability in target tissues such as the brain (Brown et al., [@B10]; Vandeputte et al., [@B67]). Further compounding this problem, the development of new antifungal is currently unable to keep pace with the urgent demand for safe and effective new drugs. Hence, there is a pressing and urgent need for novel, inexpensive, and safe antifungal drugs to combat these dangerous pathogens. The concept of drug repositioning has recently gained momentum and emerged as a viable approach to expedite anti-infective drug development (Butts and Krysan, [@B11]; Thangamani et al., [@B61],[@B64],[@B65]). For example, several reports have demonstrated that auranofin, an orally bioavailable FDA-approved drug for treatment of rheumatoid arthritis, exhibits potent antibacterial and antiparasitic activities (Jackson-Rosario et al., [@B33]; Debnath et al., [@B22]; Cassetta et al., [@B14]; Hokai et al., [@B31]; Aguinagalde et al., [@B1]; Thangamani et al., [@B62],[@B63]). This discovery led to the FDA granting auranofin Orphan Drug status for treatment of amebiasis. Auranofin is currently approved for long-term treatment of unresponsive rheumatoid arthritis; it is the first, and only, gold compound to be administered orally (Bernhard, [@B6]; Furst et al., [@B25]). Although auranofin is slightly less effective than parenteral gold compounds for treatment of rheumatoid arthritis, auranofin\'s oral bioavailability and reduced associated side effects offer significant advantages over traditional injectable gold drugs (Bernhard, [@B6]; Furst et al., [@B25]). Although auranofin has been used clinically for almost 30 years, its mechanism of action (MOA) in treating unresponsive rheumatoid arthritis is still poorly understood (Shaw, [@B57]; Berners-Price and Filipovska, [@B5]). The emergence of new anti-rheumatoid drugs with fewer side effects and faster activity has resulted in the decline of oral gold therapy clinically (Berners-Price and Filipovska, [@B5]). Nevertheless, there has been considerable research efforts employed to identify alternative therapeutic applications for auranofin, particularly in the area of infectious diseases (Shapiro and Masci, [@B55]; Lobanov et al., [@B39]; Kuntz et al., [@B34]; Bonilla et al., [@B8]; Sannella et al., [@B54]; Angelucci et al., [@B3]; Chomont et al., [@B18]; Jackson-Rosario et al., [@B33]; Lewis et al., [@B36]; Caroli et al., [@B13]; Debnath et al., [@B22]; Ilari et al., [@B32]; Chirullo et al., [@B17]; Tejman-Yarden et al., [@B60]; Sharlow et al., [@B56]). Recent studies by Fuchs et al. ([@B24]) and Stylianou et al. ([@B59]) reported that auranofin also possesses antifungal activity. However, the antifungal MOA and *in vivo* antifungal efficacy of auranofin remain unclear with several possible targets reported. Thus, the objectives of our study were to determine the antifungal activity of auranofin against clinical isolates of different fungal pathogens, to investigate the drug\'s antibiofilm activity, to deduce auranofin\'s antifungal MOA using an unbiased chemogenomic approach, and to validate the drug\'s *in vivo* antifungal efficacy in a *Cryptococcus neoformans*-infected *Caenorhabditis elegans* whole animal model. Materials and methods {#s2} ===================== Fungal strains and reagents --------------------------- Fungal strains used in this study are presented in Table [1](#T1){ref-type="table"}. Yeast peptone dextrose agar (YPD) was purchased from BD Biosciences (San Jose, CA). Auranofin (Enzo Life Sciences, Farmingdale, NY), fluconazole (Acros Organics, New Jersey), and flucytosine (TCI chemicals, Tokyo, Japan) were purchased from commercial vendors. XTT-sodium salt, menadione, RPMI powder, and MOPS were purchased from Sigma-Aldrich (St. Louis, MO). Concanavalin A--conjugated with FITC 488 dye was acquired from Thermo Fisher Scientific Inc. (Waltham, MA). ###### **MIC of auranofin and control antifungal drugs against Candida and Cryptococcus strains**. **Strains** **Description** **Auranofin (μg/ml)** **Fluconazole (μg/ml)** **Flucytosine (μg/ml)** --------------------------------- ------------------------------------------------------------------------------------------------------------------ ----------------------- ------------------------- ------------------------- *C. albicans* NR 29434 Bloodstream isolate from a person with a bloodstream infection collected in Winnipeg, Manitoba, Canada, in 2000 8 4 0.125 *C. albicans* ATCC 10231 Isolated from a man with bronchomycosis 2 2 0.25 *C. albicans* NR 29449 Is a vaginal isolate from a person with vaginitis collected in Ann Arbor, Michigan, USA, between 1990 and 1992 8 2 4 *C. albicans* NR 29435 Is a bloodstream isolate from a person with a bloodstream infection collected in Iowa City, Iowa, USA, in 2000 1 4 0.0625 *C. albicans* NR 29448 Is an isolate from a person with a bloodstream infection, collected in Arizona, USA. 4 \>64 0.0625 *C. albicans* NR 29437 Is a bloodstream isolate from a person with a bloodstream infection collected in Brussels, Belgium in 2000 4 2 0.0625 *C. albicans* NR 29446 Is a bloodstream isolate from a person with a bloodstream infection collected in Utah, USA. 16 \>64 0.25 *C. albicans* NR 29453 Is an oral isolate from an HIV+ person collected in Pretoria, South Africa 8 2 0.0625 *C. albicans* NR 29438 Is a bloodstream isolate from a person with a bloodstream infection, collected in Tel-Hashomer, Israel, in 2000. 16 2 0.0625 *C. albicans* ATCC 26790 Pulmonary candidiasis 8 2 0.0625 *C. albicans* ATCC 24433 Nail infection 8 4 1 *C. albicans* ATCC 14053 Human blood, Bethesda, MD 8 4 0.125 *C. albicans* ATCC 90028 Blood, Iowa 16 4 1 *C. albicans* NR 29366 Human isolate collected in China 16 \>64 0.0625 *C. albicans* NR 29367 Human isolate collected in China. 16 \>64 0.0625 *C. glabrata* ATCC MYA-2950 -- 8 4 0.0625 *C. glabrata* ATCC 66032 -- 8 2 0.0625 *C. tropicalis ATCC 13803* -- 16 2 0.125 *C. tropicalis ATCC 1369* -- 4 1 0.25 *C. parapsilosis ATCC 22019* Case of sprue, Puerto Rico 4 1 0.25 *C. neoformans* NR-41291 Obtained from human cerebrospinal fluid in China in July 2011. 4 1 0.5 *C. neoformans* NR-41292 Obtained from human cerebrospinal fluid in China in February 2012. 0.5 1 0.5 *C. neoformans* NR-41296 Obtained from human cerebrospinal fluid in China in February 2012. 1 2 0.5 *C. neoformans* NR-41295 Obtained from human cerebrospinal fluid in China in February 2012. 4 2 0.5 *C. neoformans* NR-41294 Obtained from human cerebrospinal fluid in China in June 2011. 0.5 4 2 *C. neoformans* NR-41297 Obtained from human cerebrospinal fluid in China in February 2012. 1 8 4 *C. neoformans* NR-41298 Obtained from human cerebrospinal fluid in China in February 2012. 1 4 2 *C. neoformans* NR-41299 Obtained from human cerebrospinal fluid in China in August 2009. 4 4 2 *C. neoformans* NR-41291 Obtained from human cerebrospinal fluid in China in July 2011. 1 4 1 *Cryptococcus gattii*---CBS1930 Isolated from a goat in Aruba prior to the outbreak in Vancouver, British Columbia, Canada. 0.5 2 2 *Cryptococcus gattii*---R265 Isolated from a human on Vancouver Island, Canada during the outbreak that began in the late 1990\'s 1 1 1 *Cryptococcus gattii*---Alg40 Progeny of a genotypic cross between *C. gattii* strains R265 and CBS1930. 0.5 2 0.5 *Cryptococcus gattii*---Alg75 Progeny of a genotypic cross between *C. gattii* strains R265 and Alg40. 8 8 8 *Cryptococcus gattii*---Alg81 Progeny of a genotypic cross between *C. gattii* strains R265 and Alg75. 4 8 4 *Cryptococcus gattii*---Alg99 Progeny of a genotypic cross between *C. gattii* strains R265 and Alg81. 4 8 4 *Cryptococcus gattii*---Alg114 Progeny of a genotypic cross between *C. gattii* strains R265 and Alg99. 8 8 4 *Cryptococcus gattii*---Alg115 Progeny of a genotypic cross between *C. gattii* strains R265 and Alg114. 8 8 4 *Cryptococcus gattii*---Alg127 Progeny of a genotypic cross between *C. gattii* strains R265 and Alg115. 4 4 4 Antifungal susceptibility testing --------------------------------- Antifungal susceptibility testing was carried out as per the National Committee for Clinical Laboratory Standards M-27A3 (NCCLS) guidelines (da Silva et al., [@B21]). Briefly, the inocula were prepared from 24 h old cultures of *Candida* spp. or 48 h old cultures of *Cryptococcus* spp. in YPD plates. Five colonies were then transferred to 5 mL of sterile 0.9% saline (PBS). The suspensions were adjusted to McFarland standard 0.5 and then diluted 1:2000 in RPMI 1640 buffered to pH 7.0 with 0.165 M MOPS (RPMI-MOPS) to yield an inoculum of 5.0 × 10^2^, −2.5 × 10^3^ CFU/mL. An aliquot (100 μL) of the resulting suspension was incubated with serially diluted fluconazole, flucytosine, and auranofin for 24 h for *Candida* spp and 72 h for *Cryptococcus* spp. The minimum inhibitory concentration (MIC) of fluconazole and flucytosine were determined as the prominent decrease (\~50%) in visible growth compared to untreated controls, as per NCCLS guidelines. Similarly the MIC of auranofin was determined as the lowest concentration resulting in 50% reduction in visible growth. All experiments were carried out in triplicate wells. Time kill assay --------------- Fungal cultures of *Candida albicans* and *Cryptococcus neoformans* were diluted approximately to 5 × 10^5^ CFU/mL and treated with 5 × and 10 × MICs of auranofin and fluconazole (in triplicate) in RPMI-MOPS, at 35°C. Samples were collected at indicated time points and serially diluted in PBS and plated onto YPD plates. Plates were incubated at 35°C for 24--48 h prior to counting fungal colony forming units (CFU), as described elsewhere (Cantón et al., [@B12]). XTT-reduction assay ------------------- *C. albicans* ATCC 10231 was grown in YPD broth at 35°C for 24 h. Cells were washed with PBS and resuspended in RPMI-MOPS at 10^6^ cells/mL (Pierce et al., [@B48]; Rane et al., [@B49]). An aliquot (100 μL) of cell suspension was transferred to wells in a 96-well tissue culture plate. After 48 h incubation (at 37°C), wells were washed with PBS and drugs (auranofin, fluconazole, and flucytosine) were added at indicated concentrations. After 24 h of incubation, the supernatant was removed and 100 μL of XTT/menadione solution was added to each well. The plates were covered with aluminum foil and incubated at 37°C for 1 h. Aliquots (75 μL) were taken from each well and the absorbance (OD~495~) was measured using a spectrophotometer. The antifungal activity of each drug was expressed as a percentage of metabolic activity of treatment groups relative to the DMSO-treated control groups. The experiment was performed using triplicate samples for each treatment regimen. Confocal imaging of fungal biofilms ----------------------------------- *C. albicans* ATCC 10231 was seeded on FBS-coated glass cover slips in 6-well tissue-culture plates and grown in RPMI-MOPS medium with 0.2% glucose at 37°C (Dongari-Bagtzoglou et al., [@B23]). After 48 h, wells were washed with PBS and drugs (auranofin, fluconazole, and flucytosine) were added at indicated concentrations. After 24 h of treatment, wells were washed with PBS and stained with concanavalin A--conjugated with FITC 488 dye (25 μg/mL in PBS) for 45 min at 37°C. After incubation, the coverslips were washed three times with PBS and mounted on glass slides. Stained biofilms were observed using Leica confocal laser scanning microscopy. Images were reconstructed using IMARIS software. Chemogenomics profiling of *Saccharomyces cerevisiae* ----------------------------------------------------- Initial testing of *Saccharomyces cerevisiae* sensitivity to auranofin was performed with the wild-type BY4743 diploid strain, the isogenic parent to the heterozygous diploid deletion collection. BY4743 was grown in YPD in 96-well plates with 1% DMSO or auranofin in concentrations ranging from 10 to 200 μM to determine a suitable level of growth inhibition. Auranofin (75 μM) was used for haploinsufficiency profiling because it delayed growth by 30% compared to the no drug control half-maximal optical density (OD). All experiments were performed at 30°C and cultures were shaken at 300 rpm. The heterozygous deletion set was purchased in a pooled format (Thermo Fisher Scientific, Waltham, MA). A frozen aliquot (200 μL) was thawed and used to inoculate 2 mL of YPD and grown for 9 h to reach an OD~600~ of 4.0. The culture was diluted to an OD~600~ of 0.13 and either 1% DMSO or 75 μM auranofin was added (three replicates each, 1 mL) and grown for 7 h. The cultures were grown again by diluting to an OD of 0.13 in 1 mL YPD with DMSO or 60 μM auranofin and grown for 8 h. Cultures were harvested and genomic DNA extracted using the YeaStar Genomic DNA kit (Zymo Research, Irvine, CA). The UPTAGs were amplified by PCR with Phusion Hot Start II High-Fidelity DNA polymerase at 0.02 U/μL (Thermo Fisher Scientific, Waltham, MA) using 0.5 ng/μL genomic DNA. Primers are listed (Table [S1](#SM2){ref-type="supplementary-material"}). The PCR reactions were electrophoresed on an agarose gel and the 267 bp product extracted using a QIAquick Gel Extraction Kit (Qiagen, Valencia, CA). Purified DNA was measured using a Qubit instrument and samples were normalized and mixed together to a final concentration of 10 nM. Strains were grown and maintained on media according to standard practices (Amberg et al., [@B2]). The pooled PCR products were sequenced using standard Illumina sequencing in a HiSeq 2500 instrument. The reads were separated based on a 5 base multiplex tag unique for each experiment and an average of 5 million reads per replicate was obtained. The UPTAG barcodes in each experimental sample were separated based on a reference database of recharacterized barcode sequences (Smith et al., [@B58]). The resulting strain counts were imported into R and analyzed with edgeR (Robinson et al., [@B51]). Sequencing library sizes were normalized using the default parameters. Only strains with one or more counts in three or more samples were analyzed further. Differential representation of strains was determined using the quantile-adjusted conditional maximum likelihood (qCML) method. False discovery rates were determined to control for multiple testing. *Saccharomyces* deletion strain haploinsufficiency validation ------------------------------------------------------------- Overnight grown yeast cells were diluted (OD~600~ \~ 0.03) and grown in the presence and absence of auranofin, at indicated concentrations. Growth was monitored using a spectrophotometer (OD~600~) at indicated time points and the results were expressed as percent growth rate for each strain compared to the untreated control group. To assess growth on solid medium, 5 μL of ten-fold diluted yeast cells were spotted onto YPD agar containing DMSO or auranofin (6.25 μg/mL). Growth of yeast strains was monitored after incubating the plates for 48 h, as described elsewhere (Gamberi et al., [@B26]). Oxygen consumption and membrane potential measurements ------------------------------------------------------ Mitochondria were purified from yeast cells grown on YPEG as described previously (Hasson et al., [@B30]). Oxygen consumption measurements with isolated mitochondria were performed using an oxygen electrode (Hansatec) as described previously (Dabir et al., [@B19]). Membrane potential measurements of purified mitochondria were performed with fluorescent 3, 3′-dipropylthiadicarbocyanine iodide dye \[DiSC3(5)\]. 1% DMSO, carbonyl cyanide *m*-chlorophenyl hydrazone (CCCP), MB-6, or MB-7 was added to mitochondria in import buffer (0.6 M sorbitol, 2 mM KH~2~PO~4~, 60 mM KCl, 50 mM HEPES-KOH, 5 mM MgCl~2~, 2.5 mM EDTA, 5 mM L-methionine, pH 7.1) for 10 min. Subsequently 0.2 μM DiSC3(5) in import buffer was added, incubated for 5 min, and fluorescence was measured at excitation and emission length of 620 and 670 nm, respectively. Purification of mitochondria ---------------------------- Mitochondria were purified from wild-type yeast or yeast overexpressing Erv1 with a hexahistidine tag (\[a2up\] Erv1) grown in YPEG as described previously (Glick and Pon, [@B28]; Dabir et al., [@B20]). Yeast cultures were kept at 25°C with vigorous shaking during growth. Mitochondria concentration was measured by BCA assay and stored at 25 mg/mL at −80°C. Mitochondria with increased levels of Erv1 were purified from a strain in which Erv1 was overexpressed from a 2-micron plasmid (Dabir et al., [@B20]). Import of radiolabeled proteins into yeast mitochondria ------------------------------------------------------- Prior to import into purified mitochondria, \[^35^S\]-methionine and cysteine labeled proteins were generated with TNT Quick Coupled Transcription/Translation kits (Promega) and plasmids carrying the genes of interest. Transcription of genes was driven by either a T7 or SP6 promoter. Import reactions were conducted as previously described (Hasson et al., [@B30]; Dabir et al., [@B19]). After frozen mitochondria aliquots were thawed and added to the import buffer at a final concentration of 100 μg/mL, 1% DMSO or the small molecule was added as indicated. A final concentration of 1% DMSO was used in all experiments. Following incubation at 25°C for 15 min, import reactions were initiated by the addition of 5--10 μl of translation mix. Aliquots were removed at intervals during the reaction time course and import was terminated with addition either of cold buffer or 25 μg/mL trypsin, or the combination. If trypsin was added to digest non-imported precursor protein, soybean trypsin inhibitor was subsequently added in excess after 15 min incubation on ice. After a final recovery of by centrifugation (12,000 × g, 6 min), mitochondria were disrupted in Laemmli sample buffer. Samples from import reaction time points were resolved by SDS-PAGE and visualized by autoradiography. For experiments to investigate the Cmc1-Mia40 intermediate, non-reducing conditions were used. The import reactions were stopped in the presence of 20 mM iodoacetamide and mitochondria disrupted in Laemmli sample buffer lacking β-mercaptoethanol. The imported products were separated by non-reducing SDS-PAGE. *Caenorhabditis elegans* (*C. elegans*) infection study ------------------------------------------------------- L4-stage worms of *C. elegans* AU37 (sek-1; glp-4) strain (glp-4(bn2) were used to examine the antifungal efficacy of auranofin as described elsewhere (Mylonakis et al., [@B42]; Thangamani et al., [@B66]). Briefly, L4-stage worms were infected with *C. neoformans* NR-41292 for 2 h at room temperature. After infection, worms were washed with M9 buffer and treated either with DMSO or drugs (auranofin, fluconazole, and flucytosine), at a concentration of 8 μg/mL. After 24 h, worms were washed with PBS and disrupted using silicon carbide particles (Thangamani et al., [@B66]). The final suspensions were plated onto YPD agar plates containing ampicillin (100 μg/mL), streptomycin (100 μg/mL), and kanamycin (45 μg/mL) to determine the colony forming unit (CFU) per worm (Li et al., [@B37]). Statistical analyses -------------------- Statistical analyses were done using GraphPad Prism 6.0 (GraphPad Software, La Jolla, CA). *P* values were calculated via the Student *t*-test and *P*-values of ≤0.05 were deemed significant. Results {#s3} ======= Antifungal activity and killing kinetics of auranofin ----------------------------------------------------- The antifungal activity of auranofin was tested against various clinical isolates of *Candida* and *Cryptococcus*. Auranofin was very active in inhibiting the growth of all strains of *C. albicans, C. glabrata, C. tropicalis*, and *C. parapsilosis* with the MIC ranging from 1 to 16 μg/ml (Table [1](#T1){ref-type="table"}). Auranofin also displayed potent activity against both *C. neoformans* and *Cryptococcus gattii* inhibiting growth of these fungal species at a concentration ranging from 0.5 to 8 μg/ml (Table [1](#T1){ref-type="table"}). A time-kill assay was employed to investigate the killing kinetics of auranofin against both *C. albicans* and *C. neoformans*. Similar to fluconazole, auranofin (at the 5 × MIC) exhibited fungistatic activity against *C. albicans* ATCC 10231 and *C. neoformans* NR-41296 (Figure [1](#F1){ref-type="fig"}). However, at the 10 × MIC, auranofin and fluconazole exhibited fungistatic activity against *C. albicans* ATCC 10231, whereas unlike fluconazole, auranofin (at the 10 × MIC) completely kills *C. neoformans* NR-41296 after 48 h of incubation (Figure [1](#F1){ref-type="fig"}). {#F1} Antibiofilm activity of auranofin --------------------------------- The antibiofilm activity of auranofin, against *C. albicans*, was evaluated using the XTT reduction assay in order to measure the metabolic activity of fungal cells post-treatment. The metabolic activity of *C. albicans* was reduced by more than 70% with the treatment of auranofin at 8 × MIC (Figure [2A](#F2){ref-type="fig"}). However, the control antifungals fluconazole and flucytosine were ineffective (less than 10% reduction observed) at reducing metabolic activity of *C. albicans* biofilm, even at a concertation equivalent to 32 × MIC when compared to the DMSO-treated control groups (Figure [2A](#F2){ref-type="fig"}). {#F2} The effect of auranofin on reducing fungal biofilm density was further evaluated using confocal microscopy. Fungal cells stained with ConA conjugated with FITC revealed that auranofin (8 × MIC) eradicates a considerable portion of *Candida* cells in comparison to the control group (Figure [2B](#F2){ref-type="fig"}). However, treatment with fluconazole and flucytosine, even at 32 × MIC, appear similar to control group (Figure [2B](#F2){ref-type="fig"}). These results correlate with the results of the XTT reduction assay. Chemogenomic profiling identifies Mia40 as a potential target of auranofin -------------------------------------------------------------------------- To investigate the MOA, we subjected auranofin to chemogenomic profiling using the *S. cerevisiae* heterozygous deletion collection. Haploinsufficiency profiling (HIP) allows for the simultaneous assessment of the sensitivity of the pooled genome-wide set of heterozygous deletion strains because each strain is uniquely identified with a synthetic DNA barcode. The method is an unbiased approach to survey the genome-wide strain set in order to identify the strains with the most sensitivity to auranofin. We first identified the concentration that reduced wild-type growth by 30% and used 75 μM to profile the pooled heterozygous strains in biological samples. PCR was used to amplify the unique UPTAG DNA barcodes located at the gene deletion site and we tracked the barcode abundance with Illumina sequencing. The resulting counts were normalized and visualized using EdgeR (Figure [3A](#F3){ref-type="fig"}). We identified 85 heterozygous deletion strains that were under-represented based on an FDR ≤ 0.1 when comparing auranofin treatment to DMSO. These 85 strains were analyzed to identify associated gene ontology cellular component annotations and found to be enriched in several categories including the mitochondrial intermembrane space and chromatin components ([Supplementary data excel file](#SM1){ref-type="supplementary-material"}). Five heterozygous deletion strains within these enriched categories (*mia40*Δ, *acn9*Δ, *coa4*Δ, *rad18*Δ, and *nsi1*Δ) were selected to validate sensitivity to auranofin using a variety of growth assays (Figure [3A](#F3){ref-type="fig"}). These strains were randomly selected on the basis that they represented genes in the highest significance gene ontology categories including "regulation of translational initiation" (*p*-value = 0.00033) and "mitochondrial intermembrane space" (*p*-value = 0.0072) ([Supplementary data excel file](#SM1){ref-type="supplementary-material"}). In addition, as outlined in the following section, the *mia40*Δ strain was tested because it was implicated as sensitive to auranofin in a previous study (Lee et al., [@B35]). ![**Auranofin targets mitochondrial protein(s). (A)** Chemogenomic profiling of *S. cerevisiae* with treatment of auranofin. The strain abundance were normalized using EdgeR and shown. **(B)** Growth curve of wild type (BY4743) and heterozygous deletion strains (*mia40*Δ, *acn9*Δ, *coa4*Δ, *rad18*Δ, and *nsi1*Δ) in the presence of indicated concentration of auranofin in YPD broth were determined. **(C)** The percent growth of yeast cells (OD~600~ after 24 h) incubated with auranofin (6.25 μg/mL) in YPD broth was determined in relation to the DMSO treatment. The results are presented as means ± *SD* (*n* = 3). Statistical analysis was calculated using the two-tailed Student\'s *t*-test. *P*-values (^\*^*P* ≤ 0.05) (^\*\*^*P* ≤ 0.01) are considered as significant. **(D)** Yeast cells grown in YPD broth overnight were serially diluted and spotted on solid YPD agar containing auranofin (6.25 μg/mL) or DMSO and the CFU were shown. **(E)** Comparison of Lee et al. ([@B35]) HIP results with our 85 strains are shown as a Venn diagram.](fcimb-07-00004-g0003){#F3} Growth of these five heterozygous deletion strains and the wild-type (BY4743) strain were monitored in the presence of different concentrations of auranofin (6.25, 12.5, and 25 μg/mL) in a liquid growth assay. The result indicated that only three heterozygous deletion strains (*mia40*Δ, *acn9*Δ, and *coa4*Δ) exhibited drug-induced haploinsufficiency under these conditions. The growth of these deletion strains was suppressed, even in the presence of low concentrations of auranofin (6.25 μg/mL) (Figure [3B](#F3){ref-type="fig"}). However, auranofin does not induce haploinsufficiency in the other two deletion strains (*rad18*Δ and *nsi1*Δ) as growth of these strains, in the presence of auranofin, mimics the pattern observed with the wild-type strain (Figure [3B](#F3){ref-type="fig"}). These two deletion strains were not affected possibly because of the concentration used in our validation studies or because they may be false positives. For each strain, the growth of cells (OD~600~ after 24 h) incubated with auranofin (6.25 μg/mL) was determined in relation to DMSO treatment. The growth of three heterozygous deletion strains (*mia40*Δ, *acn9*Δ, and *coa4*Δ) was drastically suppressed by more than 50% in the presence of auranofin (6.25 μg/mL). However, the remaining two deletion strains (*rad18*Δ and *nsi1*Δ) had a modest reduction in growth of\~10% compared to the wild-type strain (Figure [3C](#F3){ref-type="fig"}). The growth of these five deletion strains was further confirmed by spotting serial dilutions of cultures on solid agar. As shown in Figure [3D](#F3){ref-type="fig"}, growth of the wild-type and five heterozygous deletion strains was normal in agar containing DMSO. However, the heterozygous deletion strain, *mia40*Δ, exhibited a nearly two-fold reduction in colony forming units when spotted onto YPD agar containing auranofin (6.25 μg/mL). A study conducted by Lee et al. ([@B35]) previously analyzed a heterozygous deletion pool, representing essential genes, using haploinsufficiency profiling with hundreds of compounds including auranofin. Examination of theauranofin-generated data set shows that they identified 17 strains as possibly sensitive. Comparison of Lee et al.\'s results with our 85 strains showed that two strains, *rho1*Δ and *mia40*Δ, overlapped between the data sets (Figure [3E](#F3){ref-type="fig"}). An additional study by Gamberi et al. ([@B26]) specifically assessed sensitivity and resistance of haploid deletion strains involved in mitochondrial function and found them to be differentially effects to auranofin compared to the haploid wild type parental strain. Based on studies by Gamberi et al. ([@B26]) and Lee et al. ([@B35]), we next moved to examine sensitivity of the corresponding heterozygous deletion strains involved in mitochondrial function and redox homeostasis that are possibly sensitive to auranofin. Heterozygous deletion strains with genes deleted in mitochondrial function and redox homeostasis experienced a significant growth reduction when treated with auranofin (6.25 μg/mL) relative to DMSO-treated cells (Figure [4A](#F4){ref-type="fig"}--green and gray bars), including *ndil*Δ*, atp2*Δ*, citl*Δ*, sdh4*Δ*, gsh1*Δ*, gsh2*Δ*, prx1*Δ*, trr1*Δ, *erv1*Δ*, toa2*Δ*, arp7*Δ*, ydl63w*Δ*, and yjl086c*Δ. These results are in agreement with Gamberi\'s et al. ([@B26]) results in that mitochondrial function appears to be a target of auranofin. It should be noted that Gamberi et al. used haploid deletion strains sensitivity, which generally does not inform on the direct target of a compound as opposed to the heterozygous deletion strains used in our study. In addition, Gamberi et al. examined resistance of haploid deletion strains and varied media conditions, resulting in their conclusion that Pos5 was the target---however, several other haploid deletions strains also demonstrated slight resistance to auranofin and heterozygous strain sensitivity was not examined. Our results demonstrate that the heterozygous *pos5*Δ strain is not sensitive to auranofin in liquid or agar conditions suggesting Pos5 is unlikely to be the direct target of auranofin (Figures [4A,B](#F4){ref-type="fig"}). The heterozygous strains that were identified by Lee et al. ([@B35]) were generally not as sensitive in liquid growth compared to mitochondrial or redox-related strains (Figure [4A](#F4){ref-type="fig"}---brown bars). Because this set of strains contained deletions of genes with a wide variety of functions it is likely that they not specific hits from the screen and shows the importance of testing individual strains to confirm sensitivity to auranofin. These results were confirmed using the YPD agar-spotting assay in that many strains were not sensitive. Interestingly, heterozygous deletion strains involved in ROS response and redox homeostasis (*sdh4*Δ*, gsh1*Δ*, gsh2*Δ, and *prx1*Δ) which had significant growth reduction in liquid medium did not demonstrate considerable reduction in growth when spotted onto YPD agar containing auranofin (6.25 μg/mL) (Figure [4B](#F4){ref-type="fig"}). {#F4} As noted earlier, heterozygous deletion strains that encode genes required for mitochondrial function (including *ndil*Δ*, atp2*Δ*, citl*Δ, and *erv1*Δ), showed a considerable decrease in colony count (almost one-fold log reduction) when spotted onto YPD agar containing auranofin (6.25 μg/mL) (Figure [4B](#F4){ref-type="fig"}). Interestingly, a deletion strain (*erv1*Δ), which forms a complex with Mia40 (Rissler et al., [@B50]), showed considerable sensitivity to auranofin, which coincides with Lee et al.\'s findings (Figure [4B](#F4){ref-type="fig"}). Taken altogether, our results as well as the overlap with the previous haploinsufficiency profiling (Lee et al., [@B35]), supports the notion that Mia40/Erv1 is the probable antifungal target of auranofin. Auranofin inhibits the import of Mia40 substrate (Cmc1) ------------------------------------------------------- To further confirm the specific inhibition of the Mia40-Erv1 pathway by auranofin we employed several biochemical experiments using purified yeast mitochondria similar to a previous study that investigated the effect of several small molecule inhibitors of redox-regulated protein import into mitochondria (Dabir et al., [@B19]). A possible indirect mechanism of inhibition of mitochondrial function and the Mia40-Erv1 pathway is by the disruption of membrane potential or diminished oxidative phosphorylation. Maintenance of membrane potential was determined by mitochondrial uptake of DiSC3 (5) dye and subsequent quenching in the presence of membrane potential. Auranofin had no effect on the membrane potential compared to DMSO whereas the uncoupling agent, CCCP, caused a 4-fold increase in fluorescence indicating uncoupled mitochondria (Figure [5A](#F5){ref-type="fig"}). The effect on mitochondrial respiration was determined by measuring dissolved oxygen in a chamber with purified mitochondria and respiration was initiated with NADH resulting in an oxygen consumption rate (−0.45 O~2~ nmol/s) consistent with well-coupled mitochondria. The addition of DMSO did not increase respiration rate and auranofin at a concentration of 34 μg/mL only slightly increased the respiration rate (−0.64 O~2~ nmol/s) (Figure [5B](#F5){ref-type="fig"} and Table S2). As a control, the addition of CCCP resulted in a severe increase in consumption rate (−1.15 O~2~ nmol/s) suggestive of uncoupled mitochondria (Figure [5B](#F5){ref-type="fig"}). {#F5} To confirm that auranofin targets the Erv1/Mia40 pathway we measured the effect of compound on the import of mitochondrial protein substrates compared to control compounds previously identified as Erv1 inhibitors (Dabir et al., [@B19]). Radiolabeled precursor proteins were incubated with mitochondria in the presence of small molecules or DMSO and the reaction was terminated with protease and subsequently analyzed by gel electrophoresis. Protein substrates from different import pathways were assessed including the Tim23 substrate, Su9-DHFR, and the Mia40 substrate, Cmc1. Auranofin at a lower concentration of 6.8 μg/mL inhibits import of Su-DHFR to a 60% level and Cmc1 to a 25% level compared to untreated samples (Figures [5C,D](#F5){ref-type="fig"}). These results indicate the preferential activity of auranofin toward inhibiting Cmc1 import compared to Su9-DHFR, which is expected because Cmc1 is directly imported by Mia40/Erv1. Strikingly, auranofin exhibits more potent activity than control compound, MB-7 with a drastic difference in import efficiency observed between the compounds at 10 μM (6.8 μg/mL for auranofin and 8.5 μg/mL for MB-7; Figure [5D](#F5){ref-type="fig"}). Although auranofin does inhibit Su9-DHFR import at high concentrations, these results demonstrate the compound has specificity toward the Mia40 pathway and increased potency compared to previously identified inhibitors from a large-scale chemical library screen (Dabir et al., [@B19]). It is not surprising that the import of Su9-DHFR is mildly inhibited because mitochondrial import pathways are interconnected. Mia40 has previously been demonstrated to form an intermediate with Cmc1 as part of the import process (Bourens et al., [@B9]; Neal et al., [@B43]). The effect of compounds on the formation of a disulfide intermediate between Mia40 and Cmc1 was monitored *in organello*. Auranofin inhibits radiolabeled Cmc1 from interacting with Mia40 in a similar dose dependent manner to MB-7 (Figure [5E](#F5){ref-type="fig"}). The addition of another control, MB-6, causes the accumulation of the intermediate. In sum, auranofin inhibits the heterodimer formation of the Mia40-Cmc1 intermediate and is a potent inhibitor of the Mia40 import pathway. Overexpression of Erv1 in yeast mitochondria confers resistance to auranofin ---------------------------------------------------------------------------- To further validate the Mia40 pathway as a target of auranofin, import of Cmc1 was performed with mitochondria from WT and Erv1 overexpressing yeast. Erv1 overexpression is expected to maintain the Mia40 pool in an oxidized state, which is required for the interaction with substrate proteins (Mesecke et al., [@B41]; Dabir et al., [@B20]) and hence should be more resistant to auranofin inhibition. As predicted, the Erv1 overexpressing mitochondria were resistant to auranofin (3.4 μg/mL) treatment as evidenced by the increased level of Cmc1 (60%) import compared to WT (30%) mitochondria providing further confirmation of Mia40 as a target (Figure [5F](#F5){ref-type="fig"}). *In vivo* efficacy of auranofin in *C. neoformans* infected *C. elegans* model ------------------------------------------------------------------------------ To investigate if the *in vitro* antifungal activity of auranofin translates *in vivo*, the antifungal efficacy of auranofin was examined in a *C. neoformans*-infected *C. elegans* animal model. As shown in Figure [6](#F6){ref-type="fig"}, treatment of infected *C. elegans* with fluconazole, flucytosine, and auranofin, at 8 μg/mL, produced a significant reduction (*P* ≤ 0.01) in mean fungal load when compared to the untreated control groups. Strikingly, *C. elegans* treated with auranofin (8 μg/mL) generated the largest reduction in *C. neoformans* CFU (0.87 ± 0.03 log~10~), followed by fluconazole (8 μg/mL) (0.82 ± 0.03 log~10~) and flucytosine (8 μg/mL) (0.58 ± 0.11 log~10~) (Figure [6](#F6){ref-type="fig"}). {#F6} Discussion {#s4} ========== Auranofin has a well-established pharmacological and toxicological profile that has permitted it to be used for the treatment of rheumatoid arthritis for more than 30 years (Bernhard, [@B6]; Furst et al., [@B25]). Independent of its antirheumatic effect, several studies have reported the anti-infective properties of this drug against important parasitic and bacterial pathogens including *Schistosoma mansoni, Trypanosoma brucei, Plasmodium falciparum, Entamoeba histolytica, Staphylococcus aureus*, and *Streptococcus pneumoniae* (Jackson-Rosario et al., [@B33]; Debnath et al., [@B22]; Cassetta et al., [@B14]; Harbut et al., [@B29]; Thangamani et al., [@B62],[@B63]). However, a dichotomy exists regarding the antimicrobial MOA of auranofin. Debnath et al. ([@B22]) and Harbut et al. ([@B29]) reported that auranofin exhibits its antimicrobial activity through the inhibition of the thioredoxin reductase (TrxR) enzyme in both *E. histolytica* and *S. aureus*. However, a recent crystallographic study conducted by Parsonage et al. ([@B44]) revealed that auranofin most likely does not bind to the cysteine residues in TrxR of *E. histolytica*. Another study, conducted by our research group, also demonstrated that TrxR is not the primary target of auranofin in bacteria (Thangamani et al., [@B62]). We demonstrated that auranofin inhibits multiple biosynthetic pathways including DNA, protein and cell wall synthesis in bacteria (Thangamani et al., [@B62]). However, the exact molecular target of auranofin, in bacteria, still remains unclear. The FDA\'s approval of auranofin as an anti-amoebic agent opened the door for researchers to examine additional clinical applications for this drug (Debnath et al., [@B22]). Recent studies, including the present work, demonstrate that auranofin inhibits the planktonic growth of multiple species of fungi including *C. albicans, C. glabrata, C. tropicalis, C. parapsilosis, C. neoformans*, and *C. gattii with* an average MIC of 4 μg/ml (Stylianou et al., [@B59]; Fuchs et al., [@B24]). In addition to planktonic growth, fungi especially, *Candida* spp., are known to form biofilms that are recalcitrant to treatment with antifungal agents. Fungal cells encased within the biofilm are resistant to most clinically used antifungals, including azole drugs, ultimately resulting in treatment failure (Chandra et al., [@B15]; Mathé and Van Dijck, [@B40]; Chandra and Mukherjee, [@B16]). Biofilm-related *C. albicans* infections thus pose a major threat to the public (Mathé and Van Dijck, [@B40]). Therefore, we examined the effect of auranofin on adherent *C. albicans* biofilms. Our results indicate that auranofin is very effective in disrupting *C. albicans* biofilm as this drug is able to reduce the metabolic activity of fungal cells present within the biofilm by more than 75% (relative to the control groups). Further examination of *C. albicans* biofilm using confocal microscopy revealed that auranofin treatment markedly reduces the number of fungal cells present in the biofilm compared to the control and treatment groups (fluconazole and flucytosine). These findings illustrate that auranofin is a potential candidate for use in treatment of biofilm-related fungal infections. After verifying auranofin\'s antifungal activity, we proceeded to investigate the antifungal mechanism of auranofin. To examine auranofin\'s antifungal MOA, chemogenomic profiling was employed given it is a highly-specific technique to investigate the target of unknown compounds (Giaever et al., [@B27]; Roemer et al., [@B52]). This technique uses drug-induced haploinsufficiency, where it causes a strain-specific fitness defect after treatment with compounds, and thereby aids in identifying the drug target (Giaever et al., [@B27]; Roemer et al., [@B52]). In our study, chemogenomic profiling of *S. cerevisiae* with auranofin identified three heterozygous deletion strains of genes (*mia40*Δ, *acn9*Δ, and *coa4*Δ) involved in mitochondrial function were found to be highly susceptible to auranofin. On the other hand, genes (*rad18*Δ *and nsi1*Δ) involved in chromatin function were found to not be affected possibly because of the concentration used in our validation studies or because they may be false positives from the profiling data. Our results are in agreement with Gamberi et al. ([@B26]) indicating the mitochondrial protein(s) is the potential target of auranofin. We further confirmed the three heterozygous deletion strains (*mia40*Δ, *acn9*Δ, and *coa4*Δ) were sensitive to auranofin when grown in YPD liquid medium with auranofin. Interestingly, only one deletion strain, *mia40*Δ, was very sensitive to auranofin in YPD agar containing the drug. Gamberi et al. demonstrated that another mitochondrial protein, the Pos5 NADH kinase, is thought to be the likely target of auranofin because the haploid deletion strain exhibited resistance. However, our chemogenomic profiling results did not identify *pos5*Δ heterozygous deletion strain as sensitive. When the *pos5*Δ heterozygous deletion strain was examined in our study, it was found to be not sensitive to auranofin. Chemogenomic profiling by Lee et al. also did not identify the *pos5*Δ heterozygous deletion strain as sensitive to auranofin (Lee et al., [@B35]). We also examined additional genes reported in Gamberi et al.\'s study and found heterozygous deletion strains involved in mitochondrial function (*ndil*Δ*, and atp2*Δ) showed moderate susceptibility to auranofin. This aligns with the results of Gamberi et al. where they observed resistance to auranofin in haploid deletion strains for these two genes (Gamberi et al., [@B26]). Chemogenomic profiling of 6000 genes in *S. cerevisiae* is a non-biased technique and further the gene (mia40) identified by this method also overlaps with data from Lee et al. ([@B35]). Auranofin is one of 300 compounds screened using haploinsufficiency profiling and results from Lee et al.\'s study indicates that mia40 is one of the target genes for auranofin (Lee et al., [@B35]). In addition, we also observed that another deletion strain (*erv1*Δ), which forms a complex with mia40 as reported in Lee et al.\'s study, also showed considerable sensitivity to auranofin (Rissler et al., [@B50]). The Mia40 (mitochondrial intermembrane space import and assembly protein 40)--Erv1 pathway is mainly involved in oxidation of several cysteine rich proteins that enter the mitochondria from the cytoplasm (Rissler et al., [@B50]; Banci et al., [@B4]). These proteins, present in the inner mitochondrial space, are essential for cell viability and are functionally linked to the respiratory chain (Rissler et al., [@B50]; Bihlmaier et al., [@B7]). In addition, an *ERV1* mutant strain was shown to be deficient in respiration (Lisowsky, [@B38]) consistent with the metabolic shift from respiration to fermentation observed in auranofin treated cells (Gamberi et al., [@B26]). Results from the present study indicates that auranofin does not have a generalized mode of action resulting in the disruption of membrane potential or respiration and mitochondrial integrity is maintained in the presence of the compound. However, auranofin preferentially inhibits the import of mitochondrial protein substrate Cmc1 which is directly imported by Mia40/Erv1pathway. Also, overexpression of Erv1confered resistance to auranofin which is directly noticed by the increased level of Cmc1 import. Taken together, our findings support the notion that auranofin preferentially targets Mia40/Erv1pathway in yeast. It should also be taken into account the affinity of auranofin to human Mia40 protein. The central part of the human homolog of Mia40 shares high sequence identity with most of its eukaryotic analog. However, Mia40 in yeast differs from its human homolog in one major respect---yeast Mia40 lacks the N-terminal extension including a transmembrane region (Banci et al., [@B4]). Future studies are needed to examine the affinity and binding of auranofin to human Mia40 protein. It may be possible that a therapeutic window exists because human Mia40 is not accessible or affected by auranofin at the concentrations needed for antifungal activity. As reported earlier, in bacteria and parasites the thioredoxin reductase gene was proposed to be the target of auranofin (Debnath et al., [@B22]). Gamberi et al. used homozygous deletion strains and demonstrated that auranofin does not displayed resistance to both the mitochondrial thioredoxin reductase (*TRR2*) and glutathione reductase (*GLR1*) genes in *S. cerevisiae* (Gamberi et al., [@B26]). However, the effect of auranofin on cytoplasmic thioredoxin reductase (*TRR1*) gene was not explored in that study (Gamberi et al., [@B26]). Results from our investigation indicate that the heterozygous deletion strain (*trr1*Δ) behaves similar to wild type. We therefore conclude that auranofin does not target the thioredoxin reductase gene in yeast which is in agreement with a previous study (Gamberi et al., [@B26]). The genes involved in ROS response (*SOD1, GSH1, GSH2*, and *PRX1*) were also examined in this study. Our results revealed that heterozygous deletion strains encoding these genes were sensitive to auranofin when grown in YPD liquid broth but not in YPD agar containing auranofin. Gamberi et al. through various experiments demonstrated that auranofin does not elicit the production of ROS (Gamberi et al., [@B26]) but some haploid deletion strains were sensitive suggesting they are selectively important for resistance to auranofin. Taken together it appears that the ROS response enzymes are not direct targets but some enzymes do mediate resistance to inhibitory activity by auranofin that is not due to generation of ROS. The final step in our study involved investigating the *in vivo* efficacy of auranofin in a *C. neoformans*-infected *C. elegans* animal model. Auranofin significantly reduced the mean fungal load in worms compared to control groups. Future studies are needed to test the efficacy of auranofin in an appropriate mouse model of fungal infection. Altogether, results from our study suggests that auranofin, with its unique MOA and potent *in vivo* antifungal activity, warrants further investigation as an antifungal agent to combat drug-resistant fungal infections. Auranofin has advantageous qualities to be repurposed as an antifungal agent, including oral bioavailability, clinically safe, potent broad-spectrum fungicidal activity, and the ability to cross blood brain barrier. The characteristics of auranofin as an antifungal agent offer a significant improvement over current approaches for treating fungal infections and provide valuable evidence that auranofin has significant promise to be repurposed as a novel antifungal drug. Author contributions {#s5} ==================== ST, MM, and LA performed the experiments. PP analyzed the sequencing results. ST, TH, MM, CK, and MS designed the study, analyzed the data and interpreted the results. ST, TH, MM, CK, HM, and MS wrote the manuscript. All authors reviewed and discussed the results. Funding {#s6} ======= TH was partially funded by the Bindley Bioscience Center Fellow program. This research was supported by NIH GM61721 and CIRM RT307678 to CK, the Ruth L. Kirschstein National Research Service Award GM007185 to MM. Research reported in this publication was also supported by the National Institute of Allergy And Infectious Diseases of the National Institutes of Health under Award Number R56AI114861 to MS. Conflict of interest statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. We thank Professor Debkumar Pain, from The State University of New Jersey, for sharing the *pos5*Δ heterozygous deletion strain of *S. cerevisiae*. We thank Rick Westerman and the Purdue Genomics Core for help with Illumina sequencing and data processing. Supplementary material {#s7} ====================== The Supplementary Material for this article can be found online at: <http://journal.frontiersin.org/article/10.3389/fcimb.2017.00004/full#supplementary-material> ###### Click here for additional data file. ###### Click here for additional data file. [^1]: Edited by: Martin G. Klotz, Queens College (CUNY), USA [^2]: Reviewed by: Derek Sullivan, Dublin Dental University Hospital, Ireland; Joy Scaria, South Dakota State University, USA; Tavan Janvilisri, Mahidol University, Thailand

There was a typographic error in the title of this article. The correct title is: Prokaryotic Ubiquitin-Like Protein (Pup) Proteome of Mycobacterium tuberculosis. The correct citation is: Festa RA, McAllister F, Pearce MJ, Mintseris J, Burns KE, et al. (2010) Prokaryotic Ubiquitin-Like Protein (Pup) Proteome of Mycobacterium tuberculosis. PLoS ONE 5(1): e8589. doi:10.1371/journal.pone.0008589 **Competing Interests:**No competing interests declared.

Conole D, Chou Y‐T, Patsiarika A, et al. Discovery of a novel fluorescent chemical probe suitable for evaluation of neuropilin‐1 binding of small molecules. Drug Dev Res. 2020;81:491--500. 10.1002/ddr.21641 31958155 **Funding information** British Heart Foundation, Grant/Award Number: PG/10/52/28448 FP : fluorescence polarization NRP1 : neuropilin‐1 SPR : surface plasmon resonance TGF‐β1 : transforming growth factor‐β1 VEGF : vascular endothelial growth factor 1. INTRODUCTION {#ddr21641-sec-0001} =============== The ability of neuropilin‐ 1 (NRP1) to bind and augment the action of growth factors such as vascular endothelial growth factor (VEGF), transforming growth factor‐β1 (TGF‐β1), placental growth factor (PLGF), HGF (scatter factor) and Semaphorins 3A, 4F (Pellet‐Many, Frankel, Jia, & Zachary, [2008](#ddr21641-bib-0019){ref-type="ref"}) are consistent with its emerging role as a tumor promoting receptor acting by a number of mechanisms. These mechanisms can be via modulation of the immune response to tumors through affecting the function of macrophages (Nissen, Selwood, & Tsirka, [2013](#ddr21641-bib-0017){ref-type="ref"}) and regulatory T cells (T~regs~) (Delgoffe et al., [2013](#ddr21641-bib-0003){ref-type="ref"}), via angiogenesis by promotion of NRP1/VEGF‐A signaling (Pan et al., [2007](#ddr21641-bib-0018){ref-type="ref"}); through prevention of tumor cell migration by binding to NRP1 (Jia et al., [2010](#ddr21641-bib-0011){ref-type="ref"}); or a direct effect on the tumor cells (Grun, Adhikary, & Eckert, [2016](#ddr21641-bib-0005){ref-type="ref"}). The binding site on NRP1 b1 is formed by loops in the protein structure forming a "receptor" for a C‐terminal arginine residue (Jarvis et al., [2010](#ddr21641-bib-0009){ref-type="ref"}). This is variously termed in the literature as the tuftsin site, arginine receptor, or aromatic box (Y297, W301, and Y353). The peptide and small molecule antagonists reported dating all bind to this site (Peng, Bai, Zhu, Hu, & Xu, [2019](#ddr21641-bib-0020){ref-type="ref"}). Many assay systems have been developed to detect NRP1 binding ranging from classical radiolabelled formats (Jia et al., [2006](#ddr21641-bib-0010){ref-type="ref"}), surface plasmon resonance (SPR), luminescence (Powell et al., [2018](#ddr21641-bib-0021){ref-type="ref"}) to bead based systems (Huang et al., [2019](#ddr21641-bib-0007){ref-type="ref"}) and homogeneous time resolved fluorescence (HTRF) (Auriau et al., [2018](#ddr21641-bib-0001){ref-type="ref"}). These assays may be expensive and complicated to implement. The development of simple and reliable assays for NRP1 binding and function is therefore of importance. Carboxyfluorescein (Flu)---labeled RPARPAR peptide is a known NRP1 ligand probe (National Center for Biotechnology Information, [2019](#ddr21641-bib-0015){ref-type="ref"}. PubChem BioAssay Database; AID = 602438, <https://pubchem.ncbi.nlm.nih.gov/bioassay/602438>) for NRP1 fluorescence polarization (FP) experiments (Figure [1](#ddr21641-fig-0001){ref-type="fig"}). However, while this peptide was reported as suitable for single point binding analysis in our hands it exhibited poor performance in competition experiments. We therefore initiated a study to identify a fluorescent probe suitable for the evaluation of ligand binding to the b1 or b1b2 domains of NRP1 (NRP1‐b1b2) via FP‐based ligand competition experiments. {#ddr21641-fig-0001} The new probes were based on the structure of EG01377 **2**, a compound identified in our group and known to be selective for NRP1 over the closely related NRP2 receptor and possess submicromolar NRP1 antagonistic properties (Powell et al., [2018](#ddr21641-bib-0021){ref-type="ref"}). The design incorporated several combinations of polyethylene glycol (PEG) and triazole click linker units to examine whether the fluorophore could be attached without compromising the intrinsic binding of EG01377 (Figure [1](#ddr21641-fig-0001){ref-type="fig"}). Three structurally distinct probes (3, 4, and DS108, Figure [1](#ddr21641-fig-0001){ref-type="fig"}) were synthesized and evaluated against the literature RPARPAR probe. We used SPR as an orthogonal biophysical technique to assess the binding properties of these probes and to determine their respective K~D~ values. From this, we were able to quickly identify conditions in which to assess the FP dynamic range of each probe, by optimizing probe and NRP1‐b1 concentrations. Subsequent NRP1‐b1 protein titration experiments and competition assays with known NRP‐b1 ligands (EG01377 and EG03286) confirmed the utility of this assay for the medium‐to‐high throughput discovery of novel ligands for NRP1‐b1 binding. 2. METHODS {#ddr21641-sec-0002} ========== 2.1. SPR {#ddr21641-sec-0003} -------- All SPR analysis was performed on a BIAcore T200 system using series S CM5 sensor chips. The Biotage SPR is effectively a stop‐flow instrument, and dissociation from the immobilized protein is initiated by the absence of analyte (ligand) when buffer alone is perfused, this setup allows the determination of on and off rates and equilibrium binding constants. Relatively high DMSO concentrations are normal for SPR experiments to limit solubility problems and minimize nonspecific aggregation. Extensive DMSO controls are included and automatically subtracted from the sensorgrams. Sensorgrams were double referenced by subtracting the response on a reference flow cell and a blank sample. Ligands were evaluated against the NRP1‐b1 domain. NRP1‐b1 was covalently attached to a CM5 chip via amine coupling (Powell et al., [2018](#ddr21641-bib-0021){ref-type="ref"}) with a surface density of 2,000--3,000 RU. Binding of novel fluorescent probes (0.4--100 μM) were analyzed by multicycle sequential injections (30--120 s association time) followed by undisturbed dissociation (30--60 s). A regeneration step was not used. Peptide stocks were dissolved in dimethyl sulfoxide (DMSO), and the final sample solutions for kinetic affinity experiments contained 3% DMSO in 1× phosphate‐buffered saline P20 buffer (PBS‐P, Cat no 28995084, GE Healthcare Ltd.). DMSO solvent effects were corrected for with eight calibration solutions (0.5--1.8% DMSO in PBS‐P). Equilibrium constants (K~D~) were calculated using either kinetic or affinity models, assuming simple 1:1 (Langmuir) binding. Data processing and analysis were performed using BIAevaluation and OriginPro software. The theoretical R~max~ (the maximal feasible signal between a ligand---analyte pair) for each compound/protein pair was calculated using Scheme [1](#ddr21641-fig-0007){ref-type="fig"} (Marquart, [2017](#ddr21641-bib-0013){ref-type="ref"}). {#ddr21641-fig-0007} The experimentally observed R~max~ was then calculated as a percentage of the theoretical R~max~ as a quality control measure. 2.2. FP {#ddr21641-sec-0004} ------- Initial experiments were performed using PBS‐P buffer with 3% DMSO in a final volume of 80 μl. The reaction plates or tubes were kept on ice during pipetting. Saturation binding: dynamic range and protein titration experiments were performed at probe concentrations selected with guidance from in‐house SPR analysis, literature K~D~ of the untagged compound, and an FP technical resource guide (Invitrogen, [2006](#ddr21641-bib-0008){ref-type="ref"}). The NRP‐b1 concentrations ranged from 10 nM to 30 μM. Samples were prepared in the following order---NRP1‐b1 protein (serial dilution as shown in Figure X, 40 μl) in PBS‐P buffer, and FP probes (concentrations shown in Tables [2](#ddr21641-tbl-0002){ref-type="table"}, 40 μl) in PBS‐P buffer. Competition experiments: FP samples were prepared in the following order---ligand in PBS‐P buffer (20 μl), NRP1‐b1 protein (1,400 nM, 20 μl) in PBS‐P buffer, and DS108 probe (1,500 nM, 40 μl) in PBS‐P buffer. FP was measured and normalized to experiment controls (buffer + probe, buffer + probe + protein) using a BMG Labtech PHERAstar® plate reader (filter settings: 485 nm \[excitation\] and 520 nm \[emission\]). Background FP was blanked using a PBS‐P buffer only control. ### 2.2.1. Final optimized protocol for FP {#ddr21641-sec-0005} The final FP assays were realized at 50 nM of DS108, 300 nM of NRP1 b1b2 All solutions contained a buffer constituted of 10 mM HEPES and 0.5% DMSO, reflecting the low initial concentration of the molecule under investigation needed for the assay. EG00229 and EG01377 starting concentrations were 5 μM only (stock solution consisted of 10 mM of the compounds diluted in 100% DMSO). DMSO derived from DS108 (stock solution 10 mM in 100% DMSO) is at a very low concentration; 100 times lower compared to the 0.5% derived from the compounds and did not interfere with the acquisition of the competition assay curves. The final sample volume in each well was 30 μl. Data were analyzed using OriginPro software. The assays were performed using PHERAstar®. From the general settings of the PHERAstar® fluorescence polarization software the Settling time was changed from 0.3 to 0.5 s, which is important in order to enhance the assay\'s accuracy at lower probe concentrations. Data processing: raw data were processed using OriginPro curve fitting software to obtain the IC~50~s and error values shown. A dose--response curve fitting model was utilized for the competition experiments. A web based IC~50~‐to‐K~i~ converter that computes K~i~ values from experimentally determined IC~50~ values was employed (Cer, Mudunuri, Stephens, & Lebeda, [2009](#ddr21641-bib-0002){ref-type="ref"}) (<https://bioinfo-abcc.ncifcrf.gov/IC50_Ki_Converter/index.php>) Each experiment was conducted in triplicate and repeated two times. The statistical reproducibility of this assay was evaluated using the z factor equation (Scheme [2](#ddr21641-fig-0008){ref-type="fig"}). {#ddr21641-fig-0008} 3. CHEMISTRY {#ddr21641-sec-0006} ============ All starting materials were from commercial sources or synthesized by literature procedures as indicated. FAM‐RPARPAR was purchased from Peptide Protein Research Ltd, Hampshire, UK. 3.1. Fluorescent probe 3: (S)‐5‐((1‐(4‐(7‐(N‐(2‐((1‐carboxy‐4‐guanidinobutyl)carbamoyl)thiophen‐3‐yl)sulfamoyl)‐2,3‐dihydrobenzofuran‐5‐yl)phenyl)‐3‐oxo‐6,9,12‐trioxa‐2‐azatetradecan‐14‐yl)carbamoyl)‐2‐(6‐hydroxy‐3‐oxo‐3H‐xanthen‐9‐yl)benzoic acid {#ddr21641-sec-0007} ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ### 3.1.1. Stage 1 {#ddr21641-sec-0008} 1377‐PEG‐amine. (3‐((5‐(4‐(14‐amino‐3‐oxo‐6,9,12‐trioxa‐2‐azatetradecyl)phenyl)‐2,3‐dihydrobenzofuran)‐7‐sulfonamido)thiophene‐2‐carbonyl)‐L‐arginine. To the Fmoc‐PEG‐acid (44.3 mg, 0.10 mmol, 1.0 eq) in DMF was added DMAP (36.7 mg, 0.3 mmol, 3.0 eq) and PYBOP (52.0 mg, 0.1 mmol, 1.0 eq). The reaction was then stirred for 10 min. The 1,377 di TFA salt (83.1 mg, 0.1 mmol, 1.0 eq) was added in portions. LCMS showed evidence for product but with partial loss of the Fmoc group. The DMF was removed on the rotary evaporator (\~1 mmHg) when LCMS indicated Fmoc loss complete. TFA (50 μl) was added to the residue with ACN/H~2~O 50/50 (3 ml). This was applied directly to the reverse phase column, eluting with a gradient of 5--95% ACN/H~2~O 0.1% TFA and the product isolated as the TFA salt and freeze‐dried to give a solid (37.5 mg, 0.041 mmol, 41.0%). ^1^H NMR (600 MHz, Acetonitrile‐*d* ~3~) δ 7.73--7.66 (m, 2H), 7.54--7.47 (m, 2H), 7.44 (d, *J* = 5.5 Hz, 1H), 7.35 (d, *J* = 8.1 Hz, 2H), 7.30 (d, *J* = 5.5 Hz, 1H), 4.68 (td, *J* = 8.8, 1.6 Hz, 2H), 4.52 (dd, *J* = 9.5, 4.9 Hz, 1H), 4.40 (s, 2H), 3.73 (t, *J* = 5.8 Hz, 2H), 3.66--3.62 (m, 2H), 3.61--3.56 (m, 8H), 3.25 (t, *J* = 8.8 Hz, 2H), 3.15 (hept, *J* = 6.9 Hz, 2H), 3.02 (t, *J* = 5.0 Hz, 2H), 2.52 (t, *J* = 5.8 Hz, 2H), 1.83 (ddt, *J* = 13.8, 9.5, 7.5 Hz, 1H), 1.69--1.62 (m, 2H). ^13^C NMR (151 MHz, CD~3~CN) δ 164.84, 158.05, 157.45, 143.21, 139.07, 139.00, 134.10, 132.93, 130.92, 130.35, 129.95, 128.79, 128.03, 127.73, 126.05, 121.74, 121.32, 114.13, 74.85, 71.04, 70.84, 70.54, 70.25, 67.71, 67.15, 52.94, 43.39, 41.69, 40.40, 36.90, 29.46, 28.72, 25.67. ### 3.1.2. Stage 2 {#ddr21641-sec-0009} To the amine TFA salt (24.6 mg, 24.166 μmol, 1.000 eq) in DMF was added 5‐carboxyfluorescein (9.1 mg, 24.166 μmol, 1.0 eq), DMAP (11.7 mg, 95.8 μmol, 3.96 eq) and PYBOP (12.0 mg, 23.059 μmol, 0.9 eq) stirred overnight then water (0.5 ml) added and the residue applied directly to a reverse phase (C18) column and eluted with ACN/H~2~O (5--95%) containing 0.1% TFA. Fractions containing product were concentrated on a rotary evaporator and freeze‐dried. Product isolated as a yellow solid (25.6 mg, 20.281 μmol, 83.9%). LCMS ESI \[M + H\]^+^ 1,148. 3.2. Fluorescent probe 4: (3‐((5‐(4‐(1‐(1‐(3‐(3′,6′‐dihydroxy‐3‐oxo‐3H‐spiro(isobenzofuran‐1,9′‐xanthene)‐5‐carboxamido)propyl)‐1H‐1,2,3‐triazol‐4‐yl)‐17‐oxo‐2,5,8,11,14‐pentaoxa‐18‐azanonadecan‐19‐yl)phenyl)‐2,3‐dihydrobenzofuran)‐7‐sulfonamido)thiophene‐2‐carbonyl)‐L‐arginine {#ddr21641-sec-0010} -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ### 3.2.1. Stage 1 {#ddr21641-sec-0011} (3‐((5‐(4‐(3‐oxo‐6,9,12,15,18‐pentaoxa‐2‐azahenicos‐20‐yn‐1‐yl)phenyl)‐2,3‐dihydrobenzofuran)‐7‐sulfonamido)thiophene‐2‐carbonyl)‐L‐arginine. To a stirred solution of **EG01377 2** (0.026 g, 0.045 mmol, 1.2 eq.) in DMF (2 ml), was added triethylamine (0.02 ml, 0.148 mmol, 4 eq.). A solution of acetylene‐PEG~4~‐NHS ester (0.015 g, 0.037 mmol 1 eq.) in DMF (1.5 ml) was then added, and left stirring overnight at RT. The crude was concentrated for reverse phase LC purification, with 0:100--85:15 methanol: water (0.1% formic acid) gradient elution, to give **5** as a colorless oil (0.03 g, 0.03 mmol, 95.0% yield). ^**1**^ **H NMR** (600 MHz, DMSO‐*d* ~6~) δ 1.63--1.77 (m, 3H, 25), 2.39 (t, *J* = 6.5 Hz, 1H, 8), 3.45--3.56 (m, 16H, 12, 13, 15, 16, 18, 19, 21, 22), 3.64 (t, *J* = 6.4 Hz, 2H, 24), 4.13 (d, *J* = 2.4 Hz, 2H), 4.23 (s, 1H), 4.29 (d, *J* = 5.9 Hz, 2H, 28), 4.59 (dq, *J* = 8.8, 13.9 Hz, 2H), 6.87 (s, 2H, 38), 7.16 (d, *J* = 5.4 Hz, 1H, 30), 7.23 (d, *J* = 5.4 Hz, 1H, 6), 7.28--7.32 (m, 2H, 5, 31), 7.46--7.49 (m, 2H, 4, 27), 7.53 (dd, *J* = 1.1, 2.1 Hz, 1H, 52), 7.68 (d, *J* = 2.0 Hz, 1H, 51), 8.22 (s, 1H, 34), 8.39 (t, *J* = 5.9 Hz, 1H), 10.09 (s, 1H, 54).^**13**^ **C NMR** (151 MHz, DMSO‐*d* ~6~) δ 39.13, 39.28, 39.28, 40.02, 40.22, 68.52, 69.60, 69.77, 69.82, 77.17. **LCMS**: MS *m/z* 874.4 \[M + H\]^+^. ### 3.2.2. Stage 2 {#ddr21641-sec-0012} To a stirred solution of N‐(3‐azidopropyl)‐3′,6′‐dihydroxy‐3‐oxo‐3H‐spiro\[isobenzofuran‐1,9′‐xanthene\]‐5‐carboxamide (23.6 mg, 0.027 mmol 1.0 eq.) and EG01377‐PEG~4~‐acetylene **5** (12.4 mg, 0.027 mmol 1.0 eq.) in 2.0 ml of ^*t*^BuOH/water (1:1) solution was added sodium ascorbate (50 mg, 0.27 mmol 10.0 eq.) dissolved in 0.5 ml of ^*t*^BuOH/water (1:1), followed by aqueous copper (II) sulfate pentahydrate (33.7 mg, 0.135 mmol 5.0 eq.) dissolved in 0.5 ml of ^*t*^BuOH/water (1:1). The mixture was left stirring at RT for 2 hr, after which it turned to red‐orange mixture and the solid product precipitated. The solvent was evaporated and 5 ml of aqueous solution of tris(3‐hydroxypropyltriazolylmethyl)‐amine (THPTA) (58.7 mg, 0.135 mmol 5.0 eq.) was added to the dried solid. The solution was filtered through IST Phase separator frit with a layer of celite, and the solid was washed with 5 ml of water to remove excess copper. Finally, DMF wash diluted the red solid layer of product, which was then obtained after solvent evaporation (5.1 mg, 0.004 mmol, 14.2% yield). The solid was \~85% pure by LCMS ESI 1331 \[M + H\]^+^. 3.3. Fluorescent probe DS108: (3‐((5‐(4‐(1‐(1‐(3‐(3′,6′‐dihydroxy‐3‐oxo‐3H‐spiro(isobenzofuran‐1,9′‐xanthene)‐5‐carboxamido)propyl)‐1H‐1,2,3‐triazol‐4‐yl)‐17‐oxo‐2,5,8,11,14‐pentaoxa‐18‐azanonadecan‐19‐yl)phenyl)‐2,3‐dihydrobenzofuran)‐7‐sulfonamido)thiophene‐2‐carbonyl)‐L‐arginine {#ddr21641-sec-0013} ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ ### 3.3.1. Stage 1 {#ddr21641-sec-0014} (3‐((5‐(4‐(((4‐(prop‐2‐yn‐1‐yloxy)benzyl)amino)methyl)phenyl)‐2,3‐ dihydro benzofuran)‐7‐sulfonamido)thiophene‐2‐carbonyl)‐*L*‐arginine. To the 4‐(prop‐2‐yn‐1‐yloxy)benzaldehyde (Enamine, EN300‐56436) (16 mg, 0.1 mmol) in DMF (1.6 ml) was added EG01377 di‐TFA salt (162 mg, 0.2 mmol), and acetic acid (30 μl) and the reaction stirred at 50° for 1 hr. Then sodium triacetoxyborohydride (80 mg, 0.4 mmol) was added and the reaction stirred at 50° overnight. Water (1 ml) was added to the reaction mixture which was applied directly to a reverse phase column and eluted with ACN/H~2~O (5--95%) containing 0.1% TFA. The product was isolated as a solid. Yield 14.5 mg, (0.020 mmol, 19.9%). LCMS ESI \[M + H\]^+^ 731. ^1^H NMR (600 MHz, Methanol‐*d* ~4~) δ 7.63 (dd, *J* = 15.3, 1.8 Hz, 2H), 7.53 (d, *J* = 7.4 Hz, 2H), 7.44 (d, *J* = 8.2 Hz, 2H), 7.37 (d, *J* = 5.5 Hz, 1H), 7.34 (d, *J* = 8.2 Hz, 2H), 7.16 (dd, *J* = 5.5, 2.2 Hz, 1H), 6.98 (d, *J* = 8.6 Hz, 2H), 4.67 (d, *J* = 2.4 Hz, 2H), 4.65--4.58 (m, 2H), 4.52--4.47 (m, 1H), 3.19--3.11 (m, 6H), 2.87 (t, *J* = 2.4 Hz, 1H), 2.00--1.92 (m, 1H), 1.79--1.72 (m, 1H), 1.67--1.57 (m, 2H). ^13^C NMR (151 MHz, Methanol‐*d* ~4~) δ 173.30, 163.85, 158.53, 157.06, 156.59, 141.35, 140.52, 132.48, 131.61, 131.02, 130.17, 129.89, 128.47, 126.88, 125.10, 123.41, 120.62, 120.33, 115.05, 77.82, 75.51, 73.47, 55.06, 51.57, 49.98, 49.87, 40.30, 28.13, 26.53, 25.06. ### 3.3.2. Stage 2 {#ddr21641-sec-0015} N‐(2‐(2‐(2‐(2‐azidoethoxy\]ethoxy)ethoxy)ethyl)‐3′,6′‐dihydroxy‐3‐oxo‐3H‐spiro\[isobenzofuran‐1,9′‐xanthene)‐6‐carboxamide. To 1‐Amino‐11‐azido‐3,6,9‐trioxaundecane (72 mg, 65 μl, 0.33 mmol) in DMF (1 ml) was added 5,6‐carboxyfluorescein (mixture of isomers) (125 mg, 0.33 mmol) followed by PyBOP (173 mg, 0.33 mmol) and DIPEA (84 mg, 113 μl, 0.6 mmol). And the reaction stirred overnight. Water (0.2 ml) was added to the reaction mixture and this applied directly to a reverse phase C18 column eluting with 10 to 90% ACN in water containing 0.1% TFA. The product was isolated as a yellow gum. Yield (74.5 mg, 0.129 mmol, 38.9%). LCMS ESI \[M + H\]^+^ 576. ^1^H NMR (600 MHz, Acetone‐*d* ~6~) δ 8.44 (dd, *J* = 1.6, 0.7 Hz, 0.5H), 8.31 (dd, *J* = 8.0, 1.6 Hz, 0.5H), 8.23 (dd, *J* = 8.0, 1.4 Hz, 0.5H), 8.06 (dd, *J* = 8.0, 0.8 Hz, 0.5H), 7.74 (dd, *J* = 1.4, 0.8 Hz, 0.5H), 7.38 (dd, *J* = 8.0, 0.8 Hz, 0.5H), 6.77 (t, *J* = 2.4 Hz, 2H), 6.70 (d, *J* = 8.7 Hz, 2H), 6.63 (ddd, *J* = 8.7, 2.4, 1.3 Hz, 2H), 3.72--3.67 (m, 1H), 3.66--3.58 (m, 7H), 3.57--3.54 (m, 0.5H), 3.53--3.47 (m, 5H), 3.41--3.30 (m, 2H), 2.08--2.03 (m, 2H). ^13^C NMR (151 MHz, Acetone) δ 168.91, 168.83, 165.95, 165.93, 162.28, 160.38, 160.36, 158.63, 158.36, 156.01, 154.11, 153.36, 153.33, 142.06, 137.63, 135.16, 130.36, 130.26, 130.17, 129.78, 128.17, 125.62, 125.16, 124.14, 123.38, 116.86, 114.97, 113.39, 113.34, 111.22, 111.18, 103.30, 103.28, 79.22, 79.01, 78.79, 71.32, 71.25, 71.18, 71.16, 71.13, 71.11, 70.97, 70.81, 70.70, 70.67, 70.61, 70.10, 69.95, 51.36, 51.32, 51.29, 47.23, 40.63, 40.56. ### 3.3.3. Stage 3 {#ddr21641-sec-0016} To a stirred solution of EG01377‐phenoxy‐acetylene (14.9 mg, 0.020 mmol, 1 eq) and the Flu‐PEG‐azide (11.8 mg, 0.020 mmol) in DMF (0.5 ml) was added the copper complex R3 (7.6 mg, 0.02 mmol) premixed with the lutidine R4 (0.1 ml, 0.86 mmol) in DMF (0.5 ml). The reaction was stirred overnight and then water (0.5 ml) added and the mixture applied directly to a reverse phase column eluted with ACN/H~2~O (5--95%) containing 0.1% TFA. The product was isolated as a yellow gum (12.0 mg, 0.009 mmol, 45.0%). LCMS ESI (MH + H)^+^ 1307. HRMS theoretical for \[C~65~H~66~N~10~O~16~S~2~ + H\]^+^ 1307.4172, measured 1307.4137 Error:−0.19 ppm. 4. RESULTS AND DISCUSSION {#ddr21641-sec-0017} ========================= 4.1. Chemistry {#ddr21641-sec-0018} -------------- The synthesis of the EG01377 derived probes was carried out using the unprotected EG01377 as the starting material. This minimized the number of synthetic steps required. Reaction mixtures could be purified directly by reverse phase chromatography. Fluorescent probe **3** was prepared from EG1377 in two steps (Figure [2](#ddr21641-fig-0002){ref-type="fig"}). First the Fmoc‐PEG‐acid was preactivated with PyBOP in DMF and DIPEA then the EG01377 added. The Fmoc group was removed in the work‐up then the resulting amine coupled with 5,6‐carboxyfluorescein. Probe **4** was synthesized using a similar coupling of alkyne‐PEG‐acid to EG01377 to provide alkyne **5** (Figure [3](#ddr21641-fig-0003){ref-type="fig"}). This was followed by a copper mediated Huisgen click cycloaddition reaction (Rostovtsev, Green, Fokin, & Sharpless, [2002](#ddr21641-bib-0022){ref-type="ref"}) with the fluorescein alkyl azide to give the product. For DS108 we wanted to preserve the basic amine group on the EG01377 structure which was known to be important for binding. EG01377 was converted to the phenoxy alkyne **7** by reductive amination with the commercially available aldehyde **6** (Figure [4](#ddr21641-fig-0004){ref-type="fig"}). Reaction of 5‐carboxyfluorescein with the PEG amino azide **8** provided the fluorescein‐PEG‐azide. Finally, a copper mediated click reaction gave the desired DS108. {#ddr21641-fig-0002} {#ddr21641-fig-0003} {#ddr21641-fig-0004} 4.2. SPR and FP {#ddr21641-sec-0019} --------------- We have previously validated the SPR system for NRP1 and it provides similar data to other assay systems such as biotinylated‐VEGF---luciferase or radiolabelled VEGF. Fluorescently tagged probes RPARPAR, **3**, **4**, and DS108 exhibited SPR K~D~ values of 69.03 ± 22, 24.6 ± 6.3, 3.3 ± 0.2, and 2.13 ± 0.81 μM, respectively (Figure [5](#ddr21641-fig-0005){ref-type="fig"}, Table [1](#ddr21641-tbl-0001){ref-type="table"}). This was compared with the positive control EG01377, which possessed a K~D~ = 1.32 ± 0.08 μM (Powell et al., [2018](#ddr21641-bib-0021){ref-type="ref"}). All probes demonstrated reasonable equilibrium binding characteristics as evidenced by their sensorgrams and calculated R~max~ values as shown in Table [1](#ddr21641-tbl-0001){ref-type="table"}. {#ddr21641-fig-0005} ###### Summary of SPR and FP binding properties of NRP1‐b1 fluorescent probes SPR --------- ------------- ------- RPARPAR 69.03 ± 22 60.61 **3** 24.6 ± 6.3 77 **4** 3.3 ± 0.2 43 DS108 2.13 ± 0.81 76.5 Abbreviations: FP, fluorescence polarization; NRP1, neuropilin‐1; SPR, surface plasmon resonance. With steady‐state equilibrium binding dissociation constant K~D~ values from SPR analysis in hand, fluorescently tagged RPARPAR, **3**, **4**, and DS108 were also evaluated for their utility as probes in FP experiments (Table [1](#ddr21641-tbl-0001){ref-type="table"}). Based on this information and FP guidelines (Moerke, [2009](#ddr21641-bib-0014){ref-type="ref"}) we assessed the dynamic FP window for the probes via a protein titration experiment (Figure [6](#ddr21641-fig-0006){ref-type="fig"}, Table [2](#ddr21641-tbl-0002){ref-type="table"}). {#ddr21641-fig-0006} ###### FP binding properties of fluorescent probes to NRP1‐b1 protein FP --------- ------ --------------------------------------------- ------ --------------- RPARPAR 7 2.81[b](#ddr21641-note-0004){ref-type="fn"} 22.6 1.88 ± 1.4 (**3**) 1 10 30.7 4.61 ± 0.19 (**4**) 0.3 3 5.2 0.62 ± 0.098 DS108 0.75 7.5 77.8 0.248 ± 0.097 Abbreviations: FP, fluorescence polarization; NRP1, neuropilin‐1. K~D~ derived from the 50% signal maximum. Maximum NRP1‐b1 concentration attempted. FP conditions should allow for \~50% of the probe being bound to the NRP1‐b1 protein, thus maximizing the potential readout window (Du, [2015](#ddr21641-bib-0004){ref-type="ref"}; Lakowicz, [1983](#ddr21641-bib-0012){ref-type="ref"}). Interestingly, the readout window between the "probe + NRP1‐b1" and "probe only" wells differed markedly for each probe (Table [2](#ddr21641-tbl-0002){ref-type="table"}). In our hands 5‐Flu‐RPARPAR probe showed poor affinity for NRP‐b1, as assessed by SPR, and this was coupled with only a modest dynamic window observed in FP (Tables [1](#ddr21641-tbl-0001){ref-type="table"} and [2](#ddr21641-tbl-0002){ref-type="table"}). Despite probe **4** demonstrating reasonable binding properties by SPR (K~D~ = 3.3 ± 0.2 μM), only a very small dynamic window could be obtained (Table [2](#ddr21641-tbl-0002){ref-type="table"}; Figures [5a](#ddr21641-fig-0005){ref-type="fig"}, and [6a](#ddr21641-fig-0006){ref-type="fig"}). In contrast, probe **3** and DS108 showed low μM SPR binding properties, and this correlated with improved FP dynamic window readouts (Table [1](#ddr21641-tbl-0001){ref-type="table"}). DS108 in particular appeared to possess a superior readout window, and this was confirmed by running full NRP1‐b1 titration curves for each probe (Figure [6a](#ddr21641-fig-0006){ref-type="fig"}). Pleasingly, DS108 not only showed the greatest FP window but this probe was also able to elicit robust FP dynamic windows at significantly lower NRP1‐b1 protein concentrations than the literature FP probe, 5‐Flu‐RPARPAR, which is an important practical consideration for high throughput drug screening. As a check on the specificity of the probe we evaluated it in a protein titration experiment with NRP1 b1b2 and the NRP1 b1 Y297A mutated protein (Figure [6b](#ddr21641-fig-0006){ref-type="fig"}). The probe showed very similar binding to both NRP1 b1 and the more complete receptor NRP1 b1b2. The NRP1 b1 Y297A mutant changes the structure of the binding domain and has been previously shown to reduce VEGF‐A binding (Herzog, Pellet‐Many, Britton, Hartzoulakis, & Zachary, [2011](#ddr21641-bib-0006){ref-type="ref"}). DS108 showed markedly reduced binding to this mutated protein. With a useful probe (DS108) and FP conditions now established, a competition experiment was performed. DS108 was first examined against the gold standard NRP1 antagonist EG00229 **1** and its analogue compound EG01377 **2**. Concentration‐dependent displacement of the probe was observed (Figure [6c](#ddr21641-fig-0006){ref-type="fig"}). We also wanted to discover whether a structurally diverse competitor would displace this probe. We also employed the bicyclic disulfide bonded peptide, EG03287, which is derived from the C‐terminal domain of VEGF‐A165, as a competitor compound in this assay. Gratifyingly, concentration‐dependent competition of DS108 was also observed with EG03287 (Figure [6c](#ddr21641-fig-0006){ref-type="fig"}) though we observed a reduced maximal signal. It is possible that the peptide ligand EG3287 is not able to fully displace the fluorescent probe. IC~50~ values were 20.47 ± 0.091, 16.84 ± 1.25, and 5.77 ± 0.87 μM EG00229, EG01377, and EG03287, respectively. To check the relevance of this assay the inhibition constant (K~i~) was calculated. Using kinetic equations (Nikolovska‐Coleska et al., [2004](#ddr21641-bib-0016){ref-type="ref"}), inhibition constants (K~i~ values) of 9.0, 7.39, and 2.47 μM were calculated for EG00229, EG01377, and EG03287, respectively, which were in agreement (within 1 log concentration unit) with competition and/or binding assays in the literature (Table [3](#ddr21641-tbl-0003){ref-type="table"}) (Jarvis et al., [2010](#ddr21641-bib-0009){ref-type="ref"}; Jia et al., [2006](#ddr21641-bib-0010){ref-type="ref"}; Powell et al., [2018](#ddr21641-bib-0021){ref-type="ref"}). In addition, the z‐factor for this assay was determined to be 0.90, suggesting its statistical robustness. ###### Summary of fluorescence polarization (FP) competition experiment analyses and comparison to literature values Compound IC~50~ (μM) Literature IC~50~ (μM)[a](#ddr21641-note-0006){ref-type="fn"} K~i~ (μM)[b](#ddr21641-note-0007){ref-type="fn"} ---------- --------------- ------------------------------------------------------------------ -------------------------------------------------- EG00229 20.47 ± 0.091 8.0 (Jarvis et al., [2010](#ddr21641-bib-0009){ref-type="ref"}) 9.0 EG01377 16.84 ± 1.25 1.32 (Powell et al., [2018](#ddr21641-bib-0021){ref-type="ref"}) 7.39 EG03287 5.77 ± 0.87 1.20 (Jia et al., [2006](#ddr21641-bib-0010){ref-type="ref"}) 2.47 *Note:* Z‐factor for this assay was 0.90. Literature values for NRP1 inhibition using bt‐VEGF~165~ binding, reference indicated. Ki values calculated using method in reference. Final optimization of the assay was conducted by evaluating different buffers, and lower probe and DMSO concentrations. Then, 50 nM of probe concentration was chosen as it was still detectable by the plate reader without compromising the assay\'s accuracy. First, different concentrations of NRP1 b1 and b1b2 were titrated with 50 nM of probe in order to find out whether this low concentration was able to create a satisfactory assay window. As shown in Figure [6d](#ddr21641-fig-0006){ref-type="fig"} NRP1 b1b2 domains created a good assay window. From the titration curve 300 nM of NRP1 b1b2 was chosen as it increased the initial signal by 65%. The optimal increase in signal is considered to lie between 50 and 80% as below 50% the assay window is poor and above 80% competition is visualized and the competition assay curve is expected to be a straight line. Secondly, a competition assay was performed using either EG00229 or EG01377. At 0.5% DMSO and using 10 mM of HEPES buffer the assay was able to generate a competition assay curve for both EG00229 and EG01377 (Figure [6e](#ddr21641-fig-0006){ref-type="fig"}). HEPES buffer was found to be superior to PBS or PBS‐P. Miniaturization of the initial FP assay can be exploited for high throughput screening for NRP1 b1b2 inhibitors with expected Ki at the micromolar range. FP assay instruments can read a 384‐well plate at around 10 min and at the same time the miniaturized nature of the optimized protocol constitutes a cost effective alternative to the existing screening methods. Finally, the initial protocol was unable to generate a VEGF165 (natural ligand of NRP1 b1b2) competition assay curve. However, the optimized protocol revealed that the reason for this was nonspecific binding of DS108 to VEGF165, suggesting once more the sensitivity of the optimized protocol (Figure [6f](#ddr21641-fig-0006){ref-type="fig"}). 5. CONCLUSION {#ddr21641-sec-0020} ============= We have identified DS108, a novel and convenient EG01377‐based fluorescent probe for evaluation of NRP1 binding, an important growth factor receptor that is implicated in the progression of various cancers. The probe was utilized as the ligand for a competitive FP binding assay. This probe could serve as a useful addition to the technical tools available for NRP1 study. CONFLICT OF INTEREST {#ddr21641-sec-0022} ==================== None to report. AUTHOR CONTRIBUTIONS {#ddr21641-sec-0023} ==================== E.D., D.S., and C.S. synthesized the compounds, E.D., V.N., Y.C., D.C., and F.M. did the biological assays. D.S. conceived the project. D.C., A.P., S.D., and D.S. wrote the paper. Supporting information ====================== ###### **Appendix** **S1**: Supporting information ###### Click here for additional data file. The authors acknowledge the British Heart Foundation (BHF Grant PG/10/52/28448), Ark Therapeutics Ltd and Magnus Life Sciences Ltd for funding.

1. Introduction {#sec1-nutrients-12-01063} =============== Prader--Willi syndrome (PWS) is a rare genetic disorder that is caused by lack of expression of paternal inherited genes located in the chromosome 15q11-q13 region \[[@B1-nutrients-12-01063]\]. The disease is present in 1:10,000--20,000 individuals and its most frequent genetic mechanisms are deletions in the paternal chromosomal region (approximately 65%--75% of cases) followed by maternal disomy of chromosome 15 (approximately 20%--30% of cases); other less frequent causes include imprinting defects or translocations in that chromosomal region \[[@B1-nutrients-12-01063]\]. PWS is characterized by a wide range of developmental and behavioral disturbances, including hypogonadism and hormonal dysfunction, hyperphagia, hypotonia, altered body composition (higher fat and lower lean mass), short stature, as well as dysmorphic features \[[@B1-nutrients-12-01063]\]. A complex hypothalamic dysregulation is currently thought to be responsible for this phenotype. Regardless of the genetic origin, pediatric patients show a very characteristic progression through different nutritional phases \[[@B2-nutrients-12-01063]\]. Feeding difficulties usually accompanied by failure to thrive appear during the first months of life. These are followed by a period of normal-rate weight gain until approximately two years of age. After this phase, weight gain starts steadily increasing even without a change in energy intake. At around four years of age, patients start showing increased appetite and interest in food, which precede the onset of hyperphagia \[[@B2-nutrients-12-01063]\]. Together with the insatiable hunger, aberrant behaviors typically occur in this latter phase, including food seeking behavior, constant thinking about food, withdrawn-depression symptoms, and social problems \[[@B3-nutrients-12-01063]\]. Therefore, these patients require close supervision by caregivers and strict food access control, with consequent negative effects on their quality of life and important distress for both patients and families. The occurrence of both abnormal body composition (reduced lean mass) and insatiable hyperphagia leads to reduced energy expenditure and increased caloric intake, respectively, favoring energy accumulation and resulting in excess of adiposity \[[@B4-nutrients-12-01063]\]. In fact, PWS is the most common cause of genetic obesity. Obesity is the main driver of comorbidities in these patients, including respiratory difficulties, sleep apneas, hypertension, steatohepatitis, and type 2 diabetes, responsible for their premature mortality \[[@B5-nutrients-12-01063]\]. Thus, one of the main goals of therapeutic strategies is preventing or delaying the early onset of obesity in order to change its clinical progression. Current management strategies include early interventions involving physical and cognitive stimulation, growth hormone therapy, and early dietary recommendations. Based on their lower energy requirements, these recommendations include limiting caloric intake by 20%--40%, together with a well-balanced macronutrient distribution and high fiber intake \[[@B1-nutrients-12-01063],[@B4-nutrients-12-01063],[@B6-nutrients-12-01063],[@B7-nutrients-12-01063],[@B8-nutrients-12-01063]\]. However, despite noticeable improvements in care during the last decade, adequate weight control continues to be extremely challenging and patients often develop obesity and associated complications. The gut microbiome has emerged as an important player in determining host health and energy balance \[[@B9-nutrients-12-01063],[@B10-nutrients-12-01063]\]. Its composition is highly affected by diet \[[@B11-nutrients-12-01063],[@B12-nutrients-12-01063]\], and several studies have shown a causative role of microbiota in developing obesity \[[@B10-nutrients-12-01063],[@B13-nutrients-12-01063]\]. However, research on the gut microbiome in subjects with PWS is very scarce. To date, only two studies have compared the microbiome of subjects with PWS with obesity and controls also with obesity \[[@B8-nutrients-12-01063],[@B14-nutrients-12-01063]\]. In this study, we aimed to identify dietary components closely linked to weight status and adiposity in children and adolescents with PWS. Furthermore, we analyzed the gut microbiome in these subjects to shed more light into the interrelationship between diet, gut microbiome, and obesity in this population. 2. Materials and Methods {#sec2-nutrients-12-01063} ======================== 2.1. Human Subjects {#sec2dot1-nutrients-12-01063} ------------------- This study was approved by the Hospital's ethical committee (code number PIC-73-17). Participants were recruited at the Hospital Sant Joan de Déu (Barcelona, Spain) and Parc Taulí Hospital Universitari (Sabadell, Spain), from September 2017 to June 2018. Both hospitals are reference centers for this disease in Catalonia. Inclusion criteria were genetic confirmation of PWS, age between 4.5 and 18-years-old, and absence of any major cognitive disorder that could preclude participation in the study. All parents signed the informed consent form; additionally, patients older than 12 years of age provided written informed assent. All participants except for one were on growth hormone therapy. Subjects had been patients at the hospital for at least one year before the study, and parents were proficient in nutrition and care related to PWS at the time of the study. Also, given that this disease is characterized by low muscle tone and patients are significantly more sedentary than the general population, we encourage enrollment in physical activity programs, which mostly includes individual sports (especially dance classes, swimming, and tennis). 2.2. Physiologic and Metabolic Variables {#sec2dot2-nutrients-12-01063} ---------------------------------------- Fasting blood tests and anthropometric measurements were obtained at the study visit. A stool sample obtained up to 48 h before was also collected at this time. Weight and height were determined with a digital scale and stadiometer by a qualified nurse. Body composition was determined on a Prodigy Lunar^®^ DXA scanner (Lunar Corp., Madison, WI, USA). Age- and sex-adjusted body mass index standard deviation scores (BMI-SDS) were calculated using WHO AnthroPlus software (World Health Organization, Geneva, Switzerland) \[[@B15-nutrients-12-01063]\]. Blood biochemical measurements were performed at the hospital laboratory following standard procedures. Homeostatic model assessment of insulin resistance (HOMA-IR) was calculated as described previously \[[@B16-nutrients-12-01063]\]. Hyperphagia levels were assessed with the Hyperphagia Questionnaire for Clinical Trials (HQ-CT), a nine-question questionnaire (score 0-36) developed for patients with PWS. 2.3. Dietary Analysis {#sec2dot3-nutrients-12-01063} --------------------- Parents filled a 4-day food diary (three weekdays and one weekend day) during the week prior to the study visit. Food records were discussed in a personal interview with the nutritionist during the study visit, where portion sizes and cooking methods were detailed. Participants were taking no probiotic or prebiotic supplements. Reviewed records were then analyzed with DIAL software \[[@B17-nutrients-12-01063]\], providing the macro- and micro-nutrient composition of the diet, as well as the average daily intake of different food groups. Considered food groups included grains, legumes, vegetables, fruit, meat, fish, dairy, eggs, oils and fat, and beverages (sodas and juices), and were calculated as % of total calorie intake. Energy intake was calculated as daily calorie intake and as % of daily recommended caloric intake (RCI) for each patient, based on age-, gender-, and BMI-matched subjects. 2.4. Gut Microbiota Analysis {#sec2dot4-nutrients-12-01063} ---------------------------- DNA from stool samples was isolated following Yuan et al. \[[@B18-nutrients-12-01063]\] with minor modifications, with the aid of QIAamp PowerFecal DNA Kit (Qiagen, Hilden, Germany) to avoid bias in DNA purification toward misrepresentation of gram-positive bacteria. For massive sequencing, the hypervariable region V3-V4 of the bacterial 16s gene was amplified using key-tagged eubacterial primers \[[@B19-nutrients-12-01063]\] and was sequenced with a MiSeq Illumina Platform (Illumina, Sant Diego, CA, USA) following Illumina recommendations for library preparation and sequencing for metagenomics studies. An average of 85,000 reads per sample were obtained. The resulting sequences were split taking into account the barcode introduced during the PCR reaction, while R1 and R2 reads were overlapped using PEAR program version 0.9.125 providing a single FASTQ file for each of the samples. Quality control of the sequences was performed in different steps. First, quality filtering (minimum threshold of Q20) was performed using fastx tool kit version 0.013. Next, primer (16s rRNA primers) trimming and length selection (reads over 300nts) was done with cutadapt version 1.4.126. These FASTQ files were then converted to FASTA files, and UCHIME program version 7.0.1001 was used to remove chimeras that could arise during the amplification and sequencing step. Those clean FASTA files were BLAST27 against the National Center for Biotechnology Information (NCBI) 16s rRNA database using blastn version 2.2.29+. The resulting XML files were processed using a python script developed by ADM Lifesequencing (Valencia, Spain) to annotate each sequence at different phylogenetic levels. 2.5. Statistical Analysis {#sec2dot5-nutrients-12-01063} ------------------------- Data are shown as mean and standard deviation (SD) or *n* and percentage (%) for continuous and categorical variables, respectively. Student's *t*-test or Chi-square test for continuous or categorical variables were used for group comparisons. Multivariate linear least-squares regression was applied to assess associations between % fat mass or BMI-SDS (dependent variables) and dietary variables (independent variables), adjusted by sex, age, physical activity, genotype, metformin use, and energy intake. Results are shown as effect size (B) with 95% confidence intervals (CI). For microbiome analysis, DESeq2 package from R (R Core Team, 2012) was used to generate a generalized linear model with fixed effects with negative binomial family to compare operational taxonomic unit (OTU) counts between groups, using group (normal weight (NW) and overweight or obesity (OWO)) as a fixed effect. Count normalization was made by rarefaction and significance assessed with the Wald test and the Benjamini--Hochberg correction to adjust for false discovery rate (FDR). Spearman correlation was used to assess associations of physiologic and dietary variables with the selected genus. A *p* \< 0.05 was considered significant. Analyses were performed in JMP^®^ v14.3 (SAS, Cary, NC, USA). 3. Results {#sec3-nutrients-12-01063} ========== 3.1. Subject Characteristics {#sec3dot1-nutrients-12-01063} ---------------------------- Demographic and metabolic characteristics of participants are described in [Table 1](#nutrients-12-01063-t001){ref-type="table"}. Our cohort included 31 subjects (19 female and 12 male) within a range of ages (5 to 18 years-old) and BMI-SDS (−1.24 to 4.5 standard deviations). Subjects were divided into normal weight (NW, *n* = 12) and overweight or obesity (OWO, *n* = 19) groups based on BMI-SDS below or over 1 standard deviation, respectively, following the WHO criteria for overweight and obesity. Groups showed no differences in age, genotype distribution, or hyperphagia levels; a slightly higher proportion of female subjects and patients with metformin therapy was included in the OWO group (*p* = 0.075 and *p* = 0.061, respectively). Fifty-five per cent of subjects in our cohort performed more than 2 h per week of exercise, and there were no differences between NW and OWO groups. As expected, patients from the OWO group had significantly higher BMI-SDS and body fat mass than NW patients ([Table 1](#nutrients-12-01063-t001){ref-type="table"}). Although in general subjects showed a normal lipid profile (triglycerides \<150 mg/dL, low-density lipoprotein (LDL)-cholesterol \<130 mg/dL) \[[@B20-nutrients-12-01063]\], patients in the OWO group had higher LDL-cholesterol (*p* = 0.036) and a trend towards increased total cholesterol (*p* = 0.069) compared to the NW group ([Table 1](#nutrients-12-01063-t001){ref-type="table"}). Fasting glucose and HbA1c were also within the normal range in both groups (glucose \<100 mg/dL, HbA1c \<6%), while insulin levels were on the higher end of the normal range for our laboratory reference values. When comparing both groups, fasting insulin levels and HOMA-IR were two-fold higher in OWO than in NW group (*p* \< 0.05 for both; [Table 1](#nutrients-12-01063-t001){ref-type="table"}), indicating some degree of insulin resistance in children and adolescents with overweight and obesity compared to normal weight patients. 3.2. Dietary Analysis of Children and Adolescents with Prader--Willi Syndrome {#sec3dot2-nutrients-12-01063} ----------------------------------------------------------------------------- The dietary analysis revealed that energy intake was 80% the recommended caloric intake (RCI), in agreement with the recommendations for PWS ([Table 2](#nutrients-12-01063-t002){ref-type="table"}). Macronutrient distribution analysis showed a well-balanced diet within the recommended ranges \[[@B21-nutrients-12-01063]\]. Interestingly, we observed no differences in energy intake between OWO and NW groups ([Table 2](#nutrients-12-01063-t002){ref-type="table"}). Similarly, there were no major differences in macronutrient distribution or fiber intake, although diet from OWO subjects was slightly higher in fat and lower in carbohydrates than the diet from NW patients ([Table 2](#nutrients-12-01063-t002){ref-type="table"}). Regarding micronutrients, while intake of iron, folic acid, and vitamins A, B6, B12, and E achieved the recommended intake level, calcium and vitamin D were significantly below the threshold ([Figure S1a](#app1-nutrients-12-01063){ref-type="app"}). No differences in micronutrient intake were observed between NW and OWO groups ([Figure S1b](#app1-nutrients-12-01063){ref-type="app"}). Regarding intake of the different food groups, normal weight patients showed significantly higher fruit and lower meat intake than those overweight or obese ([Table 3](#nutrients-12-01063-t003){ref-type="table"}). Intake of other food groups, including grains, legumes, vegetables, dairy, fish, eggs, oils and fats, and beverages were similar between groups. 3.3. Associations between Dietary Variables and the Degree of Obesity {#sec3dot3-nutrients-12-01063} --------------------------------------------------------------------- To identify dietary determinants of the degree of obesity in these patients, we applied multivariate least-squares linear regression to assess potential associations between the different dietary variables and BMI-SDS or % of body fat mass. We observed no association between total calorie intake and BMI-SDS or adiposity ([Table 4](#nutrients-12-01063-t004){ref-type="table"}). Among macronutrients, only saturated fat intake was significantly associated with both variables ([Table 4](#nutrients-12-01063-t004){ref-type="table"}). Although markedly weaker, carbohydrate intake showed an inverse association with BMI-SDS ([Table 4](#nutrients-12-01063-t004){ref-type="table"}). Regarding food groups, intake of grain, legume, vegetable, dairy, fish, eggs, oils and fats, or beverages did not correlate to BMI-SDS or fat mass ([Table 5](#nutrients-12-01063-t005){ref-type="table"}). Notably, meat intake showed a direct association with both BMI-SDS (*p* = 0.001) and body fat mass (*p* = 0.026) in these subjects, while fruit intake was inversely associated with both variables ([Table 5](#nutrients-12-01063-t005){ref-type="table"}). In agreement with the nutritional composition of these two food groups, meat consumption was correlated to saturated fat intake (Pearson coefficient *r* = 0.41, *p* = 0.022) and fruit consumption was correlated to carbohydrate intake (Pearson coefficient *r* = 0.43, *p* = 0.017). 3.4. Gut Microbiota Composition in Children and Adolescents with Prader--Willi Syndrome {#sec3dot4-nutrients-12-01063} --------------------------------------------------------------------------------------- The analysis of the gut microbiota by 16S profiling showed higher phylogenetic diversity in normal weight patients compared to those overweight or obese (Shannon index; [Figure 1](#nutrients-12-01063-f001){ref-type="fig"}a). While there were no major changes in community structure as shown by permutational analysis of variance (PERMANOVA *p* = 0.146, [Figure 1](#nutrients-12-01063-f001){ref-type="fig"}b), we observed differential abundance of seven microbial taxa at the genus level in the categorical comparison between NW and OWO groups ([Figure 1](#nutrients-12-01063-f001){ref-type="fig"}c). Three genera had higher abundance in OWO compared to NW group, namely *Klebsiella*, *Lactobacillus,* and *Eubacterium*, while genera *Lachnoclostridium*, *Murimonas*, *Alistipes,* and *Prevotella* had lower abundance ([Figure 1](#nutrients-12-01063-f001){ref-type="fig"}c). Three of these genera (*Klebsiella*, *Murimonas*, and *Alistipes*) were still significant after correction for multiple comparisons (FDR adjusted *p* \< 0.05). OTU abundance at genus level and fold change between groups are listed in [Table S1](#app1-nutrients-12-01063){ref-type="app"}. Interestingly, we observed a gradation effect when further subdividing OWO patients into overweight (1 ≤ BMI-SDS \< 2) and obesity (BMI-SDS ≥ 2) in both phylogenetic diversity and abundance of the selected genera ([Figure S2](#app1-nutrients-12-01063){ref-type="app"}). We next assessed potential correlations between the selected genera and physiologic variables. Correlation coefficients and significance levels are depicted in [Table S2](#app1-nutrients-12-01063){ref-type="app"}. Among them, lower *Alistipes* abundance was highly associated with higher body fat mass ([Figure 1](#nutrients-12-01063-f001){ref-type="fig"}d). Furthermore, *Alistipes* was also inversely correlated to total and LDL-cholesterol levels, as well as glucose homeostasis parameters including fasting insulin levels and HOMA-IR ([Figure 1](#nutrients-12-01063-f001){ref-type="fig"}d). Among the other genera, only *Klebsiella* showed a correlation to total and LDL-cholesterol ([Figure 1](#nutrients-12-01063-f001){ref-type="fig"}d). Notably, while fruit intake was not associated to abundance of any of these genera, meat consumption was directly correlated to *Eubacterium* and inversely correlated to *Alistipes* and *Lachnoclostridium* abundance. 4. Discussion {#sec4-nutrients-12-01063} ============= Dietary management is one of the critical aspects that determine health and quality of life in patients with PWS. Based on their lower energy requirements, current dietary recommendations for PWS are based on a 20%--40% caloric restriction \[[@B1-nutrients-12-01063],[@B4-nutrients-12-01063],[@B6-nutrients-12-01063]\]. Different types of calorie-restricted diets have been proposed to improve weight management in patients with PWS \[[@B2-nutrients-12-01063],[@B6-nutrients-12-01063],[@B8-nutrients-12-01063],[@B22-nutrients-12-01063]\]. In addition to energy restrictionS, a well-balanced macronutrient distribution and high fiber content are also important for successful weight control \[[@B7-nutrients-12-01063],[@B8-nutrients-12-01063]\]. In agreement with the general recommendations, children and adolescents from our cohort had an average of 20% calorie restriction in their diets, with a well-balanced macronutrient distribution and a fairly high fiber intake. These patients also had an adequate intake of micronutrients, except for calcium and vitamin D which were below the recommended intakes. Due to the calorie restriction, dietary deficits in some micronutrients often occur in patients with PWS \[[@B23-nutrients-12-01063],[@B24-nutrients-12-01063]\]. Interestingly, both calcium and vitamin D were also found to be below the recommended intake in a US-based cohort of youth with PWS \[[@B24-nutrients-12-01063]\]. Our results showed no differences in hyperphagia scores or energy intake between normal weight subjects and those overweight or obese, indicating that total calories were not contributing to obesity in this population. These data are in agreement with a study by Miller et al. showing that a group of children with PWS that followed a well-balanced macronutrient distribution diet achieved much better weight control than another group that followed a simple energy-restricted diet, even when both diets had similar caloric content \[[@B7-nutrients-12-01063]\]. All together, these data strongly suggest that diet quality plays an important role in weight management of patients with PWS. Indeed, our results show that those patients with lower meat and higher fruit consumption showed better weight management independently of total calorie intake. From observational investigations to longitudinal cohorts and intervention studies, a myriad of studies has analyzed the dietary determinants of obesity in the general population. However, due to the complex etiology of obesity with multiple genetic and environmental factors and confounders, research has only been able to identify very few nutrients associated with weight management, including fiber, fat, fruit and vegetable intake, sugar-rich drinks, or energy-dense micronutrient-poor foods \[[@B25-nutrients-12-01063],[@B26-nutrients-12-01063],[@B27-nutrients-12-01063]\]. It is worth mentioning that the parents of children with PWS are generally extremely involved in their care and nutrition plan, as shown by the well-balanced macronutrient distribution in their diets. Furthermore, sugary drinks or high-energy foods are practically absent from their diets, and most likely do not significantly contribute to weight gain in these children. While fat intake was once considered an important dietary determinant of obesity, a number of studies have now shown only weak or no association of total fat intake with weight gain \[[@B28-nutrients-12-01063],[@B29-nutrients-12-01063],[@B30-nutrients-12-01063]\]. In the Nurses' Health Study, higher intake of saturated fat from animal origin showed a much stronger association to weight gain than total fat intake \[[@B30-nutrients-12-01063]\]. Consistent with these data, we observed that saturated fat intake was associated with both BMI-SDS and adiposity in our cohort. Saturated fat and meat intake were correlated to each other in our cohort, indicating that meat was an important source of saturated fat from animal origin. These data suggest that meat consumption might favor energy accumulation in the context of PWS, in agreement with other studies that have linked meat consumption, especially red and processed meat, with BMI in adult general population \[[@B31-nutrients-12-01063],[@B32-nutrients-12-01063]\]. More consistent in the literature is the link between fiber intake and weight status \[[@B25-nutrients-12-01063]\], and increasing dietary fiber content is a general recommendation for patients with PWS. In fact, higher dietary fiber intake is associated with better weight management and weight loss in children and adolescents with PWS \[[@B7-nutrients-12-01063],[@B8-nutrients-12-01063]\]. In our study, we observed no association of fiber intake with BMI-SDS or adiposity. This could be explained by the relatively high fiber content that was already present in their diet. Increasing fruit and vegetable intake has also been favor as a strategy for weight loss \[[@B26-nutrients-12-01063]\]. While we observed no association of obesity with vegetable intake in our cohort, lower fruit intake was correlated to higher BMI-SDS and adiposity. During the last decade, an increasingly large number of studies have implicated the gut microbiome in host energy homeostasis and body weight management \[[@B10-nutrients-12-01063],[@B13-nutrients-12-01063],[@B33-nutrients-12-01063]\]. However, only two studies have examined the gut microbiome in subjects with PWS, both comparing the microbiota from patients with PWS and obesity to controls with obesity \[[@B8-nutrients-12-01063],[@B14-nutrients-12-01063]\]. A study in children with obesity found no major differences in microbiota diversity and composition between patients with PWS and controls \[[@B8-nutrients-12-01063]\]. Conversely, a recent study in adult subjects with obesity observed higher phylogenetic diversity and differential abundance of certain taxa in subjects with PWS compared to matched controls \[[@B14-nutrients-12-01063]\]. Several factors could explain this apparent discrepancy between the studies, including different age-ranges (children versus adults), geographical localization, and ethnicity. Furthermore, in the adult study no differences in microbiota composition between subjects with PWS and their non-PWS parents were observed \[[@B14-nutrients-12-01063]\], suggesting that diet and other environmental factors could be playing an important role in determining gut microbiome structure. In our study, we compared the microbiome from normal weight patients and overweight or obese patients in the context of PWS. Notably, we observed lower phylogenetic diversity in overweight and obese patients compared to patients with normal weight. Indeed, decreased microbiota richness has been consistently associated with obesity and poor metabolic health in the general population, including impaired glucose homeostasis and increased adiposity \[[@B33-nutrients-12-01063],[@B34-nutrients-12-01063]\]. Our study has identified several microbial taxa at the genus level associated with obesity status in children and adolescents with PWS. Three of these genera sustained adjustment for multiple comparisons, namely *Klebsiella*, *Murimonas*, and *Alistipes*. Among these genera, *Alistipes* showed a highly significant inverse association with body fat mass and was also inversely correlated to cholesterol levels and insulin resistance in our cohort. These data indicate that lower *Alistipes* abundance is linked to a poor metabolic health phenotype, at least in the PWS context. Interestingly, other studies have found lower *Alistipes* abundance in individuals with obesity compared to normal weight controls \[[@B35-nutrients-12-01063],[@B36-nutrients-12-01063],[@B37-nutrients-12-01063],[@B38-nutrients-12-01063]\]. Furthermore, *Alistipes* was identified as a predictor of successful weight loss in a two-year intervention in adult subjects with obesity \[[@B39-nutrients-12-01063]\]. Together with these data, our results support a potential beneficial role of *Alistipes* in the context of obesity and metabolic health. Our results showed that lower *Alistipes* abundance was also linked to higher meat intake. In fact, diet is one of the main determinants of gut microbiota composition \[[@B11-nutrients-12-01063],[@B12-nutrients-12-01063]\]. In patients with PWS, Zhang et al. showed that a 30-day intervention with a diet rich in non-digestible carbohydrates led to changes in the gut microbiome and improved body weight and metabolic health in children with PWS \[[@B8-nutrients-12-01063]\]. By means of microbiota transplants into gnotobiotic mice, the authors demonstrated that these changes in the microbiome contributed to the beneficial effects of the intervention \[[@B8-nutrients-12-01063]\]. Thus, modulating the gut microbiome with probiotic supplementation or specific dietary patterns could provide a therapeutic strategy for PWS. Further research is warranted to determine whether meat consumption modulates intestinal abundance of *Alistipes* or other microbial taxa impacting obesity status and metabolic health. The limitations of this study include that the dietary analysis is based on parental reported food diaries; while in general the parents and caregivers are well-trained regarding the quality of the diet and are proficient in filling up food diaries, we cannot exclude potential omissions due to foods consumed without parental knowledge. Furthermore, all subjects are from a specific geographical area (Catalonia), and further studies would be necessary to determine whether meat and fruit intake are also associated with the degree of obesity in patients with PWS from other regions of the world, with different dietary and lifestyle patterns. Finally, this is a cross-sectional study that describes associations between dietary variables, gut microbial abundances, and obesity; therefore, causality cannot be inferred. 5. Conclusions {#sec5-nutrients-12-01063} ============== In summary, our results indicate that obesity status and adiposity in children and adolescents with PWS are not associated with total caloric intake but correlate to higher meat and lower fruit consumption. Further studies are warranted to determine whether limiting meat and increasing fruit consumption can provide a strategy for weight control in children with PWS. The authors wish to sincerely thank the families and children that voluntarily participated in this study. We are indebted to the "Biobanc de l'Hospital Infantil Sant Joan de Déu per a la Investigació" integrated in the Spanish Biobank Network of ISCIII for sample processing and procurement. The following are available online at <https://www.mdpi.com/2072-6643/12/4/1063/s1>, Figure S1: Micronutrient intake in children and adolescents with Prader--Willi syndrome, Figure S2: Microbiome composition in normal weight (NW), overweight (OW), and subjects with obesity (OB); Table S1: Differentially abundant operational taxonomic units (OTUs) between NW and OWO groups; Table S2: Correlations between selected taxa and the indicated physiologic and dietary variables. ###### Click here for additional data file. Conceptualization of the study, M.R.-K. and C.L.; recruitment of participants, R.C., L.I., and M.R.-K.; patient visits and data acquisition, S.G.-R. and M.A.-B.; nutritional analysis, S.G.-R. and M.L.; microbiome analysis, E.C. (Eric Climent) and E.C. (Empar Chenoll); global data analysis, S.G.-R., M.A.-B., and C.L.; writing---original draft preparation, M.A.-B. and C.L.; writing---review and editing, all authors; funding acquisition, C.L. All authors have read and agreed to the published version of the manuscript. This research was funded by the Barcelona Magic Line fundraising program to C.L. The authors declare no conflict of interest. The funder had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results. {#nutrients-12-01063-f001} nutrients-12-01063-t001_Table 1 ###### Demographic and metabolic characteristics of participants. -------------------------------------------------------------------------------------------- Variables All Subjects\ NW\ OWO\ *p*-Value ^a^ (*N* = 31) (*n* = 12) (*n* = 19) -------------------------------- --------------- ------------- ------------- --------------- Gender (Females) 19 (61%) 5 (42%) 14 (74%) 0.075 Age (years) 12.0 (4.0) 11.4 (3.9) 12.4 (4.2) 0.498 Pre-pubertal status 12 (39%) 6 (50%) 6 (32%) 0.306 Genotype (Deletions) 18 (58%) 5 (42%) 13 (68%) 0.243 Hyperphagia (HQ-CT Score) 6.8 (6.0) 7.3 (7.4) 6.6 (5.1) 0.787 Physical activity (\>2 h/week) 17 (55%) 8 (67%) 9 (47%) 0.290 Growth hormone therapy 30 (97%) 12 (100%) 18 (95%) 0.317 Metformin therapy 8 (26%) 1 (8%) 7 (37%) 0.061 BMI-SDS 1.51 (1.38) 0.22 (0.55) 2.32 (1.09) **\<0.001** Body fat mass (%) 43.2 (8.4) 35.9 (3.9) 47.9 (7.0) **\<0.001** Lipid profile   Triglycerides (mg/dL) 70 (25) 64 (26) 74 (25) 0.294   Cholesterol (mg/dL) 170 (36) 155 (34) 179 (35) 0.069   LDL-cholesterol (mg/dL) 102 (31) 88 (30) 112 (28) **0.036**   HDL-cholesterol (mg/dL) 56 (13) 55 (13) 58 (14) 0.529 Glucose metabolism   Glucose (mg/dL) 87 (9) 85 (10) 88 (9) 0.352   HbA1c (%) 5.3 (0.2) 5.2 (0.3) 5.3 (0.2) 0.692   Insulin (mU/L) 12.6 (9.2) 8.0 (7.2) 15.6 (9.3) **0.017**   HOMA-IR 2.82 (2.18) 1.76 (1.62) 3.49 (2.26) **0.020** -------------------------------------------------------------------------------------------- Data are shown as mean and standard deviation (SD) for continuous variables and *n* and percentage (%) for categorical variables. ^a^, group differences were assessed with Student's *t*-test (continuous variables) or Chi-square test (categorical variables). Bold font represents *p* \< 0.05. HQ-CT, hyperphagia questionnaire for clinical trials; BMI-SDS, body mass index standard deviation score; LDL, low-density lipoprotein; HDL, high-density lipoprotein; HOMA-IR, homeostatic model assessment of insulin resistance. nutrients-12-01063-t002_Table 2 ###### Dietary intake analysis of participants. ---------------------------------------------------------------------------------- Dietary Intake All Subjects\ NW\ OWO\ *p*-Value *N* = 31 *n* = 12 *n* = 19 -------------------------- --------------- ------------- ------------- ----------- Energy Intake (kcal/day) 1571 (349) 1525 (308) 1600 (378) 0.552 (% RCI) 80 (22) 84 (26) 78 (20) 0.493 Protein (% kcal) 17.7 (2.7) 17.3 (2.7) 18.0 (2.9) 0.530 Fat (% kcal) 32.7 (6.0) 31.1 (6.1) 33.7 (5.9) 0.256   SFA (% kcal) 9.4 (2.6) 8.6 (2.7) 9.9 (2.4) 0.170   MUFA (% kcal) 14.9 (2.7) 14.1 (2.8) 15.4 (2.6) 0.198   PUFA (% kcal) 5.1 (1.3) 5.2 (1.2) 5.0 (1.3) 0.727 Carbohydrate (% kcal) 46.6 (5.6) 48.2 (4.6) 45.5 (6.0) 0.176   Fiber (g/day) 23.8 (12.0) 24.9 (10.0) 23.1 (13.4) 0.681 ---------------------------------------------------------------------------------- MRI, macronutrient recommended intake; NW, normal weight; OWO, overweight and obesity; RCI, recommended caloric intake; SFA, saturated fatty acids; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids. Data are shown as mean and SD. Group differences were assessed with Student's *t*-test. nutrients-12-01063-t003_Table 3 ###### Food group intake analysis of participants in the study. --------------------------------------------------------------------- Food Groups All Subjects\ NW\ OWO\ *p*-Value *N* = 31 *n* = 12 *n* = 19 --------------- --------------- ------------ ------------ ----------- Grains 27.6 (7.0) 25.6 (8.2) 29.0 (6.1) 0.235 Legumes 4.4 (3.6) 4.4 (4.4) 4.3 (3.2) 0.960 Vegetables 5.4 (2.1) 6.0 (2.0) 5.1 (2.1) 0.245 Fruit 11.8 (6.5) 14.6 (5.9) 10.0 (6.3) **0.050** Dairy 16.9 (6.6) 18.4 (8.3) 16.0 (5.2) 0.394 Meat 9.9 (5.8) 7.0 (3.8) 11.7 (6.3) **0.015** Fish 3.6 (1.7) 3.9 (1.7) 3.5 (1.8) 0.499 Eggs 1.7 (1.2) 1.8 (1.1) 1.7 (1.2) 0.946 Oils and fats 11.6 (2.6) 11.9 (3.0) 11.4 (2.4) 0.589 Beverages 1.6 (2.0) 1.6 (1.9) 1.7 (1.6) 0.899 --------------------------------------------------------------------- NW, normal weight; OWO, overweight and obesity. Food group data are shown as mean and SD of % of total calorie intake. Group differences were assessed with Student's *t*-test. Bold font represents *p* \< 0.05. nutrients-12-01063-t004_Table 4 ###### Associations between dietary variables and adiposity or BMI-SDS. Dietary Variables BMI-SDS Body Fat Mass (%) ------------------------------- ------------------------------------ ------------------- ------------------------- ----------- Caloric intake (kcal/day) ^a^ 0.00 (0.00 to 0.00) 0.293 −0.01 (−0.02 to 0.00) 0.133 Protein (% kcal) 0.11 (−0.11 to 0.33) 0.302 0.11 (−1.18 to 1.41) 0.856 Fat (% kcal) 0.08 (0.00 to 0.17) 0.061 0.44 (−0.06 to 0.94) 0.084 SFA (% kcal) **0.26 (0.06 to 0.47)** **0.014** **1.49 (0.31 to 2.67)** **0.016** MUFA (% kcal) 0.16 (−0.02 to 0.35) 0.081 0.88 (−0.20 to 1.96) 0.105 PUFA (% kcal) −0.12 (−0.56 to 0.32) 0.571 −1.09 (−3.61 to 1.43) 0.379 Carbohydrates (% kcal) −**0.09 (**−**0.18 to** −**0.01)** **0.** **038** −0.39 (−0.92 to 0.14) 0.145 Fiber (g/day) −0.03 (−0.08 to 0.02) 0.168 −0.21 (−0.50 to 0.08) 0.144 Linear regression models were adjusted by sex, age, physical activity level, genotype, metformin use, and caloric intake. ^a^, model not adjusted by caloric intake. B, effect size from the multivariate regression model; CI, confidence interval; SFA, saturated fatty acids; MUFA, monounsaturated fatty acids; PUFA, polyunsaturated fatty acids. Bold font indicates *p* \< 0.05. nutrients-12-01063-t005_Table 5 ###### Associations between food group intake and BMI-SDS or adiposity. Food Groups BMI-SDS Body Fat Mass (%) -------------- ----------------------------------- ------------------- ------------------------------------ ----------- Grains 0.03 (−0.04 to 0.11) 0.360 0.19 (−0.25 to 0.63) 0.375 Legumes −0.05 (−0.23 to 0.12) 0.535 −0.19 (−1.20 to 0.83) 0.707 Vegetables −0.22 (−0.48 to 0.04) 0.098 −0.64 (−2.21 to 0.93) 0.406 Fruit −**0.12 (**−**0.2 to** −**0.04)** **0.008** −**0.64 (**−**1.14 to** −**0.14)** **0.015** Dairy −0.01 (−0.11 to 0.08) 0.751 −0.08 (−0.62 to 0.47) 0.776 Meat **0.12 (0.05 to 0.20)** **0.002** **0.52 (0.05 to 0.99)** **0.033** Fish −0.12 (−0.46 to 0.21) 0.459 −0.95 (−2.86 to 0.97) 0.318 Eggs 0.23 (−0.24 to 0.69) 0.329 1.45 (−1.24 to 4.15) 0.276 Oils and fat −0.05 (−0.27 to 0.17) 0.632 −0.03 (−1.31 to 1.26) 0.966 Beverages −0.17 (−0.44 to 0.11) 0.216 −0.38 (−2.01 to 1.25) 0.632 Model adjusted by sex, age, physical activity level, genotype, metformin use, and caloric intake. B, effect size from the multivariate regression model; CI, confidence interval. Bold font indicates *p* \< 0.05. [^1]: These authors contributed equally.

1. Introduction {#sec1-nanomaterials-09-01457} =============== Iron-oxide magnetic nanoparticles (MNPs) have been widely used in various biomedical applications, including tumor detection, imaging and therapy because of their excellent magnetic, optic and electric properties, biocompatibility and biodegradability \[[@B1-nanomaterials-09-01457],[@B2-nanomaterials-09-01457],[@B3-nanomaterials-09-01457]\]. In particular, MNPs show promising applications in magnetic hyperthermia therapy (MHT) for cancer treatment as an individual treatment or an adjuvant treatment combined with chemotherapeutic and/or radiotherapeutic agents \[[@B4-nanomaterials-09-01457],[@B5-nanomaterials-09-01457]\]. It has been shown that MNPs can convert the magnetic energy into thermal energy through hysteresis loss or Néel/Brownian relaxation process in an alternating magnetic field \[[@B6-nanomaterials-09-01457],[@B7-nanomaterials-09-01457],[@B8-nanomaterials-09-01457]\]. For superparamagnetic iron oxide nanoparticles, the energy conversion is mostly based on relaxation process \[[@B9-nanomaterials-09-01457]\]. To distinguish the relaxation process between Néel relaxation and Brownian relaxation, a gel or glycerol system which has the different viscosity was usually used \[[@B10-nanomaterials-09-01457],[@B11-nanomaterials-09-01457]\]. Alternatively, comparing the fitted effective relaxation time (obtained by SLP vs. frequency fitting) with Néel relaxation time and Brownian relaxation time (obtained by calculation) has been proposed \[[@B12-nanomaterials-09-01457]\]. In order to achieve a higher thermal energy intra tumor, MNPs in delivery systems for targeting deep or multiple tumor are requested \[[@B13-nanomaterials-09-01457]\]. To this end, linking ligands are applied, but it makes the synthesis process more tedious and ultimately results in lower productivity with higher cost. Over the past decade, a few cage-like proteins have been used for synthesis of nanosized magnetic minerals inside their cavity \[[@B14-nanomaterials-09-01457]\]. One of these MNPs is magnetoferritin, consisting of a recombinant ferritin shell and a magnetite inner core \[[@B15-nanomaterials-09-01457],[@B16-nanomaterials-09-01457]\]. Outer protein shell ferritin composed of human heavy-chain ferritin has shown a great ability to actively targeting lots of tumors in vivo and in vitro through the transferrin receptor 1 which is over expressed on most tumor cell membrane \[[@B17-nanomaterials-09-01457],[@B18-nanomaterials-09-01457],[@B19-nanomaterials-09-01457]\]. The concept of using magnetoferritin for MHT was proposed by Babincová \[[@B20-nanomaterials-09-01457]\]. Recently, Fantechi and co-workers reported that a 6.8 nm sized and 5% Co-doped iron oxide core of magnetoferritin killed \~70% B16 melanoma cells in vitro by MHT \[[@B21-nanomaterials-09-01457]\]. Massner and co-workers have shown that magnetoferritin with 4 nm iron oxide core caused \~10% mortality of human embryonic kidney 293T cells in vitro \[[@B22-nanomaterials-09-01457]\]. Nevertheless, the heating efficiency and influencing factors of magnetoferritin in aqueous solution have not been studied in detail yet. In the present work, we carried out a systematic study to investigate the influences of core-size, Fe concentration, AMF amplitude and frequency on the specific loss power (SLP) and intrinsic loss power (ILP) of magnetoferritin in aqueous. We also measured the heating efficiency in media of different viscosity (glycerol-water mixture) to identify the heat generation mechanisms of magnetoferritin. 2. Materials and Methods {#sec2-nanomaterials-09-01457} ======================== 2.1. Preparation of Recombinant Human Ferritin (HFn) {#sec2dot1-nanomaterials-09-01457} ---------------------------------------------------- The *Pichia pastoris* X-33 containing recombinant plasmid pPICZ A-HFn was constructed in the Beijing Biogeomagnetic Laboratory (Beijing, China). The expression of HFn was determined according to the EasySelect^TM^ Pichia Expression Kit user's manual (Invitrogen, Carlsbad, CA, USA). 2.2. Synthesis of Magnetoferritin (MHFn) {#sec2dot2-nanomaterials-09-01457} ---------------------------------------- The MHFn nanoparticles were synthesized according to our previous work \[[@B23-nanomaterials-09-01457]\]. Briefly, a deoxygenated solution of 0.1 M NaCl (40 mL) with HFn (0.5 mg/mL) was injected into a reaction vessel, kept the temperature at 65 °C and pH 8.5. Freshly prepared Fe (II) (25 mM (NH~4~)~2~Fe(SO~4~)~2~‧6H~2~O) and H~2~O~2~ (8.3 mM) were added to the vessel in a rate of 50 Fe/(protein‧min). To prepare MHFn nanoparticles with different mineral core sizes, different theoretical amounts of iron atoms per protein cage were added into the reaction vessel. The whole reactions were under an anaerobic atmosphere to the end. The synthesized MHFn nanoparticles were centrifuged to remove the aggregated nanoparticles. Finally, the MHFn nanoparticles were kept in an aqueous medium for further use. 2.3. Mineral Core Characterization {#sec2dot3-nanomaterials-09-01457} ---------------------------------- To characterize the mineral core's size and crystallinity, the mineralized MHFn nanoparticles in aqueous medium were dried on an ultra-thin amorphous-carbon film and analyzed by transmission electron microscopy (TEM, JEOL JEM-2100, Tokyo, Japan) with an accelerating voltage of 200 kV. The inner core size of MHFn for each sample was averaged from about 1000 separated particles. Here, we measured the area of the single grain and then calculated the diameter with D = (4‧area/π)^1/2^. High-resolution TEM (HR-TEM) images and selected area electron diffraction (SAED) patterns were taken to analyze the crystallinity of mineral core of MHFn nanoparticles. 2.4. Magnetic Measurements of MHFn Nanoparticles {#sec2dot4-nanomaterials-09-01457} ------------------------------------------------ MHFn nanoparticles were freeze dried with an Alpha 1--2 LD plus freeze dryer (Martin Christ Gefriertrocknungsanlagen GmbH, Osterode am Harz, Germany). The dried samples were placed in a non-magnetic capsule with a copper spoon. The measurements were performed on a Quantum Design MPMS SQUID magnetometer (model XP-XL5, with magnetic moment sensitivity of 5.0 × 10^−10^ A‧m^2^, Quantum Design North America, San Diego, CA, America). Briefly, the hysteresis loops were measured in fields of ± 3T at 5 K and 300 K, respectively. Zero-filed cooled (ZFC) and filed-cooled (FC) curves were measured in a filed of 1.5 mT from 5 K to 200 K, the blocking temperatures (T~B~) was determined from the maximum of the ZFC curves. The acquisition of isothermal remanent magnetization (IRM) and dc demagnetization curves were measured in a fields of 0--1 T at 5 K. 2.5. Dynamic Light Scattering (DLS) Measurements and Thermogravimetric (TG) Analysis {#sec2dot5-nanomaterials-09-01457} ------------------------------------------------------------------------------------ The hydrodynamic diameter of the samples in aqueous medium were measured by dynamic light scattering (DynaPro NanoStar, Wyatt Technology Corporation, Goleta, CA, USA). Samples (10 μL) at 1.5 mg\[Fe\]/mL were added to the holder and tested three times. Thermogravimetric analysis (TG/DTA 6300, Seiko instruments Inc., Chiba, Japan) was used to obtain the proportion of the inner mineral core of the entire MHFn nanoparticles. Samples were heated from room temperature to 800 °C at 10 °C/min under an air flow at 1 L/min. 2.6. Heating Efficiency Analyses {#sec2dot6-nanomaterials-09-01457} -------------------------------- The heating efficiency of MHFn samples have been measured with a commercial system DM100 series (nB nanoScale Biomagnetics, Zaragoza, Spain) at Shanghai Normal University (Shanghai, China). A MHFn aqueous sample (0.5 mL) in a 2 mL glass chromatography vial was set in the middle of the coil. The temperature of all samples during magnetic treatment was recorded by an optic fiber temperature probe with a response temperature of 0.1 °C. The initial temperature of each sample has been controlled and stabilized to room temperature (about 22--23 °C controlled by air condition). Two different test methods were used to evaluate repeatability of samples under H = 19.5 kA/m and *f =* 808.5 kHz at 5 mg\[Fe\]/mL. One of them, named single test method, is one by one test after cooling to room temperature and measured four times, the other named continuous test method is to turn off the AMF when temperature rise to 33 °C and turn on the AMF when temperature is low to 28 °C for four cycles. The effects of size and concentration on SLPs and ILPs have been measured at H = 19.5 kA/m and *f =* 808.5 kHz with 1.5 and 5 mg\[Fe\]/mL MHFn samples. The dependences of AMF parameters on SLPs have also been investigated. Two core sizes of MHFn samples at 5 mg\[Fe\]/mL were tested at 805.5 kHz with different AMF amplitudes (11.9, 13.9, 15.9, 17.9, 19.5 kA/m) and at 19.5kA/m with different frequencies (274.5, 405.5, 466.0, 546.5, 598.0, 737.0, 805.5 kHz), respectively. 4.3 nm MHFn sample at 2.5 mg\[Fe\]/mL and 4.8 nm MHFn sample at 1.5 mg\[Fe\]/mL in different glycerol-water mixtures (0, 50% and 70% glycerol ratio) were also measured. In this work, we set the moment that field amplitude reached the setting value as t = 0, and set the temperature (about 22--23 °C) at t = 0 as ∆T = 0. The method proposed by Kallumadil and co-workers was used to calculate the heating efficiency \[[@B24-nanomaterials-09-01457]\]: fitting the curves of field applied time vs. temperature with Box-Lucas equation \[$T\left( t \right) = a\left( {1 - e^{- {bt}}} \right)$\] to calculate the ∆T/∆t at t = 0; the heating efficiency SLP was calculated with ${\ {SLP}} = C_{water}/c_{Fe} \times \frac{\Delta T}{\Delta t}$, where *C~water~* is the heat capacity of water (4.185 J‧g^−1^‧K^−1^) and *c~Fe~* is the Fe concentration in aqueous solution; the intrinsic loss power (ILP) was calculated by ILP = SLP/(*f*‧H^2^). The Néel-Brownian relaxation time and the effective relaxation time were calculated with formula given by Rosensweig \[[@B6-nanomaterials-09-01457]\]. 3. Results and Discussion {#sec3-nanomaterials-09-01457} ========================= 3.1. Core Size and Crystallinity {#sec3dot1-nanomaterials-09-01457} -------------------------------- Low magnification TEM micrographs ([Figure 1](#nanomaterials-09-01457-f001){ref-type="fig"}A) show that the MHFn nanoparticles were well dispersed. The HR-TEM micrographs ([Figure 1](#nanomaterials-09-01457-f001){ref-type="fig"}B) show the clear lattices fringes without obvious lattice defects. The lattice fringes measured from HR-TEM micrographs matched with magnetite and/or maghemite. The diffraction rings in SAED pattern images ([Figure 1](#nanomaterials-09-01457-f001){ref-type="fig"}C) were clear and sharp, which indicates the magnetic cores were crystalline. The mean size of MHFn cores are 3.5 ± 0.6 nm, 3.6 ± 0.5 nm, 4.3 ± 0.6 nm, 4.8 ± 0.7 nm ([Figure 1](#nanomaterials-09-01457-f001){ref-type="fig"}D and [Table 1](#nanomaterials-09-01457-t001){ref-type="table"}), respectively, with a lognormal distribution (dotted lines). 3.2. Magnetic Properties {#sec3dot2-nanomaterials-09-01457} ------------------------ The saturation magnetization (Ms) measured at 5 K ([Figure 2](#nanomaterials-09-01457-f002){ref-type="fig"}A) and 300 K ([Figure 2](#nanomaterials-09-01457-f002){ref-type="fig"}B) and blocking temperature (T~B~) ([Figure 2](#nanomaterials-09-01457-f002){ref-type="fig"}C) increased with increasing MHFn core size. The decrease of R-value ([Figure 2](#nanomaterials-09-01457-f002){ref-type="fig"}D) suggests that the static magnetic interaction is slightly increased with core size. At 300 K, the magnetization hysteresis loops M (H) curves show that all MHFn nanoparticles are superparamagnetic (no measurable coercivity). The magnetic data is summarized in [Table 1](#nanomaterials-09-01457-t001){ref-type="table"}. Interestingly, the T~B~ increases linearly with the volume of MHFn ([Figure 3](#nanomaterials-09-01457-f003){ref-type="fig"}A) in this study and our previous study \[[@B25-nanomaterials-09-01457],[@B26-nanomaterials-09-01457]\]. The slope may be used to roughly calculate the effective anisotropic energy constant K~eff~ (k~eff~V = 25k~B~T~B~) \[[@B27-nanomaterials-09-01457]\]. The Ms measured at 300 K likely increases linearly with the volume of measured samples ([Figure 3](#nanomaterials-09-01457-f003){ref-type="fig"}B). 3.3. Dynamic Light Scattering (DLS) Measurements and Thermogravimetric (TG) Analysis {#sec3dot3-nanomaterials-09-01457} ------------------------------------------------------------------------------------ The hydrodynamic diameter of pure HFn is about 16 nm while the mineralized MHFn is about 26 nm, which is larger than the HFn ([Figure 4](#nanomaterials-09-01457-f004){ref-type="fig"}A, [Table 1](#nanomaterials-09-01457-t001){ref-type="table"}). The inner mineral core may be responsible for this gap. As expected, the inner core mass proportion of particles was increased as core size increases ([Figure 4](#nanomaterials-09-01457-f004){ref-type="fig"}B and [Table 1](#nanomaterials-09-01457-t001){ref-type="table"}). A clear weight loss from 200--300 °C may be related to protein shell loss. 3.4. Specific Loss Power and Intrinsic Loss Power {#sec3dot4-nanomaterials-09-01457} ------------------------------------------------- ### 3.4.1. Stability of MHFn Nanoparticles in Aqueous {#sec3dot4dot1-nanomaterials-09-01457} Samples with mean size of 4.3 nm and 4.8 nm were used for analysis. As shown in [Figure 5](#nanomaterials-09-01457-f005){ref-type="fig"}, temperature rising curves of both 4.3 nm and 4.8 nm MHFn were all well overlapped by the single test method ([Figure 5](#nanomaterials-09-01457-f005){ref-type="fig"}A,B), and the same periodical changes were obtained during the four cycles by continuous test method ([Figure 5](#nanomaterials-09-01457-f005){ref-type="fig"}C,D) for both 4.3 nm and 4.8 nm MHFn. The results indicated that the MHFn are stable in aqueous medium during the AMF treatment. ### 3.4.2. Core-Size and Fe Concentration Effects {#sec3dot4dot2-nanomaterials-09-01457} [Figure 6](#nanomaterials-09-01457-f006){ref-type="fig"} shows the effects of core-size and Fe concentration on the temperature rise of MHFn samples. In this study, two Fe concentrations were used, 1.5 mg\[Fe\]/mL ([Figure 6](#nanomaterials-09-01457-f006){ref-type="fig"}A) and 5 mg\[Fe\]/mL ([Figure 6](#nanomaterials-09-01457-f006){ref-type="fig"}B). The SLPs fitting results have been summarized in [Table 2](#nanomaterials-09-01457-t002){ref-type="table"}. The temperature rise (∆T) at 6 min increases from 2.5 to 4.2 °C with core size increases at 1.5 mg\[Fe\]/mL ([Figure 6](#nanomaterials-09-01457-f006){ref-type="fig"}A and [Table 2](#nanomaterials-09-01457-t002){ref-type="table"}), while increases from 6.1 to 13.8 °C at 5 mg\[Fe\]/mL ([Figure 6](#nanomaterials-09-01457-f006){ref-type="fig"}B and [Table 2](#nanomaterials-09-01457-t002){ref-type="table"}). SLP also increases from 32.7 to 51.3 W/g\[Fe\] at 1.5 mg\[Fe\]/mL ([Figure 7](#nanomaterials-09-01457-f007){ref-type="fig"}A and [Table 2](#nanomaterials-09-01457-t002){ref-type="table"}) and from 23.2 to 67.7 W/g\[Fe\] at 5 mg\[Fe\]/mL ([Figure 7](#nanomaterials-09-01457-f007){ref-type="fig"}A and [Table 2](#nanomaterials-09-01457-t002){ref-type="table"}). Our results show that SLPs of MHFn samples were increased with increasing Fe concentration. However, previous works have shown that SLPs were increased \[[@B28-nanomaterials-09-01457],[@B29-nanomaterials-09-01457]\] or decreased \[[@B30-nanomaterials-09-01457],[@B31-nanomaterials-09-01457]\] with the increasing Fe or nanoparticle concentration. Furthermore, some works have shown that SLPs were independent of Fe or nanoparticle concentration \[[@B12-nanomaterials-09-01457],[@B32-nanomaterials-09-01457]\]. These inconsistent findings may indicate a complicated mechanism of Fe concentration effects on SLP for different MNPs. Nevertheless, for most MNPs, it is believed that increasing the concentration will increase the dipolar interaction while increasing the magnetic anisotropy energy, which results in the SLP increasing gradually to a maximum and then starting to decrease \[[@B33-nanomaterials-09-01457]\]. Generally, as the concentration increases, SLP increases for the smaller particles and decreases for the lager particles with the increasing Fe concentration \[[@B34-nanomaterials-09-01457]\]. Studies on MHFn samples in this work suggest the increasing of SLP with increasing Fe concentration to some extent. ### 3.4.3. Magnetic Field Effects and the Néel Relaxation {#sec3dot4dot3-nanomaterials-09-01457} Due to the better temperature rising effect of 4.3 nm and 4.8 nm sized MHFn, we mainly focused on this two samples in the following studies. The effect of the AMF parameter on temperature rise of 4.3 nm and 4.8 nm MHFn samples were carried out at 5 mg\[Fe\]/mL in aqueous medium. As AMF amplitude increases from 11.9 to 19.5 kA/m under a frequency of 805.5 kHz, the ∆T of 4.3 nm MHFn at 6 min increased from 7.7 to 12.0 °C ([Figure 8](#nanomaterials-09-01457-f008){ref-type="fig"}A and [Table 2](#nanomaterials-09-01457-t002){ref-type="table"}), and from 6.8 to 13.8 °C for 4.8 nm ([Figure 8](#nanomaterials-09-01457-f008){ref-type="fig"}B and [Table 2](#nanomaterials-09-01457-t002){ref-type="table"}), respectively. SLPs also increase from 35.6 to 56.3 W/g for 4.3 nm MHFn and from 31.2 to 67.8 W/g for 4.8 nm MHFn ([Figure 7](#nanomaterials-09-01457-f007){ref-type="fig"}B and [Table 2](#nanomaterials-09-01457-t002){ref-type="table"}). Based on linear response theory (LRT), we fitted the SLP vs. amplitude data with the equation SLPs = a*f*H^2^ + b and found that the data were well characterized with a R^2^ \> 0.99 ([Figure 7](#nanomaterials-09-01457-f007){ref-type="fig"}B). Note that the slope *a* is directly the ILP value proposed by Kallumadil \[[@B24-nanomaterials-09-01457]\]. [Figure 9](#nanomaterials-09-01457-f009){ref-type="fig"}A,B and [Table 2](#nanomaterials-09-01457-t002){ref-type="table"} show the effect of AMF frequency on temperature rise for this two MHFn samples. The ΔT at 6 min increased about six-fold (from 1.9 °C to 11.5 for 4.3 nm MHFn and from 2.2 to 14.2 °C for 4.8 nm MHFn) when AMF frequency increases from 274.5 to 805.5 kHz under amplitude 19.5 kA/m ([Figure 9](#nanomaterials-09-01457-f009){ref-type="fig"}A,B). It was noted that the SLP also significantly increased with increasing frequency for both 4.3 nm (from 7.9 to 56.3W/g) and 4.8 nm (from 9.5 to 68.6W/g) samples ([Figure 7](#nanomaterials-09-01457-f007){ref-type="fig"}C and [Table 2](#nanomaterials-09-01457-t002){ref-type="table"}). Here, we fitted the SLPs vs. frequency curve with the formula below \[[@B12-nanomaterials-09-01457]\]:$$SLP\left( f \right) = A \times \frac{\tau_{fit} \times \left( {2\pi f} \right)^{2}}{1 + \left( {\tau_{fit} \times 2\pi f} \right)^{2}}$$ where *A* is a parameter about the magnetic property of sample, *τ~fit~* is fitted effective relaxation time. As shown in [Figure 7](#nanomaterials-09-01457-f007){ref-type="fig"}C, the data were well described by the equation with R^2^ \> 0.99. The fitted relaxation time τ~fit~ were shown in [Table 3](#nanomaterials-09-01457-t003){ref-type="table"}. The theoretical Néel and Brownian relaxation timew were calculated with the equations developed by Rosensweig \[[@B6-nanomaterials-09-01457]\]. Comparing the relaxation time between fitted relaxation time (τ~fit~), calculated Néel relaxation time (τ~N~), Brownian relaxation time (τ~B~) and effective relaxation time τ~eff~ ([Table 3](#nanomaterials-09-01457-t003){ref-type="table"}), we found both τ~eff~ and τ~fit~ is comparable to τ~N~. Additionally, we also measured the temperature rise of MHFn samples in different glycerol ratio solutions ([Figure 9](#nanomaterials-09-01457-f009){ref-type="fig"}B for 4.3 nm MHFn and [Figure 9](#nanomaterials-09-01457-f009){ref-type="fig"}B for 4.8 nm MHFn) to simulate different viscosities. Although the coefficients of viscosity were increased from 1.0 to 22.5 mPa‧s (at 20 °C) \[[@B35-nanomaterials-09-01457]\], SLPs were basically unchanged for both 4.3 nm and 4.8 nm samples (low than 10%) ([Figure 7](#nanomaterials-09-01457-f007){ref-type="fig"}D and [Table 2](#nanomaterials-09-01457-t002){ref-type="table"}), which is obviously different from previous studies where the SLPs would significantly decrease with increasing viscosity based on a Brownian mechanism \[[@B10-nanomaterials-09-01457],[@B36-nanomaterials-09-01457]\]. Combined the results of theoretical calculated relaxation time and actual measurements under different viscosity conditions, we propose that heat generation mechanism for both 4.3 nm and 4.8 nm MHFn samples is dominated by Néel relaxation. ### 3.4.4. Intrinsic Loss Power (ILP) {#sec3dot4dot4-nanomaterials-09-01457} Like SLP, ILP increases with size or volume and increasing Fe concentration ([Figure 10](#nanomaterials-09-01457-f010){ref-type="fig"}A and [Table 2](#nanomaterials-09-01457-t002){ref-type="table"}). [Figure 10](#nanomaterials-09-01457-f010){ref-type="fig"}C shows the effect of AMF frequency on ILP. We note that the ILP increases with increasing frequency, which indicates that SLPs are not linear with frequency for the measured samples, consistent with a previous study \[[@B37-nanomaterials-09-01457]\]. However, ILP may be independent of frequency for those MNPs whose SLPs are linear with frequency \[[@B38-nanomaterials-09-01457]\]. As we mentioned in [Section 3.4.3](#sec3dot4dot3-nanomaterials-09-01457){ref-type="sec"}, ILP can be obtained by fitting SLPs vs. amplitude. We found the ILP decreases with increasing amplitude for both 4.3 nm and 4.8 nm MHFn samples ([Figure 10](#nanomaterials-09-01457-f010){ref-type="fig"}B), which is consistent with previous studies \[[@B39-nanomaterials-09-01457],[@B40-nanomaterials-09-01457]\]. Nevertheless, in the formula ILP = SLP/*f*H^2^, if we consider the intercept b in SLP (SLP = a*f*H^2^ + b), the ILP could be calculated with ILP = (a*f*H^2^ + b)/*f*H^2^ = a + b/*f*H^2^. Obviously, the relationship between ILP and amplitude is closely related to the value of intercept b. If b = 0, ILP is independent of amplitude; if b \> 0, ILP will decrease with increasing amplitude; oppositely, if b \< 0, ILP will increase with increasing amplitude. In this work, ILP decreases with increasing amplitude, and the reason may be the positive intercept b obtained for 4.3 nm MHFn (b = 14.0) and for 4.8 nm MHFn (b = 10.6). Considering these effects on ILP, it may not advisable to calculate the ILP of the particles directly under a single AMF parameter test. 3.5. Physical Model {#sec3dot5-nanomaterials-09-01457} ------------------- The equation of motion for spherical MHFn particles in a viscous medium is given by:$$I\overset{˙}{\mathsf{\omega}} = f_{r}\mathsf{\omega} + \mu_{0}\mu \times H + \mathsf{\xi}_{B}$$ where *I* is the moment of inertia, ω is the angular velocity vector, *f~r~* is the rotational viscous friction coefficient, μ is the magnetic moment, H is the applied field, and ξ~B~ is the random perturbation torque associated with Brownian motion. The maximum magnitude of the magnetic torque relative to the viscous torque can be evaluated assuming $f_{r} = 8\pi\eta a_{h}^{3}$ for a sphere with hydrodynamic radius ah in a medium with dynamic viscosity *η*, and $\mu = 4\pi\mu_{s}a_{m}^{3}/3$ for the magnetic moment of the MHFn's magnetic core with spontaneous magnetization *μ~s~* and radius *a~m~*. Equating $\omega = 2\pi f$ with the frequency of the applied field, the ratio between magnetic and viscous torques becomes:$$\frac{\xi_{m}}{\xi_{h}} \leq \frac{\mu_{0}\mu_{s}H}{12\pi\eta f} \times \left( \frac{a_{m}}{a_{h}} \right)^{3}$$ Near-unit values are obtained with maghemite-like (i.e., *μ*~s~ = 380 kA/m), uncoated (i.e., *a~m~ = a~h~*) particles in water (*η* = 0.001 Pa⋅s) at the lowest frequency (274.5 kHz) used in these experiments. However, the much smaller magnetic core in MHFn ensures that the magnetic torque is at least two orders of magnitude smaller than the viscous drag, meaning that the magnetic moment is affected only by Néel's relaxation in the applied field, as for with static magnetic measurements of [Figure 2](#nanomaterials-09-01457-f002){ref-type="fig"} for mechanically blocked particles. Heat generation is thus related to the energy loss ∆*U* of the hysteresis loop *M(H)* of mechanically blocked particles by:$$P = f\Delta U = - \mu_{0}\oint M\left( H \right)dH$$ where the integral represents the area enclosed by the loop. It is therefore possible to compare Equation (4) with the generated heat power if room-temperature hysteresis loops measured at the same frequency and same maximum field as the heat-generating field were available. Because the hysteresis opening is proportional to the amount of blocked particles, one can use Néel's magnetic relaxation theory to find measurement time *τ* and temperature *T* combinations that yield the same fraction of blocked particles and thus the same hysteresis opening. These combinations obey the simple relation:$$\frac{T_{2}}{T_{1}} = \frac{\ln\left( {\tau_{1}/\tau_{0}} \right)}{\ln\left( {\tau_{2}/\tau_{0}} \right)}$$ with *τ*~0~ ≈ 0.1 ns being the atomic reorganization time. Using *T*~1~ ≈ 300 K and $\tau_{1} = \left( {2\pi f} \right)^{- 1}$ for the room-temperature heating experiments, one gets the temperature:$$T_{2} = - T_{1}\frac{\ln\left( {\tau_{0}f} \right)}{\ln\left( {\tau_{2}/\tau_{0}} \right)}$$ of equivalent static hysteresis measurements such as those of [Figure 2](#nanomaterials-09-01457-f002){ref-type="fig"}, where *τ*~2~ is the total time needed for the measurements. For instance, *f* = 805.5 kHz and *τ*~2~ = 6 h (the typical measurement time required by a MPMS SQUID magnetometer) yields *T*~2~ = 85 K. A hysteresis measurement performed at this temperature using the same maximum field *H*~0~ of the heating experiments can be used to estimate the hysteresis loss power with Equation (4), after correcting for the temperature dependence of *Ms*, that is:$$p\left( {T_{1},H_{0}} \right) = - \mu_{0}f\frac{M_{S}\left( T_{1} \right)}{M_{S}\left( T_{2} \right)}\oint M\left( {H,T_{2},H_{0}} \right)dH$$ This model will likely explain the dependence of SLP on $H_{0}^{2}$ (the opening of Rayleigh loops is proportional to $H_{0}^{2}$), as well as the quadratic dependence of SLP on *f*, since the loop opening depends on the fraction of blocked particles, which increases at larger frequencies. 4. Conclusions {#sec4-nanomaterials-09-01457} ============== Based on this study, the following conclusions can be drawn: (1)The heating efficiency SLP of MHFn nanoparticles increased with the core-size, Fe concentration, AMF frequency, and amplitude.(2)The fitted relaxation time τ~fit~ and calculated effective relaxation time τ~eff~ were closer to calculated the Néel relaxation time τ~N~, and the SLP of 4.3 nm and 4.8 nm MHFn basically unchanged in different viscosity glycerol-water mixtures. This indicates that the AMF heating generation mechanism of MHFn nanoparticles is dominated by Néel relaxation.(3)The calculated ILP of MHFn nanoparticles under different AMF frequency and amplitude conditions varied.(4)The max temperature rise on 6 min measured is 14.2 °C for 4.8 nm MHFn at 5 mg\[Fe\]/mL under 805 kHz and 19.5 kA/m. This suggests that MHFn nanoparticles may be applied to medical treatments in future for tumor magnetic hyperthermia, heat-triggered drug release with consideration of good biocompatibility and active targeting to various tumors. We thank Shiping Yang, Ping Zhou and Guang Deng at Shanghai Normal University for help in the heating measurement, and Yao Cai for help improving the manuscript. We thank Ramon Egli and the anonymous reviewer for their constructive comments on the manuscript and special thanks to R.E. for providing the physical model in [Section 3.5](#sec3dot5-nanomaterials-09-01457){ref-type="sec"}. H.X. and Y.P. designed the project. H.X. performed the experiment, and analyzed the data. H.X. and Y.P. wrote this manuscript. This work was supported by the National Natural Science Foundation of China (no.: 41621004) and Key Program of Chinese Academy of Sciences (QYZDJ-SSW-DQC024). The authors declare no conflict of interest. {#nanomaterials-09-01457-f001} {#nanomaterials-09-01457-f002} {#nanomaterials-09-01457-f003} {#nanomaterials-09-01457-f004} ![Repeatability of heating efficiency of the MHFn in aqueous under AMF treatment. (**A,B**) Single measurement of temperature rise for four times for 4.3 and 4.8 nm samples, respectively. (**C,D**) Continuous measurement of temperature rise on heating to 33 °C and cooling to 28 °C for four cycles. All the measurements were performed at c = 5 mg\[Fe\]/mL, magnetic field H = 19.5 kA/m and *f* = 805.5 kHz.](nanomaterials-09-01457-g005){#nanomaterials-09-01457-f005} ![Heating efficiency of MHFn nanoparticles with different core sizes at (**A**) 1.5 mg\[Fe\]/mL and (**B**) 5 mg\[Fe\]/mL. All the measurements were performed in a magnetic field H = 19.5 kA/m and *f* = 805.5 kHz. Dotted lines are the fitting results of Box-Lucas equation.](nanomaterials-09-01457-g006){#nanomaterials-09-01457-f006} ![The Special Loss Power (SLP) vs. different factors. (**A**) SLP vs. MHFn particles volume. Measurements were performed at H = 19.5 kA/m and *f* = 805.5 kHz. (**B**) SLP vs. amplitude. Measurements were performed at c~(4.8\ nm)~ = c~(4.3\ nm)~ = 5 mg\[Fe\]/mL, *f* = 805.5 kHz. Dotted lines were linear fitting with SLP = a*f*H^2^ + b. (**C**) SLP vs. frequency. Measurements were performed at c~(4.8\ nm)~ = c~(4.3\ nm)~ = 5 mg\[Fe\]/mL, H = 19.5 kA/m. Lines were represent fits results of equation (1). (**D**) SLP vs. glycerol ratio. Measurements were performed at c~(4.8\ nm)~ = 1.5 mg\[Fe\]/mL, c~(4.3\ nm)~ = 2.5 mg\[Fe\]/mL, H = 19.5 kA/m and *f* = 805.5 kHz.](nanomaterials-09-01457-g007){#nanomaterials-09-01457-f007} ![Magnetic field amplitude dependence on heating efficiency of 4.3 nm (**A**) and 4.8 nm (**B**) MHFn nanoparticle samples at 5 mg\[Fe\]/mL. All the measurements were performed at 805.5 kHz. Dotted lines are the fitting results of Box-Lucas equation.](nanomaterials-09-01457-g008){#nanomaterials-09-01457-f008} ![Effects of magnetic field frequency and viscosity on heating efficiency of 4.3 nm (**A**,**B**) and 4.8 nm (**C**,**D**) samples. Dependence of AMF frequency on heating efficiency of (**A**) 4.3 nm and (**C**) 4.8 nm MHFn nanoparticles with concentration of 5 mg\[Fe\]/mL, measured at 19.5 kA/m. Temperature rise measurement in different glycerol-water mixture of (**B**) 4.3 nm sample at 2.5 mg\[Fe\]/mL and (**D**) 4.8 nm sample at 1.5 mg\[Fe\]/mL performed at magnetic field H = 19.5 kA/m and *f* = 805.5 kHz. Dotted lines are the fitting results of Box-Lucas equation.](nanomaterials-09-01457-g009){#nanomaterials-09-01457-f009} ![The intrinsic loss power (ILP) vs. different factors. (**A**) ILP vs. particles volume. Measurements were performed at H = 19.5 kA/m and *f* = 805.5 kHz. (**B**) ILP vs. frequency. Measurements were performed at c~(4.8\ nm)~ = c~(4.3\ nm)~ = 5 mg\[Fe\]/mL, H = 19.5 kA/m. (**C**) ILP vs. amplitude. Measurements were performed at c~(4.8\ nm)~ = c~(4.3\ nm)~ = 5 mg\[Fe\]/mL, *f* = 805.5 kHz.](nanomaterials-09-01457-g010){#nanomaterials-09-01457-f010} nanomaterials-09-01457-t001_Table 1 ###### Core size, magnetic properties of the synthesized MHFn nanoparticles. Sample Core Size (nm) D~h~(PDI) ^a^ (nm) T~B~ (K) R value 5 K-Hysterisis 300 K-Hysterisis TG (%) -------- ---------------- -------------------- ---------- --------- ---------------- ------------------ -------- ------ A 3.5 ± 0.6 16.8(0.10) 37.2 0.42 22.6 20.9 17.0 36.5 B 3.6 ± 0.5 26.5(0.07) 49.5 0.40 34.6 21.9 27.0 56.1 C 4.3 ± 0.6 27.1(0.10) 62.5 0.39 42.3 21.1 34.0 60.6 D 4.8 ± 0.7 25.4(0.06) 84.5 0.38 49.5 21.8 40.9 61.5 ^a^ D~h~: hydrodynamic diameter, PDI: polydispersity index. nanomaterials-09-01457-t002_Table 2 ###### Summary of temperature rise, SLP and ILP under different Fe concentration, medium and AMF parameter measurement condition. Core Size (nm) Fe Concentration (mg\[Fe\]/mL) Solution Frequency (kHz) Amplitude (kA/m) ΔT at 6 mins (°C) SLP (W/g\[Fe\]) ILP (nHm^2^/kg) ---------------- -------------------------------- ------------------------------ ----------------- ------------------ ------------------- ----------------- ----------------- 4.8 1.5 Water 805.5 19.5 4.2 51.3 0.17 4.3 1.5 Water 805.5 19.5 3.7 46.8 0.15 3.6 1.5 Water 805.5 19.5 2.6 32.7 0.11 3.5 1.5 Water 805.5 19.5 2.5 33.7 0.11 4.8 5 Water 805.5 19.5 13.8 67.7 0.22 4.3 5 Water 805.5 19.5 12.0 56.3 0.18 3.6 5 Water 805.5 19.5 8.2 35.1 0.11 3.5 5 Water 805.5 19.5 6.1 23.2 0.08 4.3 5 Water 805.5 13.9 7.7 35.6 0.23 4.3 5 Water 805.5 15.9 9.2 42.4 0.21 4.3 5 Water 805.5 17.9 10.7 50.5 0.20 4.3 5 Water 805.5 19.5 12.0 56.3 0.18 4.8 5 Water 805.5 11.9 6.8 31.2 0.27 4.8 5 Water 805.5 13.9 8.5 40.4 0.26 4.8 5 Water 805.5 15.9 10.7 50.1 0.25 4.8 5 Water 805.5 17.9 12.3 58.7 0.23 4.8 5 Water 805.5 19.5 13.8 67.8 0.22 4.3 5 Water 274.5 19.5 1.9 7.9 0.08 4.3 5 Water 405.5 19.5 3.3 12.0 0.08 4.3 5 Water 466.0 19.5 4.0 15.7 0.09 4.3 5 Water 546.5 19.5 6.0 25.1 0.12 4.3 5 Water 598.0 19.5 7.0 29.5 0.13 4.3 5 Water 737.0 19.5 9.9 42.9 0.15 4.3 5 Water 805.5 19.5 11.5 56.3 0.18 4.8 5 Water 274.5 19.5 2.2 9.5 0.09 4.8 5 Water 405.5 19.5 4.2 17.2 0.11 4.8 5 Water 466.0 19.5 5.2 23.3 0.13 4.8 5 Water 546.5 19.5 7.8 35.7 0.17 4.8 5 Water 598.0 19.5 9.3 42.9 0.19 4.8 5 Water 737.0 19.5 12.6 59.4 0.21 4.8 5 Water 805.5 19.5 14.2 68.6 0.22 4.3 2.5 **100% water** + 0 glycerol 805.5 19.5 5.5 48.7 0.16 4.3 2.5 **50% water** + 50% glycerol 805.5 19.5 7.2 52.6 0.17 4.3 2.5 **30% water** + 70% glycerol 805.5 19.5 7.6 54.1 0.18 4.8 1.5 **100% water** + 0 glycerol 805.5 19.5 4.2 51.5 0.17 4.8 1.5 **50% water** + 50% glycerol 805.5 19.5 4.6 54.3 0.18 4.8 1.5 **30% water** + 70% glycerol 805.5 19.5 4.7 49.3 0.16 nanomaterials-09-01457-t003_Table 3 ###### Calculated relaxation time. Sample τ~B~ (s) τ~N~ (s) τ~eff-cal~ τ~fit~ -------- -------------- -------------- -------------- -------------- 4.3 nm 6.5 × 10^−6^ 7.3 × 10^−8^ 7.2 × 10^−8^ 1.1 × 10^−8^ 4.8 nm 6.5 × 10^−6^ 4.0 × 10^−7^ 3.8 × 10^−7^ 7.5 × 10^−8^

Introduction to the myofibroblastic phenotype ============================================= Myofibroblasts were first described in healing skin wounds, where it was hypothesized that they were responsible for the phenomenon of wound contraction.[@b1-ccid-7-301] Since then, cells morphologically similar to myofibroblasts have been described in many tissues, predominantly in pathological states where their sustained presence is generally a marker of fibrosis and scarring.[@b2-ccid-7-301] Early studies identified myofibroblasts on the basis of their ultrastructural morphology, with prominent microfilament bundles in their cytoplasm distinguishing them from "normal" quiescent tissue fibroblasts. Myofibroblasts also possessed fibronexus junctions between cells and the surrounding extracellular matrix (ECM), thus in some ways appearing to share some of the morphological characteristics of smooth muscle (SM) cells.[@b3-ccid-7-301] Many tissues and pathologies have been described in which myofibroblasts have been identified, including hypertrophic and keloid scars in the skin, fibrotic liver as seen in liver cirrhosis and other liver pathologies, renal fibrosis, and idiopathic pulmonary fibrosis.[@b4-ccid-7-301] More recently, cells with phenotypic features of myofibroblasts have also been found in and around a number of epithelial tumors, where they have been termed cancer-associated fibroblasts or stromal myofibroblasts.[@b5-ccid-7-301]--[@b7-ccid-7-301] The role of myofibroblasts in driving fibrotic diseases, and the recent finding that cancer-associated myofibroblasts likely influence tumor growth and correlate with poor clinical prognosis, has increased our interest in their cellular origins, their regulation, and their role in repair and disease.[@b8-ccid-7-301] After early studies that defined myofibroblasts on the basis of their ultrastructural morphology, later research using antibodies and immunohistochemistry resulted in myofibroblasts and their microenvironment being more clearly defined.[@b9-ccid-7-301] It is now accepted that myofibroblasts go through a precursor stage of expressing large stress fibers that are not seen in quiescent fibroblasts, prominent bundles of microfilaments that permit some contraction and pre-stressing and remodeling of the surrounding ECM.[@b10-ccid-7-301] Later, fully differentiated myofibroblasts show expression of the usually SM-specific cytoskeletal protein, α-SM actin, which is now often used to define the myofibroblast phenotype.[@b9-ccid-7-301],[@b11-ccid-7-301] The presence of a splice variant form of fibronectin (ED-A fibronectin) in the microenvironment adjacent to the myofibroblast is also a defining feature and appears to be required for their differentiation.[@b12-ccid-7-301] De novo expression of osteoblast (OB) cadherin has also been reported to be found on the surface of differentiated myofibroblasts, and is not seen on α-SM actin-negative fibroblasts.[@b13-ccid-7-301] The other defining feature of myofibroblasts is that they fail to express the full repertoire of SM cell markers, allowing myofibroblasts to be distinguished from SM cells. Specifically, myofibroblasts in most cases are negative for SM cell markers such as SM myosin heavy chain,[@b14-ccid-7-301] n-caldesmon,[@b15-ccid-7-301] and smoothelin.[@b16-ccid-7-301] Desmin has also been used as a negative marker of myofibroblasts. Generally, SM cells express desmin and vimentin as well as SM myosin, while myofibroblasts express only vimentin. However, some situations have been reported in the literature where myofibroblasts in some pathologies have been found to be desmin positive.[@b17-ccid-7-301] Distinguishing myofibroblasts from pericytes is perhaps more problematic since pericytes can closely resemble myofibroblasts in being α-SM actin positive, vimentin and desmin positive, but SM myosin negative.[@b15-ccid-7-301],[@b18-ccid-7-301] Indeed, pericytes may in some cases be a source of myofibroblasts in some conditions, including wound repair, where myofibroblasts may represent a pericyte that has lost some phenotypic features such as desmin expression.[@b19-ccid-7-301] Similarly, SM cells from the media of an injured blood vessel may lose late differentiation markers such as desmin, smoothelin, and SM myosin and acquire a myofibroblast phenotype.[@b20-ccid-7-301] Lastly, myofibroblasts show both fibronexus junctions with other cells and specialized junctional complexes with the ECM; these large mature focal adhesions allow them to make strong attachments, contract and remodel the ECM, and provide a means of transducing mechanical force in the tissue.[@b21-ccid-7-301],[@b22-ccid-7-301] The contractile nature of myofibroblasts has some similarities to SM cells, despite the differences in expression of cytoskeletal features. For example, the calcium signaling in myofibroblasts appears to be similar to that in SM cells, and the arrangement of cells into something resembling that of SM cells in tissues with single-unit SM is also similar, that is, with cells having junctional complexes and connections (including gap junctions) that allow the spread of contraction signals through the tissue. Contraction of myofibroblasts seems to be possible through Ca^2+^-dependent mechanisms that are similar to those present in SM cells, with increased free Ca^2+^ regulating phosphorylation of myosin light chain. This may explain the rapid contractile responses of myofibroblasts in vitro to agonists such as angiotensin II or endothelin. Slower and more sustained contraction, which is perhaps more typical of what occurs during slow retractile contraction of connective tissue in granulation tissue, involves activity of the guanosine triphosphate (GTP)-ase RhoA and activation of its downstream target Rho-associated kinase (ROCK). This results in more continued phosphorylation of myosin light chain and thus sustained contraction.[@b23-ccid-7-301] Role in wound healing ===================== Immediately after wounding, the healing process commences, leading to (partial) restoration of injured tissue. Wound healing proceeds in three interrelated dynamic phases that temporally overlap ([Figure 1](#f1-ccid-7-301){ref-type="fig"}). Based on morphological changes over the course of the healing process, these phases are defined as the inflammatory phase, the proliferative phase (the development of granulation tissue), and the regeneration phase, including maturation, scar formation, and re-epithelialization.[@b24-ccid-7-301] The inflammatory phase begins with damage of capillaries, triggering the formation of a blood clot consisting of fibrin and fibronectin. This provisional matrix fills in the lesion and allows a variety of recruited cells to migrate into the lesion. Platelets present in the blood clot release multiple chemokines, which participate in the recruitment of inflammatory cells, neutrophils, and macrophages, but also in chemotaxis and recruitment of fibroblasts and endothelial cells. The second stage of wound healing is the proliferative phase. Angiogenesis, which is critical for the wound healing process, allows new capillaries to deliver nutrients to the wound, and contributes to the proliferation of fibroblasts. Initially the wound is hypoxic due to the loss of vascular perfusion, but with the development of a new capillary network, vascular perfusion is restored. Regulating wound angiogenesis in itself may represent a means for improving healing in some cases, particularly where delayed or defective angiogenesis is implicated in healing impairment.[@b25-ccid-7-301] In the granulation tissue, fibroblasts are activated and acquire α-SM actin expression and become myofibroblasts. These myofibroblastic cells synthesize and deposit the ECM components that eventually replace the provisional matrix ([Figure 2](#f2-ccid-7-301){ref-type="fig"}). These cells exhibit contractile properties, due to the expression of α-SM actin in microfilament bundles or stress fibers, playing a major role in the contraction and maturation of the granulation tissue.[@b26-ccid-7-301] Presently, it is accepted that the myofibroblastic modulation of fibroblastic cells begins with the appearance of the protomyofibroblast, whose stress fibers contain only β- and γ-cytoplasmic actins. Protomyofibroblasts generally evolve into differentiated myofibroblasts, the most common variant of this cell, with stress fibers containing α-SM actin (for review, see Tomasek et al[@b27-ccid-7-301]). Myofibroblasts can, depending on the experimental or clinical situation, express other SM-related contractile proteins, such as SM myosin heavy chains or desmin; however, the presence of α-SM actin represents the most reliable marker of the myofibroblastic phenotype.[@b27-ccid-7-301] The third phase of healing, scar formation, involves a progressive remodeling of the granulation tissue. During this remodeling process, proteolytic enzymes, essentially matrix metalloproteinases (MMPs) and their inhibitors (tissue inhibitor of metalloproteinases \[TIMPs\]) play a major role.[@b28-ccid-7-301] The synthesis of ECM is not totally stopped, but considerably reduced, and the synthesized components are modified as the matrix is remodeled. Progressively, collagen type III, the major component of the granulation tissue, is replaced by collagen type I, which is the main structural component of the dermis. Lastly, elastin, which contributes to skin elasticity and is absent in the granulation tissue, also reappears. In the resolution phase of healing, the cell number is dramatically reduced by apoptosis of both vascular cells and myofibroblasts.[@b29-ccid-7-301] To date, it is not known whether myofibroblasts can reacquire a quiescent phenotype, that is, return to a normal dermal fibroblast phenotype with no expression of α-SM actin ([Figure 2](#f2-ccid-7-301){ref-type="fig"}). Origin of wound myofibroblasts ============================== It is generally accepted that the major source of myofibroblasts are local connective tissue fibroblasts that are recruited into the wound.[@b30-ccid-7-301] Dermal fibroblasts located at the edges of the wound can acquire a myofibroblastic phenotype and participate in tissue repair.[@b31-ccid-7-301] However, important heterogeneity in fibroblastic cell subpopulations has also been observed. These subpopulations reside in different locations within the skin and have specific activation and deactivation properties. At least three subpopulations have been identified in the dermis: superficial (or papillary) fibroblasts (papillary dermis is around 300--400 μm deep and is arranged as a ridge-like structure), reticular fibroblasts, which reside in the deep dermis (made of thick collagen and elastin fibers arranged parallel to the surface of the skin), and fibroblasts associated with hair follicles. These cell subpopulations can be isolated and exhibit, depending of the nature and age of the original skin, distinct phenotypic differences when cultured separately.[@b32-ccid-7-301],[@b33-ccid-7-301] Recently, the involvement in tissue repair of local mesenchymal stem cells has been increasingly raised. These progenitor cells have been described in the dermal sheath that surrounds the outside of the hair follicle facing the epithelial stem cells. They are involved in the regeneration of the dermal papilla and can also became wound healing (myo)fibroblasts after a lesion or injury.[@b34-ccid-7-301],[@b35-ccid-7-301] Recent data have also implicated circulating cells, dubbed fibrocytes, in the tissue repair process. Fibrocytes enter injured skin together with inflammatory cells and may then acquire a myofibroblastic phenotype.[@b36-ccid-7-301] In postburn scars, fibrocytes are recruited to the site of the lesion where they stimulate the local inflammatory response and produce ECM proteins, thus contributing to hypertrophic scar formation.[@b37-ccid-7-301] Another type of circulating cell originating from bone marrow has also been suggested to play a role in tissue repair. Mesenchymal stem cells are bone marrow-derived non-hematopoietic precursor cells[@b38-ccid-7-301] that contribute to the maintenance and regeneration of connective tissues through engraftment.[@b39-ccid-7-301] Indeed, they have the capacity to seed into several organs and to differentiate into wound-healing myofibroblasts. This engraftment in injured organs is regulated by the severity of the damage.[@b40-ccid-7-301] Finally, differentiated (or malignant) epithelial and endothelial cells can undergo a phenotypic conversion that gives rise to matrix-producing fibroblasts and myofibroblasts (through epithelial- and endothelial-to-mesenchymal transition processes).[@b41-ccid-7-301] This mechanism is increasingly recognized as an integral part of tissue fibrogenesis after injury, but seems to play a limited role during normal tissue repair. Overall, mesenchymal stem cells, fibrocytes, bone marrow-derived cells, and cells derived from epithelial-and endothelial-to-mesenchymal transition processes may represent alternative sources of myofibroblasts when local fibroblasts are not sufficient to carry out tissue repair and remodeling. Role of myofibroblasts in diseases (excessive scarring/fibrosis) ================================================================ Myofibroblasts are implicated in many fibrotic and scarring diseases, where they carry out the important process after initial injury of providing mechanical support and integrity to the tissue. In normal physiological conditions, they are then lost via apoptosis, generally when the tissue integrity has been sufficiently restored to be mechanically coherent.[@b9-ccid-7-301],[@b29-ccid-7-301] Thus, in normal physiological situations like skin wound healing, myofibroblasts disappear in a prominent wave of apoptosis, leaving a markedly less cellular scar. However, it is now assumed that, in many fibrotic and scarring conditions, as well as in the stromal response to tumors, myofibroblasts fail to undergo cell death, persist, and thus in turn lead to ongoing pathology and scarring ([Figure 3](#f3-ccid-7-301){ref-type="fig"}). An example of reduced or inhibited apoptosis leading to scarring is in a model of hypertrophic scarring, where mechanical loading increases survival of myofibroblasts and was found to lead to greater scar formation.[@b42-ccid-7-301],[@b43-ccid-7-301] A better understanding of the control and signaling that governs apoptosis (or autophagy) in myofibroblasts may lead to more targeted approaches to combatting fibrosis and scarring. It has been demonstrated that elevated nicotinamide adenine dinucleotide phosphate (NADPH) oxidase 4-derived hydrogen peroxide, supported by concomitant decreases in nitric oxide signaling and reactive oxygen species scavengers, are central factors in the molecular pathogenesis of fibrosis.[@b44-ccid-7-301] Redox signaling could therefore represent an interesting target to restore the physiological fibroblast--myofibroblast ratio. Apoptosis in myofibroblasts is thought to be regulated by a reduction in the local growth factors that drive and sustain myofibroblast differentiation. In particular, local concentrations of transforming growth factor (TGF)-β1 and endothelin-1 play a role in stimulating myofibroblast survival via protein kinase B (AKT) activation.[@b45-ccid-7-301] However, changes in mechanical signaling such as unloading of mechanical force likely also plays a role (discussed below). The importance of myofibroblasts in causing fibrosis in internal organs and the skin (hypertrophic scars), and the role that persistence of stromal myofibroblasts appear to play in tumor growth and spread, makes the (down)regulation of myofibroblasts and the potential regulation of myofibroblast disappearance through apoptosis of increasing interest ([Figure 4](#f4-ccid-7-301){ref-type="fig"}) (for review, see Hinz and Gabbiani[@b46-ccid-7-301]). Regulation of myofibroblasts by mechanical forces ================================================= Mechanical signals have been shown to play a role in myofibroblast differentiation as the ECM that surrounds the fibroblasts in damaged tissue changes its composition and its stiffness as tissue repair proceeds.[@b47-ccid-7-301] The early ECM present in damaged tissue, or provisional matrix, is rich in fibrin and has been estimated to be very compliant. Fibroblasts cultured in compliant ECM such as soft three-dimensional (3D) collagen gels, show little development of stress fibers. These fibroblasts then form only small adhesions with the ECM.[@b48-ccid-7-301],[@b49-ccid-7-301] Fibroblasts grown in stiffer collagen matrices have been shown to form stress fibers and mature focal adhesions, though they are still negative for the myofibroblast marker α-SM actin. Lastly, the stiff matrix found either in 3D cultures using stiff (higher concentration) collagen matrix or in vivo in granulation tissue and fibrotic tissues is able to induce full myofibroblast differentiation in concert with growth factor stimulation from TGF-β1.[@b22-ccid-7-301] The contractile nature of myofibroblasts itself leads to an increase in stiffness and mechanical stress of the ECM as healing progresses, leading to a positive-feedback loop where increased stress signals myofibroblast differentiation and also increases myofibroblast survival.[@b42-ccid-7-301] For this reason, mechanical feedback is considered to be important in driving pathological conditions such as contractures post-injury. The role of mechanical stress in stimulating myofibroblast activity has also been shown in experiments where dermal wounds in mice are mechanically stressed by stretching or splinting the wound, where increased myofibroblast activity results in increased scar formation, to some extent mimicking hypertrophic scarring that is seen in humans.[@b42-ccid-7-301] In cancer biology, matrix stiffness can be used as a diagnostic indicator of the risk of malignancy and appears to correlate with increased invasiveness of tumors, for example in breast cancer where density of tissue on mammography correlates strongly with the risk of cancer formation. Recent studies have suggested this may be due to increased cell proliferation of epithelial and mesenchymal cells on stiffer matrices. Conversely, releasing mechanical stress or reducing stiffness has been shown to induce both apoptosis and a reduction in α-SM actin expression and contractility in myofibroblasts.[@b50-ccid-7-301],[@b51-ccid-7-301] Mechanical signaling and stress modulate myofibroblast differentiation via a number of pathways and mechanisms. Stress may directly activate transcription of the α-SM actin gene, since application of force across integrin binding sites has been shown to up-regulate α-SM actin promoter activity.[@b52-ccid-7-301] As mentioned above, mechanical force alone is not generally sufficient to induce myofibroblast differentiation and other factors are needed to act in concert, specifically TGF-β1. Mechanical signaling and TGF-β1 stimulation also increase collagen gene expression by fibroblasts, emphasizing the role these factors play in stimulating a pro-fibrotic phenotype as is shown by activated myofibroblasts. TGF-β1 also favors deposition of ECM proteins over degradation by up-regulating TIMPs while decreasing expression of the MMPs themselves.[@b53-ccid-7-301] Stimulation of myofibroblasts by TGF-β1 itself is affected by mechanical forces within the damaged or fibrotic tissue. TGF-β1 released from a variety of inflammatory cells and platelets in the microenvironment of damaged or fibrotic tissue is in a latent form. Indeed, myofibroblasts themselves release latent TGF-β1 complexed with latency-associated peptide (LAP). Together with a binding protein, TGF-β1 is bound to ECM proteins, providing a reservoir of latent TGF-β1 that can be activated as healing and scar formation progress.[@b54-ccid-7-301],[@b55-ccid-7-301] Myofibroblasts express integrins that can bind to the LAP, and mechanical stress applied to the integrins, either by mechanical stress on the matrix and/or via myofibroblast contraction, can effectively activate TGF-β1 without cleaving the LAP and allow its binding to cell membrane receptors.[@b51-ccid-7-301] Thus, both increased mechanical stress and contraction can further increase myofibroblast contractile and matrix synthetic activity. This mode of activation provides another possible pathway for regulating myofibroblast activity by blocking integrin binding to latent TGF-β1, for example by blocking the integrin involved in latent TGF-β1 activation, αvβ5.[@b56-ccid-7-301] Inhibition of other integrin-binding sites has also been shown to inhibit myofibroblast development, including blocking of integrins α3β1,[@b57-ccid-7-301] α11β1,[@b58-ccid-7-301] αvβ5,[@b59-ccid-7-301] or β1.[@b60-ccid-7-301] Hypoxia ======= Tissue oxygenation or hypoxia may play a role in both normal healing and pathological situations. In normal wound healing, the wound is transiently hypoxic as vascular perfusion is disrupted by the initial injury. Staining for the hypoxia-induced transcription factor, hypoxia inducible factor (HIF)-1α shows both areas of the early granulation tissue and the overlying migrating keratinocytes to be hypoxic. During normal healing, this hypoxia is resolved within a few days of injury and expression of HIF-1α declines. Hypoxia signaling can induce a number of growth factors that are beneficial to the healing process, prominent amongst them being vascular endothelial growth factor (VEGF) and, thus, acute hypoxia likely plays a beneficial role in healing. However, the same may not be true for chronic hypoxia and chronic hypoxia signaling. Hypoxia has been reported to reduce myofibroblast activation and reduce collagen synthesis and α-SM actin expression.[@b61-ccid-7-301] In vivo studies using HIF-1-deficient mice showed that reducing HIF-1 availability during healing resulted in reduced collagen synthesis and delayed myofibroblast differentiation, suggesting that, overall, in vivo acute hypoxia during healing was normally compensated by induction of genes that allow tissue to adapt to transient hypoxia, such as VEGF.[@b62-ccid-7-301] In fact, fibroblasts that show reduced HIF-1α expression during hypoxia show inhibited migration and collagen synthesis in vitro. It is possible that, in some organs where there is pathological fibrosis and scarring, hypoxia and HIF signaling, possibly during more chronic hypoxic states, actually drives fibrosis, which has been suggested in the case of renal fibrosis. In some cells at least, there is cross talk between hypoxia signaling and TGF-β signaling that may exacerbate matrix synthesis and thus fibrosis.[@b63-ccid-7-301],[@b64-ccid-7-301] Therapies ========= Anti-fibrotic and anti-scarring therapies have proven to be a difficult and elusive area for research, with relatively few advances until quite recently. As the growth factor TGF-β is central to many of the mechanisms of pathological scarring and fibrosis, it has been the target of some therapeutic strategies. Some positive results have been reported with the drug pirfenidone, particularly in lung fibrosis, where the drug lowered TGF-β expression and both tissue and lavage fluid levels of the protein.[@b65-ccid-7-301] Interfering with activation of latent TGF-β is another potential target for anti-scarring therapies, and the role of integrin binding in TGF-β activation makes integrin-blocking antibodies a potential therapy for lowering TGF-β activation and thus downstream signaling.[@b66-ccid-7-301],[@b67-ccid-7-301] Specific inhibitors of TGF-β signaling have also been suggested as possible treatments for scarring and fibrosis. TGF-β exerts its pro-fibrotic effects through transcription factor signaling Smad3, and selective inhibition of Smad3 phosphorylation and inhibition of Smad3 interaction with Smad4 has been shown to reduce fibroblast activation to the myofibroblast phenotype and also reduce ECM synthetic activity of the cells.[@b68-ccid-7-301] Other molecular targets include tyrosine kinases, and the drug imatinib mesylate has been reported to be anti-fibrotic through inhibiting downstream molecules that are required for the TGF-β response while having an additional anti-fibrotic role by also inhibiting platelet-derived growth factor signaling.[@b69-ccid-7-301] Interestingly, HMG-coA reductase inhibitors such as statins have been shown to have anti-fibrotic effects, likely through inhibition of ROCK.[@b70-ccid-7-301] The widespread use of, and low rate of side effects associated with, these drugs may make them promising as anti-fibrotic therapies in the future. Conclusion ========== Since their first description in the early 1970s, our knowledge about myofibroblast biology has increased greatly, and our interest in the biology of myofibroblasts has also increased, as this cell has been implicated in many pathological situations in addition to their role in normal wound repair. Despite major advances in our understanding of the origins of myofibroblasts and the factors that regulate their differentiation and activity, it remains a challenge and a major goal of researchers to find ways in which to regulate their activity to improve healing and scarring in the clinic. It is interesting that the skin is a highly sensitive organ. It is densely innervated by different sensory nerve fiber subtypes that react to tissue injury, temperature variation, and tactile stimuli ([Figure 5](#f5-ccid-7-301){ref-type="fig"}). Cutaneous sensory nerve fibers are endings of dorsal root ganglia (or spinal ganglia) neurons that carry signals from sensory organs toward the appropriate integration center of the brain or of the spinal cord. Several clinical observations indicate that damage to the peripheral nervous system influences wound healing, sometimes resulting in chronic wounds within the affected area. Patients with cutaneous sensory defects due to spinal cord injury or diabetic neuropathy have an increased risk of developing ulcers that fail to heal. In addition, in aged patients, cutaneous repair processes are less efficient and this could be partly due to a deterioration of the peripheral nervous system at the skin level. Interestingly, factors that are required for sustaining peripheral nerves, the neurotrophin network, have also been shown to have direct effects on dermal fibroblasts in regulating ECM secretion, fibroblast differentiation, and tensile strength via effects on myofibroblasts.[@b71-ccid-7-301] Understanding the role of innervation during wound healing and myofibroblastic differentiation therefore represents an interesting domain. In addition, cutaneous innervation is certainly necessary to provide good hemostasis and to maintain the mechanical and cosmetic properties of the skin.[@b72-ccid-7-301] In conclusion, the recent advances made in understanding control of differentiation, proliferation, and survival of myofibroblasts will hopefully lead to new therapeutic approaches to limit scarring and improve healing in the not-too-distant future.[@b73-ccid-7-301] This work was supported in part by the French Armaments Procurement Agency (DGA, No 2013 94 0903). Betty Laverdet was recipient of a fellowship from the French Armaments Procurement Agency. The authors thank Régine Baudet (LVMH Recherche) for the production of the figures. **Disclosure** The authors report no conflicts of interest directly relevant to the content of this paper. {#f1-ccid-7-301} {#f2-ccid-7-301} {#f3-ccid-7-301} {#f4-ccid-7-301} {#f5-ccid-7-301}

Introduction ============ Polychlorinated biphenyls (PCBs) consist of two linked benzene rings in which one or more of the hydrogen atoms have been replaced by chlorine atoms. Ten positions are available for substitution; there are 209 PCB congeners defined by the number and placement of the chlorine atoms. Mixtures of PCB congeners were manufactured in the United States from 1929 to 1977 for use as coolants and lubricants in transformers, capacitors, and other electrical equipment \[[@r1]\]. They were also used in building construction materials as additives to elastic sealants, caulking, grouts and paints, and flame retardant coatings of acoustic ceiling tiles ([@r6]; [@r10]). Unfortunately, the unique chemical properties of PCBs that made them useful in commercial applications (e.g., thermal stability, resistance to acids and bases, low water solubility) also contributed to their resistance to degradation, bioaccumulation in terrestrial and aquatic food chains, long-range transport, and toxicity ([@r1]). Although the commercial manufacture of PCBs was banned in the United States in 1979, they continue to be released into the environment through the use and disposal of PCB-containing products. In 2004, [@r11] conducted a study of 24 buildings in and around Boston, Massachusetts. Exterior caulk samples from eight buildings contained PCB concentrations high enough \[range, 70--36,200 ppm (parts per million); mean, 15,600 ppm\] to require the material to be treated as PCB bulk product waste \[[@r39]\]. [@r21] reported on an elementary school with PCB-containing caulk (range, 1,830--29,400 ppm) that had a mean PCB indoor air concentration \> 500 ng/m^3^: orders of magnitude greater than typical background concentrations in ambient urban air \[1--10 ng/m^3^ ([@r1])\]. Similar indoor air PCB concentrations have been reported for other buildings constructed with PCB-containing caulk ([@r10]; [@r16]; [@r41]). Caulk is only one of several potential sources of indoor air PCB contamination ([@r33]). Additional primary sources (i.e., those that were manufactured containing PCBs or had PCBs added during construction) that might currently be found in buildings include window glazing, fluorescent light ballasts, ceiling tile coatings, and other materials such as paints or floor finishes. Secondary sources of PCBs may also contribute to elevated indoor air PCB concentrations. Secondary sources are defined here as materials that become contaminated through absorption from direct contact with primary PCB sources, or through absorption of PCBs in the indoor air that have been emitted by primary sources. Secondary sources may include paints, mastics, ceiling tiles, flooring, and wall boards. Caulk and other building materials containing PCBs were widely used from the 1950s through the 1970s. PCBs in fluorescent light ballasts were discontinued in the late 1970s, but many buildings throughout the United States that were constructed before 1979 may still have fluorescent lighting fixtures that contain PCBs and/or PCB residues from leaking or previously burned-out ballasts ([@r38]). Thus, human inhalation exposure to PCBs may be more widespread than previously assumed. Discussion ========== Humans can be exposed to PCBs via ingestion, inhalation, or dermal contact. Consumption of contaminated food has historically been considered the major route of exposure among the general population, with fatty foods (i.e., fish, meat, dairy products) being the major contributors to dietary exposure ([@r1]). Based on data from the Food and Drug Administration Total Diet Study ([@r25]), which evaluates ready-to-eat foods to determine the dietary intake of selected contaminants, dietary exposures to PCBs have declined over the last several decades (e.g., from 27 ng/kg-day in 1978 to 2 ng/kg-day in 1997 for adults) ([@r1]). Inhalation may be an important contributor to total PCB exposure. To evaluate its contribution to total exposure, we estimated potential total and route-specific background PCB indoor and outdoor exposures (see Supplemental Material, Table S1). We estimated exposure for ingestion of soils and dusts, ingestion of food, dermal contact, and inhalation for children (i.e., ages 2--3 years and 6--12 years) and adults using average PCB exposure concentrations reported in the scientific literature ([@r9]; [@r27]) and exposure factors from the U.S. EPA's *Exposure Factors Handbook* ([@r36]). For children ages 2--3 years and 6--12 years and adults, respectively, total PCB exposures were 6.9, 4.7, and 3.2 ng/kg-day (see Supplemental Material, Table S1). The estimates suggest that the extent of inhalation exposure to PCBs in some indoor settings may be at least as large as typical dietary exposure. Inhalation of indoor air was estimated to account for 60.8, 50.5, and 34.6% of total exposure, whereas diet accounted for 28.9, 42.7, and 62.8% of total exposure for children ages 2--3 years and 6--12 years and adults, respectively. The contributions to total exposure from all other pathways, including outdoor air, were relatively small. The exposures estimated here were based on background PCB concentrations. Total exposures in PCB-contaminated buildings would be expected to be higher, and the contributions from the various exposure pathways would likely differ from those reported here, with inhalation potentially accounting for an even greater proportion of the overall exposure. These results suggest the need to consider inhalation when evaluating the overall human health risk from exposure to PCBs in some environments. To characterize the health risk associated with PCB exposure, information is needed on relevant health outcomes and their exposure--response relationships. This information may come from human or laboratory animal studies. Many such studies have shown that chronic oral PCB exposure is associated with both cancer and noncancer health effects ([@r1]; [@r18]), but there is a profound lack of data to support exposure and exposure--response assessment for inhaled PCBs. Three intermediate-duration studies in animals provide information on the effects of inhaled PCBs. [@r34] observed histopathologic lesions in livers of rats, mice, rabbits, and guinea pigs exposed to 1.5 × 10^6^ ng/m^3^ of a mixture of PCB congeners volatilized from Aroclor 1254 for 7 hr/day, 5 days/week for a total study duration of 213 days. [@r3] exposed adolescent male Sprague-Dawley rats to an atmosphere containing 900 ng/m^3^ of a PCB mixture volatilized from Aroclor 1242 via whole-body inhalation for 23 hr/day for 30 days. Effects included significant histopathological changes in the thyroid and thymus, increases in serum thyroid hormone concentrations, and a significant decrease in normal exploratory behavior. Notably, the inhalation exposure level tested was within the range of concentrations observed in some public buildings. [@r14] observed minimal toxicity in female Sprague-Dawley rats exposed via nose-only inhalation to 5.2 × 10^5^ ng/m^3^ of a PCB mixture developed to mimic the congener profile of air in Chicago, Illinois, for an average of 1.6 hr/day, 5 days/week for 4 weeks. Many toxicological end points were investigated, but only minimal responses were observed. Although relative thymus weight was 12.5% lower in the PCB-exposed group, the difference was not statistically significant. In contrast with [@r3], [@r14] did not report histopathological effects in the thyroid or measure changes in serum thyroid hormone concentrations or neurobehavioral end points. Differences in the exposure methods used by these two studies (i.e., whole-body vs. nose-only), the PCB congener mixtures tested, and the number of hours per day that the rats were exposed (i.e., 23 vs. 1.6) make it difficult to interpret differences in the results. Of the three toxicological studies of PCB inhalation exposure described above, two ([@r3]; [@r34]) reported notable health effects, identifying inhalation exposure to PCBs as a potential health hazard. However, for human health risk assessment, preferred exposure--response data are those which demonstrate a clear, quantifiable association between exposure and response measured across multiple exposure levels ([@r35]). In the [@r34] study, several experimental and control-group animals died from extraneous causes, including an epidemic of pneumonia, which could confound the association between PCB exposure and liver histopathology. Also, there may be some uncertainty in the exposure estimate reported by [@r3] because of this study's use of whole-body inhalation exposure (vs. nose-only). As stated by the study authors, whole-body exposure was chosen over nose-only exposure to prevent stress to the animals that would result from restraint for 23 hr/day; such stress could affect the behavioral, endocrine, and immunological end points evaluated. However, a disadvantage of whole-body exposure is that possible deposition of test article on animal fur or skin may result in additional unquantified oral exposure to the animals as a result of grooming behavior. Furthermore, [@r3] did not report complete exposure information including breathing zone concentrations and evidence supporting a uniform distribution of chemical within the exposure chamber. Although studies investigating health effects from inhaled PCBs are limited, the collective results of those studies suggest potential to result in adverse health effects. Robust exposure--response information is important to characterize the risk of inhaled PCBs; thus, additional research is critical to support exposure--response assessment. Needed studies include epidemiological and/or laboratory animal-based investigations with well-characterized PCB exposure by inhalation. Initial studies might focus on health end points known to be sensitive to disruption by PCB exposure (e.g., developmental neurotoxicity, immunotoxicity, and changes in thyroid hormone levels) ([@r1]). Examples of the studies needed, including important study design considerations, are provided as Supplemental Material (Appendix S1). Because PCBs exist in the environment as mixtures, careful consideration should be given to which congeners should be measured to monitor exposure. In some cases, biomonitoring data or information on PCB-containing products manufactured or used in a specific study setting can guide this choice. However, PCB body burden reflects exposure integrated over all routes and pathways (including inhalation and diet), over a prolonged time interval. Consequently, it is not generally possible to determine how much of a person's body burden may be attributable to exposure only from inhalation during a specified period of time. Furthermore, as reviewed by [@r7], many epidemiological studies have assessed exposure based on the analysis of limited subsets of PCB congeners selected because of their toxicological similarity to dioxin, their relative occurrence in biological samples, and/or the ease with which they can be detected using a given analytical method. In most cases, these subsets favor congeners found at high levels in food chains, neglecting congeners found in indoor and outdoor air. And, many of the congeners commonly found in air have relatively short biological half-lives; therefore, even if these congeners are analyzed, associations between exposure and effect may be missed if exposure assessment is based solely on long-term PCB body burden. Despite these limitations, in some cases where one exposure source is dominant, it may be possible to disentangle the pathways of exposure contributing to PCB body burden. Several studies have used biomonitoring to evaluate PCB exposure for persons spending time in buildings where PCBs are known to be present (e.g., [@r12]; [@r24]). In general, patterns of exposure differ between "exposed" and "unexposed" persons, with exposed persons having higher proportions of lower-chlorinated congeners (i.e., congeners containing ≤ 4 chlorines). Physiologically based pharmacokinetic models describing the kinetic properties of inhaled PCB congeners with varying numbers of chlorines may facilitate the use of biomonitoring data for assessing inhalation exposures. Human and animal studies can support the development of such models by measuring individual PCB congeners and metabolites in biological samples taken at various time points relative to exposure. There has also been recent progress in characterizing metabolites of some PCB congeners found predominantly in air ([@r5]). After careful validation, it may be possible to use measures of certain PCB congeners or their metabolites as useful biomarkers of inhalation exposure to PCBs; but in the meantime, it is important for epidemiological studies to characterize exposure based on PCB concentrations in air using analytical methods that can detect a wide range of congeners (e.g., [@r13]). In a general population setting, one straightforward approach would be to have study participants use personal air monitors to measure PCB exposure levels. Another approach would be similar to those described above (see Supplemental Material, Table S1) and in a report by [@r33], in which information on PCB levels in air in different locations (e.g., inside an office building, inside a residence, outside) was combined with information on each individual's time spent in monitored areas (e.g., a person's work schedule and time spent at home). Adjustments for inhalation rate may be made for factors such as sex, age, and exertion level to more accurately quantify the degree of exposure ([@r36]). It may also be useful to gather data to compare dietary PCB exposures in inhalation-exposed and reference populations. Initially, epidemiological studies should investigate associations between PCB inhalation exposure and key health effects identified from studies of PCB body burden, including the following: changes in serum thyroid hormone levels, increased susceptibility to infection, decreased antibody responses to immunization, and cognitive effects in children ([@r2]). Populations of interest include adults and children directly exposed to PCB-contaminated air, and children born to women exposed to PCBs by inhalation before or during pregnancy and lactation. Repeated exposure assessment at various time points in both children and their mothers during (or even before) pregnancy may reveal associations between key health effects and PCB exposures during sensitive periods of development ([@r22]). PCB exposures before pregnancy may be important because many PCB congeners accumulate in body lipids and may be transferred to an infant during pregnancy or through breastfeeding. As described by [@r19], this toxicokinetic property of PCBs may also have important implications for developmental exposure studies in animals, which may benefit from the inclusion of an exposure period before mating and gestation. Relevant end points (e.g., changes in serum thyroid hormone levels, neurodevelopmental effects, and immunosuppression) can be measured in both humans and animals. For example, relevant neurodevelopmental end points can be tested using assays of schedule-controlled behavior, attention and associative processing, spatial learning, and behavioral inhibition ([@r28]). The most relevant assays for determining potential immunotoxicity hazard are those measuring disease resistance or immune function (e.g., antibody response to antigen, and natural killer cell activity) ([@r15]). Gathering data across species and showing similar patterns of effects for concordant outcomes provides important support for the use of animal exposure--response data in human health risk assessment. A chronic or developmental exposure study in animals exposed to PCBs by inhalation would be an improvement over the intermediate-duration data that are currently available to support exposure--response assessment for inhalation exposure to PCBs. Because offspring exposure occurs by maternal transfer first via the placenta and then through milk, assessment of total offspring exposure can be facilitated by measuring PCB body burden directly in the offspring ([@r19]). This is especially important because of the above-mentioned tendency of PCBs to accumulate in adipose tissue before pregnancy; these established maternal PCB stores can contribute to offspring exposure during pregnancy and lactation, resulting in higher exposures to the offspring than would occur only with concurrent maternal PCB exposure by inhalation or other routes. Important data might also be gleaned from a developmental study using oral exposure if *a*) the administered PCB mixture were patterned after a congener profile observed in air (e.g., [@r42]), and *b*) toxicokinetic data were collected to support route-to-route extrapolation. An additional factor important to consider in PCB inhalation study design is the specific method used to deliver PCBs to the animals (i.e., whole-body or nose-only exposure). As mentioned above as a potential point of concern with the study by [@r3], whole-body exposure to volatilized PCBs may allow for some additional unquantified oral exposure to occur as a result of animal grooming behavior. However, there are also disadvantages of nose-only exposure. The restraint required for nose-only exposure is associated with stress-related responses that might interfere with the detection of exposure-related responses. Also, to prevent undue stress to the animals, nose-only exposure must be limited to only a few (e.g., 4--6) hours per day. And nose-only exposure systems accommodate a limited number of animals at one time, which may reduce the total number of animals that can be used in a single study. Another improvement to the existing exposure--response data could come from a study testing three or more PCB concentrations to support benchmark dose modeling, which is preferred over the NOAEL/LOAEL (no observed adverse effect level/lowest observed adverse effect level) approach to exposure--response assessment ([@r37]). Although a study of chronic duration may be preferred when using data to support human health risk assessment for chronic exposures, even a study of subchronic duration would be an important contribution to the existing PCB inhalation database if it were to measure health effects at three or more exposure levels. In addition to the study design elements described above, there are additional factors to consider when planning studies of PCB inhalation exposure. A PCB mixture's physicochemical properties and toxicity are related to the degree of congener chlorination and the specific congeners that constitute the mixture. For example, lower-chlorinated congeners are much more volatile than higher-chlorinated congeners. Thus, the PCB congener profile found in air may not match the congener profile found in source material (e.g., caulk) because of differences in congener volatility. For exposure assessment, it is important to determine the congener composition of the PCB mixture in air to which humans or animals are exposed. In an inhalation study conducted in animals, one goal might be to use a volatilized PCB mixture with a congener profile most similar to a "typical" human inhalation exposure. However, although some progress has been made in characterizing the congener profiles of PCBs in air from different geographical regions and in different contexts, a great deal of variability has been observed ([@r26]). For example, congener profiles of air in Chicago were dominated by congeners with ≤ 5 chlorines ([@r42]), but congener profiles of outdoor air in Rice Creek, New York, contained large percentages of congeners with 6 or 7 chlorines ([@r4]). In Germany, researchers noted that rooms contaminated with PCB-containing caulk contained a much lower proportion of "dioxin-like" congeners \[i.e., PCBs 77, 81, 105, 114, 118, 123, 126, 156, 157, 167, 169, and 189 ([@r40])\] in air than rooms with acoustic ceiling tiles treated with PCB-containing flame retardants ([@r10]). Thus, different congener patterns in indoor air may exist in buildings with different PCB sources and conditions. To support both exposure and exposure--response assessment, additional work is needed to measure specific congeners in the air of buildings with PCB sources to better characterize and understand the range of relevant inhalation mixtures and specific congener concentrations. It is likely that toxicological data from a number of different mixtures of inhaled PCBs will be needed to adequately assess human health risk across a number of relevant exposure scenarios. Preferably, both epidemiological and toxicological studies of PCB inhalation would characterize PCB congener profiles in the exposure atmosphere to facilitate comparison of results across studies. For toxicological testing, additional work is needed to develop and characterize PCB mixtures in air. Previous studies of the health effects of environmental PCBs have used synthetic mixtures developed from individual PCB congeners (e.g., [@r29]) or extracts of contaminated environmental media (e.g., [@r20]). More recently, researchers have found some success in formulating environmental PCB mixtures by combining multiple commercial mixtures ([@r17]; [@r42]). Most notably, [@r42] developed a PCB mixture resembling the average congener profile in Chicago air by combining Aroclors 1242 and 1254. An atmosphere generated from this mixture was used by [@r14] to investigate PCB inhalation toxicity in female rats. Although these studies reveal a promising initial approach for developing and testing inhaled PCB mixtures, more work is needed. For example, achieving vapor-phase levels of higher-chlorinated PCB congeners in a laboratory setting that match those in environmental air samples has proven to be difficult ([@r14]). Furthermore, some PCB congeners present in environmental media, including air, are not found in commercial mixtures (e.g., PCB-11) ([@r13]), suggesting the possible need to supplement the test mixtures used in toxicological studies with any missing congeners. The relationship between a PCB mixture's congener content and its toxicity has been shown to vary depending on the specific health effect under investigation. Based on oral PCB exposure studies, mixtures with larger percentages of both higher-chlorinated and dioxin-like congeners are the most potent for inducing hepatocellular neoplasms, hepatic histopathological changes, and thyroid follicular-cell hyperplasia in rodents ([@r23]). In contrast, these mixtures have not been found to be more potent than lower-chlorinated PCB mixtures in immunotoxicity and neurotoxicity assays ([@r8]; [@r32]). Because the same type of complexity is expected for PCB inhalation, investigating a wide range of health effects will be important before drawing conclusions regarding the relative toxicity of a particular inhalation exposure mixture. PCB mixtures that have been extensively tested in studies of oral exposure may be very different in congener composition than mixtures found in air. Compared with higher-chlorinated congeners, the lower-chlorinated PCBs commonly found in air are more likely to metabolize to reactive hydroquinones or quinones, which may contribute to genotoxicity and other forms of cellular damage ([@r30]). Some lower-chlorinated congeners, or their metabolites, have been reported to be active in neurotoxicity, estrogenicity, and genotoxicity assays ([@r7]; [@r30]). The congeners most potent at depleting dopamine in pheochromocytoma cells (PCB-4) and increasing phorbol ester binding in cerebellar granule cells (PCB-52) are lower-chlorinated congeners detected at high levels in Chicago air ([@r7]; [@r42]). Another lower-chlorinated congener (PCB-3) has been shown to increase the mutation frequency in the liver of male Big Blue rats, an observation supported by *in vitro* studies in which metabolites of this congener have induced genetic mutations, chromosome breaks, chromosome loss, and polyploidization ([@r30]). There may also be differences in tissue distribution or metabolism of PCBs between the inhalation and oral routes of exposure. For these reasons, it is quite possible that inhalation studies will reveal additional toxicological end points that are particularly susceptible to PCB exposure by inhalation. Conclusions =========== Highly elevated indoor air PCB concentrations have been observed in some buildings. PCB-containing caulk, fluorescent light ballasts, or other PCB sources may be found in any building, public or private, constructed before 1979 (e.g., schools, office buildings, hospitals, houses of worship, residences, and many others), contributing to elevated indoor air PCB concentrations ([@r24]; [@r31]). However, because so few studies of PCB inhalation toxicity have been published, the ability to confidently assess health risk to the occupants of these buildings is limited. Additional inhalation exposure and toxicity information for humans and/or animals exposed to PCBs by inhalation would better inform the assessment of PCB risks from this exposure route. Supplemental Material ===================== ###### Click here for additional data file. We are grateful to K. Thomas, G. Woodall, L. Burgoon, R. Sams, J. Vandenberg, K. Deener, L. Flowers, and D. Walsh (U.S. EPA) for providing a thorough review of the draft manuscript. The views expressed in this article are those of the authors and do not necessarily reflect the views or policies of the U.S. EPA. The authors declare they have no actual or potential competing financial interests.

Introduction ============ The coronavirus strains have been known since 1960 and usually cause up to 15% of common colds in humans each year, mainly in mild forms. Previously two variants of coronavirus have caused severe illnesses: severe acute respiratory syndrome (SARS) in 2002, with severe acute respiratory distress, resulting in 9.6% mortality; and Middle East respiratory syndrome in 2012, with a higher mortality rate of 34.4% \[[@ref1]-[@ref3]\]. The novel coronavirus (SARS coronavirus 2), the seventh coronavirus known to infect humans, is a positive single-stranded RNA virus that probably originated in a seafood market in Wuhan in December 2019 \[[@ref4],[@ref5]\]. Since then, the coronavirus disease (COVID-19), named by the World Health Organization, has affected more than 2 million people worldwide, killing more than 130,000 of them \[[@ref6]\]. The COVID-19 pandemic coincided with the launch and development of the 5G mobile network. Compared to the current 4G networks, 5G wireless communications provide high data rates (ie, gigabytes per second), have low latency, and increase base station capacity and perceived quality of service \[[@ref7]\]. The popularity of this technology arose because of the burst in smart electronic devices and wireless multimedia demand, which created a burden on existing networks. A key benefit of 5G is that some of the current issues with cellular networks such as poor data rates, capacity, quality of service, and latency will be solved \[[@ref7]\]. Although there is no scientific proof, the technology is suggested to negatively affect health on certain social media channels \[[@ref8]\]. In the first week of January, some social media users pointed to 5G as being the cause of COVID-19 or accelerating its spread. The issue became a trending topic and appeared visible to all users on Twitter within the United Kingdom. Since then, multiple videos and news articles have been shared across social media linking the two together. The conspiracy has been of such a serious nature that, in Birmingham and Merseyside, United Kingdom, 5G masts were torched over concerns associating this technology and the spread of the disease according to the British Broadcasting Corporation \[[@ref9]\]. More concerningly, Nightingale hospital in Birmingham, United Kingdom had its phone mast set on fire \[[@ref10]\]. This is unwelcome damage especially at a time when hospitals are required to operate with maximum efficiency. The independent fact-checking website Full Fact noted that the conspiracy was not true and concluded that the theories given to support the 5G claims were flawed \[[@ref11]\]. The National Health Service Director, Stephen Powis, noted in a press conference that the 5G infrastructure is vital for the wider general population who are being asked to remain at home. He noted that: "I\'m absolutely outraged and disgusted that people would be taking action against the infrastructure we need to tackle this emergency" \[[@ref10]\]. The origin of this theory demonstrates the transnational dimension to the new media landscape and the way that fake news and conspiracy theories travel. Previous research has traced the emergence of the conspiracy theory to comments made by a Belgian doctor in January 2020, linking health concerns about 5G to the emergence of the coronavirus \[[@ref12]\]. From April 2-6, 2020, it is estimated that at least 20 mobile phone masts were vandalized in the United Kingdom alone \[[@ref13]\]. Social media is an important information source for a subset of the population, and previous seminal research has noted the potential of Twitter for providing real time content analysis, allowing public health authorities to rapidly respond to concerns raised by the public \[[@ref14]\]. During the unfolding COVID-19 pandemic, recent research has found that platforms such as YouTube have immense reach and can be used to educate the public \[[@ref15]\]. Furthermore, recent research has also called for more understanding of public reactions on social media platforms related to COVID-19 \[[@ref15]\]. The aim of this study was to analyze the 5G and COVID-19 conspiracy theory. More specifically, the research objectives were to answer the following questions: (1) who is spreading this conspiracy theory on Twitter; (2) what online sources of information are people referring to; (3) do people on Twitter really believe 5G and COVID-19 are linked; and (4) what steps and actions can public health authorities take to mitigate the spread of this conspiracy theory? Methods ======= The data set used in this article consists of 6556 Twitter users whose tweets contained the "5Gcoronavirus" keyword or the \#5GCoronavirus hashtag, or were replied to or mentioned in these tweets from Friday, March 27, 2020, at 19:44 Coordinated Universal Time (UTC) to Saturday, April 4, 2020, at 10:38 UTC. Users were included in the data set if they sent a tweet during the time the data was retrieved or were mentioned or replied to in these tweets. This specific keyword and hashtag were selected, as this was the most popular and briefly became a trending topic on Twitter within the United Kingdom in early April. The network consists of a total of 10,140 tweets, which are composed of 1938 mentions, 4003 retweets, 759 mentions in retweets, 1110 replies, and 2328 individual tweets. The data was retrieved using NodeXL (Social Media Research Foundation) and the network graph was laid out using the Harel-Koren Fast Multiscale layout algorithm \[[@ref16]\]. In interpreting the network graph, the results build upon previous seminal research, which has identified six network shapes and structures that Twitter topics tend to follow \[[@ref17]\]. These network shapes can consist of broadcast networks, polarized crowds, brand clusters, tight crowds, community clusters, and support networks. A computer running Microsoft Windows 8 was used to retrieve data in Microsoft Excel 2010 using the professional version of NodeXL (release code: +1.0.1.428+). NodeXL uses Twitter's search application programming interface (API). URLs were automatically expanded within NodeXL. A number of techniques were drawn upon. First, the study used the 5Gcoronavirus keyword, which retrieved mentions of both "5Gcoronavirus" and "\#5Gcoronavirus." Second, influential users, topics, and web sources were studied, and a social network analysis of the discussion was conducted with NodeXL, a validated methodology used in previous research \[[@ref18],[@ref19]\], which provided an understanding of the shape of the conversation. The graph's vertices were grouped by cluster using the Clauset-Newman-Moore algorithm. Third, a manual content analysis \[[@ref20]\] of Twitter data was conducted by removing a 10.00% sample of individual tweets (n=233/2328). Coding categories were created by exploring the data and the extracted sample was read and coded. In our content analysis, mentions were not examined because they are typically conversations between users, and retweets were excluded to avoid overpopulating the sample with similar messages. Retweets and mentions were only removed for the manual content analysis and all other analysis in the study includes them. Only English-language tweets were coded. The coding was confirmed by another author and any disagreements were discussed and resolved, which led to a 100% agreement. Individual users have been anonymized in-line with widely cited best practices for research on Twitter \[[@ref21]\]. Results ======= Social Network Analysis ----------------------- [Figure 1](#figure1){ref-type="fig"} groups Twitter users in social network graph clusters. Each small color dot represents a user and a line between them represents an edge. Groups were formed around this topic based on how frequently users mentioned each other. There is an edge for each "replies-to" relationship in a tweet, an edge for each "mentions" relationship in a tweet, and a self-loop edge for each tweet that is not a "replies-to" or "mentions." The size of the nodes has been ranked by their betweenness centrality score (BCS) \[[@ref22]\], which measures the influence of a vertex over the flow of information between all other vertices under the assumption that information flows over the shortest paths among them. {#figure1} [Figure 1](#figure1){ref-type="fig"} highlights that a number of different groups were formed, but two large groups stand out within the cluster, which are labelled as "*Group 1 - Isolates Group*" and ""*Group 2 - Broadcast Group*." The network shape "*Group 1 - Isolates Group*" displays users who were tweeting without mentioning one another. Isolates groups are a common network structure found in Twitter networks. Large brands, sporting events, and breaking news stories tend to have a large isolates network structure. During a sports event, for example, a large number of users may offer their view or opinion toward a team without mentioning or replying to other users forming an isolates cluster. Group 2 (labeled Group 2 - *Broadcast Group)* contains a number of Twitter accounts who would tweet that there was a link between 5G and COVID-19, which attracted retweets, giving rise to a broadcast network. Within this group, a number of influential user accounts can also be seen toward the center of the group and a circle of accounts around these. The broadcast network structure is often found in the networks for news accounts and journalists because their tweets are retweeted with high frequency. Celebrities with large followings will also tend to have a broadcast network shape. Group 4 contains the label "*activism account*" because it contained an account with the name "5gcoronavirus19," which was set up to spread the conspiracy theory on Twitter that is further discussed in the next section. User Analysis ------------- [Table 1](#table1){ref-type="table"} has ranked influential users by betweenness centrality and has provided a description of the user account. The rank column orders users by their betweenness centrality score, the account description provides an outline of the type of account, the betweenness centrality column provides the raw score for each user, the follower column lists the number of followers an account had, and the NodeXL group column identifies which group Twitter users belonged to in [Figure 1](#figure1){ref-type="fig"}. The follower count is based on the amount of followers the users had during this time period. ###### Influential users ranked by their betweenness centrality score. Rank Account description Betweenness centrality score Followers, n Network group in [Figure 1](#figure1){ref-type="fig"} ------ ----------------------------------------------- ------------------------------ -------------- ------------------------------------------------------- 1 Citizen 3,059,934.33 432 7 2 Citizen 3,042,916.47 12 2 3 Citizen 2,926,695.58 546 3 4 Writer 2,655,235.44 1874 2 5 5G and coronavirus dedicated activism account 2,637,433.23 383 4 6 Citizen 2,577,072.58 14 6 7 Citizen 2,354,744.84 175 2 8 Citizen 2,066,430.77 51 2 9 YouTuber 2,003,753.23 130 5 10 Donald Trump 1,380,314.74 75,916,289 4 The majority of influential users tweeting about 5G and COVID-19 consisted of members of the public sharing their views and opinions or news articles and videos supporting their cause. A key feature of the accounts was that they were actively engaged in sharing conspiracy theories; their bios included words such as "uncover" and "truth." User accounts ranked 1-3 appeared to be citizens who were tweeting during this time. The fourth most influential user was a writer who had over 1874 followers. Interestingly, results show that the fifth most influential account was a dedicated propaganda account (created on January 24, 2012), whose sole purpose was to raise awareness of the link between COVID-19 and 5G, and the account was named "5gcoronavirus19." This account was in the group labeled "*activism account*" in [Figure 1](#figure1){ref-type="fig"}. The account creation date appears to be 2012, which suggests that a previously created account was converted, as Twitter allows users to change their user handle and username. The account has since been removed; however, the account bio description was "*5G causes our immune system to lower and we become more susceptible to viruses. Wuhan was the FIRST FULL 5G city! \#Coronavirus caused by 5G*." This user was in group 4 in the network graph outlined in [Figure 1](#figure1){ref-type="fig"} and had sent a total of 303 tweets in the 7-day time period studied in this paper. Group 4 contained a total of 408 Twitter accounts. At tenth place, the president of the United States, Donald Trump, appears as an influential user; however, unlike other users in the network, Trump did not directly tweet about the link between COVID-19 and 5G. Trump appears because he is mentioned by other Twitter users related to general policy and discussion surrounding 5G. All other users within the network were actively tweeting during this time period. The analysis reveals that there was a lack of an authority figure who was actively combating such misinformation. Twitter users that have the highest number of mentions during this time period are shown in [Multimedia Appendix 1](#app1){ref-type="supplementary-material"}. The two most mentioned users were members of the public, and the third most mentioned user was the dedicated COVID-19 activism account mentioned previously and ranked the fifth most influential user in [Table 1](#table1){ref-type="table"}. [Multimedia Appendix 2](#app2){ref-type="supplementary-material"} lists the users that were replied to the most during this time period. The second most mentioned user was again the dedicated coronavirus activism account. This shows how the account linking 5G to COVID-19 also stimulated debate on Twitter and held power over the network because the account was both highly influential and mentioned. Influential English-Language Websites ------------------------------------- In order for the conspiracy theory to spread the public needs information and a reference source. This study has identified which sources were influential during this time. [Table 2](#table2){ref-type="table"} highlights the most influential websites relating to this topic during this time period. The most popular web source shared on Twitter during this time was the website known as *InfoWars*, which is a popular conspiracy theory website based in the United States. The article itself linked to several videos in which "top scientists" revealed how 5G could weaken a population's immune system. The rank column refers to the ranking of each website based on the count column. It can be seen that the majority of the websites can be argued as being "fake" or "alternative" news websites. The websites and information shared on Twitter can also shed light on the types of sources social media users were drawn toward. [Multimedia Appendix 3](#app3){ref-type="supplementary-material"} shows the most frequently occurring domains within the network. This analysis is different to that conducted in [Table 2](#table2){ref-type="table"}, as it identifies overall web domains that were most used in tweets rather than specific websites, showing, for instance, that YouTube appears ranked as the second most popular domain. ###### Influential web sources. Rank Website title Source Tweets, n ------ -------------------------------------------------------------------------------------------------------------------------------------------------------------------------- ------------------------------------------------ ----------- 1 "WATCH LIVE: CORONAVIRUS HOME SCHOOL SPECIAL & ASK THE EXPERTS! Prestigious doctors & scientists confirm 5G weakens the immune system to all viruses including Covid-19" InfoWars \[[@ref23]\] 38 2 "VIDEO: Former President Of Microsoft Canada, Frank Clegg: 5G Wireless IS NOT SAFE" RayGuardNJ Electrosmog Protection \[[@ref24]\] 31 3 "There's A Connection Between Coronavirus And 5G" Stillnessinthestorm \[[@ref25]\] 20 4 "BREAKING NEWS: Slovenia Stops 5G Due to Health Risks" 5gcrisis website \[[@ref26]\] 18 5 "CAN YOU BELIEVE THIS??" Jeff Censored! YouTube channel. \[[@ref27]\] 18 Content Analysis ---------------- From the overall data set, a 10.00% sample of tweets (n=233/2328) that did not mention or reply to another user were extracted. Content analysis revealed that, of the 233 sample tweets, 34.8% (n=81) of individual tweets contained views that 5G and COVID-19 were linked, 32.2% (n=75) denounced the conspiracy theory, and 33.0% (n=77) were general tweets not expressing any personal views or opinions. [Table 3](#table3){ref-type="table"} below displays the results of this coding alongside examples of tweets. The focus was to identify the percent of pro- and anticonspiracy themes. Any other tweets would be classified as "general tweets." It was found that 32.2% (n=75/233) of tweets were views against the conspiracy theories that were being shared. They either attacked or ridiculed those sharing such views with humor. The second category contained tweets that were general in nature and used the "5G" keyword or hashtag in their tweets as highlighted in [Table 3](#table3){ref-type="table"}. This occurred in 33.0% (n=77/233) of tweets. Users may have used the keywords and hashtags for additional exposure. This theme also contained general news articles related to 5G and COVID-19. This is not surprising, as other Twitter users attempt to "link bait" on Twitter by flooding popular topics with content to obtain more viewers for their own tweets or web links. The next category consisted of tweets that were clearly expressing views against the conspiracy or were intending to be humorous toward those linking 5G and COVID-19. The largest category of users, with 34.8% (n=81/233) of the tweets, were engaging with and spreading information that linked COVID-19 and 5G. Anonymized tweet extracts for this theme are provided in [Textbox 1](#box1){ref-type="boxed-text"}. Thus, 65.2% (n=152/233) of tweets derived from nonconspiracy theory supporters, which suggests that, although the topic attracted high volume, only a handful of users genuinely believed the conspiracy. It is also worth noting that on April 4, 2020, the media began to report that a number of 5G masts had been set on fire \[[@ref8]\]. This coincides with the final day that we collected data, and we observed users actively encouraging other users to destroy 5G towers, as highlighted by the final three anonymized tweet extracts in [Textbox 1](#box1){ref-type="boxed-text"}. ###### Content analysis of individual tweets (n=233). Category Theme Example Tweets, n (%) ---------- ------------------------------------------------- -------------------------------------------------------------------------------------- --------------- 1 5G and the coronavirus disease are linked "*5G Kills! \#5Gcoronavirus - they are linked! People don't be blind to the truth!*" 81 (34.8) 2 General tweets not expressing a view or opinion "*I have a 10AM Skype Chat on Monday, COVID-19 \#5Gcoronavirus*" 77 (33.0) 3 Anticonspiracy theories or humor "*5G is not harming or killing a single person! COVID-19 \#5Gcoronavirus*" 75 (32.2) ###### Anonymized tweet extracts from category 1. **Tweets** "5G is the one and only Coronavirus! Radiation from it will easily wipe out the world population. Think! Why did China get rid of their 5G towers? This is why they are now free from the Corona." "5G volumes peaked and infected COVID-19 cases in Italy also peaked, no coincidence!" "People must open their minds and see the truth that 5G kills!" "I didn't believe in all of this stuff until I read this article! \[URL\] Folks, please educate yourselves!" "Make sure to SMASH THOSE 5G masts up!! \#5Gcoronavirus" "5G Towers are burning \[link to video\] - now what should we do with the others?" "Hope we can see some more go down" Discussion ========== Academics have been alarmed at the rate of fake news and misinformation across social media \[[@ref28]-[@ref32]\]. Initially, social media platforms had been praised for their ability to spread liberal messages during events such as the Arab Spring \[[@ref22]\] and during the initial launch of WikiLeaks \[[@ref33]\]. False information has been a genuine concern among social media platforms during COVID-19, and Facebook has implemented a new feature that will inform users if they have engaged with false information \[[@ref34]\]. One method of counteracting fake news is to be able to detect it rapidly and address it head-on at the time that it occurs. In the specific influencer analysis (in the User Analysis section), there was a lack of an authority figure who was actively combating such misinformation. This study found that a dedicated individual Twitter account set up to spread the conspiracy theory formed a cluster in the network with 408 other Twitter users. This account, at the time of analysis, had managed to send a total of 303 tweets during this specific time period before it was closed down by Twitter. In hindsight, if this account would have been closed down much sooner, this would have halted the spread of this specific conspiracy theory. Moreover, if other users who were sharing humorous content and link baiting the hashtag refrained from tweeting about the topic and instead reported conspiracy-related tweets to Twitter, the hashtag would not have reached trending status on Twitter. As more users began to tweet using the hashtag, the overall visibility increased. Public health authorities may wish to advise citizens against resharing or engaging with misinformation on social media and encourage users to flag them as inappropriate to the social media companies. Many social media platforms provide users with the ability to report inappropriate content. A further method of counteracting misinformation is to seek the assistance of influential public authorities and bodies such as public figures, government accounts, relevant scientific experts, doctors, or journalists. A further key point to make is that the fight against misinformation should take place on the platform where it arises. This is because people will not go to a website to read the counteracting report, but they will watch a video or a memo voice sent via WhatsApp or posted on a social media platform. Public TV, newspapers, and radio stations could also seek to devote regular programs to counteract fake news by discussing conspiracy theories that were spreading at the time. It could also be argued that it is important to analyze the context of the fake news and why it is spreading. Are people afraid? Does the theory propose a risk? Any content that aims to correct misinformation should aim to dispel people's fears. This research set out to address four research questions that are now discussed. In regard to identifying how the conspiracy was spreading on Twitter, this article shows that a number of citizens who believed the conspiracy theory were actively tweeting and spreading it (as highlighted in [Table 1](#table1){ref-type="table"}). A dedicated account that was set up for the sole purpose of spreading the conspiracy theory was identified. We also identified the "humor effect" in the sense that even those users who joined the discussion to mock the conspiracy theory inadvertently drew more attention to it. In addressing the second research objective, this paper identifies a number of influential online sources that created content aiming to show a link between COVID-19 and 5G (as highlighted in [Tables 2](#table2){ref-type="table"} and [3](#table3){ref-type="table"}). These consisted of the website *InfoWars,* a commercial organization selling products that protect against electromagnetic fields. A website dedicated to linking 5G to COVID-19 was also identified. Specific YouTube videos and the YouTube domain itself were also found to be influential. The third research objective was to identify whether people really believed 5G and COVID-19 were linked, as Twitter is known to contain humorous content \[[@ref16]\]. It was found that 34.8% (n=81/233) of individual tweets contained views that 5G and COVID-19 were linked. Although it is a low percentage, there are indeed users who genuinely believe COVID-19 and 5G are linked. In regard to the fourth research objective, this article sought to identify and discuss potential actions public health authorities could take to mitigate the spread of the conspiracy theory. Specifically, this study found that an individual account had been set up to spread the conspiracy theory and was able to attract a following and send out many tweets. Based on our analysis of this conspiracy theory on Twitter, its spread could have been halted if the accounts set up to spread misinformation were taken down faster than they were. Public health authorities should also aim to focus on these types of accounts in combating misinformation during the current COVID-19 pandemic. In addition, an authority figure with a sizeable following could have tweeted messages against the conspiracy theory and urged other users that the best way to deal with it is to not comment on, retweet, or link bait using the hashtag. This is because when users joined the discussion to dispel, ridicule, or piggyback on the hashtag, the topic was raised to new heights and had increased visibility. A strength of this study is that it has identified the drivers of the conspiracy theory, the content shared, and the strategies to mitigate the spread of it. Our results are likely to be of international interest during the unfolding COVID-19 pandemic. A further strength of our study is that our methodology can be applied to other conspiracy topics. A limitation of our study is that the Search API can only retrieve data from public facing Twitter accounts. Previous research has noted that certain Twitter topics are likely to contain automated accounts known as "bots" \[[@ref35]\]; for instance, in the case of electronic cigarette (e-cigarette) tweets, research has found that social bots could be used to promote new e-cigarette products and spread the idea that they are helpful for smoking cessation \[[@ref35]\]. A limitation of our study is that we did not identify social bot accounts; however, influential accounts in our study did not appear to display bot behavior (eg, high number of tweets posted) and appeared to display characteristics of genuine accounts. This could be inferred because certain accounts linked to their profile on other platforms such as YouTube. However, future research could seek to identify the ratio of bots to individual accounts related to conspiracy theories. A further limitation is that our content analysis was conducted on English-language tweets, and further research could seek to examine tweets in other languages. Furthermore, a limitation to our study is that, as we retrieved data using a specific keyword, our data may have excluded tweets from users who tweeted about the conspiracy during this time without using our target keyword or hashtag. The COVID-19 pandemic has been a serious public health challenge for nations around the world. This study conducted an analysis of a conspiracy theory that threatened to potentially undermine public health efforts. We discussed key users and influential web sources during this time, and discussed potential strategies for combating such dangerous misinformation. The analysis reveals that there was a lack of an authority figure who was actively combating such misinformation, and policymakers should insist in efforts to isolate opinions that are based on fake news if they want to avoid public health damage. Future research could seek to conduct a follow-up analysis of Twitter data as the COVID-19 pandemic evolves. Conflicts of Interest: None declared. Top mentioned users. Top replied-to users. Top domains. BCS : betweenness centrality score COVID-19 : coronavirus disease e-cigarette : electronic cigarette SARS : severe acute respiratory syndrome UTC : Coordinated Universal Time

Introduction {#s1} ============ Pigs are a very important domesticated species in China. The extreme climate and geographical conditions in China have contributed to the development of 88 indigenous species, many of which have unique characteristics ([@B46]). With an area of more than 15,000 square kilometers and a diverse terrain, Shandong province is a main center for pig production in China. Many pig breeds have been developed to meet the needs of the local people. Native breeds are characterized by strong breeding and lactation ability, good meat quality, strong adaptability, and strong resistance to disease. However, to cater to the growth performance and lean meat content associated with live pig production, modern commercial varieties were introduced in the early 20th century and crossed with local pigs ([@B43]). Indigenous breeds from Shandong province, as well as other Chinese indigenous pig breeds, have been in direct competition with select Western breeds, primarily Duroc (DD), Yorkshire (YY), and Landrace (LL), and many of them are considered to be rare species in danger of becoming extinct. Existing indigenous pig breeds in Shandong are Heigai (HG), Dapulian (DPL), Wulian Black (WL), Laiwu (LW), Yantai Laizhou Black (YTL), Yantai Wendeng Black (YTW), Licha Black (LC), and Yimeng black pig breed (YM). Assessing the genetic variation and population structure in Chinese indigenous pig breeds is critical in animal genetics research and government germplasm conservation. After years of effort, genetic variation and phylogenetic relationships in Chinese indigenous pig breeds have been explored morphologically, cytogenetically, and biochemically ([@B41]). With the rapid development of molecular biology technology, research on the nuclear DNA diversity of pigs has increased, including RAPD, microsatellite DNA markers, AFLP, DNA fingerprinting, and mitochondrial DNA sequencing ([@B27]; [@B5]; [@B48]; [@B35]). However, the majority of genetic data are still poorly characterized, except in a few very famous breeds. Of the seven pig breeds in Shandong province, only LW, YM, and LC have been investigated by microsatellite or mtDNA markers ([@B47]; [@B8]; [@B22]). A comprehensive assessment of their genetic diversity and relationships with Western pig breeds is therefore an important step in the development of protection and improvement programs. Large-scale genotyping plays an important role in research on genetic diversity. High-throughput sequencing technology provides new opportunities for exploring gene functioning. Given its low cost, genotyping accuracy, and labeling efficiency, our study used specific-locus amplified fragment sequencing (SLAF-seq) for genotyping. This strategy, which is used in evolutionary studies on domestication, focuses on the discovery of single nucleotide polymorphisms (SNPs). SLAF-seq is generally suitable for assessing a variety of species and populations. The purpose of this study was to use SLAF-seq to determine the phylogenetic relationships among seven pig breeds in Shandong province, China, and to what extent they were influenced by Western pig breeds (DD, YY, and LL). Materials and Methods {#s2} ===================== Ethics Statement {#s2_1} ---------------- All animal care and treatment procedures were approved by the Animal Ethics Committee of Shandong Agricultural University, China, and performed in accordance with the Committee\'s guidelines and regulations (Approval No.: 2004006). Samples Collection and Genomic DNA Extraction {#s2_2} --------------------------------------------- Ear-punch samples of 80 individuals representing 10 pig breeds were used in the study. Eight pigs were sampled from each breed except HG (n = 7), WL (n = 7), and LC (n = 10). Samples for the seven Shandong breeds were collected from conservation farms, and those for the three Western breeds were collected from a pig breeding farm in Shandong province. The number of samples per breed was determined according to the lineage of the breed. To ensure that samples covered the entire lineage of each pig population, we collected a boar from each lineage. We extracted total genomic DNA from ear-punch samples using the standard phenol-chloroform method ([@B30]). SLAF Library Preparation for Sequencing {#s2_3} --------------------------------------- Analyses of genomic DNA were based on SLAF-seq ([@B31]). The pig genome was selected as the reference genome for electronic enzyme digestion prediction. Our study relied on reduced representation library sequencing, which can reduce the complexity of the genome. The optimal digestion scheme was determined according to four principles: the ratio of SLAFs in the repeat sequence was as low as possible, the digested SLAFs were distributed as evenly as possible across the genome, the length of the SLAFs was highly consistent with the specific experimental system, and the number of SLAFs obtained matched the expected number of tags. These requirements elevated the efficiency of digestion. A combination of *Rsa*I and *Hae*III restriction enzymes was ultimately used (Thermo Fisher Scientific, Waltham, MA, USA) to digest pig genomic DNA into 1,010,652 SLAF tags 314 to 364 bp in length. Then, a single nucleotide (A) overhang was added to the 3′ region of the SLAF tags. To ligate dual-index sequencing adapters to the A-tailed tags, we performed restriction-ligation reactions with T4 DNA ligase (NEB). Then polymerase chain reaction (PCR) amplification was performed that included the diluted restriction-ligation DNA samples of each pig breed, dNTP, Taq DNA polymerase (NEB), and primers containing barcode 1. The PCR products were purified and run out on a 2% agarose gel. Fragments 314 to 364 bp (with indexes and adaptors) were separated. Barcode two was then added to the isolated fragments by PCR amplification. These products were purified in gel and then diluted for pair-end sequencing using an Illumina HiSeq 2500 system (Illumina, San Diego, CA, USA) according to the manufacturer\'s instructions. Read Mapping, SNP Calling, and Filtering {#s2_4} ---------------------------------------- SLAF-seq data were processed by computer. We mapped all raw SLAF pair-end reads to the pig reference genome (Sscrofa 11.1) using BWA ([@B17]). In general, SLAF groups generated by reads were mapped to the same location. If the restriction enzymes only partially digested an accession, a number of reads mapped to the reference genome may have overlapped with two SLAF tags. As mentioned previously, these reads were allotted to both of the SLAF tags in the same accession. We performed SNP calling using both GATK and samtools analysis ([@B18]; [@B21]). If a locus was called by both packages, we defined it as an SNP. PLINK v1.0774 was used to filter high-quality SNPs for subsequent analysis. Minor allele frequency (MAF) evaluation was used to define alleles in each SLAF. Briefly, SNPs with integrity lower than 0.5 and MAF lower than 0.05 were excluded. A total of 314,243 highly consistent population SNPs representing 10 breeds were obtained for analyses of genetic differentiation. To avoid false positives and assess the accuracy of the digestion experiments objectively, we used rice genomic data as a control to evaluate the sequence error rate. Population Differentiation and Genetic Evolution Basics Analysis {#s2_5} ---------------------------------------------------------------- The divergence index, F-statistics (Fst), is a measure of population differentiation based on genetic polymorphism data ([@B13]). We used the PopGen package in BioPerl to evaluate Fst based on 100 kb sliding windows in 10 kb steps ([@B40]). We calculated the average pairwise divergence within a population (π) using a HIERFSTAT package for R. Neighbor-joining trees were constructed using MEGA 5 based on Kimura 2-parameter ([@B33]). The bootstrap analysis (n = 1000 replications) was performed with a DISPAN package ([@B24]) and then beautified *via* the ITOL website (<http://itol.embl.De/>). The population structure was constructed with admixture v1.22 ([@B12]). We hypothesized that the cluster number (K value) would be between 1 and 10. We performed PCA using the smart program of EIGENSOFT ([@B26]), which can assist in evolutionary analysis. To visualize the results of the algorithm analysis, we further analyzed a scatterplot of the first and second principal components. The linkage disequilibrium (r^2^) between pairwise SNPs was computed with Haploview ([@B3]). It is generally true that r^2^ between SNPs separated by large and small genetic distance represents recent and ancient effective population sizes (Ne), respectively ([@B32]). In addition, we inferred the gene flow between the three Western commercial breeds and the seven Chinese indigenous pig breeds in Shandong province. We plotted migration events among the populations using TreeMix, assuming nine migration events ([@B40]). The covariance matrix was calculated based on 314,243 unlinked sites with a window size of an SNP. Detection of Selective Sweep Regions {#s2_6} ------------------------------------ To identify the regions that were selected during domestication, we combined the three Western pig breeds (DD, LL, and YY) into a single domestic gene pool called DLY. We used a genome-wide sliding window method to scan the selected regions with the greatest differences in genetic diversity (π log-ratio DLY/DPL and DLY/LW) and extreme divergence in allele frequency between Western commercial and Chinese indigenous pig breeds in Shandong province. To evaluate nucleotide diversity (π) and Fst, we calculated π and Fst using 100 kb windows with 10 kb steps among genomes. The top 5% π and Fst values ([**Supplementary Figure S1**](#SM1){ref-type="supplementary-material"}) between DLY and DPL and between DLY and LW in each 100 kb sliding window with a 10 kb step were used to determine potential selective-sweep regions from significantly differentiated regions. To examine whether the candidate selective scanning regions had excessive singleton polymorphisms, we calculated Tajima\'s D for each pig breed using the sliding window method mentioned previously. Genes in these selective regions were identified through the Sscrofa 11.1 assembly (<ftp://ftp.ensembl.org/pub/release-91/fasta/sus_scrofa/dna/>) and NCBI database. In addition, we used Panther bioinformatics ([www.Pantherdb.Org](www.Pantherdb.Org)) for the gene function enrichment analysis. Results {#s3} ======= Specific-Locus Amplified Fragment Sequencing and SNP Discovery {#s3_1} -------------------------------------------------------------- Three of the 80 genotype samples (two HG, one YTW) were removed because the sample call rate did not meet criteria (\< 95%). Nest quality control was performed on the remaining 77 samples. Of the 77 samples analyzed, 53 were from Shandong province, and 24 represented pig breeds originating outside of China ([**Figure 1**](#f1){ref-type="fig"}). {#f1} The Sscrofa genome served as the reference genome for predicting electron enzymes and identified a restriction fragment length of Rsal + EcoRV-HF from 314 to 364 bp, which was ultimately defined as a SLAF tag. To acquire the actual SLAF marker analyzed in our study, we performed SLAF-seq in seven WL, eight DPL, eight LW, five HG, eight YTL, seven YTW, 10 LC, eight YY, eight LL, and eight DD using the same enzyme combination as in the silico restriction experiment. As shown in [**Table 1**](#T1){ref-type="table"}, a total of 441.19 million reads were gained from all individuals, and average Q30 and GC content was 96.26% and 39.45%, respectively. The higher Q30 value represented a lower base error rate, which means that the results of the tested sequences were reliable. Similar to the number of expected SLAFs, the total number of SLAF tags was 1,010,652. The average numbers of SLAFs and sequencing depth were 311,266.29 (13.25×), 293,802.00 (15.82×), 268,606.88 (16.82×), 251,058.20 (13.99×), 292,233.75 (16.59×), 297,446.71 (17.89×), 299,237.70 (17.49×), 271,012.13 (17.58×), 293,646.25 (18.49×), and 291,879.88 (18.70×) in WL, DPL, LW, HG, YTL, YTW, LC, YY, LL, and DD, respectively ([**Table 1**](#T1){ref-type="table"} and [**Supplementary Table S1**](#SM1){ref-type="supplementary-material"}). In addition, we mapped the SLAF tags to the reference genome using BWA ([@B17]). Moreover, when *Oryza sativa indica* was used as a control for evaluating the sequencing data, the efficiency of paired-end comparison and digestion of control were 96.97% and 95.54%, respectively, which indicates that the process was normal and available. ###### Characteristics of SLAF-seq between Shandong indigenous pig breeds and three Western pig breeds. Item WL DPL LW HG YTL YTW LC YY LL DD --------------------- ----------------- ----------------- ----------------- ----------------- ----------------- ----------------- ----------------- ----------------- ----------------- ----------------- Statistics of reads Total Reads 2990944-5966079 4019083-6070323 2975990-6381053 2650499-4682311 3867880-6470358 4454048-7384759 4208772-7544960 4391496-5628749 3713911-6903758 4557758-7187571 Average of Reads 4329274.14 4839942.63 4712400.25 3919117 5023752.88 5514933.71 5388685.20 4890713.86 5600453 5596310 GC percentage (%) 37.84-40.55 38.10-40.00 37.99-39.11 37.84-42.99 38.57-39.68 38.52-39.41 38.45-39.71 37.76-39.64 38.58-42.41 38.52-39.59 Q30 percentage (%) 94.87-95.66 94.91-95.65 95.04-95.68 95.17-95.45 95.05-95.74 94.81-95.82 94.54-95.20 94.49-95.21 94.39-94.87 94.68-95.26 Statistics of SLAFs SLAF number 235261-350679 273735-316650 218020-292516 191031-292759 266302-311520 273951-327892 275906-322120 239010-289822 252186-324777 265283-322884 Average of SLAFs 311266.29 293802 268606.88 251058.20 292233.75 297446.71 299233.70 271012.13 293646.25 291879.88 Average depth 13.25 15.82 16.82 13.99 16.59 17.89 17.49 17.58 18.49 18.70 Statistics of SNPs SNP number 797081-1160118 950380-1087829 749375-1008772 656017-1026786 923071-1053079 946837-1122484 943594-1073169 826568-977905 863697-1076420 904408-1053875 Average of SNPs 1050908.57 1013398.63 928014.88 870682.4 1002405.5 1024125.14 1007319.2 925445.25 986074.5 980454.5 Integrity (%) 48.03 46.31 42.41 39.79 45.81 46.80 46.03 42.29 45.06 44.81 Heter ratio (%) 10.52 9.83 7.86 9.03 9.20 10.28 9.81 6.92 7.27 5.84 SNP markers were defined based on the greatest depth of the sequence type as a reference sequence in each SLAF tag. A total of 8,615,537 SNPs were obtained from data from 77 samples and two wild boars downloaded from the database (ERR173221: <http://www.ebi.ac.uk/ena/data/view/SAMEA1557437>; ERR173222: <http://www.ebi.ac.uk/ena/data/view/SAMEA1557421>). The average integrity of SNPs was 44.73% (range = 29.98%--53.02%). A previous study involving SNPs (MAF \< 0.05) was biased in quantifying genetic connectivity in the study of population genetic evolution ([@B28]). To reduce any sequencing errors, eliminate baseline differentiation, and evaluate accuracy, we filtered 314,243 SNPs with MAF \> 0.05 from 1,010,652 SLAFs in the 77 samples. The average number of SNPs in each breed was identified, ranging from 870,682.40 in HG to 1,050,908.57 in WL ([**Table 1**](#T1){ref-type="table"}). In addition, the number of SNPs in each sample was determined separately ([**Supplementary Table S2**](#SM1){ref-type="supplementary-material"}). Analyses of heterozygous loci revealed that SNP heterozygosity differed markedly by population. DD, which originated in the eastern United States, had a heterozygous ratio of only 5.84%, whereas the highest heterozygous ratio (10.52%) was found in WL. Genetic Variation and Genetic Distance Among Populations {#s3_2} -------------------------------------------------------- The observed heterozygous number (Ho), expected heterozygous number (He), Nei diversity index (Nei), polymorphism information content (PIC), and MAF were used to determine genetic diversity. As shown in [**Table 2**](#T2){ref-type="table"}, all five measures revealed larger values for the Shandong indigenous breeds than the Western breeds, except for LC. Of the seven indigenous breeds in Shandong, HG had the highest genetic diversity (Ho = 0.2926, He = 0.3752, Nei = 0.3814, PIC = 0.2988, and MAF = 0.2852), whereas LW had the lowest genetic diversity (Ho = 0.2443, He = 0.37453, Nei = 0.33743, PIC = 0.2771, and MAF = 0.2585). In contrast, genetic diversity did not vary significantly among the Western pig breeds. As can be seen from [**Table 2**](#T2){ref-type="table"}, the average Ho ranged from 0.2249 in YY to 0.2403 in DD, and the average He ranged from 0.3259 in LL to 0.3378 in DD. It is interesting that all five calculations were higher than in the other two introduced breeds. In general, these results indicate that diversity was greater in the Shandong indigenous pig breeds than the Western breeds. Pairwise comparisons of Fst among five populations described by Weir and Cockerham are shown in [**Supplementary Table S3**](#SM1){ref-type="supplementary-material"} ([@B39]); obvious interpopulation genetic variation appeared between DD and DPL (0.4393), between DD and LW (0.5144), and between DPL and LW (0.2337), which indicates that DPL and LW have a closer genetic relationship with each other than with DD. The same was found among the YY, LL, DPL, and LW populations. ###### The genetic variation of 7 Shandong indigenous and 3 Western pig breeds. Breed Ho He Nei PIC MAF ------- -------- -------- -------- -------- -------- WL 0.2596 0.3497 0.3814 0.2894 0.2597 DPL 0.2711 0.3449 0.3721 0.2767 0.2579 LW 0.2443 0.3453 0.3743 0.2771 0.2585 HG 0.2926 0.3752 0.4281 0.2988 0.2852 YTL 0.2502 0.3447 0.3723 0.2771 0.2553 YTW 0.2602 0.3429 0.3735 0.2761 0.2535 LC 0.2634 0.3348 0.3548 0.2696 0.2476 YY 0.2249 0.3356 0.3625 0.2706 0.2476 LL 0.2283 0.3259 0.3511 0.2639 0.2378 DD 0.2403 0.3378 0.3639 0.2719 0.2504 Ho stands for observed heterozygous number; He, expected heterozygous number; Nei, nei diversity index; PIC, polymorphysm information content; MAF, minor allele frequency. The following is the same. Population Genetic Structure and Gene Flow {#s3_3} ------------------------------------------ To explore genome-wide relationships and the divergence between the Shandong indigenous breeds and Western breeds, we constructed neighbor-joining trees based on pairwise genetic distance ([**Figure 2A**](#f2){ref-type="fig"}). The trees showed that individuals of the same breed generally clustered together, which signified that they possessed unique breed identities. The results did not distinctly differentiate the Shandong and Western breeds into various clusters. Two European breeds (YY and LL) first formed a distinct outgroup, then joined with DD, and finally joined with YTL. It is noteworthy that LW and DPL were grouped together. We utilized a genetic admixture analysis to search the population structure with increasing numbers of inferred population groups (from k = 2 to k = 10). We determined based on the minimum CV error that the optimal number was K = 2 ([**Supplementary Figure S2**](#SM1){ref-type="supplementary-material"}), which was also the number of origins collected in this study. In panels with K = 2 inferred clusters, Shandong indigenous breeds and Western breeds were differentiated. When K = 9, six local breeds in Shandong (LW, HG, YTL, YTW, LC, and DPL) formed a distinct cluster, whereas the remaining breeds appeared in clusters mentioned previously in different proportions ([**Figure 2B**](#f2){ref-type="fig"}). As expected, PCA and neighbor-joining tree analyses clearly differentiated the Shandong indigenous breeds and Western breeds ([**Figure 2C**](#f2){ref-type="fig"}), and no distinct clusters were discovered that sustained the efficaciousness of the phylogenetic tree. Many of the geographically close breeds resembled one another. To determine the genetic contribution among the different breeds, we performed migration modeling using TreeMix. As [**Figure 2D**](#f2){ref-type="fig"} shows, we found relatively strong gene flow between WB and a variety of indigenous and introduced breeds and relatively weak gene flow between YTW and YY, between DD and HG, and between YTL and LL. {#f2} Accessing of Linkage Disequilibrium by r^2^ {#s3_4} ------------------------------------------- The r^2^ for pairs of loci was counted for each pig breed. As physical distance increased, a similar downward trend in average r^2^ was noted in each population ([**Figure 3**](#f3){ref-type="fig"}). As expected, LD patterns revealed that the distance of LD decay was higher in DD than DPL and LW. This shows that DPL and LW are probably subject to higher selection pressure than DD. {#f3} Screening for Selective Sweeps in the Two Indigenous Breeds {#s3_5} ----------------------------------------------------------- To confirm the genomic regions involved in domestication, genetic diversity (π) and SNP-based Fst estimation were calculated to determine positively selected regions, defined as regions with shared differential selection signals among populations. We combined the three Western pig breeds (YY, LL, and DD) into a single gene pool and compared them to DPL and LW. A total of 162 differentially selected regions (DSRs) were shared between DPL and the commercial breeds. A further 841 unique known genes were found that contained a large number of genes related to immunity, such as *SBNO2, PIK3AP1*, and so on ([**Figures 4A**](#f4){ref-type="fig"} and [**5A**](#f5){ref-type="fig"}). Gene functional analyses indicated that biological processes in DPL and the three Western pig populations were enriched in T cell homeostasis, the production of tumor necrosis factor superfamily cytokines, the toll-like receptor 7 signaling pathway, and the inflammatory response ([**Table 3**](#T3){ref-type="table"}). {#f4} {#f5} ###### The autosomal differentially selected regions (DSRs) among DPL, LW, and three Western commercial pig breeds. The partial candidate genes were given within the SNPs of the top 5% for each region. DSRs between DPL and three Western pig breeds -------------------------------------------------- --------------------- --------------- ------------------ ----------------------------------------------------------------------------- **Chr** **Position** **Candidate** **GO accession** **Description** 2 77240001-77340000 SBNO2 GO:0050727 regulation of inflammatory response 3 40350001-40490000 TMEM204 GO:0030947 regulation of vascular endothelial growth factor receptor signaling pathway 3 41450001-41630000 POLR3K GO:0051607 defense response to virus 3 59060001-59160000 GNLY GO:0050832 defense response to fungus 3 76690001-76790000 ACTR2 GO:0038096 Fc-gamma receptor signaling pathway involved in phagocytosis 3 126840001-126950000 ADAM17 GO:0032722 positive regulation of chemokine production 4 83570001-83710000 CD247 GO:0050852 T cell receptor signaling pathway 5 21260001-21380000 RAB5B GO:0030100 regulation of endocytosis 6 29430001-29590000 GNAO1 GO:0007212 dopamine receptor signaling pathway 6 37750001-38030000 MYLK3 GO:0002528 regulation of vascular permeability involved in acute inflammatory response 6 40110001-40240000 URI1 GO:0009615 response to virus 6 51600001-51800000 EXOC3L2 GO:0006887 exocytosis 6 53320001-53470000 EHD2 GO:0006897 endocytosis 6 72100001-72240000 TNFRSF8 GO:0042108 positive regulation of cytokine biosynthetic process 6 79380001-79480000 ECE1 GO:0001921 positive regulation of receptor recycling 6 81890001-81990000 IFNLR1 GO:0050691 regulation of defense response to virus by host 9 33220001-33340000 MMP7 GO:0050830 defense response to Gram-positive bacterium 13 23860001-23970000 CX3CR1 GO:0070098 chemokine-mediated signaling pathway 14 31430001-31660000 P2RX7 GO:0006954 inflammatory response 14 108030001-108220000 PIK3AP1 GO:0050727 regulation of inflammatory response 18 24840001-24960000 CADPS2 GO:0045921 positive regulation of exocytosis **DSRs between LW and three Western pig breeds** **Chr** **Position** **Candidate** **GO accession** **Description** 3 80750001-80850000 PEX13 GO:0001561 fatty acid alpha-oxidation 4 61560001-61660000 JPH1 GO:0007517 muscle organ development 4 63870001-63980000 EYA1 GO:0014706 striated muscle tissue development 5 17230001-17410000 NR4A1 GO:0035914 skeletal muscle cell differentiation 7 75050001-75180000 FITM1 GO:0034389 lipid particle organization 8 43640001-43740000 MSMO1 GO:0006633 fatty acid biosynthetic process 11 68810001-68940000 ZIC5 GO:0030154 cell differentiation 13 108990001-109100000 SKIL GO:0007519 skeletal muscle tissue development 13 197680001-197790000 SLC5A3 GO:0015798 myo-inositol transport 14 121360001-121490000 ADRA2A GO:0050995 negative regulation of lipid catabolic process 14 131210001-131320000 FGFR2 GO:0051150 regulation of smooth muscle cell differentiation Genome-wide scans of LW and the three commercial breeds are shown in [**Figure 4B**](#f4){ref-type="fig"}. A total of 157 DSRs were discovered in these groups. A further 707 unique known genes were discovered, such as *NR4A1, MSMO1*, and so on ([**Figure 5B**](#f5){ref-type="fig"}). Gene enrichment analyses revealed that DSRs in LW and the three foreign breeds were involved in the regulation of smooth muscle cell differentiation, muscle organ development, lipid particle organization, and negative regulation of lipid catabolic processes ([**Table 3**](#T3){ref-type="table"}). Discussion {#s4} ========== In this research, we used SLAF-seq to explore population structure and genetic diversity in Shandong local pig breeds and modern commercial pig breeds. SLAF-seq uses reduced representation library sequencing for *de novo* SNP and precise genotyping ([@B17]). Compared to traditional mark development methods, SLAF-seq has several significant advantages: there is no need to reference the genomic sequence and polymorphic information; it is highly efficient; it is inexpensive; and it saves time. In addition, research methods, such as mitochondrial genome sequencing, focus on a limited range of molecular markers of the genome. SLAF-seq, in contrast, has better resolution and provides more genetic information based on high-density and uniform coverage of the entire gene. Genomic data have provided new insights into plant domestication ([@B15]; [@B11]; [@B45]). In recent years, genomic data have been used to screen positive selection signals in dogs and chickens ([@B2]; [@B37]). However, to date, there has been little research on genetic relationships between local pig breeds in Shandong province and outside breeds involving large-scale analyses of genomic data. In this study, we investigated seven pig breeds collected from all growing regions of Shandong province, ensuring adequate representativeness of the sample. With an average sequencing depth of 16.93-fold for all individuals, 8,615,537 SNPs were identified from 10 pig populations by SLAF-seq. Previous studies have shown that the accuracy of evolutionary analysis can be guaranteed when the sequencing depth reaches 5-fold ([@B19]). This result is several times the number of SNPs obtained from Illumina PorcineSNP60 BeadChip genotyping. To more fully understand genetic diversity and population structure in Taihu indigenous pigs, researchers identified a total of 105,550 SNPs with MAF ≧ 0.05 using GGRS ([@B36]). By comparison, 314,243 filtered SNPs were available for analysis in our study. Therefore, SLAF-seq can produce more information on genomic variation. The YY, LL, and DD pig breeds originated in the United Kingdom, Denmark, and the United States, respectively. This indicates that the geographical distance of indigenous pig breeds in Shandong province is generally greater than that of Western breeds. As anticipated, both population structure and cluster analyses demonstrated that the Western breeds were genetically distant from the Shandong indigenous breeds ([**Figure 2B**](#f2){ref-type="fig"}). Furthermore, pairwise Fst estimation showed that DD was more genetically distant than DPL and LW (Fst = 0.4386 for DPL, Fst = 0.5134 for LW), which indicates tremendous genetic differentiation between DD and other local pig breeds. Our sequencing results are consistent with data from other genetic markers, such as mitochondrial DNA and microsatellite markers ([@B43]; [@B44]), which have also revealed that European and Chinese indigenous pig breeds are genetically distant. LW and DPL were grouped in a separate cluster in the neighbor-joining trees ([**Figure 2A**](#f2){ref-type="fig"}) with a close genetic distance of 0.2337, which is consistent with the results of previous mitochondrial DNA sequencing ([@B34]; [@B35]). Firstly, the main distribution areas of Laiwu and Dapulian are contiguous. Although the establishment of conservation farms has been relatively perfect, there is still a small cross between them. Secondly, there is frequent gene flow among them. As we know that in some conservation farms of them, phenotypically alike indigenous pigs of other breeds are sometimes introduced to reduce the inbreeding of their population. These results show that Shandong indigenous pig breeds have great genetic variety. We speculate that the Western pig breeds are subject to strong human selection pressure. It also may be due to the fact that there were more indigenous pigs than commercial pigs. To summarize, this study supplies a theoretical basis for elucidating patterns in genetic diversity and population structure in local pig breeds and Western pig breeds. When we examined the migration patterns of different pig breeds using TreeMix ([@B25]), we discovered different degrees of genetic contribution between indigenous breeds in Shandong and Western breeds. The analyses showed obvious gene flow from YTW to YY and from YTL to LL. Yantai is located in the northeastern Shandong Peninsula, where the economy and culture are relatively advanced and Western pig breeds have long been introduced. In contrast, Laiwu and Dapulian have been relatively less influenced by Western pig breeds. The two indigenous breeds are distributed mainly in the middle of Shandong province, where traffic and economic development are lagging. Si et al. recorded that LW were influenced by YY in 1950, but beyond that, DPL was hybridized with Middle Yorkshire in the 1950s and 1960s. In other words, LW and DPL are more protected and less influenced by the Western pig breeds. Shandong indigenous pig breeds are a good model for identifying meaningful signatures of selection on genes reflecting economic traits in pig at the genome level. DPL is distributed mainly in Jining, Shandong province, China. Historically, DPL was well known for properties related to resistance and immunity. This study used the thresholds of Fst and π to identify DSRs. Fst is more convincing for identifying complicated events, such as detection of significant variation ([@B14]). A large number of key genes potentially related to immunity, such as *SBNO2, IFNLR1, CX3CR1, ADAM17, URI1*, and *PIK3AP1*, were found in the DSR data sets. Previous results indicated that *SBNO2* participates in anti-inflammatory response by inhibiting the NF-kB pathway ([@B7]). However, little is known about the function of *SBNO2* in porcine resistance to disease. Our results demonstrate that *SBNO2* may play an important role in porcine resistance to inflammation, but this hypothesis requires further investigation and validation. *IFNLR* belongs to the type II cytokine receptor family. It can bind with type III interferons (IFN-λs) to participate in signal transduction and induce antiviral and anti-tumor functions ([@B38]). Recently, more and more research has indicated that *IFNLR1* is also widely expressed in immune cells. In viral infection or acute inflammation, IFN-λ can directly or indirectly act on immune cells to play a role in immunoregulation ([@B1]; [@B16]). Therefore, our data might provide insight into the role of candidate genes in porcine immunity. LW is representative of a typical Shandong black pig. As it has fresh meat and a high proportion of intramuscular fat ([@B6]), it serves as a perfect model for tracking selection for fat deposition in muscle. We therefore investigated DSRs under selection between LW and the Western pig breeds. Gene annotation of the selected regions revealed a series of functional genes for fat deposition and muscle development. It is interesting to note that we discovered some candidate genes (i.e., *EYA1, NR4A1, MSMO1, FITM1*, and *JPH1*). *EYA1*, a homolog of Drosophila eyes absent, is expressed in the mesenchyme ([@B42]). *EYA1* may play a key role in the genetic hierarchy of kidney organogenesis ([@B29]). Moreover, the synergistic effect of *SIX1* and *EYA1* can promote transformation of slow-twitch to fast-twitch phenotype in the muscles of adult mice ([@B9]). As skeletal muscle fibers affect the tenderness of pork, *EYA1* may be a special gene regulating the quality of pork. Our data also indicate that *NR4A1* might contribute to regulating fat metabolism. It has been reported that *NR4A1* affects glucose and lipid metabolism in muscle ([@B20]; [@B4]). At present, most studies indicate that *NR4A1* has an inhibitory effect on fatty buildup. However, the specific and complete regulatory mechanisms remain to be verified. In a word, our data provide clues for identifying genetic differences between LW and Western pig breeds. LC, YTL, and YTW have a very close genetic relationship. At the genome level, the genetic difference between LC and Western pig breeds is still largely unknown. Results of KEGG pathway analyses have shown that most genes are involved in the chemokine signaling pathway, the Ras signaling pathway, and tyrosine metabolism. Some key genes (i.e., *SLC9A1* and *TIAM1*) in the regulation of actin cytoskeleton, and a novel gene designated 2-acylalycerol O-acyltransferase 2-A in fat digestion and absorption, were discovered in pathway analyses. Other studies have shown that the DH domain of *TIAM1* plays an important role in cell deformation and cytoskeletal reorganization ([@B10]; [@B23]). Our results might provide new insights into their function in affecting pig growth and development. To date there is still a lack of conservation of WL, which continues to be crossbred with Western pig breeds. Nonetheless, our findings indicate that many positive selection genes are involved in the MAPK signaling pathway in WL, such as *MAP3K12, RPS6KA1, RAPGEF2*, and *MKNK2*. MAPK is very important for mediating inflammation and cytokine production. In addition, through sequencing, we verified that HG has its own unique genetic characteristics and is an independent indigenous pig population. Similarly, we found that *KDR* is involved in the PI3K-AKT pathway in HG. Our data suggest that *MAP3K12* and *KDR* might regulate the porcine inflammatory response, but additional studies are needed to confirm this. In sum, the genetic structure and relationships between Shandong indigenous pig breeds and three Western breeds were revealed by SLAF-seq. Our results provide additional information explaining differences in economic characteristics between the breeds. Our research therefore not only provides a cost-effective strategy for pig genome-wide screening but also lays a genetic foundation for further research on the candidate gene functions of DSRs. Data Availability Statement {#s5} =========================== The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation, to any qualified researcher. Ethics Statement {#s6} ================ All animal care and treatment procedures were approved by the Animal Ethics Committee of Shandong Agricultural University, China. All methods were carried out in accordance with the guidelines and regulations of the Animal Ethics Committee of Shandong Agricultural University (Approval number: 2004006). Author Contributions {#s7} ==================== WC and YZ conceived and designed the research. WC, MQ, CL, and ZL performed the research. MQ, WC, and CL analyzed the data. YZ contributed reagents and materials. MQ wrote the manuscript. WC and YZ provided substantial comments and revised the manuscript. All authors read and approved the final version of the manuscript. Funding {#s8} ======= This study was supported financially by the National Natural Science Foundation of China (No. 31401055), the Funds of Shandong "Double Tops" Program (No. SYL2017YSTD12), the Shandong Provincial Modern Pig Technology and Industry System Project (No. SDAIT-08-02), Shandong Provincial Natural Science Foundation (No. ZR2018BC046, ZR2019MC053). Conflict of Interest {#s9} ==================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. We are grateful to Tao Chen and Lixia Ma for their assistance in sample collection and laboratory analyses. Supplementary Material {#s10} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fgene.2019.01351/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: Susana Seixas, University of Porto, Portugal [^2]: Reviewed by: Salvatore Mastrangelo, University of Palermo, Italy; Qishan Wang, Shanghai Jiao Tong University, China [^3]: This article was submitted to Evolutionary and Population Genetics, a section of the journal Frontiers in Genetics

In the previous SMR Editorial we highlighted in the broadest sense the widening gap between pharmacy practice activities in the Global North and South \[[@ref1]\]. In this Editorial we take the opportunity to think about the ethnic-minorities residing in high-income countries. In many ways these groups are similar to populations in low-income countries. The struggles that challenge health systems in developed countries with respect to innovative high-cost medicines are some of the challenges the Global South face when accessing essential medicines. Asthma is one such example in which access to essential medicines may not be a big problem in high-income countries but that compliance and optimal use may be similar to low-income countries, for some cohorts of the population. When comparing North and South, one might expect all to be well in the developed world with regard medicine access and use. This may not be the case, and populations such as Australian Aboriginals, Canadian Inuit, American Native Indians and many other ethnicities come to mind. Taking New Zealand as an example may help to extrapolate to the indigenous peoples previously outlined. Several cohort populations come to mind when thinking about potential disparities in access and health outcome. Firstly, the indigenous peoples of the land ¨C the original owners and occupiers who settled from the Pacific in a pre-European era. In New Zealand these folk are the Maori people. For a number of complex and inter-related reasons, Maori die on average 10 years younger than non-Maori; even within a developed health care system \[[@ref2]\]. In general their presentation and access to health care services is at a lower rate than the Caucasian population. Availability of essential medicines has been addressed through the policies of the Pharmaceutical Management Agency of New Zealand (PHARMAC). However, Maori are over represented in the lower socio-economic cohort of the New Zealand population and so issues of medicine affordability and health literacy arise \[[@ref3], [@ref4]\]. Robust evaluation of affordability is yet to be undertaken across various lower socio-economic populations and/or high need ethnic groups. Another example is the people of the pacific; those folk who migrated from Polynesian islands in the 1950s and continue to do so today. They are represented through a very large population grouping in the south of Auckland; New Zealand¡¯s largest city. There are now more peoples of pacific island origin in Auckland than in the pacific islands per se. The challenges with pacific peoples are not dissimilar to Maori in that uptake of some health services is less than optimal and that much of the population falls into the high need category in terms of socio-economic and health care \[[@ref4]\]. The other important sub-group are recent immigrants from countries in greater Asia, Middle East and Africa, to name a few. These populations have significant risk factors for the development of chronic diseases and this is reflected in the high prevalence of cardio-vascular disease and diabetes to name two. Aside from the medical risk there are misconceptions about the system and rules due to culture and language barriers and as a result the system relating to medicines is confusing. There are different perceptions with respect to quality of available medicines as well as variability in community pharmacy service provision to these communities. Financial barriers including lack of affordability, particularly for over the counter (OTC) therapies have been reported \[[@ref5]\]. All of these factors add to a complex picture of medicines use which may potentially impact on health outcomes for these sub-populations. A New Zealand-based study conducted by Bassett-Clarke et al (2012) also found ethno-cultural barriers with regards to medicines use \[[@ref6]\]. They found that the conceptualisation of medicine, self medication and the seeking of further information varied amongst NZ European, Pacific peoples and Maori and folk from Asia and South Asia. We have taken New Zealand as our exemplar of some of the issues faced by ethnic minorities accessing and using pharmaceuticals. There are however, many ¡°like¡± examples including Hispanic and African races and the Hawaiian people of the United States (USA) \[[@ref7], [@ref8]\], Canadian Inuit \[[@ref9]\], Australian aboriginal\[[@ref10]\] and Pakistanis in Scandinavia \[[@ref11]\], to name a few. The European Union has created a huge melting pot of races and movement from Eastern European countries to the continent per se and there is huge potential for research in this area. This is not an exhaustive list but with common examples from high-income countries, it is fair to suggest that for some populations, the gap between Global North and South may be less than we give credit for. In-order to complement work in the Global South, we encourage authors to forward papers related to medicines policy and use in ethnic subpopulations and migrant groupings within high income-countries. It is expected that by thinking about ethnic minorities and placing ¡°culture¡± at the center of social pharmacy practice and research, local and international disparities in access to, and use of medicines, as well as pharmacy service provision will be better addressed \[[@ref12]\].

{#sp1 .vi}

{.673} {.674} {.675} {.676} {.677} {.678} {.679} {.680} {.681} {.682}

Background ========== Early detection of progressive glaucomatous optic disc damage is essential in the management of patients with glaucoma. Accurate assessment of the optic disc can reveal early structural changes that precede field changes. In spite of recent developments in imaging techniques of the optic disc -- such as the scanning laser tomography and optic coherence tomography-, digital stereoscopy of the optic disc is still considered the gold standard against which other technologies are evaluated \[[@B1]\]. The use of the stereo photographic images of the optic disc is common in glaucoma clinics in the United Kingdom. The images are usually displayed simultaneously on high-resolution computer screens and viewed using a hand held stereo viewer or liquid-crystal shutter goggles. These techniques have their limitations as the viewer has to wear- or be close to -- some device to separate the left and right views, together with limited head freedom, and dimness of the displayed stereo image. Autostereoscopic displays represent a relatively new technology that do not require the observer to wear any device to separate the left and right views and instead present them directly to the correct eye offering potential benefits in ease of use in clinic settings This study was designed to assess whether the stereoviewing performance of an autostereoscopic screen would provide equivalent clinical and diagnostic accuracy as compared to using liquid crystal shutter goggles working in synchronisation with a view switching display, when viewing stereoscopic optic disc images Methods ======= Sixty optic disc stereo-images were randomly selected from the database of patients who had attended the glaucoma clinic at Sunderland Eye Infirmary- patients with glaucoma, suspect glaucoma and normals were included. The images were 512 × 512 digital monochromatic sequential stereo-photographic optic disc images captured by the DISCAM optic disc camera (Marcher Enterprises Ltd, Hereford) following pharmacological mydriasis as part of the patients\' routine care. As the main aim of this study was to compare the quality of stereo viewing using the two modalities, poor quality images due to poor illumination or images with large vertical shift on the screen between the stereo pairs, were excluded. Ethical committee approval was obtained. The images were displayed in two ways, using a custom designed program written in Borland C++ Builder using the Open GL graphics libraries. The first modality used a shuttering goggle system with a Dell Ultra-scan P1110 21 inch monitor, an Oxygen GVX1 video card and a pair of Stereo-graphics Crystal Eye CE-3 polarized liquid crystal shutter goggles \[[@B2]\]. The Oxygen GVX1 card, in conjunction with Open GL, was able to display sequential stereos at half the maximum refresh rate, by interleaving presentations of the left and right images and synchronously controlling the stereo goggles through a mounted infrared emitter. Given that the P1110 monitor is capable of refresh rates up to 120 Hz, but at 100 Hz with the screen resolution used, we achieved a 50 Hz refresh in stereo mode. The result was satisfactory from the point of view of flicker, and avoided the 50% reduction in vertical resolution suffered by some other shuttering goggle stereo set ups. However, it did still suffer from other problems inherent in the shuttering goggle approach -- in particular; variable cross talk between left and right images peaking at 15% \[[@B3]\]; and a 68% reduction in brightness caused by the shuttering effect \[[@B2]\]. Secondly, the stereo-images were displayed on an autostereoscopic screen; a Dimension Technology Inc. 2015 XLS virtual window 15-inch 2D/3D screen \[[@B4]\]. The auto-stereo screen is a flat-panel LCD with a rear illumination optical system that when switched to 3D mode allows a stereoscopic pair to be presented to the viewer at a specific position without requiring the viewer to wear any special glasses. When the observer views the screen at the display\'s optimum viewing position, \"sweet spot\", they see half the pixels from the display in the left eye and half the pixels in the right eye \[[@B5]\]. The 2015 XLS display was driven by placing the stereoscopic image pair side- by side in a single image; this required each image to be shrunk by 50% horizontally. Electronics in the display then interleaved the two images in real time to achieve the required alternate column, interleaving pattern for the 2015 XLS 3D mode. The interleaved images were then visible as completely separate left and right images to the viewer at the display\'s sweet spot. The display does have an unpublished fixed level of cross talk, although this is at a lower level than the peak of the Crystal-Eyes display. The driver video rate for the 2015 XLS is 60 Hz and although the LC material in the display will not respond as fast as this, for this application using still images, this was not relevant. As a consequence unlike the shutter goggles there was no flicker problem with the 2015 XLS. (Figure [1](#F1){ref-type="fig"}) {#F1} The stereoscopic and auto-stereoscopic displays clearly have differing physical and optical characteristics and our evaluation sought to determine if the differences between the two modalities made qualitative or quantitative differences to performance in assessing various characteristics and different measurements in stereoscopic optic disc images. This was assessed in two ways; Firstly, quantitative optic disc parameters were analyzed in both modalities using a mouse-controlled cross-shaped cursor and custom designed software. The program allowed the user to first adjust the vertical and horizontal offsets between the two images to achieve a good stereo effect and then a three dimensional cursor was used to mark up the images. The cursor was displayed using the standard \"exclusive-OR\" approach, where the intensity of grey-scale pixels in the cursor was set to the value 255-g, where g is the pixel value in the covered pixel. Generally speaking this was effective, although we did occasionally observe that, where the left and right image\' intensities varied significantly (because of lighting variations), significant differences in the cursor colour could evoke a disassociation, so that two separate crosses were perceived. The cursors depth could be adjusted (changing the disparity between left and right display of the cursor): after adjustment the cursor appeared confined within a chosen plane. This gave a good, consistent stereo effect provided that the cursor was kept in a position where it was correctly perceived to be at the retinal surface. In each setting the cursor depth was first adjusted to the plane of the outer edge of the optic disc (optic disc rim) whilst stereo viewing. The optic disc edge was then traced with the cursor, clicking to place a contour, which was displayed in the cursor plane. The mouse controlled cursor\'s depth was then adjusted to the plane at which the disc vessels first deviated posteriorly, and the edge of the optic cup was then traced at that plane, placing a second contour. After familiarization with these assessment techniques using an additional training set of 10 images, the test images were presented randomly to two observers. The two observers assessed the selected images independently on two separate occasions two months apart (at each time point half the images were viewed using the autostereoscopic system and half using the goggles system to avoid bias) In most cases, the observers needed to adjust the vertical and horizontal offset of the stereo pairs to ensure optimum quality of stereo viewing. The observers were masked to their previous/and other observer\'s assessments. Once saved, the computer system then counted the exact pixel number included within each plotted disc and cup areas. The overall cup/disc area ratio was determined as the ratio between the two plotted areas. The vertical cup/disc ratio was determined by calculating the ratio between the height of the disc and cup plotted areas in the vertical meridian. The areas were evaluated for agreement and overlapping between the two viewing modalities. Secondly, a third independent assessor graded the optic disc appearances on a 5-point scale for the probability of glaucoma as described by Greaney et al \[[@B6]\]. Each optic disc was thus graded as 1(definitely normal), 2(probably normal), 3(undecided), 4(probably glaucoma), and 5(definitely glaucoma). Criteria for allocation to these categories was based on the presence of neuroretinal rim thinning, notching, undermining of optic disc cup edge, nerve fibre layer defects and optic disc haemorrhages \[[@B7]\]. Each disc was also graded in a dichotomous manner as glaucomatous or non-glaucomatous. The overall optic disc classification grading together with the quantitative disc parameters\' scores were then analyzed to assess inter-modality variability. Statistical Analysis -------------------- Basic statistical analysis was undertaken using SPSS Version 14.0. Observer agreement was assessed via StatXact Version 6.0, whilst S-Plus Version 6.2 professional was used for graphing. Distributions were confirmed as normal via the Shapiro-Wilk test prior to basic parameters (means and standard deviations) being calculated. Agreement was assessed via the weighted kappa test (with 95% confidence intervals) and via Bland Altman plots. Results ======= A total of 60 optic disc stereo-images were included (38 Right and 22 Left). There were 36 males and 24 females. Their age ranged from 45 to 72 years. (Mean = 62 years). Patients were diagnosed with glaucoma (29), ocular hypertension (11), or glaucoma suspect (20). The inter-modality agreements for observers were assessed using the weighted Kappa coefficient. Landis and Koch \[[@B8]\] offer the following guidelines for interpretation of the weighted kappa coefficient: \<0.20 = poor agreement, 0.21 -- 0.40 = fair agreement, 0.41 -- 0.60 = moderate agreement, 0.61 -- 0.80 = good agreement, 0.81 -- 1.0 -- very good agreement. The estimates for the overall cup/disc ratio (CDR), and vertical cup/disc ratio (vCDR) were compared. The inter-modality agreement was good for each observer; the average kappa coefficient was 0.78 for observer 1 (p-value \<0.0001), and 0.81 for observer 2 (p-value \<0.00001). (Table [1](#T1){ref-type="table"}) ###### Inter-modality agreement for cup-disc ratios Observer 1 (Autostereoscopic screen versus Shutter Goggles) Observer 2 (Autostereoscopic Screen versus Shutter Goggles) ------------------------------------------------------------- ------------------------------------------------------------- ------------------------- --------------------------------- ------ ---------------- ------------------------- --------------------------------- Weighted Kappa 95% Confidence Interval Pearson Correlation Coefficient Weighted Kappa 95% Confidence Interval Pearson Correlation Coefficient CDR 0.78 0.65 -- 0.91 0.84 CDR 0.83 0.73 -- 0.92 0.88 vCDR 0.82 0.73 -- 0.91 0.86 vCDR 0.86 0.78 -- 0.94 0.88 CDR = overall cup/disc ratio. vCDR = Vertical cup/disc ratio Using Bland Altman Plots the inter-modality agreement fell within two standard deviations for both observers in more than 95% of cases. Importantly there was no evidence of fixed or proportional bias. (Figure [2](#F2){ref-type="fig"}) {#F2} The plotted cup and disc areas were evaluated for agreement and overlapping between viewing modalities. Observer 1 scored an agreement matching percentage of 97% and 91% for the optic disc and cup areas respectively, while Observer 2 scored an inter-modality agreement of 94% and 89% for the same areas (Table [2](#T2){ref-type="table"}) ###### Inter-modality agreement for area measurements expressed as Pearson correlation coefficient (p value) Observer 1 (Autostereoscopic screen versus Shutter Goggles) Observer 2 (Autostereoscopic screen versus Shutter Goggles) ----------- ------------------------------------------------------------- ------------------------------------------------------------- Disc Area 0.97 (p \< 0.001) 0.71 (p \< 0.001) Cup Area 0.92 (p \< 0.001) 0.89 (p \< 0.001) Using Bland Altman Plots, in more than 95% of cases the inter-modality agreement fell within two standards. Again there was no evidence of fixed or proportional bias. The inter-modality kappa coefficient for the dichotomous grading of glaucoma/non glaucoma was perfect at 1.0. The weighted Kappa coefficient for the five-point grading was very good at 0.97 (95% CI 0.95 to 0.99) with 51 of the 60 cases having exact agreement, i.e. there was exact agreement 85% of the time (95% CI 73% to 93%). None of the nine cases where the five point scale differed had a greater than one point difference. In five patients the grade was higher when using the autostereoscopic screen and in four it was higher using the goggles system suggesting no systematic bias. Discussion ========== Careful and detailed assessment of the optic nerve head is still considered to be the most important sign for diagnosis of glaucomatous progression \[[@B9]\]. In early glaucoma, up to 20--50% of the optic nerve can be lost before any reproducible visual field defect is identified \[[@B10]\] and retinal nerve fiber layer RNFL defects can be only detectable after a 50% loss of neural tissue in a given area \[[@B11]\]. Despite the recent advances of computerized optic disc analysis, fundal biomicroscopic examination or stereo viewing of stereoscopic photographic images remains a gold standard for evaluation and monitoring of the optic disc. Various studies in the past attempted to compare stereoscopic and non- stereoscopic cup/disc C/D assessments. Results have been equivocal; some studies have shown the stereoscopic C/D ratio measurements to be larger \[[@B12]-[@B14]\], equal \[[@B15]\] or smaller than non-stereoscopic images \[[@B1]\]. However recently, Morgan et al reported lower estimates of neuroretinal rim width and higher levels of inter-observer agreement with stereoscopic assessments as compared to monoscopic assessments \[[@B16]\] reinforcing the value of stereoscopic assessment In this study our aim has been to compare one example of the newly emerging auto-stereoscopic 3D display technologies, DTI 2015 XLS, with an existing stereoscopic display, CrystalEyes CE-3 shutter goggles. Auto-stereoscopic displays offer the potential for \"easier\" clinical use, without requiring the user to wear goggles. When assessing any new technique which offers improved clinical convenience it is important to ensure that clinical and diagnostic accuracy is not impaired. Hence we designed the study to ascertain whether the detection of glaucomatous optic disc characteristics, and assessment of quantitative optic disc parameters was equivalent between autostereoscopic screen viewing and a commonly used existing method of stereo viewing. This study was an initial step in the assessment of this new technology. We did not attempt to ascertain the accuracy of auto-stereoscopic displays in diagnosing glaucoma and hence did not perform a sensitivity and specificity analysis on this diagnosis, which relies on other investigations including visual fields. The three observers, who participated in the study, noted a distinct and definite improvement in the subjective perception of the quality of stereopsis in appreciating the optic cup depth together with an increase in the clarity and contrast of the optic disc rim, and the disc vessels in the viewed stereo images when using the auto stereoscopic display as compared to the shutter goggle modality. The reduced brightness, residual flicker and high peak cross-talk levels, as well as the physical \'discomfort\' associated with the shutter goggles are all possible reasons. We achieved a 50 Hz refresh in stereo viewing mode for the shutter goggle system. The result was satisfactory from the point of view of flicker, and avoided the 50% reduction in vertical resolution suffered by some other shuttering goggle stereo set-ups but did have a 68% reduction in brightness caused by the shuttering effect. Unlike the shutter goggles there was no flicker problem with the auto stereoscopic screen set up together with relatively less reduction in brightness. We acknowledge that the study had a potential bias, as we were unable to mask the observers to the stereo viewing modality used -- unfortunately any masking would have affected the differences in the observed stereoscopic effect. Previous studies have assumed that different methods of stereo viewing are equivalent and this is the first study that we are aware of in the ophthalmic literature to compare viewing methods. Analysis of our results demonstrated a good level of inter- modality agreement of cup disc ratio measurements with an average weighted kappa coefficient of 0.8 for the C/D ratios measurements. There was perfect agreement for categorization of the disc to glaucomatous or non-glaucomatous between the two modalities and very good agreement on a five point glaucomatous grading scale. This suggests that viewing stereoscopic optic disc images with the autostereoscopic screen used do not alter the ability the ability of the observer to assess changes in the optic disc as compared to the shutter goggle system used. A number of other general issues arose during the study concerning the use of any stereoscopic viewing technique for this task, namely cross talk, the marking of feature locations, such as the edge of the optic cup, and image alignment. We believe these deserve further investigation before clinical use of digital stereoscopic imaging can be fully optimized. In the shutter goggles system used, inter-channel stereoscopic cross-talk peaks at 15%. The auto stereoscopic display used does have an unpublished fixed level of cross talk and although this is at a level lower than the peak of the Crystal- Eyes display both display modalities have significant cross-talk. This could easily lead to errors in judgment of region boundaries, particularly where the edge is initially of low contrast. It may be that recently available zero cross talk stereoscopic or auto-stereoscopic systems will prove more suited to clinical tasks. These should certainly be used in any comparisons with Gold standard results. Other studies have reported differences in marking between monoscopic and stereoscopic modes. We believe this may be partly due to imaging issues such as cross talk. For example this could wash out an edge and result in consistent under-estimation of feature size. Alternatively it could extend an edge and result in consistent over-estimation of feature size. This may explain conflicting results in previous studies. Again studies are needed using a modern zero cross- talk display system to investigate this and several other more detailed aspects that arise from mismatches between the geometry of the picking cursor and the camera used to capture the original images It became clear during the study that the alignment of images from the camera was often poor vertically and horizontally, to the extent that images could be uncomfortable to fuse without manual adjustment. This will significantly affect task performance, particularly in comparative imaging of one patient over time. Auto image alignment is now becoming feasible using software but it may be that this issue is best resolved by manufacturers improving camera design. Conclusion ========== Auto-stereoscopic display technology is rapidly evolving and improving and we anticipate this will lead to additional quantitative benefits using the latest high contrast, low cross-talk, displays. Our results show that viewing of stereoscopic optic disc images using an autostereoscopic screen provides comparable diagnostic and clinical assessment to liquid crystal shutter goggles. Furthermore we believe that autostereoscopic displays have significant clinical usability benefits over goggle based stereoscopic displays. Competing interests =================== The authors declare that they have no financial or any other non-financial competing interests in any of the products mentioned in this manuscript. The authors confirm that this work received no public or organizational funding Authors\' contributions ======================= MSH and DV and DHWS carried out the independent viewing processes with the 2 modalities. MSH participated in the study design and drafted the manuscript. NSH, AH and JAL provided the stereoviewing technology and designed the needed computer software. AH performed the statistical analysis. DHWS, the lead investigator, conceived the study, participated in its design and coordination and helped to draft and revise the manuscript. All authors read and approved the final manuscript Pre-publication history ======================= The pre-publication history for this paper can be accessed here: <http://www.biomedcentral.com/1471-2415/8/13/prepub>

Background {#Sec1} ========== Cardiovascular disorders are today by far the leading global cause of death \[[@CR1]\]. As the prevalence of these diseases increases, so does the need for diagnosis. Echocardiography is the most commonly used imaging modality within clinical cardiology, as the information it provides is extensive and immediate as well as cost-effective, non-invasive and accessible. The disadvantage is that it is user-dependent and requires extensive practice to master \[[@CR2]--[@CR4]\]. The development of medical ultrasound, since its origin 70 years ago, has been extensive. Modalities such as M-Mode, Doppler, 2D as well as 3D are now an integral part of a routine examination \[[@CR5], [@CR6]\]. The ultrasound practitioner needs an understanding of the physics behind these modalities, as well as adequate knowledge of the anatomy, physiology and pathophysiology of the heart. In addition, satisfactory technical competence and practical skill are required in order to perform an examination that is both useful and clinically relevant \[[@CR3]\]. Echocardiography education worldwide is very different depending on the profession subject to the learning process and on the purpose \[[@CR2], [@CR4], [@CR7]\]. However, the initial challenges of handling the transducer and connecting it to the anatomy of the heart, as well as understanding the basics of ultrasound physics are the same. It can be challenging to obtain and assess information from a two-dimensional image in a three-dimensional organ. During training for basic level, when the practitioner is expected to be able to perform a full routine examination followed by writing a report, the European Association of Echocardiography recommends a minimum of six months of active full-time training as well as performance of a minimum of 350 transthoracic echocardiographic examinations \[[@CR4]\], putting into perspective how demanding the examination can be to master. To make use of the time spent training and to design a curriculum fully favouring the students' learning, it would be beneficial to explore what the students perceive as challenging in the echocardiography learning process as well as what they feel is helpful. In addition, such results could serve as a foundation in the development of new teaching material, such as a virtual laboratory (vLAB) within echocardiography. Previous studies on vLABs \[[@CR8], [@CR9]\] have shown to increase students' knowledge, intrinsic motivation and self-efficacy and could therefore serve as a meaningful pedagogical tool when learning a knowingly complex examination like echocardiography. Studying the students' experiences in learning echocardiography in order to use it to develop curriculums or vLABs would be a way of inviting them as partners in the development process. Several studies have found numerous positive effects of using students as partners, such as increased student engagement, motivation and ownership of learning as well as the development of newer or better teaching or curriculum material \[[@CR10]\]. Even when developing new medical equipment, it is not unusual to take the users' perspectives into consideration. According to Shah and Robinson \[[@CR11]\] users (in this case, students) can provide meaningful input in different stages of the development of new medical equipment. There are several studies showing the benefits of using simulations, mannequins or e-learning within echocardiography education \[[@CR12]--[@CR16]\]. These studies often highlight and comment on the difficulties of learning echocardiography. However, they do not show or study *why* or *what* the learning difficulties might be. Therefore, the purpose of this study is to gain greater understanding of the experience behind learning echocardiography, focusing on the challenges as well as what may favour learning. Method {#Sec2} ====== A qualitative research design, utilising focus group interviews as data collection method, was employed as a means to explore the students' subjective experience of learning echocardiography. Focus group interview was chosen due to its ability to instantly provide information on similarities and differences in perspectives between participants who share experiences, compared to individual interviews where such conclusions must be drawn after the interviews by analysing separate statements from the participants \[[@CR17]\]. Study population and data collection {#Sec3} ------------------------------------ The study was performed with students at the biomedical laboratory programme at Malmö University. The curriculum consists of a three-year bachelor programme, the aim of which is to form a foundation to independently be able to plan, implement and interpret analyses and examinations. For students studying clinical physiology in the final year, echocardiography is one of these examinations. At the end of the course the students are expected to be able to perform a basic ultrasound examination including rudimental assessment of its information. In order to be able to do this they need to be able to explain ultrasound physics and describe ultrasound methodology. The curriculum is based on the principle of constructive alignment \[[@CR18]\] and the teaching is both theoretical and practical. The practical teaching is performed by teachers with several years of clinical experience within echocardiography. The students are given an introduction, beside the ultrasound machine, on how to work the machine and how to handle the transducer in order to obtain the various projections. They are also given a document that specifies the projections and methods to be used in a standardised examination. The students are then given a certain amount of training time at the machine, with a teacher available when questions and difficulties arise. The presence of teachers is greater at the beginning, when introducing the method and they are thereafter available when called on. In addition, students have free access to a training laboratory where they can practise without the teacher supervision, to allow increased student independence. The practical part is assessed by a practical examination where the students need to show that they understand how to handle the machine and the transducer, as well as which projections and methods are relevant when wanting to obtain certain information. The theoretical part is assessed by a computer-based exam in which they are given a case (including echocardiography loops in motion) where they are to assess the diagnostic information as well as answer questions concerning the related ultrasound physics. We used convenience sampling inviting all year three students enrolled in the clinical physiology course in October 2017 (*n* = 15). Out of these students, eight voluntarily accepted to participate in the study. None of the students had previous experience of ultrasound prior to the course. The study was performed after the course had been completed and the students had been informed of their grades. They were divided into two groups (4 + 4). The author who also was one of the course teachers moderated both interviews and an observer sat in on the interview. The observer, a teacher who had not participated in the course, did not participate actively in the interview but took notes on what was discussed during the interviews. The students were encouraged to share experiences, thoughts and memories connected to their learning of echocardiography. The interviews started with one open or broad question "To start off I would like every one of you to briefly share your experiences of learning echocardiography" which started a discussion among the students. Supporting questions such as "What did you mean by..." or "Could you please explain a bit more..." were then asked only when the discussion was led into areas needing clarification. In order to further stimulate the discussion, remove focus from the moderator and instead direct focus onto the topic various stimulus material related to the topic discussed can be helpful \[[@CR19]\]. In this study the handouts from the teaching were available during the interviews, key echocardiographic words were written on the boards and an echocardiography apparatus was stationed in the room. Initially everyone was invited to briefly and very openly share their experiences from their learning. They were encouraged to let the discussion circulate around difficulties in learning echocardiography as well as what was helpful, or ideas of what could have helped their learning. The students were encouraged to write key words from their discussions on the board. The interviews took place at the university, lasted 64 and 74 min respectively and were performed using audio recording. Analysis {#Sec4} -------- As a starting point for the analysis the interviews were transcribed by the first author, followed by independent reading by the three authors (AD, EC, PG). The text was reread several times to obtain a sense of the whole. Two of the authors were familiar with echocardiography, representing an internal perspective and one of the authors was not, representing an external perspective. Based on the transcribed text, the notes of the observer and what the students had written on the whiteboard, a manifest content analysis was conducted. The inductive approach was performed with a focus on the challenges of learning echocardiography as well as what favoured learning. Meaning units where identified and thereafter sorted and coded. The codes where then grouped into categories, reflecting the central message in the texts \[[@CR20]\]. To ensure credibility the preliminary analysis was returned to the students for respondent validation. Ethical considerations {#Sec5} ---------------------- In accordance with the Declaration of Helsinki \[[@CR21]\] and the Swedish Research Council's requirements regarding information and consent, informed consent was obtained prior to the interviews. All participants were given oral and written information concerning the study purpose, assurance of confidentiality and the right to withdraw from the study at any time. The students were assured that participation in the study would not compromise their professional relationship to the teacher, nor affect their final grade. However, all students had passed the course and were informed of this prior to the interviews. Results {#Sec6} ======= Our findings illustrate students' experiences of learning echocardiography with a focus on challenges in learning and what was expressed as support as well as their ideas of what could have further aided their learning. The results are a compilation of thoughts and experiences expressed within the two focus groups and are illustrated by two categories -- *practical skills* and *bridging the theory-practice gap*, presented below. Subcategories were identified within each category and findings are illustrated by quotes from the students (S1--8) from the two groups (F1--2) respectively. After taking part of the initial analysis and suggestions for development, all students considered it to be representative of what they had expressed during the interviews. Practical skills {#Sec7} ---------------- Initially, the students thought it was difficult to learn the projections. From a theoretical point of view, the projections were hard to link and relate to where in the heart the ultrasound beam cut. From a practical point of view, it was hard to know where the transducer should be placed, angled and turned in order to obtain certain projections. ### Immediate feedback from the teachers {#Sec8} The students wished for more immediate response at the beginning while learning the projections in order to know whether or not they were doing it correctly. A recurring suggestion was for the teachers to be present at the start of the students' practical training. Another suggestion was for an independent assessor to be present, correcting the students and asking them questions as they were practising. One of the students mentioned it would have helped if the ultrasound machine had some kind of indicator, showing them whether or not the transducer was placed correctly. "*F1S3 "Yes, exactly. That's what I thought too. The teachers could have been more present, just so they.... So they could say "this is what you should do, this is how it should be". But apart from that I thought it was a good course. Actually. And that we had a lot of opportunity to practise as much as we wanted to."*" "*F2S2: "When one was wondering "Where am I? What chamber am I looking at? Where is the regurgitation?"\... and then... I don't know... it had... I think it would have been good if you at the same time were given an indication on\... WHICH diseases can I find here, connected to THIS valve?"*" ### Immediate response from fellow students {#Sec9} The students highlighted the advantage of working together and helping one another at the ultrasound machine. They saw great advantages in being three at the machine at the same time -- one operator, one patient and one standing at the side -- asking questions and comparing the projections on the display with the projections in the literature. They appreciated the fact that there was someone continuously asking questions and encouraging them to reflect on what they were doing and why. "*F1S2: "Yes, exactly. And then I realised quite early on that you have to be at least three. In echocardiography, to get anything out of it. That there is one who asks the one performing the echo... when you see this projection, what do you want to measure? And then tells you..."*" ### Access to pedagogical equipment {#Sec10} Something that helped the students learn the projections was their textbook on echocardiography which contains clear drawings of the projections and where the ultrasound beam cuts the heart in order to obtain each projection. The anatomical dummy of the heart helped the students to make a connection between the anatomy of the heart and the placing of the probe and location of the beam, while they were practising by the machine. "*F1S4: And the echocardiography book too... I used it quite a lot, the orange book. And it was good that it was accessible during the lab too. So we could look in it and compare the projections we got with the pictures in the book.*" "*F2S7: Yeah, it was really...it helped me too, just to take the heart like that... how the beam cuts... short axis, long axis... F2S6: Exactly, hold it, and LOOK at it like this... yeah...*" ### Practical experience {#Sec11} The students thought it was helpful that there was a lot of practical training and that the ultrasound machine was so accessible. A recurring word during the interviews was "play"; they felt it was helpful to "play" with the ultrasound machine. It was positive that nothing could go wrong, there was no button that was "off limits". They also wanted the teacher to hold her hand over the student's while the student held the transducer in order to get an even better feel for handling the transducer. "*F2S8: "Mm, and then while practising echo it was like... I wanted to have someone beside me to be sure if I was holding the probe correctly... But then... no, I have to do this by myself to get more experience and to get a feel for it. Sometimes I did it the wrong way.... I placed the transducer wrongly just in order to see -- what projection will I obtain? What will I see? What positions? What will the heart look like? Like that. But... it was easier for me to "play", like you said, to find, and to practise on YOURSELF and it was easier".*" "*F1S1: "Well, it was really good to have the training laboratory where we were able practice. Really good."*" Bridging the theory-practice gap {#Sec12} -------------------------------- The students found it challenging to integrate knowledge from class into practice with the machine. As previously mentioned, they found it hard to both theoretically and practically learn the projections but they also found it challenging to understand how and when certain measurements and concepts should be used in clinical setting. Also, they found it challenging to link the theory of ultrasound physics to practical performance with the machine. ### Alternating between theory and practice {#Sec13} Generally, the students would have liked to alternate between theory and practice. Some students wanted practice prior to theory in order to first know *how* and then to know *why,* while other students wanted more theory before practice in order to better understand what they were doing when holding the transducer and practicing. "*F1S2: "Yes, because... I myself felt that -- when I actually placed the probe... well then, I... eh\... got a direct understanding of WHAT I saw on the screen. Because on the Wednesday when I went through the cases I just felt like "Ok, the right chamber is there and the left chamber is there, ok, ok." But when I actually placed the transducer myself it actually became logical, why the chambers were where they were\...So... I couldn't connect the placing of the transducer with the projections until I placed the transducer myself".*" "*F2S2: "Mm, one somehow wanted to connect the visual to the theory... at the same time..."*" "*F2S1: And THAT really helped a lot -- the times we could book the training laboratory and examine each other. For example, when we read something, we could also put it into practice at the same time. And it helped... when we worked together.*" The students thought it was hard to link the ultrasound physics to the ultrasound machine and to practical performance. There were suggestions about connecting the practical learning with theory by teaching physics while using the ultrasound machine. If the theoretical ultrasound teaching had been conducted with an ultrasound machine from time to time it would have helped with the learning of the functions of the machine buttons as well as how to optimize the projections. Seeing many echocardiographic images was a help. Something that both groups talked about was "visual and practice/theory" and the wish to see more images during the physics teaching, as well as when learning measurements, concepts and how to obtain projections. ### Measurements and concepts {#Sec14} The students thought it was difficult to understand why, in what context and how certain measurements and concepts are used. Initially it was also hard to understand the meaning of the concepts. A suggestion was put forward to give students a list of the most frequently used measurements and concepts, including explanations of meaning but also why and when they are used. A manual for *how* to measure correctly would have helped. The students learned the concepts through the lectures and by studying the literature. YouTube films were also a support when it came to learning why and when certain measurements are used. The students also said that it would have helped to understand which measurements are used and when, if they had not only examined healthy persons but also those with heart disease. In addition, they desired more cases during lectures. "*F1S3: Well... it was the meaning of the concepts and what they are used for and why they are so important and... using it later when... well, when measuring.F1S4: And how to measure them,F1S3: YesF1S4: That was important too...F1S3: That was the hard part to understand.*" In addition to the difficulty of learning measurements and concepts and making the connection to the projections was the concentration needed to obtain a projection and that it was so easy to lose a projection when the student had to think about measurements. "*F2S5: Yes, one was so concentrated on not losing the probe position which made it all even more difficult.*" Discussion {#Sec15} ========== Echocardiography is a clinically useful, important and globally used diagnostic examination of the heart's anatomy and function. It is however notoriously difficult to learn. This study reveals several difficulties in the process of learning echocardiography -- the projections, handling of the probe, why and how the measurements are executed and the practical application of the theory of ultrasound physics, to mention a few. Things that favour learning in echocardiography can be instant feedback, practical training (including "playing" with the ultrasound machine) and easily accessible pedagogical tools such as course literature, anatomical dummy as well as video lectures. Feedback {#Sec16} -------- This study confirms the pedagogical benefits of feedback. The students wanted more instant feedback, especially at the beginning of the learning process, to receive confirmation about whether or not they were placing the probe correctly as this was one of the more challenging parts of the learning process. Hattie and Timperley \[[@CR22]\] showed that feedback represents a central and effective part of influencing students' learning. The students in this study suggested more teacher presence as a way of increasing instant feedback. Teachers, however, only have a limited amount of time for their teaching and increased teacher presence would result in increased costs. Furthermore, studies have shown that feedback from teachers, no matter how detailed it may be, in some cases do not lead to progression in learning. One of the reasons for this could be that students do not understand the feedback given to them \[[@CR23]\]. Topping \[[@CR24]\] showed that students often consider it easier to receive feedback from peers than from teachers. In this study the students highlighted the advantage of working together and helping each other at the ultrasound machine and acting as coaches for each other. The pedagogical advantage they experienced through this could be that they not only performed the examination but could also put forward their case about what they were doing and why. One way to meet the students wish for more feedback could be to design a vLAB in such a way that it gives the student instant feedback \[[@CR8], [@CR25]\]. This might result in students being more open to feedback from the vLAB than from the teacher, in the same way as students are more open to feedback from their peers. To support them in their learning, the students in the current study wanted an independent assessor. By this they meant an assessor who was not involved in the course or in the final assessment of the students. A vLAB could perhaps act as this independent assessor. The feedback from an echocardiographic vLAB could include how the probe is placed and angled, where in the heart the ultrasound beam cuts and why, and also the quality of the measurements done. All of these points were problems that arose in the study. A vLAB provides the possibility to simultaneously show a 3D image of the heart and where the ultrasound beam cuts, while at the same time displaying a corresponding echocardiographic projection. That is something that may help in understanding projections. When relevant measurements on projections are to be done, a vLAB can indicate whether or not the measurement is correct, as well as present the user with fact boxes as to why and how the measurement is used. This feedback could be of value to the students who in the category "bridging the theory-practice gap" expressed difficulties in understanding why, when and how certain measurements are used. The vLAB cannot completely replace practical training but it can bridge the theory-practice gap. According to Dickson \[[@CR26]\], interactive technologies can bridge the theory-practice gap within ultrasound teaching through the recurring feedback it can give students. A study showed that using a vLAB in the field of gynaecological ultrasound was good preparation for clinical practice \[[@CR25]\]. The students' desire for more feedback could also partly be satisfied by encouraging them to work three at a time at the ultrasound machine -- one acting as "patient", one operating the machine and one standing at the side. The one standing at the side can have access to pictures of the projections while asking the one doing the examination to find a certain projection and state what cardiac chambers are seen in the projections. The projections on the screen can be compared to the pictures of the projections and thus instant feedback can be given. Practical training and "playing" {#Sec17} -------------------------------- In order to perform a clinically relevant, useful examination within echocardiography, one needs technical as well as practical skill. Handling the transducer is something often referred to as one of the more challenging parts of the examination \[[@CR3]\]. The students were positive to being able to "play" with the machine and to the fact that there was no machine button that was "off limits", hence it was a risk-free environment. Teaching methods can be split into two categories: guided and active/explorative training. In guided training the student is seen as a passive recipient of instructions while in active training the student is seen as an active participant in the learning process \[[@CR27]\]. During practical training, active/exploratory learning demands minimal external guidance and, compared to guided training, it seems to promote metacognitive qualities in the student (ibid). Perhaps it was this active/exploratory training the students referred to when they said it was good to "play" with the machine. This "playing" can perhaps be developed further. One study showed that an effective way to promote learning is to deliberately integrate errors and mistakes in the active/exploratory training and that it was even more effective than "only" having active/exploratory training \[[@CR28]\]. The students appreciated the possibility to practice on the machine in a risk-free environment where mistakes did not have serious consequences. Perhaps this can indicate that the students felt that the possibility of making mistakes was an important part of their learning. Since diagnostical mistakes can have serious consequences, the tradition in medical teaching has been to focus on how to avoid mistakes. However, it has been shown that diagnostical mistakes can be an important and positive part of medical learning and that the risk of making mistakes over a long time increases if they are not included in the learning process \[[@CR29]\]. Instead of following the tradition in medical education of avoiding mistakes, students can be encouraged to make mistakes as a part of their learning. Dyre et al. \[[@CR30]\] performed a study on medical students who had no previous experience of ultrasound and who were to learn transabdominal ultrasound by using a simulator. The students were divided into two groups -- one group was encouraged to "play" with the machine and mistakes were regarded as positive, and one group was encouraged to follow the instructions from the simulator and make as few mistakes as possible. The results showed that after training with the simulator, the first group showed a better transfer of learning in a clinical environment. By "playing" with the machine, making mistakes and actively learning echocardiography, the students develop their metacognitive ability as well as ability to solve problems. This can be used in future echocardiography teaching by consciously and actively encouraging students to make deliberate mistakes, find "new" projections and challenge one another to change the adjustments of the machine and then ask their fellow students to change them back etc. The limitation of the machine, however, is that it is fixed to a specific physical location and the students cannot access it whenever they wish. Also, it does not give any indication as to whether the transducer is placed correctly, whether the image quality is optimal or whether the measurements are acceptable, all of which a vLAB could be designed to do. A vLAB designed for giving feedback could therefore be combined with a function for "playing" with the machine. The feedback could be designed with the mindset that "mistakes are positive". This kind of vLAB could therefore give the students more possibility to "play" and make mistakes but also receive instant and positive feedback in response to their mistakes, all in a stress-free, risk-free environment. Mistakes could help the students to identify their need of further learning within a certain area \[[@CR29]\]. Mistakes made in a vLAB would not put patients at risk but rather prevent them from occurring in a clinical setting. Mistakes made within healthcare can be emotionally charged for those who make them. One further effect of teaching the mindset "mistakes are positive" is that the students not only learn from their actual mistakes but also learn how to handle their mistakes, leading to less stress if the mistakes are made in a clinical setting \[[@CR31]\]. In this study the students expressed a desire to alternate their learning between theory and practice. Some students wished for practical training prior to theory in order to learn *how* and thereafter *why,* whereas some students wanted to know what they were doing before they held the transducer. One way to meet this wish is to use video lectures, coupled with free accessibility to the ultrasound machine. Our study reveals the benefits of video lectures and the students did suggest that the teachers could record some of the lectures. There are previous studies showing the appreciation of video lectures to traditional lectures when learning echocardiography. Tainter et al. \[[@CR32]\] designed and evaluated a flipped classroom curriculum for teaching point-of-care echocardiography for residents rotating in the surgical ICU. Although the study was not designed to compare the traditional to the flipped classroom model, residents' feedback expressed a preference for the flipped classroom model, compared to the traditional model. Previous studies have shown that watching video lectures prior to going into the laboratory increases the students' knowledge of, confidence in and experience with certain laboratory techniques \[[@CR33]\]. According to Makransky et al. \[[@CR34]\], vLABs can work just as well as "face-to-face" lectures when preparing for a lab, suggesting a combination of virtual and physical labs within science education. Hence a vLAB perhaps can partly satisfy the students' wish for more video lectures. However, one study assessed the impact of different teaching interventions when learning basic practical echocardiography skills by trying four different methods, one of which was video lectures. They found that even though the group learning through video lectures demonstrated assessment results similar to those of the other groups they still would have preferred a different teaching method altogether. One reason for this could be the fact that the other three groups in the study (peer-teaching, peer-teaching using Peyron's four-step approach and team-based learning) all received feedback when learning the projection, unlike to the video group who only received technical support \[[@CR35]\]. This may also endorse the findings of the benefits of feedback in this study. Limitations {#Sec18} ----------- This study has some limitations that may restrict or limit the transfer of findings. We acknowledge the limited number of participants, even though the two groups discussed similar themes, there might be a risk that not all possible themes have been covered. Also, only one educational programme from one university has been included in the study. Echocardiography is used by several professions and for different purposes making the curriculums, way of teaching and perhaps time spent learning different from one profession to another. However, this study focuses on the initial challenges when learning echocardiography- handling the probe and understanding the basics of ultrasound physics are the same for any profession learning echocardiography. Another possible limitation in this study was that the first author and the study participants had a previous examiner-student relationship outside the study. This may have led to the participants not feeling able to freely share their experiences from the course as they may have felt it could have a negative effect on their relationship with the main author. In order to avoid this, they were assured that the author would not allow the result of the study to affect their professional relationship and all students had passed the course and were informed of this prior to the interviews. On the other hand, the fact that the students were interviewed by someone they knew and who knew their echocardiographic background may well have made it easier and even more comfortable for them to share their experiences. Conclusions {#Sec19} =========== This study shows the challenges when learning echocardiography and what might be helpful during the learning process. Main challenges when initially learning echocardiography are the projections and handling the probe as well as connecting ultrasound physics and measurements to practical application. Objective pedagogical suggestions for what might be helpful in the learning process such as; "playing" with the machine, immediate feedback, video lectures, the importance of swiftly alternating between practice and theory as well as the benefits of learning by mistakes in a risk-free environment are proposed. A vLAB within this field, designed to enhance that which is already helpful in the learning processes, could help to minimize or solve the difficulties in learning echocardiography. Practical implications {#Sec20} ---------------------- Based on the findings of this study, a vLAB within echocardiography could be designed to:Help students understand how to steer the transducer and how the beam should cut the heart in order to obtain a certain projection.Contain a virtual anatomical 3D dummy.Give indications as to whether the user's measurements and projections are correct and of sufficiently high quality.Include fact boxes and questions concerning the physics and different measurements.Comprise various cases so that students would receive a greater understanding of pathology and what findings to expect.Help create a set-up where it is possible to swiftly alternate between practical experience and theory by using the vLAB together with hands-on practising with the ultrasound machine. vLAB : Virtual laboratory **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. To the contributions of Anna Gustafsson to this project for acting as an observer during the interviews and also drafting the manuscript. To the students who participated and voluntary consented to take part in the study. AD designed the study and collected, interpreted and analysed the data and wrote the manuscript. PG participated in the design of the study, interpreted and analysed the data and critically reviewed the manuscript. EC participated in the design of the study, interpreted and analysed the data and critically reviewed the manuscript. All authors read and approved the final manuscript. Authors' information {#FPar1} ==================== AD and PG have several years of experience of working clinically and teaching within echocardiography, as well as experience of pedagogical development work within the area. EC has extensive experience of educational research and qualitative methods. Funding was provided by the Educational committee at Malmö University. The role of the committee is to promote pedagogical development at the University and had no involvement in the study concerning the design of the study, the data collection, analysis and interpretation of the data or writing of the manuscript. The dataset used and/or analysed during the current study are available from the corresponding author on reasonable request. The dataset is in Swedish. In accordance with the Declaration of Helsinki and the Swedish Research Council's requirements regarding information and consent, informed consent was obtained prior to the interviews. According to Swedish Law (SFS 2003:460) and the local ethical guidelines of the university, no written consent is necessary for studies that do not explore sensitive issues (e.g., political, sexual or religious). Not applicable. Two of the authors (AD, PG) are currently developing a vLAB within echocardiography.

{#sp1 .77} {#sp2 .78}

INTRODUCTION ============ Over the past decade there has been increasing interest in using the open cardiovascular system of *Drosophila melanogaster* as an animal model of human cardiovascular disease (Bier and Bodmer, 2004). The *D. melanogaster* cardiovascular system ([Fig. 1A](#f1-0040411){ref-type="fig"}) is open because it lacks discrete, closed return vasculature ([@b42-0040411]). Vertebrates, by contrast, have a well-developed system of return vasculature represented by the venous tree. Previous research has demonstrated genetic, molecular, cellular and tissue function homology between the *D. melanogaster* and vertebrate cardiovascular systems. However, the degree of global physiological homology remains an open question. For example, it is unclear whether commonly measured physiological parameters such as cardiac output and aortic velocity, when normalized to measures of body size, are similar between *D. melanogaster* and important vertebrate species, including humans. Normalization can provide a basis for comparison across different size scales in addition to providing meaningful measures of physiological performance. In both pediatric ([@b23-0040411]) and adult (Kern and King, 2008) cardiology, it is standard practice to normalize cardiac output to body size to control for variation in cardiac output that is attributable to body size (Mohrman and Heller, 2006). In rodent (e.g. mouse, rat) cardiovascular research, it also is commonplace to normalize cardiac output to body size (e.g. [@b17-0040411]; [@b37-0040411]). Aortic velocity normalized to body length provides a measure of global capacity for cardiac-driven, convection-limited transport of nutrients and signaling molecules across scales that are impractical for diffusion-mediated transport. Therefore, the comparison of normalized measures of cardiovascular performance among different species is a physiologically and metabolically relevant approach for establishing physiological homology. The existence of cardiovascular physiological homology between vertebrates and *D. melanogaster* requires similarities not only in cardiovascular performance but also in cardiac-driven global mass transport dynamics. In vertebrates, an extensive vascular network provides for blood delivery throughout the body. Flow through this network is fast and mass transport in the vascular compartment is typically convection limited. In *D. melanogaster*, however, once hemolymph leaves the vascular compartment, delivery of hemolymph occurs throughout the body in an open manner in the extracellular-extravascular space. Therefore, convection-limited transport in *D. melanogaster* requires both rapid intravascular flow rates compared with body length as well as open return flow rates in the extracellular-extravascular compartment that can outpace diffusion-mediated transport. To date, these open mass transport dynamics are unknown and are often assumed to be slow (Kirby and Schachat, 2007). In addition to basic measures of physiological performance, it is important to demonstrate that mutant *D. melanogaster* hearts can reproduce subtle functional heart defects that are observed clinically in humans. In particular, there is growing clinical interest in quantifying myocardial motion and function in terms other than pump efficiency (e.g. fractional shortening, ejection fraction) (Maeder and Kaye, 2009). In particular, there is a growing literature using local measures of myocardial function such as strain rate imaging ([@b45-0040411]; [@b27-0040411]) and Doppler-based measurements of heart wall velocity (Ho and Solomon, 2006; [@b27-0040411]; Maeder and Kaye, 2009). In addition, there is recent imaging work characterizing local myocardial function in diseased murine hearts ([@b33-0040411]; [@b35-0040411]). Therefore, the relevance of *D. melanogaster* to human cardiovascular research would be enhanced if methods and mutants were available to demonstrate quantitative heart wall motion defects given their relevance to clinical medicine. {#f1-0040411} In this study, we demonstrated physiological homology between *D. melanogaster* and vertebrate cardiovascular physiology. Flow measurements were obtained using a novel dye angiography technique that allows for the real-time microscopic imaging of hemolymph flow in the open circulation of wild-type (OreR) pre-pupal *D. melanogaster*. The pre-pupal stage was chosen because: (1) the organism is relatively transparent, (2) it is stationary and does not require anesthesia or immobilization for imaging and (3) this stage represents the anatomic end-point of embryonic and larval development before entry into pupation. In conjunction with real-time structural and Doppler optical coherence tomography (OCT) ([@b16-0040411]; [@b40-0040411]; [@b6-0040411]; [@b44-0040411]) imaging of the pre-pupal heart and aorta, we demonstrated that the open cardiovascular system of *D. melanogaster* has fast, cardiac-cycle-dependent, transport of hemolymph. Moreover, we found that several normalized measures of cardiovascular function, including cardiac output, aortic flow velocity and intracardiac flow velocity, were similar among *D. melanogaster*, humans and other vertebrates. We demonstrated that the *D. melanogaster* heart is a pulsatile pump that has intracardiac hemolymph velocities consistent with vertebrate embryo levels of intracardiac shear stress. We present evidence that convective mass transport can dominate over diffusive mass transport and that extracellular-extravascular flow is not isotropic but is partially directed. Finally, using a commercially available Doppler OCT system, we demonstrated impaired systolic and diastolic heart wall velocities in the pre-pupal, *held-up2* (*hdp2*) troponin I mutant (Beall and Fyrberg, 1991; [@b44-0040411]); a finding made possible by detailed physiological phenotyping. Interestingly, heart wall velocity defects occur in the context of a non-dilated phenotype with only mildly depressed fractional shortening. Moreover, this finding complements a growing understanding that clinical heart failure can result from biomechanical defects in systolic and/or diastolic myocardial functioning, which require physiological characterization beyond ejection fraction assessment (Maeder and Kaye, 2009). Taken together, our results provide a physiological rationale for using *D. melanogaster* as an animal model of human cardiovascular disease, a rationale that is complementary to existing genetic and molecular evidence. RESULTS ======= Cardiac physiology ------------------ We visualized the pre-pupal cardiac cycle using dye angiography and OCT ([Fig. 1](#f1-0040411){ref-type="fig"}; supplementary material Movies 1, 2). From an angiographic perspective, injection of dye adjacent to the heart allowed for characterization of several physiological aspects of the cardiac cycle. Systole was characterized by rapid ejection of dye out of the heart into the aorta and diastole by heart filling through the ostia ([Fig. 1E](#f1-0040411){ref-type="fig"}--H; supplementary material Movie 1). There was rapid accumulation, then clearance of dye in the heart ([Fig. 1J](#f1-0040411){ref-type="fig"}; supplementary material Movie 1). The accumulation and clearance dynamics are clearly evident in the low-pass-filtered version of the heart lumen pixel intensity signal (black curve in [Fig. 1J](#f1-0040411){ref-type="fig"}) because the filtering minimizes beat-to-beat variability. Large beat-to-beat variability of intracardiac dye concentration was observed ([Fig. 1J](#f1-0040411){ref-type="fig"}). The average periodicity of this variability corresponded well with the heart rate ([Fig. 1K](#f1-0040411){ref-type="fig"}). Diastolic intake of hemolymph allowed for rapid clearance of injected extracellular-extravascular dye, whereas systolic ejection allowed for rapid transport of dye through the aorta and to the head ([Fig. 1I](#f1-0040411){ref-type="fig"} and see below; supplementary material Movies 1 and 8). There was no observed retrograde flow from the aorta to the heart during diastole and diastolic filling, suggesting that this previously characterized anatomic valve ([@b7-0040411]) does indeed function as a hemodynamic valve. These microangiographic findings were corroborated by OCT imaging of the heart-aorta junction (see below; supplementary material Movies 1 and 5). This junction opens at the onset of systole, indicating that pressure generated in the heart is very rapidly transmitted to the heart-aorta junction, resulting in the opening of the aortic valve. Moreover, there was closure of the aortic valve at the end of systole when forward flow falls to zero and cardiac pressure equilibrates with that in the extracellular-extravascular space. We obtained echocardiographic parameters of the heart using OCT imaging. In the sagittal plane (*n*=5 organisms), the average diastolic long (anterior-posterior) and short (dorsal-ventral) axis diameters were 427±33 *μ*m (mean ± s.d.) and 95±6.6 *μ*m, respectively. The average heart rate was 135±23 beats per minute. Assuming a prolate spheroid model of the heart, we calculated an average end-diastolic volume (*V*~ED~) of 2.0±0.3 nL (*V*~ED~=4/3π*a*^2^*b; a*, short-axis radius; *b*, long-axis radius). Fractional shortening (i.e. percentage change from diastolic diameter during systole) of the short-axis of the anterior and posterior aspects of the heart were 82±26% and 93±16%, respectively. We did not observe shortening of the long-axis of the heart, consistent with previous histological observations that muscle fibers run only circumferentially and not longitudinally in P1 pupae ([@b7-0040411]). Using a conservative estimate of a 75% ejection fraction, and assuming a pre-pupal body mass of 2 mg ([@b10-0040411]), we estimated a cardiac output normalized to body mass of 100 ml/minute/kg. For comparison, the normalized cardiac output in humans is ∼70 ml/minute/kg assuming a heart rate of 80 beats per minute, a stroke volume of ∼100 ml ([@b21-0040411]), an ejection fraction of ∼60% ([@b21-0040411]), and a weight of ∼70 kg ([@b20-0040411]); in dogs, ∼150 ml/minute/kg (Berne and Levy, 1950); in 2 day post-fertilization (d.p.f.) zebrafish embryos, ∼30 ml/minute/kg ([@b26-0040411]) assuming a weight of 700 *μ*g ([@b15-0040411]); and in 5 d.p.f. zebrafish embryos, ∼30 ml/minute/kg ([@b26-0040411]) assuming a weight of 1.2 mg ([@b15-0040411]). Weight-adjusted cardiac output in the mouse appears to be higher than that of pre-pupal *D. melanogaster*, but there is enormous variability in this measure in the literature ([@b17-0040411]) (209--1300 ml/ minute/kg). In terms of previous measurements in open circulations, the cockroach cardiac output can been estimated to be in the ∼100 ml/minute/kg range (Birchard and Arendse, 2001) and several molluscs and arthropods have estimated cardiac outputs in the ∼100 ml/minute/kg range (D. D. Jorgensen, Cardiac output and its distribution in the intertidal mollusc, *Haliotis cracherodii*, PhD thesis, Iowa State University, 1981). Intracardiac shear rates and forces ----------------------------------- We used M-mode Doppler OCT to measure early-systolic intracardiac flow velocities. We measured flow velocities by imaging individual hemocytes (i.e. blood cells) as they underwent acceleration during the early moments of systole ([Fig. 2](#f2-0040411){ref-type="fig"}). We calculated intracardiac velocities on the order of one body length per second (∼2--4 mm/second). This method underestimated peak hemocyte velocity because an individual hemocyte leaves the M-mode field of view before reaching peak velocity. Moreover, the hemocytes are not necessarily located in the mid-lumen, another factor that can underestimate flow velocity. For comparison, human blood velocities at the mitral and tricuspid valves are in the order of 0.6--1.3 m/second and 0.3--0.7 m/second, respectively (DeMaria and Blanchard, 2007), both of which are less than one body length per second (adult human body length is ∼1.7 m) ([@b20-0040411]). Assuming that hemolymph flow in the *D. melanogaster* heart is viscous and non-turbulent, and assuming that there is insufficient time for a parabolic flow profile to develop, the shear rate can be estimated from the quotient of the midline velocity and the vessel radius ([@b14-0040411]), yielding shear rates of ∼40--60 second^--1^. Since the Reynolds number (*N~R~*) of pre-pupal intracardiac flow is ∼0.1 (*N~R=~ρvdμ*^--1^; *ρ*, fluid density; *v*, fluid velocity; *d*, vessel diameter; *μ*, fluid viscosity), the viscous, non-turbulent assumption is valid. These shear rates are comparable to those in a 37 hour post-fertilization (h.p.f.) zebrafish embryo ([@b14-0040411]). {#f2-0040411} There are several reasons to assume that intracardiac flow profiles are not necessarily parabolic. First, immediately before the onset of systole, there is no flow in the heart, so the flow profile is flat. Even if the intracardiac flow profile develops into a parabolic shape during mid or late systole, the profile cannot immediately transition from flat to parabolic at the onset of systole. Second, the derivation of the governing equations for parabolic Hagen-Poiseuille flow ([@b32-0040411]) makes several assumptions that are violated in our experiments. Hagen-Poiseuille flow assumes steady-state flow (i.e. no flow velocity changes with respect to time) under a constant driving pressure and constant vessel diameter. These assumptions cannot be plausibly made during active systolic heart contraction. Third, from a microfluidics perspective, the time needed to dissipate unsteady inertial forces is *τ~i~*∼ρ*d^2^* μ^--1^ (Squires and Quake, 2005). Using parameters described in this section, *τ~i~*∼4 milliseconds. [Fig. 2](#f2-0040411){ref-type="fig"} demonstrates that flow and wall motion time constants are shorter than *τ~i~*, which also makes a quasi-steady-state assumption of well-developed flow challenging. In sum, our assumption of a linear flow profile in calculating shear rates provides a first-order, lower-limit estimate with the minimal assumptions that (1) the flow profile peak is midline and (2) higher-order flow profile shape (e.g. parabolic) will yield shear estimates that are higher, not lower, than the linear estimate. It also should be noted that these calculations are relevant to flow dynamics in vertebrate embryo hearts (e.g. human, zebrafish, *Xenopus*) with size scales of ∼100 μm. The constant of proportionality between shear rate and shear stress is dynamic viscosity. Unfortunately, the dynamic viscosity of *D. melanogaster* hemolymph has not been characterized. However, after assuming that the relatively acellular hemolymph has a viscosity that is similar to acellular human plasma at a comparable shear rate \[∼2.5 centipoise ([@b34-0040411])\], we estimated intracardiac shear stress of ∼1--2 dyne/cm^2^, which is similar to the intracardiac shear stress present in a 37 h.p.f. zebrafish embryo ([@b14-0040411]). Systolic and diastolic heart wall dysfunction demonstrated using Doppler heart wall velocimetry of normal and *hdp2* mutant hearts ---------------------------------------------------------------------------------------------------------------------------------- It is evident from [Fig. 2](#f2-0040411){ref-type="fig"} that M-mode Doppler OCT is sensitive to Doppler shifts generated by heart wall motion. We exploited this sensitivity to perform tissue Doppler-imaging-based heart wall velocimetry ([Fig. 3](#f3-0040411){ref-type="fig"}), in analogy with clinical, ultrasound-based tissue Doppler imaging of human heart wall velocities (Ho and Solomon, 2006; [@b27-0040411]). As shown in [Table 1](#t1-0040411){ref-type="table"} and [Fig. 4](#f4-0040411){ref-type="fig"}, we demonstrated that the *hdp2* mutants have a statistically significant slower peak systolic and diastolic ventral wall velocity compared with OreR wild-type hearts. Thus, just as human hearts can demonstrate dysfunction beyond pump efficiency, mutant *D. melanogaster* hearts can have homologous physiological dysfunction that is evident only through sophisticated biomechanical phenotyping. In contrast to prior research demonstrating a dilated phenotype in adult *hdp2* mutants ([@b44-0040411]), our results indicate that the pre-pupal *hdp2* heart is not dilated ([Table 1](#t1-0040411){ref-type="table"}). Similarly to the adult *hdp2* heart ([@b44-0040411]), the pre-pupal *hdp2* heart has depressed systolic function as assessed with fractional shortening. However, compared with the wild-type OreR pre-pupa, the *hdp2* fractional shortening is only mildly depressed (73% versus 66%). However, the differences in systolic ventral wall velocities, another measure of systolic function, are large (730 *μ*m/second versus 532 *μ*m/second). To compare shortening fraction and systolic wall velocity as diagnostic tests, we generated receiver operator characteristic (ROC) curves (Hanley and McNeil, 1982; [@b46-0040411]) for each measure ([Fig. 4](#f4-0040411){ref-type="fig"}). Since each ROC curve is well above the 45° line (i.e. sensitivity=\[1--specificity\]), each is a useful discriminator of mutant and wild-type genotypes. However, systolic wall velocity had an area under the curve of 0.85±0.066 (mean ± s.e.) and fractional shortening had an area under the curve of 0.62±0.097, indicating that systolic wall velocity is a stronger discriminator. Aortic physiology ----------------- Our focused imaging of the aorta during the cardiac cycle revealed aortic wall behavior that is different from that of the human aorta. In humans, the diameter of the aorta is not critically sensitive to variation in intraluminal pressures during the cardiac cycle. Using OCT, we showed that the pre-pupal *D. melanogaster* has an aortic diameter that is cardiac-cycle dependent. During diastole, the aorta collapses and, during systole, the aortic diameter considerably increases ([Fig. 5](#f5-0040411){ref-type="fig"}; supplementary material Movies 5 and 6). The systolic increase in diameter occurs in a posterior-to-anterior direction away from the heart towards the head, consistent with the transmission of a pressure wave from the contracting heart down the aorta towards the head. As systole ends, the aorta collapses in a posterior-to-anterior direction. The absence of aortic dye during diastole is also consistent with a time-varying diameter that is at its minimum during diastole. {#f3-0040411} {#f4-0040411} We used dye angiography and digital subtraction angiography to image aortic flow dynamics. This imaging showed the rapid transport of a bolus of fluid ejected from the heart, through the aorta, and into the extracellular-extravascular space in the head ([Fig. 6](#f6-0040411){ref-type="fig"}; supplementary material Movie 7). The time-varying diameter of the aortic lumen was also evident from dye angiography movies. Using this imaging we calculated aortic velocities ranging from 1.40 to 2.55 body lengths per second (*n=*5 organisms; mean ± s.d., 1.79±0.46). These velocities are consistent with vertebrate intracardiac flow velocities. For comparison, peak human aortic velocity is on the order of 1.0--1.7 m/second (DeMaria and Blanchard, 2007), which corresponds to ∼1 body length per second. In dog, peak aortic velocity is ∼1.5 body length per second ([@b43-0040411]), assuming a body length of ∼1 m; in mouse, ∼4 body lengths per second ([@b11-0040411]) assuming a body length of 9 cm ([@b41-0040411]); in 2 d.p.f. zebrafish embryos, ∼1 body length per second assuming a body length of ∼3.5 mm (Muller and van Leeuwen, 2004); and in 5 d.p.f. zebrafish embryos, ∼1 body length per second assuming a body length of ∼4 mm ([@b36-0040411]). Extracellular-extravascular flow dynamics and mass transport ------------------------------------------------------------ We used dye angiography in the *D. melanogaster* pre-pupa to gain insight into global flow dynamics and extracellular-extravascular flow dynamics ([Figs 1](#f1-0040411){ref-type="fig"} and [6](#f6-0040411){ref-type="fig"}; supplementary material Movies 2--4, 8--9). Transport of dye away from an injection site was driven by cardiac cycle-dependent convective transport. [Fig. 1I](#f1-0040411){ref-type="fig"} demonstrates very little diffusion of dye in an anterior direction away from the injection site. There also was convective transport of dye anterior to the injection only after dye accumulation at the head, suggesting that cardiac-cycle-driven convection can dominate over diffusion. Put another way, the cardiac transport of dye from the heart to the head towards the injection was more efficient that diffusion driven by local concentration gradients. The dependence of mass transport on the cardiac cycle was most evident upon resumption of the cardiac cycle following a period of asystole that occasionally occurs after puncture of the cuticle and injection of dye ([Fig. 6](#f6-0040411){ref-type="fig"}; supplementary material Movie 8). During asystole, there was little transport of dye away from the injection site, despite a strong concentration gradient that favors diffusion. Upon resumption of the cardiac cycle, transport of dye away from the injection site was rapid. These findings support the hypothesis that convection has a central role in bulk mass transport in the pre-pupal *D. melanogaster*. Dye angiography also provided evidence for fast, non-isotropic, extracellular-extravascular, cardiac-cycle-dependent mass transport. Dye injected far from the heart tube returned to the heart via preferred flow channels, despite the absence of discrete return vessels ([Fig. 6](#f6-0040411){ref-type="fig"}; supplementary material Movies 8, 9). One of these channels is adjacent to the large tracheal tubes present on either side of the pre-pupae. Another one of these channels is dorsal to the aorta itself. The existence of a dorsal return channel was supported by our OCT imaging of the aorta that showed an anterior-to-posterior stream of hemocytes dorsal to the aorta (supplementary material Movie 5). DISCUSSION ========== The results presented here advance the study of the open cardiovascular system of *D. melanogaster* and support its use as an animal model of complex human cardiovascular disease. We demonstrated physiological homology between the pre-pupal *D. melanogaster* and vertebrate cardiovascular systems. This conclusion is supported by (1) similarities in heart performance measures, (2) similar rates of flow in the great vessels, (3) the presence of cardiac-dependent, convection-limited bulk mass transport, (4) non-isotropic return flow to the heart, and (5) the existence of complex functional heart defects that are evident only through detailed functional characterization. ###### *hdp2* hearts have normal fractional shortening but diminished systolic and diastolic wall velocities compared with wild-type pre-pupal hearts  {#f5-0040411} The wild-type pre-pupal *D. melanogaster* has vigorous, cardiac-cycle-dependent convective mass transport that can transport a nanoliter-scale stroke volume an entire body length within a few hundred milliseconds. Of particular note, we identified four vigorous flow streams along the anterior-posterior axis, the longest dimension of the organism. One stream has a posterior-to-anterior direction (aortic flow), whereas the other three streams (two para-tracheal streams and one stream dorsal to the aorta) have an anterior-to-posterior direction. The presence of vigorous, anti-parallel flow streams along the longest dimension of the animal suggests a fairly well organized flow circuit across distances that are prohibitively difficult for diffusion to traverse. Moreover, weight-adjusted cardiac output and other physiological measures are similar to those in humans. Just as diffusion has a role in mass transport from the vascular to the extracellular-extravascular compartment in vertebrates ([@b5-0040411]), the present study does not exclude the role of diffusion over small spatial scales in the *D. melanogaster* extracellular-extravascular space. However, the presence of vigorous intravascular and extracellular-extravascular flow should dramatically increase diffusion rates over those spatial scales by maintaining a relatively constant 'source' for diffusion-mediated delivery and a 'sink' for diffusion-mediated removal. Although our results support physiological homology between *D. melanogaster* and vertebrate cardiovascular systems, aortic wall imaging did highlight a difference between the *D. melanogaster* aorta and vertebrate aortas: the pre-pupal *D. melanogaster* aortic diameter varies dramatically during the cardiac cycle, whereas vertebrate aortic diameters tend to be relatively constant throughout the cardiac cycle. Previous aortic imaging was limited in its ability to assess the dynamic behavior characterized by OCT. The time-varying diameter of the aorta is predominantly manifest along its superior-inferior axis. This spatial dimension is difficult to assess through en face cross-sectional imaging (e.g. confocal microscopy) and through traditional widefield microscopy (which essentially produces a projection image over the depth of field along the superior-inferior axis). In addition, the pressure-sensitive opening and collapse of the aorta during the cardiac cycle is not immediately obvious from prior histological and anatomical studies. Similarly to the heart itself, the aorta is composed of two layers, each of which is one cell thick: an inner layer of myocardial cells and an outer layer or pericardial cells (Small and Krieg, 2002). The myocardial cells of both the heart and aorta contain muscle-specific markers such as polymerized actin ([@b22-0040411]), muscle myosin heavy chain ([@b24-0040411]) and heart-specific markers such as myocyte enhancer factor-2 ([@b24-0040411]). Even though the heart and aorta are both composed of myocytes, there is growing evidence that subdifferentiation of myocytes into the heart and aorta is driven by distinct genetic programs ([@b22-0040411]). This subdifferentiation appears to have a physiological correlate: the heart acts as a pulsatile, contractile fluid pump whereas the aorta acts as a non-pumping vessel with a pressure-dependent diameter. Furthermore, the pressure-sensitive behavior of the pre-pupal *D. melanogaster* aorta shares features with the phenomenon of critical closing pressure. Critical closing pressure refers to microvascular collapse secondary to steady-state active muscular vascular tension in the presence of vascular elasticity ([@b31-0040411]). Collapse can be overcome by a threshold pressure that is determined by the degree of active wall tension and vessel wall elasticity. Given that the aorta contain cells of myocardial origin and actively express muscle-specific markers, it is possible that these cells generate the active tension necessary for clinical closing pressure. The existence of threshold behavior dictates that a vessel governed by critical closing pressure has a hemodynamic resistance that decreases dramatically in the face of a threshold forward (i.e. posterior to anterior) pressure. Of note, valves obey a similar nonlinear rule. That is, in the face of a certain threshold forward pressure, the resistance across a valve decreases dramatically. Although a vessel governed by critical closing pressure is insensitive to the direction of an applied pressure wave, in pre-pupal *D. melanogaster* there is no apparent mechanism to generate a backward (i.e. anterior to posterior) pressure. Thus, the aorta itself appears to have valve-like properties. {#f6-0040411} Ultrasound-based Doppler heart wall velocimetry in humans is an increasingly important imaging modality in the biomechanical characterization of failing hearts (Ho and Solomon, 2006; [@b27-0040411]; Maeder and Kaye, 2009). One important clinical insight from sophisticated clinical imaging techniques is that functional defects exist beyond pump efficiency (e.g. fractional shortening, ejection fraction). Our results using OCT-based Doppler heart wall velocimetry show impaired systolic and diastolic heart wall velocities in the *hdp2* troponin I mutant, arguing for physiological homology in heart dysfunction and failure. ROC curve analysis suggests that, in the setting of a genotype with borderline changes in structural echocardiographic parameters (e.g. chamber diameters), defects in systolic function may be more readily identified by Doppler wall velocimetry. OCT-based tissue Doppler imaging also opens up possibilities to specifically study diastolic dysfunction in the *D. melanogaster* animal model of human heart failure. Likewise, by focusing primarily on global physiological characterization, several cardiovascular phenotypes have been identified (e.g. weight-adjusted cardiac output, intracardiac shear forces, aortic flow velocity) that can form the basis for future studies, such as the role of biomechanical forces in *D. melanogaster* heart development. METHODS ======= *D. melanogaster* culture and pre-pupal selection ------------------------------------------------- OreR and *hdp2 D. melanogaster* were raised on reconstituted dry medium (Carolina Biological Supply, Burlington, NC) and were exposed to ambient light-dark cycles. Files were cultured and imaged at 24°C. Pre-pupae, in contradistinction to late third instar wandering larvae, were defined as (1) being stationary, (2) having a lack of whole body motion, (3) being unresponsive to gentle touch stimuli, (4) having a cuticle, as ascertained by gently pushing on the animal with a blunt instrument, (5) being untanned in white light illumination. Before imaging, pre-pupae were carefully washed with a normal saline solution and dried with the gentle application of tissue paper. Structural and Doppler optical coherence tomography (OCT) imaging ----------------------------------------------------------------- Two OCT systems were used. Baseline structural and Doppler characterization of OreR hearts ([Figs 1](#f1-0040411){ref-type="fig"}, [2](#f2-0040411){ref-type="fig"}, [3](#f3-0040411){ref-type="fig"}, [5](#f5-0040411){ref-type="fig"} and echocardiographic parameters in the "Cardiac physiology" subsection) was performed using a custom-built 1300 nm OCT system with A-line rates of 10 or 20 kHz ([@b40-0040411]). Quantitative comparisons of OreR and *hdp2* hearts ([Table 1](#t1-0040411){ref-type="table"} and [Fig. 4](#f4-0040411){ref-type="fig"}) were made using a commercially available OCT system (Thorlabs, Newton, NJ) with a 1325 nm center wavelength and an A-line rate of 16 kHz. Post-processing and chamber measurements were performed using MATLAB (Mathworks, Natick, MA) and ImageJ (Wayne Rasband, National Institutes of Health, Bethesda, MD). End systolic and diastolic diameters for an individual specimen were reported as the average diameters measured for two distinct heartbeats. Fractional shortening was calculated as (EDD--ESD)/EDD. Fractional shortening for an individual specimen were reported as the average of fractional shortening measurements made on two distinct heart heats and not from average values of EDD and ESD. For Doppler imaging, conversion from measured Doppler shift (*f*~d~) to velocity (v) used the standard Doppler equation for echo-based Doppler velocimetry (*f*~d=~2vcos(θ)*nν*/*c*; θ, Doppler angle; *n*, tissue optical index; *ν*, optical frequency; *c*, free-space vacuum speed of light). For Doppler imaging of hemocyte velocity, a Doppler angle of 65° was assumed based on end-diastolic geometry of the heart during sagittal plane imaging. For [Figs 2](#f2-0040411){ref-type="fig"} and [3](#f3-0040411){ref-type="fig"} the background Doppler signal was thresholded based on the structural M-mode image. OCT-based tissue Doppler wall velocimetry used a semi-automated processing algorithm. M-mode Doppler data, processed using the Kasai algorithm ([@b28-0040411]) with a 1 millisecond × 5.7 μm window, was displayed using standard Doppler color maps in ImageJ. The area of subjective maximum systolic wall velocity for a particular beat was estimated by drawing a 2 millisecond × 5.7 μm region of interest and calculating the mean Doppler signal over that region of interest. This measurement was performed for two systolic beats for each M-mode recording. It was assumed that wall motions were parallel to the optical axis (i.e. Doppler angle of 0°). One-sided, unpaired, heteroscedastic *t*-tests were used to compare mean heart wall velocities between OreR and *hdp2* hearts. The assumption of heteroscedasticity was based on *F*-test values of \<95% for all comparisons. Receiver-operator characteristic (ROC) curve statistics ------------------------------------------------------- ROC curves were generated using a non-parametric approach ([@b46-0040411]) with custom MATLAB scripts. For a given measure of systolic performance (i.e. either fractional shortening or systolic wall velocity), a data array was generated that represented the sorted values across both genotypes (i.e. OreR and *hdp2*). The array had 34 elements, representing the combined sample size (16 *hdp2* and 18 OreR samples). Sensitivity and specificity datapoints were calculated using each element of this array as a cut-off value. A true positive (negative) was defined as an OreR (*hdp2*) measurement being larger than (less than) a given cut-off value. ROC curves were plotted using these 34 estimates of sensitivity and specificity. Area under the curve (AUC) and standard error of AUC was calculated as per Hanley and McNeil (Hanley and McNeil, 1982). Dye angiography --------------- The microinjection system used pressurized nitrogen, a pressure controller, and custom-pulled micropipettes. Glass capillary tubes were pulled with a vertical pipette puller. The tapered ends were manually broken to create a jagged, sharp end with an outer diameter between ∼50--150 micrometers. The micropipette was connected to the microinjection system and was filled with green dye (water, propylene glycol, FD&C yellow 5, FD&C blue 1, propylparaben; McCormick, Sparks, MD) via application of negative pressure. The cuticle was punctured by the micropipette. If needed, after initial puncture, the micropipette was withdrawn to eject any hemolymph that, via diffusion and capillary effect, had mixed with the dye. Dye was introduced by a few brief (30--60 milliseconds) pressure bursts at 3--6 p.s.i. Dye angiography movies were recorded at 320 (vertical) × 240 (horizontal) pixel resolution at 15 frames per second (f.p.s.) or with high definition 1080i (1920 vertical × 1080 horizontal) pixel resolution at an interlaced frame rate of 29.97 f.p.s. using a Nikon CoolPix 995 camera or a Sony HDR-HC9 camcorder, respectively. The camera was mounted to the trinocular port of an Olympus SZ-H10 stereomicroscope. Dye angiography image analysis and digital subtraction angiography ------------------------------------------------------------------ After dye angiography sessions, digital microscopy movies were transferred to a personal computer (MacBook Pro, Apple, Inc, Cupertino, CA) and analyzed using ImageJ and MATLAB. For time-varying pixel intensity measurements, the red channel of the RGB image was used. The red channel showed the most contrast between dye and the surrounding tissue. Time-varying pixel intensity plots were generated by (1) loading the red channel of a dye angiography movie into ImageJ as an image stack, (2) selecting a square region of interest (ROI), and (3) using the 'Plot Z-axis Profile' tool to generate the average pixel intensity within the ROI for each image in the stack. For aortic velocity measurements, the time-varying pixel intensity was acquired for a proximal and distal ROI on the aorta. As the dye bolus traverses past each ROI, the pixel intensity drops to a maximum and then return to baseline. The number of frames between the pixel intensity minimum for each ROI was counted. Dye velocity was calculated using this time delay and the measured distance between the two ROIs. Each reported velocity is the average of three such measurements for an individual animal. ###### TRANSLATIONAL IMPACT ### Clinical issue Advances in our understanding and treatment of heart failure have been underpinned by detailed physiological characterization of normal and failing hearts. Physiological characterization has also shown that measures such as heart wall velocity, when combined with traditional measures of heart pump efficiency (e.g. fractional shortening, ejection fraction), are important for diagnosing and classifying heart failure. The importance of subtle and robust characterization raises two important questions regarding the use of *Drosophila melanogaster* as an animal model of human heart failure. First, are baseline measures of cardiovascular physiology similar between vertebrates and *D. melanogaster*? Second, can subtle defects in heart function be demonstrated in *D. melanogaster* hearts? ### Results Using optical imaging techniques, the authors show that baseline measures of cardiovascular physiology in pre-pupal *D. melanogaster* are similar to those in vertebrates. Comparisons made on the basis of normalized measures (e.g. weight-adjusted cardiac output, body-length-adjusted aortic flow velocities) suggest there is physiological homology in normal functioning. Using Doppler optical coherence tomography (the optical analogue of ultrasound Doppler imaging), the authors show that the troponin mutant *hdp2* has a pre-pupal heart with impaired systolic and diastolic wall velocities with only mildly depressed fractional shortening. As troponin mutations are a cause of familial heart disorders in humans, the presence of subtle functional defects as well as quantitative diastolic dysfunction in *hdp2* flies suggests homology in heart dysfunction between *D. melanogaster* and humans. ### Implications and future directions The similarities observed here between vertebrate and *D. melanogaster* cardiovascular systems support the use of *D. melanogaster* as an animal model of complex cardiovascular disease. Optical imaging in the partially transparent pre-pupa, allowing quantification of physiological performance, will be a robust basis for future studies in mutant *D. melanogaster* hearts. This technique will also provide a basis for identifying genotype-phenotype relationships that might yield insight into complex clinical disease states such as diastolic heart dysfunction and heart failure with normal left ventricular ejection fraction. Digital subtraction angiography (DSA) was performed in ImageJ using deinterlaced HVD 1080i movies that have a native interlaced frame rate of 29.97 f.p.s. The deinterlaced movie frame rate was 59.94 f.p.s. The red color channel was used for DSA processing. For the images and movies shown in this manuscript, the background image was based on the arithmetic mean image from the 56 frames before onset of systole. Using the image calculator in ImageJ, this background image was subtracted from the subsequent 60 frames. Deinterlacing of 1080i video ---------------------------- Deinterlacing was performed using custom MATLAB code that assigned odd and even numbered lines in a frame to sequential odd and even frames, respectively. To maintain the aspect ratio and pixel density of the deinterlaced frame, each line *i* from the native interlaced frame served as lines *i* and *i*+1 if *i* was odd and as lines *i*--1 and *i* if *i* was even. This non-motion-compensated, linear-deinterlacing method is called line repetition (De Haan and Bellers, 1998). The jitter of stationary objects is a well-recognized artifact of this deinterlacing technique. Movie compression ----------------- All movies were compressed using MPEG-4 compression in QuickTime Pro (Apple, Cupertino, CA). Supplementary Material ====================== ###### Supplementary Material This work was generously supported by the Frederick H. Lovejoy Fund, Children's Hospital Boston (M.A.C.) and by an American Medical Association Foundation Seed Grant (M.A.C.). We additionally acknowledge funding support from NIH R01HL076398 and NIH R01CA103769 (development of the imaging system) and from the Department of Defense Air Force Office of Scientific Research, Medical Free Electron Laser Program FA9550-04-1-0079. We thank Thorlabs their loan of an OCT system. We thank James Vigoreaux for supplying *hdp2 D. melanogaster*. **AUTHOR CONTRIBUTIONS** M.A.C. conceived, designed and conducted the experiments, performed the data analysis and wrote the manuscript. G.J.T. contributed to experimental design. B.J.V., B.E.B. and G.J.T. designed the custom OCT system and software and M.J.S. built the custom OCT system. B.J.V. and M.J.S. provided technical support. B.E.B. and G.J.T. supervised the research. All authors edited and reviewed the final manuscript. **SUPPLEMENTARY MATERIAL** Supplementary material for this article is available at <http://dmm.biologists.org/lookup/suppl/doi:10.1242/dmm.005231/-/DC1> [^1]: Present address: Department of Diagnostic Radiology, Yale University School of Medicine, New Haven, CT 06520-8043, USA

Pulmonary arterial hypertension (PAH) affects predominantly women, yet women with PAH have better survival than men with PAH. Women have a more robust response to endothelin receptor antagonism, and treatment-associated improvements in right ventricular (RV) function account for survival differences between men and women with PAH ([@bib1], [@bib2]). Although female sex has been linked to the development of PAH, response to therapy, and better RV function, the role of estrogen in PAH is not completely understood (a comprehensive review of animal and human studies in this area is discussed by Hester and colleagues \[[@bib3]\]). Higher levels of circulating 17β-estradiol (E2), the most potent estrogen, are associated with increased risk for PAH and more severe disease in both men and postmenopausal women ([@bib4]--[@bib6]), and variations in E2 metabolism influence the penetrance of heritable PAH ([@bib7]). Although these human observations consistently demonstrate sexual dimorphism in PAH and suggest a critical role for E2 in pulmonary vascular disease, they have generated more questions than answers about the effect of E2 across the cardiopulmonary interface. Experimental models of pulmonary hypertension (PH) have provided fundamental mechanistic insight; however, as in humans, these studies fall short in crossing the translational divide, and the estrogen puzzle of pulmonary vascular disease remains unsolved. In contrast to humans, female sex in hypoxia-induced and monocrotaline-induced PH (MCT-PH) models is protective. Endogenous and exogenous E2 has been shown to prevent, mitigate, and reverse PH in these models ([@bib8], [@bib9]). In MCT-PH, E2 is associated with increased nitric oxide and prostacyclin levels and decreased macrophage infiltration ([@bib10]). In contrast, transgenic and Sugen-hypoxia (SuHx) models of PH demonstrate a female bias similar to humans, and as in humans, SuHx females have better RV function and improved survival compared with males ([@bib11]--[@bib14]). In the SuHx model, E2 promotes a proinflammatory and proangiogenic response but inhibits RV fibrosis, decreases collagen deposition in the myocardium, and increases proximal pulmonary artery (PA) compliance ([@bib12], [@bib13]). These studies demonstrate that E2 may have differential effects on the pulmonary vasculature as compared with the RV, but the effects of E2 on RV afterload and on the mechanics of the pulmonary vasculature have not been studied. In this issue of the *Journal*, Philip and colleagues (pp. [371--374](10.1164/rccm.201906-1217LE)) describe the effect of endogenous and exogenous E2 on the prevention of SuHx PH ([@bib15]). Pulmonary vascular mechanics were compared in female rats with intact cyclical endogenous E2 and ovariectomized rats with and without exogenous E2 supplementation. Rats that received continuous exogenous E2 were protected from PH with similar levels of RV systolic pressure and intermediate and distal PA impedance compared with the intact and ovariectomized rats. Exogenous E2 prevented an increase in the transpulmonary gradient, preserved distal PA distensibility, and was associated with a 60% reduction in PA wall remodeling. Treatment with a rho kinase inhibitor in the absence of continuous exogenous E2 demonstrated that SuHx-induced increases in RV afterload were driven by vasoconstriction. This carefully executed study not only adds new knowledge in its use of pulsatile pulmonary vascular mechanics as surrogates of RV afterload but also provides critical insight into the protective role of E2 on impedance, distensibility, and remodeling in the distal PA. The distinction between biologic endogenous E2 exposure in intact animals versus exogenous E2 treatment in ovariectomized animals is an important one, as is isolating vasoconstriction. The findings of this study mirror limited observational data in humans that exogenous hormone therapy may prevent PH in postmenopausal women with systemic sclerosis ([@bib16]). Similarly, higher levels of E2 in hormone therapy users have been associated with better RV function in postmenopausal women without clinical cardiovascular disease ([@bib17]). What pieces remain to solve the estrogen puzzle in PAH? Results from the preventative strategy used here will need to be replicated in rescue experiments and compared in male animals and ideally in young and aged animals. Downstream genomic effects of E2 on biomechanics and cardiopulmonary function may also differ from nongenomic vasodilatory effects. Numerous human and experimental observations implicate female sex and E2 (as well as other sex hormones, their metabolism, receptor signaling, and sex chromosomes) in the pathogenesis of PAH. This study is a strong contribution to accumulating evidence of a pleiotropic role of E2 on tissues, which may explain the observed contradictions in animal models and human studies to date. The biggest challenge remains bench-to-bedside and bedside-to-bench translation as hormones are tested as therapeutic targets to improve the lives of women and men living with PAH. Supported by NIH/NHLBI grants K23 HL141584 and R01 HL141268. Originally Published in Press as DOI: [10.1164/rccm.201910-1982ED](http://dx.doi.org/10.1164/rccm.201910-1982ED) on November 5, 2019 [**[Author disclosures](http://www.atsjournals.org/doi/suppl/10.1164/rccm.201910-1982ED/suppl_file/disclosures.pdf)**]{.ul} are available with the text of this article at [www.atsjournals.org](http://www.atsjournals.org).

An avalanche of literature exists on the history of neurosurgery in Africa from its evolution in the ancient Egyptian epoch through pre-colonial era till today\'s modern neurosurgery. A worldwide neurosurgeon to population ratio is 1:230,000; in Africa, it is 1:1,352,000 (1:338,000 in the north, 1:620,000 in the south, and 1:6,368,000 in sub-Saharan Africa). Even the blind can see glaringly that this marked discrepancy across Africa warrants a dire need for "an African Solution to an African Problem" -- "this together we can provide!" In this editorial, we will elucidate on the current challenges of neurosurgery in most African countries using the success story of our department of neurosurgery as a simple illustrative model that can be replicated, adapted, and adopted in any resource-limited setting. The history of neurosurgery dates as far back as the ancient Egyptian time. Situated at the horn of Africa, Egypt stands at the cross-roads between two great continents that in recent years have witnessed a simultaneous admixture of the epidemiologic and cultural transitions compounded by political turmoil that has undermined health care funds, creating hindrances toward the development of neurosurgery. Its strategic socioeconomic and geopolitical position explains why Egypt has long been enjoying a leading role in the development of neurosurgery in its vicinities. In the last 50 years, we have witnessed major advances and transitions in neurosurgery, which was once an appendage of general surgery but has now evolved into a full-fledged specialty of its own right and even furthermore into complex subspecialties which are on a continuous refinement. Despite the financial and material limitations, our huge manpower in a continent where the practice of neurosurgery is almost a dream in some countries has been our major strength. The high continental demand coupled with local pressing needs pushed us to surpass these impediments by setting up a simple training model. The promotion of adequate training for both nationals and continental trainees to be optimally acquainted to use the available technology and providing as much as possible state-of-the-art practice was our target. Currently our department runs with a staff of 20 full professors, 10 associate professors, 20 lecturers, and 17 assistant lecturers. Our present trainees include 14 nationals, 1 Cameroonian, 2 Yemenis, and 2 Libyans. The training program is in two stages: A masters in neurosurgery for 5 years with minors in neurology or general surgery and doctorate in neurosurgery for 3-4 years. The provision of fellowships, short training courses, regular trimester, annual conferences, and well-planned daily activities is followed by log books given to the trainees, and then regular valuation exams. The integration of the training of nurses and paramedical staff in our faculty program was parallel. Divisions and subspecialties of skull base, (endo) vascular, functional, pediatric, spine, trauma, and radiosurgery, with their integration into the national health system, improved both our service and training system. The latter warranted the upgrading of our operating theaters from an initial two to seven well-equipped rooms with up-to-date microscopes, endoscopes, ultrasound, etc. The funding has mainly been from NGOs and government subsidies; seeing our sincerity, they shared in the progress. We equally augmented the ward capacity to above 120 beds and 20 neurosurgical intensive care facilities with patient care shared with intensive care physiciansOne of our strengths has been the establishment of a good working atmosphere with related specialties like the ENT, Ophthalmologists, Radiologists, Pathologists, etc., which has strengthened our team and potentiated our outcome and outputTo achieve a modern proper patient follow-up and to build up a data bank, we revised and upgraded our filing system from a paper archive to an electronic databaseTo facilitate early exposure of our trainees to research and updating them with technologic advances, a high-speed internet has been set up in the department through which they have free access to most journals and virtual librariesA cadaveric dissection laboratory to shorten the learning curve for our young trainees and stimulate the development of new surgical techniques is in the pipelineThe ultimate demand for the progress of our field was carried by the senior neurosurgeons and the Egyptian Society of Neurological Surgeons (ESNS)In 1998, we had only 6 centers; now we have 15 centers and the number of qualified neurosurgeons has increased from 165 to 520. We are proud to have two gamma knife centers in Cairo, sharing in treating and training. It will be gross pretense if we claim that we can do it all alone. The support of established societies like the Asian, European, and American Neurosurgical Societies is indispensable to complement and supplement the training we provide. Trainees in developing countries should at the end of their course be granted scholarships in different centers to allow them gain firsthand exposure to components not present in their home or regional training centers. Young neurosurgeons should be encouraged to attend international meetings and have sessions to present their activities and also their hopes. This is already being done; however, much drive is still needed. These achievements are not an easy walk in the park. We have been limited mainly by insufficient resources, but with a post-revolution Egypt and a rapidly sensitized and advancing Africa, our stakes are high and our dream to transform neurosurgery in Egypt and our continent as a whole to advanced levels is an unfading objective. While we commit our resources for training, the burden and challenges of developing neurosurgery across the continent remains a major responsibility of the individual countries as well as a challenge for the entire neurosurgical community.

###### Strengths and limitations of this study - This is the first study to assess prevalence of restless legs syndrome (RLS) in patients with clinically established diagnosis of functional movement disorders (FMD) based on positive diagnostic signs of inconsistency and incongruence. - Case--control study design with relatively large sample of patients with FMD and matched controls was used. - Updated International RLS Study Group criteria for the diagnosis of RLS were applied. - Objective actigraphic assessment of PLM was performed using validated methodology to control for false-positive RLS cases. - The main limitation of the study results from introduction of new polysomnographic criteria for PLM in the course of the study which could not be implemented in this actigraphic study. Introduction {#s1} ============ Functional (psychogenic) movement disorders (FMD) are common neurological manifestations, clinically defined by abnormal movement control that is significantly altered by distraction or non-physiological manoeuvres and that is clinically incongruent with movement disorders known to be caused by neurological disease.[@R1] Patients with FMD frequently present with a complex and variable motor phenotype. In addition, they often report sensory symptoms and pain in multiple body regions, mood disorders, fatigue and sleep problems.[@R2] Restless legs syndrome (RLS) is defined by an urge to move a body part (usually the lower limbs) typically accompanied by a wide range of sensory symptoms.[@R4] Higher prevalence of RLS has been reported in numerous and pathophysiologically heterogeneous conditions. However, a recent systematic review of evidence confirmed higher prevalence of RLS only in kidney disease and iron deficiency and possible association in some cardiovascular diseases in women, diabetes (and neuropathy), migraine and dopaminergic treatment in Parkinson's disease.[@R5] RLS prevalence in patients with FMD has not been studied thus far. We hypothesised that patients with FMD may have a higher prevalence of RLS than is reported in the general population, and some of the sensory and motor phenomena observed in patients with FMD may be due to unrecognised RLS. The main objective of this study was to assess the prevalence of RLS in a group of patients with FMD and/or functional weakness, compared with a matched control group. RLS was defined according to the current diagnostic criteria.[@R4] In addition, the presence of periodic limb movements (PLM) that are considered as an objective biomarker of RLS was assessed by actigraphy.[@R4] Illness beliefs and unexplained somatosensory input may play an important role in the development of functional neurological symptoms.[@R7] To find out whether RLS could contribute to the later development of FMD, we also aimed to analyse the time relationship between the onset of FMD and RLS. Additionally, factors that can affect RLS prevalence (age, organic comorbidities including migraine, depression, anxiety and medication) were also considered in the analysis.[@R8] Furthermore, we studied the relationship between the FMD phenotype, the prevalence of RLS/PLM and non-motor symptoms such as daytime sleepiness, fatigue and sensory symptoms and pain in lower limbs. Methods {#s2} ======= Subjects {#s2a} -------- We recruited 115 consecutive patients (94 females, mean age (SD) 44.8 (13) years) diagnosed with clinically definite FMD according to the diagnostic criteria of Gupta and Lang between October 2014--December 2016.[@R11] The diagnosis of FMD was based on detailed clinical interviews and examination by an experienced movement disorders specialist (TS) finding positive signs of functional weakness and/or abnormal movements inconsistent in time and incongruent with known movement disorders.[@R1] FMD duration was recorded. During the same time period, 76 unrelated sex-matched and age-matched control subjects were recruited (66 females, mean age (SD) 44.2 (11) years) from database of healthy subjects willing to participate in clinical studies. In all controls, a complete medical history was recorded, and full neurological examination was performed by TS or MS. Only controls without neurological symptoms or signs of nervous system disorder affecting motor function (except for those corresponding to RLS) were included in the study. Both patients and controls with known kidney disease, iron deficiency or pregnancy were not included in the study.[@R5] Motor symptoms {#s2b} -------------- All motor symptoms present during the examination in each patient with FMD were recorded and phenomenologically classified as functional weakness, tremor, dystonia, myoclonus, gait disorder, parkinsonism, speech disturbance or eye movement abnormalities. The frequency of each phenotype was calculated. The presence or absence of any motor symptoms in lower limbs was recorded from face to face interview and physical examination. Organic multimorbidity and medication {#s2c} ------------------------------------- To evaluate the cumulative effect of organic comorbidities on RLS frequency, we used a Comorbidity Index by Szentkiralyi *et al* which was calculated as a sum of the following conditions: diabetes, hypertension, myocardial infarction, obesity, stroke, cancer, renal disease, anaemia, thyroid disease, migraine and depression. One point was assigned for each present condition.[@R9] Obesity was defined as current body mass index \>30 kg/m^2^. The presence of general medical conditions included in the Modified Comorbidity Index was based on patients' medical reports and clinical interview. Migraine was diagnosed from clinical interview according to current diagnostic criteria.[@R14] The subjects were considered to have depression either if they scored in Beck Depression Inventory (BDI-II) ≥14 or scored lower with a proven history of depression treated with antidepressants.[@R15] In each subject, we recorded presence or absence of medication possibly interfering with RLS and PLM prevalence. Use of any antidepressants, for example, selective serotonin reuptake inhibitors (SSRI), serotonin-norepinephrine reuptake inhibitors (SNRI), trazodone, mirtazapine or tricyclics in monotherapy or combination, was scored 1 if present, 0 if absent. Similarly, use of medication that may reduce RLS and PLM symptoms, for example, gabapentinoids (pregabalin and gabapentin), dopaminergic drugs (dopamine agonists and levodopa) and opioids, was scored as 1 if the medication was present, 0 if absent. Additionally, we also recorded the use of betablockers that have been reported as related to RLS and PLM.[@R10] The history of antidepressants administration initiated prior to RLS onset was taken into account for evaluation of RLS prevalence while the use of antidepressants in the time of actigraphy was considered for PLM prevalence analysis. Neurological comorbidities other than migraine were recorded and stratified according to their possible association with RLS and Neurological Comorbidity Index was calculated as follows[@R5]: the presence of a condition that has been associated with an increased frequency of RLS regardless of the level of evidence (ataxia, multiple system atrophy, multiple sclerosis, Parkinson's disease, peripheral neuropathy, polyneuropathy and radiculopathy, muscle disorder and stroke)---1 point; neurological comorbidity without association with RLS or no known neurological comorbidity (pure FMD)---0 points. RLS and PLM evaluation {#s2d} ---------------------- All the patients and control subjects were interviewed for the presence of RLS. Subjects were classified as RLS positive (RLS+) if they met all five revised criteria of the International Restless Legs Syndrome Study Group: (1) an urge to move the legs usually accompanied by uncomfortable and unpleasant sensations, (2) symptoms beginning or worsening during periods of rest or inactivity, (3) partial or total relief by movement, (4) occurrence or worsening in the evening or night and (5) occurrence of the above features is not solely accounted for as symptoms primary to another medical or a behavioural condition. All known RLS mimics were considered and excluded.[@R4] Subjects with inconsistent reports on RLS symptoms and evident suggestibility during interview were classified as functional mimics and labelled as RLS negative (RLS−). Cases with atypical presentation of RLS symptoms and/or with possible confounds were additionally interviewed by an experienced sleep disorder specialist (KŠ, DK) and if there were any diagnostic doubts, the case was labelled as RLS−. In RLS+ subjects, the duration of their RLS symptoms, a family history of RLS in first-degree relatives and the severity of their RLS according to the International Restless Legs Scale (IRLS) were recorded.[@R16] We determined the number of RLS+ patients in whom motor symptoms localised in lower limbs were attributable exclusively to RLS, that is, were characterised by an urge to move the legs, beginning or worsening during periods of rest or inactivity, relieved by movement, with occurrence or worsening in the evening or night. For detection of PLM, all control subjects and 96 patients with FMD, regardless of RLS status, underwent actigraphy from big toes for three consecutive nights at home. In 19 patients with FMD, actigraphic assessment was not available because of their refusal to participate or due to complicated logistics. The actigraphic measurement and assessment were performed as published previously.[@R17] We have evaluated the total number of limb movements and the total number of PLM. The PLM index (PLMi, number of individual movements per hour) was calculated by dividing the total number of PLM by number of hours in bed according to the subjects' diary. For statistical analysis, we considered the highest PLMi value out of three nights in each subject. A cut-off PLMi ≥22.5/hour (corresponding to PLMi ≥15/hour in polysomnography) consistent with the International Classification of Sleep Disorders 3rd edition criteria was used for clinically relevant PLM positivity.[@R6] Sensory symptoms and pain in lower limbs {#s2e} ---------------------------------------- The presence or absence of sensory disturbances (ie, paresthesias or dysaesthesias) in lower limbs was recorded from face to face interview and physical examination in each subject. We also determined the number of RLS+ subjects with sensory symptoms in lower limbs exclusively attributable to RLS, that is, accompanying the urge to move the legs, beginning or worsening during periods of rest or inactivity, relieved by movement, with occurrence or worsening in the evening or night. The presence of pain in lower limbs was recorded if subjects marked lower limb region on a PainDetect body figure as location of an important pain.[@R18] Additionally, the mean pain intensity in relation to the body parts marked on the body figure within the last 4 weeks was evaluated using a Visual Analogue Scale (0=no pain, 10=maximum pain). Questionnaires for non-motor symptoms {#s2f} ------------------------------------- Along with the BDI-II, the subjects also completed self-administered questionnaires to determine the prevalence of anxiety traits (State-Trait Anxiety Inventory, STAI X2), chronic fatigue (Fatigue Severity Scale) and daytime sleepiness (Epworth Sleepiness Scale, ESS).[@R19] Statistics {#s2g} ---------- Statistical analyses were performed using Dell Statistica V.13 (Software.dell.com, Dell, 2016). Testing for normality was performed using the Shapiro-Wilks W test. For descriptive statistics, continuous variables were summarised as mean±SD or median±IQR, whereas categorical variables were summarised as number of subjects and percentage. Continuous variables were compared using Student's t-test or Mann-Whitney U test as appropriate, while categorical variables were compared using χ^2^ statistics. We used logistic regression models to assess the association between demographic, clinical and actigraphic variables and PLMi ≥22.5/hour. Statistical significance was defined as alpha 0.05, and the Bonferroni correction for multiple testing was used where necessary. Patient and public involvement {#s2h} ------------------------------ No participants were involved with the study design. Results {#s3} ======= Clinical characteristics {#s3a} ------------------------ Demographic and clinical characteristics from 96 patients with FMD and 76 control subjects who completed the study including actigraphy are summarised in [table 1](#T1){ref-type="table"}. Nineteen patients did not complete actigraphy due to personal, technical or logistical reasons and were not included in the statistical analysis. ###### Demographic and clinical characteristics ------------------------------------------------------------------------------------------------------------------------------------------------------------------ FMD patients vs controls RLS+ vs RLS− FMD patients ---------------------------------------------------- -------------------------- --------------------------- -------------- ------------- ------------- ----------- **Subjects**  Age (years) (SD) 45.0 (13) 44.2 (11) 0.36 47.7 (11) 42.9 (14) 0.17  Females 83.3% 86.8% 0.52 88.1% 79.6% 0.27  Mean FMD duration (years) (SD) 6.0 (5) -- -- 7.0 (6) 5.1 (5) 0.03 **Motor symptom (present/present as predominant)**  Weakness 51.0%/31.3% -- -- 52.4%/28.6% 50.0%/33.3% 0.98/0.17  Gait disorder 53.1%/32.3% -- -- 76.2%/42.9% 59.3%/24.1%  Tremor 41.7%/22.9% -- -- 45.2%/19.0% 38.9%/25.9%  Dystonia 29.2%/9.4% -- -- 23.8%/9.5% 33.3%/9.3%  Myoclonus 10.4%/4.1% -- -- 4.8%/0.0% 14.8%/7.4%  Speech disturbance 9.4%/0.0% -- -- 11.9%/0.0% 7.4%/0.0% \-  Convergence spasm 5.1%/0.0% -- -- 8.2%/0.0% 2.9%/0.0% \- **Comorbidities**  **Comorbidity Index**   0 17.7% 44.7% 0.0008\* 9.5% 24.1% 0.33   1 35.4% 35.5% 42.9% 29.6%   2 33.3% 15.8% 35.7% 31.5%   3 7.3% 4.0% 7.1% 7.4%  \>3 6.3% 0.0% 4.8% 7.4%   Mean score (SD) 1.50 (1.1) 0.79 (0.9) 0.000009\* 1.55 (0.9) 1.48 (1.3) 0.56  **Migraine** 35.4% 15.8% 0.0039\* 45.2% 28.8% 0.08  **Depression** 56.3% 11.8% P\<0.00001\* 61.9% 51.9% 0.32  **Modified Comorbidity Index**   0 61.5% 60.5% 0.82 69.1% 55.5% 0.48   1 25.0% 27.6% 19.0% 29.6%   2 9.4% 10.5% 9.5% 9.3%   3 3.1% 1.4% 2.4% 3.7%  \>3 1.0% 0.0% 0.0% 1.9%   Mean score (SD) 0.57 (0.9) 0.53 (0.7) 0.93 0.45 (0.8) 0.67 (0.9) 0.27   **Neurological Comorbidity Index**   0 87.5% 97.4% 0.019 85.7% 88.9% 0.64   1 12.5% 2.6% 14.3% 11.1% **Medication**  Antidepressants 50.0% 6.7% P\<0.00001\* 54.8% 46.3% 0.41  Medication suppressing\ 39.6% 0.0% P\<0.00001\* 47.6% 33.3% 0.16  RLS/PLM  Betablockers 13.5% 5.3% 0.07 9.5% 16.7% 0.31 ------------------------------------------------------------------------------------------------------------------------------------------------------------------ All intergroup comparisons were performed using Mann-Whitney U test for quantitative parameters and χ^2^ for qualitative ones, p values stand for nominal uncorrected results. \*Significant results after correcting for three comparisons among demographics and for seven comparisons among comorbidities and medication (p\<0.05). FMD, functional movement disorder; PLM, periodic limb movements; RLS, restless legs syndrome. There was a difference between the two groups in the Comorbidity Index which was higher in the patients group. This difference in the Comorbidity Index was due to significantly higher prevalence of migraine and depression in patients with FMD. Therefore, to control for somatic comorbidities we calculated a Modified Comorbidity Index (without including migraine and depression). No difference between patients with FMD and controls was found for the Modified Comorbidity Index, as well as for the rates of each of the somatic organic comorbidities (ie, diabetes, hypertension, myocardial infarction, obesity, stroke, cancer, anaemia and thyroid disease), and for the Neurological Comorbidity Index. Without considering migraine, 72 out of 96 patients had pure FMD, while in 24 patients (25.0%) we observed a comorbid organic neurological condition. Two control subjects had incidental peripheral neuropathy without an impact on motor function. See [table 1](#T1){ref-type="table"} and online [supplementary material](#SP1){ref-type="supplementary-material"} for details. Higher percentage of patients than controls was taking both antidepressants and drugs suppressing RLS/PLM. No difference was found in the use of betablockers. 10.1136/bmjopen-2018-024236.supp1 RLS and PLM {#s3b} ----------- Results from the clinical RLS assessment and actigraphy are shown in detail in [tables 2 and 3](#T2 T3){ref-type="table"}. RLS was diagnosed in 43.8% of patients and in 7.9% of controls (p\<0.0001, corrected). The mean IRLS score was higher in the patients group. Sixty per cent of patients but no control subject had RLS symptoms present for more than 4 days in a week, 37.5% of patients but no control subjects had RLS symptoms duration longer than 3 hours per 24 hours. The prevalence of RLS, clinically relevant PLM (PLMi ≥22.5/hour) and combination of both were all significantly higher in patients with FMD than in controls. Patients with FMD had higher mean PLMi than controls (p=0.0003). RLS− controls had lower PLMi than RLS− patients with FMD (p=0.03 uncorrected, Mann-Whitney U test). There was no significant difference in the proportion of subjects with PLMi ≥22.5/hour between all controls and RLS− patients with FMD. ###### RLS assessment FMD patients vs controls ------------------------------------------------------------- -------------------------- --------------- ------------- **RLS+ (N)** 42 6 \<0.00001\*   % (95% CI) 43.8 (34 to 54) 7.9 (3 to 17) **Mean IRLS score in RLS+ (SD)** 23.6 (7) 7.0 (6) 0.00002\* **Frequency of RLS (number of days in 1 week)**  ≤1 12.5% 60% 0.03  2--3 27.5% 40%  4--5 30% 0%  6--7 30% 0% **Duration of RLS symptoms (number of hours per 24 hours)**  \<1 20% 80% 0.04  1--3 42.5% 20%  3--8 25% 0%  ≥8 12.5% 0% **Positive family history of RLS in RLS+†** 30.4% 33.3% 0.92 **Mean RLS duration (years) (SD)**  7.5 (8) 5.8 (6) 0.64 **Mean age at RLS onset (years) (SD)** 40.2 (13) 34.5 (9) 0.30 **RLS vs FMD onset**   RLS before FMD 26.8% -- --   Concurrent onset 26.8% -- --   RLS after FMD 46.3% -- -- \*Significant results (p\<0.05) after correcting for seven comparisons. †During interview, 22 patients did not know if RLS was present in their family history, and 18 patients had no family history of RLS. P values stand for nominal uncorrected results. FMD, functional movement disorder; IRLS, International Restless Legs Scale; RLS, restless legs syndrome. ###### Actigraphy FMD patients vs controls RLS+ vs RLS− FMD patients ------------------------------- -------------------------- --------------------------- ----------------- ----------------- ----------- ----------- **Mean PLMi (PLM/hour) (SD)** 21.2 (19) 11.8 (14) 0.00005\* 28.3 (19) 15.7 (17) 0.00004\* **PLMi ≥22.5/hour (N)** 30 9 0.003\* 20 10 0.002\*   % (95% CI) 31.3 (23 to 41) 11.8 (8 to 21) 47.6 (33 to 62) 18.5 (10 to 31) **RLS+/PLMi ≥22.5/hour (N)** 20 2 0.0002\* -- -- --   % (95% CI) 20.8 (14 to 30) 2.6 (1 to 9) -- -- -- \*Significant results after correction for five comparisons (p\<0.05). All intergroup comparisons were performed using Mann-Whitney U test for quantitative parameters and χ^2^ for qualitative one. P values shown are nominal uncorrected results. FMD, functional movement disorder; PLM, periodic limb movements; PLMi, periodic limb movement index; RLS, restless legs syndrome. The duration of FMD symptoms did not significantly differ between the RLS+ and RLS− subgroup, although there was a trend towards longer FMD duration in the RLS+ group. In 2 out of 10 patients with positive family history, RLS preceded FMD onset. Regarding the motor phenomena present in lower limbs in 42 RLS+ patients, 27 patients (64.3%) presented with functional gait disorder and/or functional weakness but no abnormal movements in lower limbs; 11 patients (26.2%) presented with abnormal movements, that is, functional tremor, dystonia or myoclonus in the lower limbs; 4 patients (9.5%) had no functional motor symptoms in lower limbs and their motor symptoms in the lower limbs were exclusively related to RLS. Sensory symptoms and pain in lower limbs {#s3c} ---------------------------------------- Results of the assessment of sensory symptoms and pain in lower limbs are shown in [table 4](#T4){ref-type="table"}. Both sensory symptoms and pain were more frequently found in patients with FMD than in controls. Sensory symptoms related exclusively to RLS were present in 12.9% of all patients with FMD (ie, in 30.4% of RLS+ patients). ###### Sensory symptoms and pain FMD patients vs controls RLS+ vs RLS− FMD patients --------------------------- -------------------------- --------------------------- ------------- --------- --------- ---------- **Sensory symptoms**  Number of subjects 96 76 -- 42 54 --  Lower limbs 83.3% 9.2% \<0.0001\* 97.6%† 72.2% 0.0005\*  Exclusively to RLS 12.9% 7.9% 0.087 30.4% -- -- **Pain**  Number of subjects 90 76 -- 38 52 --  Lower limbs 62.3% 11.8% \<0.0001\* 75.0% 48.2% 0.009\*  Mean VAS in 4 weeks (SD) 5.6 (3) 1.4 (2) \<0.00001\* 6.1 (2) 5.3 (3) 0.40 All intergroup comparisons were performed using Mann-Whitney U test for quantitative parameters and χ^2^ for qualitative ones, p values shown are nominal uncorrected results. \*Significant results after correction for multiple testing for four comparisons (p\<0.05). †Only one RLS+ patient did not report sensory symptoms as part of the RLS symptomatology, nevertheless he fulfilled the RLS criteria and his maximum PLMi was 33. FMD, functional movement disorder; PLMi, periodic limb movement index; RLS, restless legs syndrome; VAS, Visual Analogue Scale. Questionnaires for non-motor symptoms {#s3d} ------------------------------------- Patients with FMD suffered from significantly more severe depression, anxiety, fatigue and daytime sleepiness than control subjects. The scores are shown in [table 5](#T5){ref-type="table"}. ###### Non-motor symptoms subjective questionnaires FMD patients vs controls RLS+ vs RLS− FMD patients ------------------------------ -------------------------- --------------------------- ------------- ----------- ----------- ------- **BDI-II (depression)**   Total (N) 91 76 39 52  ≥14 53.9% 10.5% \<0.0001\* (64.1%) (46.1%) 0.044   Mean score (SD) 18.3 (14) 6.0 (6) \<0.00001\* 22.1 (14) 15.8 (13) 0.047 **STAI-X2 (anxiety trait)**   Total (N) 90 76 40 50   Mean score (SD) 47.0 (12) 37.8 (10) \<0.00001\* 49.8 (12) 44.7 (12) 0.044 **FSS (fatigue)**   Total (N) 90 76 38 52   Mean score (SD) 5.5 (1) 3.0 (1) \<0.00001\* 5.8 (1) 5.2 (2) 0.17 **ESS (daytime sleepiness)**   Total (N) 88 76 37 51   Mean score (SD) 10.6 (5) 6.5 (4) \<0.00001\* 11.8 (5) 9.7 (5) 0.07 All intergroup comparisons were performed using Mann-Whitney U test for quantitative parameters and χ^2^ for qualitative ones, p values shown are nominal uncorrected results. \*Significant results after correction for multiple testing for four comparisons (p\<0.05). BDI-II, Beck Depression Inventory II; ESS, Epworth Sleepiness Scale; FMD, functional movement disorder; FSS, Fatigue Severity Scale; RLS, restless legs syndrome; STAI-X2, State/Trait Anxiety Inventory X2. Comparison between RLS+ and RLS− patients with FMD {#s3e} -------------------------------------------------- In the FMD group, no differences in demographic or clinical parameters were found between RLS+ and RLS− groups. No significant association between a specific motor phenotype and RLS status was observed ([table 1](#T1){ref-type="table"}). On actigraphy, the RLS+ group showed a higher mean PLMi and a higher proportion of PLMi ≥22.5/hour than the RLS− group showed. Out of 96 patients with FMD who underwent actigraphy, 20 patients had both RLS and a PLMi ≥22.5/hour. Both sensory symptoms and/or pain in lower limbs were more frequent in RLS+ than in RLS− patients. No difference in mean pain intensity in last 4 weeks was found in RLS+ and RLS− groups. No difference between the RLS+ and RLS− patients group was found in scales rating depression, anxiety and sleepiness and fatigue. Additional analysis {#s3f} ------------------- Logistic regression among assessed clinical motor and non-motor parameters did not find any significant predictors of RLS+ status in patients with FMD (p=0.5). All above presented results remained unchanged after patients with FMD and organic neurological comorbidities were excluded. Discussion {#s4} ========== This is the first case--control study to assess the prevalence of RLS in patients with FMD. Forty-three per cent of patients with FMD met the diagnostic criteria for RLS while in the control group RLS was found in 8%, similarly to previous findings in population studies.[@R22] The RLS presence was not associated with medication or organic comorbidities in patients with FMD. However, only 21% of patients with FMD had both RLS and clinically relevant PLM detected by actigraphy. This result suggests there might be false-positive cases either due to suggestibility or due to functional symptoms mimicking RLS. Known mimics included in the new RLS criteria were carefully excluded. Nevertheless, suggestibility inherent in an interview on symptoms that are not reported spontaneously and/or general tendency of patients with FMD to over-report symptoms might be a source of a diagnostic bias. Additionally, and perhaps more importantly, functional motor and sensory phenomena may mimic RLS symptoms in this group of patients. However, even if we take into account only RLS cases with comorbid clinically relevant PLM, our findings still suggest FMD is associated with a twofold higher prevalence of RLS compared with general population. Also in healthy controls, we detected both RLS and clinically relevant PLM only in a small proportion (2.6%) of cases. RLS diagnosis was further supported by a higher occurrence of actigraphically detected PLM in RLS+ patients in comparison to both RLS− patients and controls. The mean PLMi was higher in RLS− patients than in the control group, suggesting an association of PLM and FMD independent of RLS occurrence. Alternatively, some false-RLS− cases could be also included in this group, especially in the subgroup with PLMi ≥22.5/hour. However, PLM may be associated with antidepressants intake which was higher in the FMD group and thus could contribute to PLM expression.[@R10] Notably, no differences between RLS+ and RLS− patients were found in organic neurological and/or somatic comorbidities, medication use or gender, suggesting that these factors do not affect RLS expression in patients with FMD. The rate of familial RLS in our patients with FMD and control subjects was rather low compared with previous reports; however, in majority of cases the family history could not be reliably established. Low rate of familial cases would suggest stronger role of genetic factors in the development of RLS can be supposed only in a minority of patients with FMD.[@R4] We did not find a relationship between the FMD phenotype and the presence of RLS or PLM. This result is in line with the current neurobiological model of FMD which suggests there may be common mechanisms acting in FMD, regardless of motor phenotype.[@R7] To our surprise, in two-thirds of the RLS+ patients, RLS started concurrently with or after the FMD onset. In these patients, the burden resulting from a functional disorder might have contributed to RLS development rather than vice versa. However, FMD duration did not appear as a significant risk factor for RLS development. The finding of the association between FMD and RLS/PLM suggests these conditions might share some common pathophysiological mechanisms. The genetic and molecular basis of FMD is still largely unknown. In RLS some pathophysiological genetic and molecular factors including dysfunction in iron homeostasis and iron--dopamine interaction, role of inflammation and hypoxia have been already elucidated.[@R5] In RLS and PLM, clinical symptoms and also some neurophysiological abnormalities, such as short intracortical inhibition (SICI) in RLS and lower threshold for the production of spinal flexor reflexes in PLM, can be improved with dopaminergic drugs.[@R25] A loss of SICI has also been demonstrated at least in functional dystonia.[@R27] In Parkinson's disease, where there is also loss of SICI, the deficit can be improved with dopaminergic drugs.[@R28] Hence, a common state of reduced inhibition might be postulated due to dopamine deficiency. Additionally, we may assume, similar environmental factors that trigger the manifestation of RLS symptoms, for example, cumulative number of diseases in a single individual, could also increase the risk of FMD in vulnerable individuals.[@R5] Indeed, organic neurological comorbidity was higher in patients with FMD than in controls. Migraine was the most frequent comorbidity. Headache seems to be a common problem in patients with FMD, and migraine can even play a role as precipitating factor in functional motor symptoms development.[@R29] Twenty-five per cent of patients were also diagnosed with other organic neurological condition. The co-occurrence of organic and functional disorders has been previously detected in patients diagnosed with psychogenic movement disorders and also in cohorts of patients with organic neurological disorders including movement disorders.[@R2] RLS severity according to the IRLS scale including reported weekly frequency and daily duration of the symptoms was high in the patients group. Therefore, we believe RLS should be taken into account as a serious comorbid condition, despite the general assumption that patients with FMD tend to over-report symptoms severity. In line with previous studies, sensory symptoms and pain were present in the vast majority of patients with functional neurological disorders.[@R29] Patients with RLS reported more sensory symptoms and pain in lower limbs than patients without RLS. Sensations in lower limbs clinically attributable exclusively to RLS were only found in a minority of RLS+ patients. RLS sensory symptoms thus constituted a part of a wider spectrum of sensory disturbances in majority of patients. Similarly, in the vast majority of patients there was an anatomical overlap in distribution of the functional motor symptoms and motor symptoms related to RLS. However, the majority of patients with FMD presented with abnormal patterns of gait and weakness in lower limbs which are easily distinguishable from motor phenomena related to RLS. Association between RLS and pain in the lower limbs is difficult to interpret. In this study, we did not consider the aetiology and detailed characteristics of pain. The uncomfortable sensations are described as painful in up to 30%--50% patients with RLS.[@R4] Therefore, at least a part of what has been previously considered as chronic pain actually could be attributable to RLS in patients with FMD. In accordance with the previous studies on FMD and other somatoform disorders, patients with FMD had higher scores of depression, anxiety, fatigue and daytime sleepiness than controls.[@R3] Epidemiological evidence also suggests that there is an association between mood disorders and both RLS and functional symptoms.[@R3] The causality of these factors is, however, impossible to determine. Our study has limitations. Currently, polysomnography is the gold standard and only clinically acceptable means of quantifying periodic limb movements in sleep.[@R6] We used a big toe-worn actigraphy to objectively assess nocturnal leg movements because of the logistic limitations of polysomnography use in this cohort size. However, the same actigraphic methodology was previously validated against polysomnography on the same night in our own laboratory which increases the reliability of this method.[@R17]Another limitation is that new polysomnographic criteria for PLM have been recommended in the course of the study.[@R35] However, these new criteria could not be implemented in this actigraphic study using validated methodology based on older criteria for PLM detection.[@R17] Additionally, considering the low absolute number of RLS+ subjects in the control group, comparisons of parameters related to RLS (such as positive family history or RLS duration) suffer from low statistical power. Finally, apart from the ESS, we did not measure sleep quality using other subjective or objective measures. Sleep disturbance or restriction often exacerbates or triggers RLS, and so it may be that FMD RLS+ patients have poorer sleep than RLS− patients. RLS would be then associated with poor sleep related to psychobehavioural profiles rather than being directly associated with FMD, and those with an underlying predisposition to RLS reach the threshold of clinical RLS. Further studies are needed to determine the RLS and PLM prevalence in FMD. The diagnosis of RLS seems to be challenging. Controlling for inner consistency in reporting RLS symptoms over time may help to differentiate RLS from functional symptoms. The efficacy of dopamine agonists would represent a supportive feature for the diagnosis and should be tested in clinical practice. However, evidence for dopamine agonists efficacy in this group should be studied in randomised control trials given the possible strong placebo effect in this group of patients. On the other hand, based on these findings, functional motor and sensory symptoms could be considered as additional RLS mimics. Conclusions {#s5} =========== In conclusion, we found increased prevalence of RLS in patients with FMD. The presence of RLS was associated with a higher proportion of clinically relevant PLM and more frequent sensory symptoms and pain in lower limbs. However, further studies are needed to confirm these findings and determine the rate of false-positive diagnosis, that is, RLS mimics in this group of patients. This finding could have several implications. RLS may be under-recognised in patients with FMD. Correct diagnosis and appropriate treatment of comorbid RLS may be important in patients with FMD as RLS is a common medical reason for sleep disturbance and impaired quality of life.[@R36] Additionally, association between FMD and RLS/PLM might be potentially relevant for future research of molecular and genetic factors in the pathophysiology of FMD which is still largely unknown. Supplementary Material ====================== ###### Reviewer comments ###### Author\'s manuscript We thank Dana Fialová, Irena Stárková and Dana Suchá for administrative support. **Patient consent for publication:** Not required. TS and MS contributed equally. **Contributors:** All authors meet the ICMJE criteria for authorship. TS was responsible for the concept and design of the study, clinical data acquisition, data analysis and interpretation, and drafting of the original report which was reviewed and revised by all co-authors. MS was responsible for actigraphic data collection, all actigraphic analysis, data interpretation and drafting of the original report. DK was responsible for study design, clinical data acquisition, data analysis including statistics and interpretation and drafting of the original report. KŠ was responsible for clinical data acquisition, data analysis and interpretation. MH was responsible for data analysis and interpretation. ER was responsible for data analysis and interpretation. Funding was obtained by TS and ER. **Funding:** This study was supported by Ministry of Health of the Czech Republic, grant AZV ČR 16-29651. MH is supported by the NINDS Intramural Program. **Competing interests:** None declared. **Ethics approval:** Ethics Committee of the General University Hospital in Prague (approval number 26/15 Grant). **Provenance and peer review:** Not commissioned; externally peer reviewed. **Data sharing statement:** There are no additional unpublished data from the study.

Introduction {#sec1-1} ============ Reactive oxygen species (ROS), such as superoxide (O2^-^) and hydrogen peroxide (H2O2), are frequently generates during metabolic pathways in all living organisms. Expose to normal physiological situation, cellular ROS generation is reverse equilibrium by the action of antioxidant enzymes and other redox molecules. On the other hand, high value of ROS production will cause to cellular damage, such as disruption to DNA, protein, and lipid membrane. Dangerous effects of high ROS level must be quickly inhibited from the cells by different antioxidant defense systems. The superoxide dismutase (SOD) enzyme is one of them, which catalyzes the detoxification of superoxide anion into H~2~O~2~ and molecular oxygen, is one of the most valuable antioxidative enzyme. The H~2~O~2~ is a potentially dangerous side effect of the metabolism of oxygen, a process that occurs in most living organisms after many metabolic pathways (Cingoz et al., 2014). By traditional agriculture, producing of biologically active secondary metabolites for commercial purposes is not a sufficient process. Generally, methods related with *in vitro* culture protocols have been investigated to develop the generation of these important medicinal compounds. Whensoever numerous abiotic and biotic factors were applied on plant tissue cultures, amount of secondary compounds was enhanced in a number of scientific reports (Cingoz et al., 2014). The H~2~O~2~ is one of the most important molecules which are caused to increase in plants by biotic and abiotic stress conditions. Higher levels of H~2~O~2~ generally terminate in toxicity to cell membrane system and damages to plant cells (He et al., 2009). However, the raised data evidence the biological activity of H~2~O~2~ as a stress signal molecule in plants (Hung et al., 2005) and H~2~O~2~ signaling functions significantly in plant development and defense system against environmental stresses (He et al., 2009; Cingoz et al., 2014). The H~2~O~2~ affects plant defense system and induced biosynthesis of glutathione-S transferase, CAT, SOD, and phenylalanine ammonia lyase (PAL) as antioxidant enzymes, transcription factors and defense proteins (Kovtun et al., 2000; Hung et al., 2005). Nonetheless, with specific elicitor, there are no data on *B. perennis* callus cultures in which improvements in phenolic production have been accomplished yet. Cingoz et al., (2014) showed that exogenous H~2~O~2~ treatment increased cardiotonic glycoside production in *Digitalis* species. Until now, there is no too much information on the use of H~2~O~2~ as an elicitor producing medicinally important secondary metabolites. *Bellis perennis* L. (common daisy) is a medicinal perennial herb in the family Compositae (Panda, 2004). Aerial parts with flowers of plant material have been used in the remedy of many types of wounds (Karakaş et al., 2012), headache (Uzun et al., 2004), common cold (Cakilcioglu et al., 2010), rheumatism, muscular pain (Morikawa et al., 2011), expectorant and anti-inflammatory (Siatka and Kasparova, 2010) in folk medicine for ancient times. Aerial part of the plant material include phenolic compounds (Karakas and Turker, 2013), triterpenoid saponins (Yoshikawa et al., 2008; Karakas et al., 2014), and essential oils (Kavalcioglu et al., 2010). The plant extracts obtained from *B. perennis* have many biological activities such as wound healing (Karakaş et al., 2012), spatial memory opener (Karakas et al., 2011), antitumor (Karakas et al., 2014), antimicrobial (Kavalcioglu et al., 2010), antihyperlipidemic (Morikawa et al., 2010), antioxidant (Siatka and Kasparova, 2010) and cytotoxic activities against some human cancer cell lines (Li et al., 2005; Karakas et al., 2015). Despite these scientific endeavors in the literature, the phenolic compounds accumulation and some activities such as the CAT, SOD, total phenolic, total flavonoid, proline activity in *in vitro*-propagated callus cultures of *B. perenn* is are still open to investigation because of a lack of the experimental research in the literature. Thus the aim of the present study was to identify how the phenolic compounds accumulation and the CAT, SOD, total phenolic, total flavonoid, proline activity in *B. perenn* is callus cultures affected abiotic stress factor such as oxidative stress by using H~2~O~2~ as an elicitation. Material and Methods {#sec1-2} ==================== Plant Material and Culture Conditions {#sec2-1} ------------------------------------- Fresh and wild-grown pedicel parts with flowers of *B. perennis* were collected from Abant Izzet Baysal University Campus, Bolu, Turkey in May. Surface sterilization of the plant materials was made according to Karakas and Turker (2013). The sterile pedicel explants for callus production were transferred to petri plates containing MS medium with 0.5 mgL^-1^ thidiazuron (TDZ) and 0.5 mgL^-1^ indole-3-acetic acid (IAA) for 30 days (Karakas and Turker, 2013). The calli (5 × 15 petri) of both treatments that have same sizes were put into the MS medium supplement with zero (control) and 10 mM H~2~O~2~ for forming oxidative stress for 10 hours. Petri dishes containing untreated (control) and oxidative stress treated calli were maintained at 23 ± 2 °C under a 16 h photoperiod from cool white fluorescent lamps. Preparation of Extracts {#sec2-2} ----------------------- Untreated (control) and oxidative stress treated calli were collected from *in vitro*-cultured common daisy. The freeze-dried callus materials were powdered with grinder and weighted. 100 mg of callus material of *B. perennis* was put into the plastic centrifuge tube including 2 mL of 80% methanol (MeOH) for 16 h at dark and room temperature. After 30 min application in an ultrasonic bath at 40 °C, it was centrifuged at 10,000 rpm for 12 min. The supernatant was filtrated and transferred to new centrifuge tube. The extract solutions of stress-treated and untreated (control) calli were kept at -80 °C deep-freeze until using in all analyses. Determination of Total Phenolic Content {#sec2-3} --------------------------------------- The amount of total phenolic content was determined in 80% methanol extracts of calli obtained from *B. perennis* using the Folin-Ciocalteu assay (Slinkard and Singleton, 1977), with some modifications. Therefore, 20 μL of each calibration solution or sample or blank, 100 μL Folin-Ciocalteu reagent (Sigma^®^) and 1.58 mL of deionized water were transferred in a 10 mL test tube and mixed thoroughly. After 2 minutes, 300 μL of 20% Na2CO3 solution was added in the test tube. The solutions were incubated at 20 ± 2 °C for 2 hours and measured the absorbance of each solutions at 765 nm against the blank (the "0 mL" solution) using the spectrophotometer (Hitachi U-1900, UV-VIS Spectrophotometer 200V, JAPAN). Gallic acid was used as a standard (0-500 mg/L). The total phenol content of 80% methanolic extracts from callus was expressed as mg gallic acid equivalents (GAE)/ g dried weight (dw). Three measurements of one sample were performed at same time. Determination of Total Flavonoid Content {#sec2-4} ---------------------------------------- Total flavonoid content of methanolic extracts of *B. perennis* calli was measured by aluminum chloride (AlCl3) colorimetric assay (Chang et al., 2002). Catechol was used as a reference flavonoid. The 0.0125 g catechol was dissolved in 25 mL of 80% methanol and this stock solution was adjusted to concentration as 500 mg/mL. In order to obtain calibration curve of catechol, 20, 40, 60, 80 and 100 mg/ mL concentrations were prepared. 500 μL of extract solution or standard solution of catechol was added to a 10 mL test tube containing 2 mL distilled water. At zero time, 150 μL of 5 % sodium nitrate (NaNO2) was added to the test tubes. After 5 min, 150 μL of 10% AlCl3 was added. At 6 min, 1000 μL of 1M sodium hydroxide (NaOH) was added to the mixture. Immediately, the reaction tube was diluted to volume 5 l with the addition of 1200 μL distilled water and thoroughly mixed. Absorbance of the mixture, pink in color, was determined at 510 nm versus a blank (Chang et al., 2002). Samples were analyzed in three replications. The total flavonoid content of methanolic extracts of *B. perennis* was expressed as mg catechol equivalents (CE)/ g dw. Enzyme Extraction and Protein Determination {#sec2-5} ------------------------------------------- Freeze-dried untreated (control) and oxidative stress treated calli (0.1 g) of *B. perennis* were grinded and mixed in 2 mL of ice cold 50 mM phosphate buffer (pH 7.0) which included 2 mM sodium ethylenediaminetetraacetic acid (Na--EDTA) and 1% (w/v) polyvinyl-polypirrolidone (PVP). Well mixed materials were centrifuged at 12,000 rpm and 4 °C for 10 min. Then, the callus extracts of *B. perennis* were kept at -80 °C for antioxidant enzyme analyses of the SOD and the CAT activity. Lowry method (Lowry et al., 1951) was used for the detection of the soluble protein content in oxidative stress treated and untreated callus extracts. The amount of the soluble protein in extracts was calculated with standard curve of bovine serum albümin (R^2^ = 0.997). Catalase (CAT; EC 1.11.1.6) activity {#sec2-6} ------------------------------------ CAT activity of the freeze-dried callus materials was described by using the method of Lartillot et al. (1988). Shortly, amount of the CAT activity was measured at 240 nm (using a specific absorption coefficient at 0.0392 cm^3^ mmol^-1^ H2O2) spectrophotometrically and the CAT activity was showed as mmol H~2~O~2~ decomposed/mg protein/min. Superoxide dismutase (SOD; EC 1.11.1.1) activity {#sec2-7} ------------------------------------------------ SOD activity of callus enzyme extracts of *B. perennis* was determined by photochemical nitroblue tetrazolium (NBT) test method (Sun et al. 1988). The SOD activity required inhibition of NBT reduction was detected using xanthine-xanthine oxidase that as a generator of superoxide anions (O2^-^) (Sun et al., 1988). A unit of the SOD was determined as the quantification of protein that prevents the ratio of NBT decreasing by 50%. Proline analysis {#sec2-8} ---------------- The proline analysis was detected with respect to the method of Bates et al. (1973). The proline content of callus materials was measured spectrophotometrically at 520 nm as μηκ)\^ dw against standart proline curve (R^2^ = 0.99). Quantification of the selected phenolic compounds {#sec2-9} ------------------------------------------------- The quantification of the chosen 20 phenolics (apigenin, caffeic acid, *p*-coumaric acid, gallic acid, genistein, kaempferol, luteolin, myricetin, procyanidin-C1, quercetin, rutin hydrate, vanillic acid, ferulic acid, salicylic acid, sinapic acid, chlorogenic acid, hesperedin, naringenin, rosmarinic acid and isorhamnetin) in 80% MeOH extracts of oxidative stress treated and untreated (control) groups was determined using Liquid Chromatography- Electro Spray Ionization-Multi Stage/Mass Spectrometry (LC-ESI-MS/MS) method. Analysis was made by "METU Central Laboratory, Molecular Biology-Biotechnology Research and Development Center, Mass Spectroscopy Laboratory, Ankara, Turkey", with Agilent 6460 Triple Quadrupole System (ESI+Agilent Jet Stream) coupled with Agilent 1200 Series HPLC. All standard compound solutions and samples were kept at -20 °C through the lab and bench work. The amount of standard phenolic molecules in callus extracts were detected from the peak areas by using the equilibrium for linear regression taken from the calibration curves (R^2^; 0.99) (Karakas and Turker, 2013). Statistical analysis {#sec2-10} -------------------- All experiments were performed in three different sets, with each set in triplicate. The results were expressed as mean values ± standard deviation (SD). Statistically significant differences between untreated (control) and oxidative stress treated groups were identified by Paired-Samples *t*-test using statistical software SPSS Version 22.0 program (SPSS Inc., Chicago, IL, USA). Differences were considered statistically significant at *p* \< 0.05. Results {#sec1-3} ======= In this study, calli were efficiently produced from pedicel explant of field-grown *B. perennis* within two weeks when cultured on MS medium supplemented with 0.5 mgL^-1^ IAA and 0.5 mgL^-1^ TDZ. After four weeks, callus cultures were maintained on MS medium with or without 10 mM H~2~O~2~ for 10 hours to investigate the effect of H~2~O~2~ on phenolic accumulation, and enzymatic and nonenzymatic antioxidant activities. The Quantification of the Selected Phenolic Compounds {#sec2-11} ----------------------------------------------------- The quantification of the chosen twenty phenolic molecules in H~2~O~2~ treated and untreated (control) callus extracts of *B. perennis* was determined using LC-ESI-MS/MS analysis. Retention times of the standard phenolic compounds and the amount of the standard compounds in callus culture of *B. perennis* were shown in [Table 1](#T1){ref-type="table"}. ###### Amount of the chosen phenolic molecules in oxidative stress treated (H~2~O~2~ treatment) and untreated (control) callus cultures of *B. perennis*. Values are means ± SD of three measurements. Nd; not detected.  Whenever the 80% MeOH extracts of oxidative stress treated and untreated control groups were matched as per the results of LC-ESI-MS/MS analysis, high total phenolic content was detected in H~2~O~2~ treatment group (285.36 μg/g of dw) ([Table 1](#T1){ref-type="table"}). The considerable amounts of the chlorogenic acid (276.25 μg/g of dw), rutin hydrate (6.39 μg/g of dw), luteolin (0.35 μg/g dw), rosmarinic acid (0.14 μg/g dw), isorhamnetin (0.08 μg/g of dw), quercetin (0.07 μg/g of dw), myricetin (0.05 μg/g of dw), and kaempferol (0.02 μg/g of dw) were detected in methanol extract of H~2~O~2~ treated group ([Table 1](#T1){ref-type="table"}). However, the caffeic acid (2.64 μg/g of dw), ferulic acid (0.19 μg/g of dw), apigenin (0.11 μg/g of dw) and *p*- coumaric acid (0.08μg/g of dw) had high value in untreated (control) group. Both callus extracts obtained from oxidative stress treated and untreated groups included chlorogenic acid, rutin hydrate, caffeic acid, ferulic acid, rosmarinic acid, luteolin, apigenin and *p-* coumaric acid as dominant compounds in our study. Gallic acid, genistein, procyanidin-*C1*, vanillic acid, hesperidin, salicylic acid, sinapic acid and naringenin were lower than the limit of detection in both groups (data not shown). Total phenolic content ranged from 220.79 μg/g of dw for MS-control (untreated callus) to 285.36 μg/g dw for H~2~O~2~ treated group. Remarkably, we detected the effect of oxidative stress on chlorogenic acid ranged from 212.08 μg/g of dw for control group to 276.25μg/g of dw for H~2~O~2~ treatment group ([Table 1](#T1){ref-type="table"}). While 8 phenolic compounds were detected in untreated callus (control) group, 12 of 20 phenolic compounds were detected in H~2~O~2~ treated callus group ([Table 1](#T1){ref-type="table"}). Isorhamnetin, quercetin, myricetin and kaempferol were detected only in H~2~O~2~ treated group. Antioxidant Enzymes {#sec2-12} ------------------- ### Catalase (CAT) and Superoxide Dismutase (SOD) Activity {#sec3-1} The exogenous oxidative stress treatment caused a certain rise in CAT activity in callus cultures of *B. perennis* ([Figure 1](#F1){ref-type="fig"}). The CAT activity was ranged from 32.52 mmol H2O2/min/mg protein for control group to 42.57 mmol H2O2/min/mg protein for H~2~O~2~ treated group. The oxidative stress treatment in callus cultures of *B. perennis* increased SOD activity from 0.56 (untreated group) to 0.72 U/mg protein (H~2~O~2~ treated group) ([Figure 2](#F2){ref-type="fig"}). Soluble protein concentration was affected by H~2~O~2~ treatment in callus cultures ([Figure 1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"}). {#F1} {#F2} ### Proline, Total Phenolic and Flavonoid Contents {#sec3-13} Antioxidant activity of methanol extracts of oxidative stress treated and untreated (control) calli was assessed. Treatment of callus samples with 10 mM H~2~O~2~ for 6 hours drastically increased proline, total phenolic and flavonoid contents. The exposure of H~2~O~2~ treatment noticeably increased the proline content from 0.086 mmol/g dw (control) to 0.139 mmol/g dw (H~2~O~2~ treatment) ([Figure 3](#F3){ref-type="fig"}). {#F3} Total phenolic content of callus extracts was assessed using the Folin-Ciocalteu assay (Slinkard and Singleton, 1977). The results were expressed as mg gallic acid equivalents (GAE)/g dry weight of callus ([Figure 4](#F4){ref-type="fig"}). The methanolic extracts of calli obtained from oxidative stress treated and control groups contained 17.65 and 12.56 mg GAE/ g dried weight, respectively. The results presented in [Figure 4](#F4){ref-type="fig"} indicated that methanolic extract of H~2~O~2~ treated callus contained higher concentrations of phenolic content than untreated callus. There was a significant positive correlation among total phenolics, total flavonoids and LC-ESI-MS/MS results. Total flavonoid content in methanolic extracts of oxidative stress treated and untreated calli were determined spectrophotometrically by aluminium chloride method. The content of flavonoids was expressed as mg catechol equivalents (CE)/g dried weight. The results presented in [Figure 5](#F5){ref-type="fig"} indicated that methanolic extract of H~2~O~2~ treated group (14.51 mg CE/g dw) contained higher concentration of flavonoid content than the control group (9.07 mg CE/g dw) ([Figure 5](#F5){ref-type="fig"}). {#F4} {#F5} Discussion {#sec1-4} ========== Abiotic stimulants or signal molecules increase the production or induce de novo synthesis of medicinally important secondary metabolites in *in vitro* plant tissue cultures. The various types of elicitors have been widely used for the enhancement of secondary metabolite production in cultures of plant cell, tissue and organ (Ramakrishna and Ravishankar, 2011). The analysis of phenolic profile points out that *B. perennis* mainly includes phenolics containing the kaempferol, isorhamnetin, quercetin and apigenin (Nazaruk and Gudej, 2001; Karakas and Turker, 2013). An increased rate of the phenolic molecules can be observed under many environmental stresses such as temperature, high light, UV-irradiation, pathogen attack, nutrient deficiencies and herbicide treatments (Ramakrishna and Ravishankar, 2011). Some studies showed that exogenous application of H~2~O~2~ was capable of increasing secondary metabolites, such as capsidiol production in pepper fruits *(Capsicum annuum* L.) (Areceli et al., 2007), monoterpenoid oxindole alkoloids production in *Uncaria tomentosa* (Huerta-Heredia et al., 2009) and cardenolides production in some *Digitalis* species (Cingoz et al., 2014). Furthermore, Cingoz et. al (2014) reported that H~2~O~2~ increased phenolic content in different *Digitalis* species. The existence of essential phenolic molecules in callus of *B. perennis*, such as chlorogenic acid, rutin hydrate, caffeic acid, luteolin, rosmarinic acid, ferulic acid, isorhamnetin, quercetin, myricetin, apigenin, *p*-coumaric acid and kaempferol in present study can be beneficial for further investigation of *B. perennis*. The SOD and CAT are basic antioxidant enzymes required for the direct inhibition of ROS. The SOD enzyme catalyzes the detoxification of superoxide anions and CAT enzyme catalyzes the decrease of H~2~O~2~ and prevents tissues of organism from largely reactive hydroxyl radicals. The CAT and SOD are essential defense enzymes against oxygen radicals in biotic and abiotic stress conditions (Cingoz et al., 2014). Some investigations demonstrated the increased CAT and SOD antioxidant enzyme activities in callus of *Digitalis* species (Cingoz et al., 2014), wheat (Li et al., 2011) and *Triticum aestivum* (He et al., 2009) seedlings with exogenous application of H2O2. Similar result was also found in our study that CAT (42.57 mmol H2O2/min/mg protein) and SOD activity (0.72 U/mg protein) significantly increased in H~2~O~2~ treated callus cultures of *B. perennis*. This study recommended that H~2~O~2~ pretreatment could induce the activation of antioxidant enzymes and concerned genes as an abiotic stress trigger molecule. The accumulation of proline under different stress conditions, such as salt, drought and heavy metals has been demonstrated in many plant species (Kishor et al., 2005). Yang et al. (2009) found that exogenous H~2~O~2~ treatment led to a significant accumulation of proline in coleoptiles and radicles of maize seedlings. Similar result was also observed by He et al. (2009) that exogenous H~2~O~2~ pretreatment significantly increased free proline content in *T. aestivum* seedlings. Ozden et al. (2009) also showed that H~2~O~2~ treatment enhanced in endogenous proline production in grapevine leaves. Similarly, our study apparently demonstrated that exogenous application of H~2~O~2~ triggered a rapid production of proline in *B. perennis* callus cultures. Cingoz et al. (2014) showed that treatment of H~2~O~2~ enhanced the total phenolic contents and antioxidant enzyme production. Similar result was also detected in our present experiment that treatment of oxidative stress induced phenolic production and antioxidant activity in callus cultures of *B. perennis*. Our data recommended that the exogenous pretreatment of oxidative stress for 10 hours on callus culture of *B. perennis* might be a powerful method to enhance the phenolic compounds, enzymatic and non-enzymatic antioxidant activities. Moreover, it was found that CAT, SOD, total phenolic, total flavonoid and proline activity had a significant positive correlation with the phenolic accumulation under H~2~O~2~ pretreatment. Generation of medicinally bioactive secondary metabolites by classic agricultural methods is an insufficient strategy. The *in vitro* culture methods under controlled laboratory conditions have been broadly investigated to develop the production of specific plant originated bioactive molecules. In this study, we can conclude that application of oxidative stress as an effective elicitor is a very useful method for the enhancement of phenolics in *B. perennis*. In this study, enhancement of phenolic compounds with the application of oxidative stress was demonstrated with callus culture. Generally, callus culture can be used for the starting material of cell suspension cultures for the large scale production of phenolic compounds. The antioxidant potential of this plant can provide an advantage in food, medicine and cosmetic industries. This study was supported by the Abant Izzet Baysal University Research Foundation (Project No: 2013.03.01.625). The authors would like to thank Prof. Dr. Ekrem Gurel for the laboratory equipment support in Department of Biology, AIBU.

The transcriptome data have been deposited in NCBI\'s Gene Expression Omnibus and is accessible through GEO Series accession number GSE91187 (<http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE91187>). All other relevant data are within the paper and its Supporting Information files. Introduction {#sec001} ============ Visceral leishmaniasis (VL) is a neglected tropical disease caused by the protozoan parasite *Leishmania donovani* or *L*. *infantum* (= *L*. *chagasi*). Approximately 90% of infected individuals experience asymptomatic infection without overt evidence of disease. The remainder of infected individuals develop a chronically progressive infection that principally involves the spleen, liver and bone marrow. The outcome of infection is primarily determined by the host immune response, which is influenced by genetic makeup and environmental factors such as malnutrition \[[@pone.0169496.ref001], [@pone.0169496.ref002]\]. Active VL is the most serious form of leishmaniasis, accounting for approximately 500,000 cases annually \[[@pone.0169496.ref003], [@pone.0169496.ref004]\]. It is usually fatal if not treated, and even with treatment, the mortality rate can be as high as 20% \[[@pone.0169496.ref003], [@pone.0169496.ref005]\]. Most cases of VL are found in India, Bangladesh, Ethiopia, Sudan and Brazil. Clinical symptoms include chronic fever, loss of appetite, cachexia, and enlarged liver and spleen. Patients with progressive VL have pancytopenia, and loss of antigen-induced lymphoproliferative and Th1 cytokine responses in peripheral blood mononuclear cell cultures \[[@pone.0169496.ref006], [@pone.0169496.ref007]\]. In experimental models of VL, control of parasite replication requires an early and strong Th1 response with production of IL-12 and IFNγ \[[@pone.0169496.ref008], [@pone.0169496.ref009]\]. In human VL, there is a strong Th1 cytokine response (IFNγ, IL-1, IL-6, IL-12 and TNFα) at the site of infection \[[@pone.0169496.ref010], [@pone.0169496.ref011]\], but this is inexplicably unable to control the infection. Anti-inflammatory and type 2 cytokines (IL-4, IL-5, IL-10 and IL-13) are also increased in serum and spleen \[[@pone.0169496.ref011]--[@pone.0169496.ref013]\], and IL-10 in particular appears to have a dual effect of limiting inflammation and promoting permissiveness to parasite replication \[[@pone.0169496.ref014]\]. To further understand the pathogenesis of this disease, it is crucial to identify the cellular sources of protective and non-protective cytokines and to determine how T cells interact with infected macrophages to restrict or promote infection. The golden Syrian hamster (*Mesocricetus auratus*) is an advantageous model to study the pathogenesis of VL because it mimics the chronic and progressive nature of human disease \[[@pone.0169496.ref015], [@pone.0169496.ref016]\]. Similar to humans, hamsters infected with *L*. *donovani* experience weight loss, hepatosplenomegaly, progressive parasite replication and ultimately death \[[@pone.0169496.ref017]\]. While it is clear that active VL is associated with a failure in cellular immunity to control parasite replication, the mechanisms behind this are unclear. As in humans, hamsters show increased splenic expression of the type 1 cytokines (IL-2, IL-12, IFNγ, TNFα) and the type 2 cytokines (IL-4, IL-10, IL-13, IL-21) \[[@pone.0169496.ref011], [@pone.0169496.ref017], [@pone.0169496.ref018]\]. The studies presented here focus on the nature and role of splenic CD4^+^ T cells in the hamster model of chronic, progressive VL. Transcriptional profiling of the infected spleen tissue identified a number of markers of T cell activation. A mixed cytokine response in spleen tissue was also evident in splenic CD4^+^ T cells. CD4^+^ T cells from chronically infected hamsters had the capacity to activate macrophages and induce parasite killing, but this was marginally effective relative to the killing induced by classical macrophage activation stimuli. Increased expression of T cell inhibitory markers, identified by transcriptional profiling of spleen tissues, led us to explore this as a potential contributor to suboptimal T cell effector function. We discovered that the splenic CD4^+^ T cell and macrophage populations expressed inhibitory receptors and ligands, respectively. Blocking PD-L2 led to a significant decrease in parasite burden in a splenic explant culture, revealing a pathogenic role for the PD-1 pathway in chronic VL. Materials and Methods {#sec002} ===================== Ethics statement {#sec003} ---------------- The animals used in this study were handled in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The protocol was approved by the Institutional Animal Care and Use Committee of the University of Texas Medical Branch, Galveston, Texas (protocol number 1101004). Animals were anesthetized during procedures with inhaled isoflurane and were euthanized by CO~2~ inhalation. Parasites {#sec004} --------- *Leishmania donovani* (MHOM/SD/001S-2D) promastigotes were cultured in M199 media supplemented with 0.1 mM adenine (in 50mM HEPES), 5 g/mL hemin (in 50% triethanolamine), 20% heat-inactivated fetal bovine serum (FBS), 100 U/mL penicillin, 100 mg/mL streptomycin at 26°C. Metacyclic promastigotes were isolated from early passage 7-day cultures by peanut agglutination as previously described \[[@pone.0169496.ref019]\]. Promastigote infectivity was maintained by regular *in vivo* passages through Syrian golden hamsters. Hamsters and *in vivo* infections {#sec005} --------------------------------- Outbred Syrian golden hamsters (*Mesocricetus auratus*) were obtained from Harlan Laboratories and were maintained and used according to the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. 6--8-week old hamsters were infected by intracardiac injection of 10^6^ metacyclic *L*. *donovani* promastigotes in 50 μL Dubelcco's Modified Eagle's Medium (DMEM) or Phosphate Buffered Saline (PBS). For co-culture experiments, an inbred Chester Beatty hamster colony was maintained in the animal resource center at the University of Texas Medical Branch. Inbred hamster litters were weaned at 3 weeks old and male or female hamsters used at 4--6 weeks of age. Experiments were set up using cells from sex-matched hamsters. Transcriptional profiling by RNA sequencing {#sec006} ------------------------------------------- Next generation sequencing of uninfected and 28-day infected spleen tissue (n = 5 hamsters per group) was performed. In short, total RNA was used to construct libraries for deep sequencing using the Illumina TruSeq RNA Sample Preparation Kit. Agilent Bioanalyzer confirmed the quality of the library and Truseq SBS kit v3 was used to sequence paired-end 50 base reads on an Illumina HiSeq 1000. Reads that aligned to the *L*. *donovani* BPK282A1 genome were removed and *de novo* assembly of a complete hamster transcriptome was performed with Trinity and BRANCH software using the Texas Advanced Computing Center (TACC) at the University of Texas at Austin. A false discovery rate (FDR) cutoff of \<0.01 and fold change (FC) cutoffs of ≥2 or ≤-2 were used to identify differentially expressed transcripts. Gene identification was achieved by BLAST alignment to the *Rattus norvegicus* and *Mus musculus* reference genomes. The transcriptome data have been deposited in NCBI\'s Gene Expression Omnibus \[[@pone.0169496.ref020]\] and is accessible through GEO Series accession number GSE91187 (<http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE91187>). To explore the biological context of the differentially expressed genes, the upregulated (≥2 FC) or downregulated (≤-2 FC) transcripts (4360 total) were uploaded into WEB-based GEne SeT AnaLysis Toolkit (WebGestalt) online software using the *Mus musculus* reference genome and matched gene symbols \[[@pone.0169496.ref021]\]. WikiPathways Enrichment Analysis was run using default settings and a significance level cutoff of 0.01. Entrez Gene IDs were translated back to Gene symbols using UniProt resource database \[[@pone.0169496.ref022]\]. Isolation of splenic CD4^+^ T cells {#sec007} ----------------------------------- Spleens from uninfected and infected hamsters were collected in ice-cold RPMI 1640 medium supplemented with Glutamax (Gibco), 10% heat inactivated FBS, 0.5 mM EDTA (Gibco) and 0.6 μg DNase (Sigma). Spleens were digested for 10 minutes at 37°C by injecting with collagenase D (Roche) at 2 mg/mL in buffer containing (150 mM NaCl, 5 mM KCl, 1 mM MgCl~2~, 1.8 mM CaCl~2~, 10 mM Hepes pH 7.4). The tissue was further minced and strained through a 100μm cell strainer to obtain a single cell suspension that was plated in large tissue culture flasks for 30 minutes at 37°C in 5% CO~2~ in 10% FBS complete DMEM culture medium to remove adherent cells. The non-adherent cell population was collected and the procedure repeated. The cells were washed once in 10% FBS complete RPMI and resuspended in 1x red blood cell lysis buffer (0.2 mM NH~4~Cl, 0.01M NaHCO~3~, 0.1mM EDTA, pH 7.4) for 10 minutes and washed again with 10% FBS complete RPMI. Cells were labeled with anti-mouse CD4^+^ magnetic particles (clone GK1.5), resuspended in ice-cold separation buffer (1x PBS, 0.5% bovine serum albumin, 2mM EDTA, pH 7.2), and separated using a BD magnet following manufacturer's protocol (BD iMag Cell Separation System, BD Biosciences). The enriched CD4^+^ cell population was resuspended in 10% FBS complete RPMI and counted for FACs staining and RNA isolation. Flow cytometry {#sec008} -------------- Single cell suspensions of enriched CD4^+^ splenocytes obtained from 28-day infected and uninfected control hamsters were adjusted to a concentration of 5--10 × 10^5^ cells per 100μL of blocking buffer containing 2% normal mouse serum and 2% normal rat serum in PBS. Cells were stained with APC-Cy^™^7 conjugated rat anti-mouse CD4^+^ or isotype control (BD Biosciences) and FITC conjugated rat anti-human CD3 (clone CD3-12) or isotype control (AbD Serotec) for 30 minutes in the dark at 4°C followed by washing in PBS with 2% FBS and 0.1% sodium azide. For intracellular staining, cells were fixed/permeabilized using Foxp3/transcription factor staining buffer set and stained with PE-Cy5 conjugated anti-mouse/rat Foxp3 (clone FJK-16S) or isotype control, and eFluor 660 conjugated anti-human/mouse T-bet (clone 4B10) or isotype control (eBioscience). All flow cytometric analyses were performed on a Stratedigm SE520EX6 flow cytometer using software CellCapTure v3.1.0. Data were analyzed using FlowJo v10.0.7 (Treestar). Determination of gene expression by real-time RT-PCR {#sec009} ---------------------------------------------------- Spleen tissue or isolated CD4^+^ T cells from uninfected and 7-, 14-, 21- and/or 28-day infected hamsters were collected and RNA was isolated using the Qiagen RNeasy Mini Kit (for concentrations larger than 1 μg) or Ambion RNAqueous-Micro Total RNA Isolation Kit (for concentrations less than 1 μg RNA). Total RNA was DNase treated with Life Technologies Turbo DNA-Free kit and reverse transcribed into cDNA according to manufacturer's protocol (High-Capacity cDNA Reverse Transcription Kit, Life Technologies). Primer sequences were designed using Genscript Primer Design Tool and the National Center for Biotechnology Information (NCBI) Primer-BLAST, which provided wider selection parameters. The Ensembl genome database was used to map exons and introns according to the mouse genome, and each primer set was designed to span an intron on the hamster target gene. Primer fidelity was confirmed by analysis of dissociation curves. The target genes for which primers were designed are detailed in [S1 Table](#pone.0169496.s001){ref-type="supplementary-material"}. Gene expression was determined in total spleen tissue, baby hamster kidney cells (BHK fibroblast cell line) and CD4^+^ splenocytes by SYBR green PCR on ViiA 7 Real-Time PCR System. Data was analyzed using comparative Ct method relative to uninfected BHK controls or uninfected hamster controls and using the 18S ribosomal RNA gene as the normalizer. Isolation of hamster bone marrow-derived macrophages {#sec010} ---------------------------------------------------- Femurs from uninfected hamsters were collected and bone marrow cells were flushed using GlutaMax RPMI 1640 culture medium (Gibco) supplemented with 10% heat inactivated FBS, 50 μM β-mercaptoethanol, 100 U/mL penicillin, 100 mg/mL streptomycin and 20 ng/mL recombinant human macrophage-colony stimulating factor (M-CSF) (eBioscience). Cells were adjusted to 8 x 10^6^/mL and cultured for 3 days at 37°C in 5% CO~2~ after which the culture medium was replenished and cells were allowed to differentiate for 3 more days and purity was determined by microscopy as previously described \[[@pone.0169496.ref023]\]. Cells were then washed with PBS and detached with Trypsin/EDTA (Gibco). IFNγ-induced macrophage activation and priming {#sec011} ---------------------------------------------- Bone marrow-derived macrophages were plated and allowed to adhere overnight in 2% FBS complete RPMI medium. The next day, culture medium was replenished and cells were primed with IFNγ (10% v/v in supernatants from CHO cells expressing recombinant hamster IFNγ) \[[@pone.0169496.ref024]\] for 1--2 hours and then LPS (20 ng/μL) was added. For *in vitro* infections, stationary phase *L*. *donovani* promastigotes from 6--7 day old cultures were used at a parasite to cell ratio of 5:1 in 2% FBS complete RPMI medium. Parasites were allowed to be phagocytized for 4 hours and then the monolayer was carefully washed with pre-warmed medium to remove extracellular parasites. Media was replenished and cells were incubated for 48 hours at 37°C in 5% CO~2~. T cell---macrophage co-cultures {#sec012} ------------------------------- CD4^+^ T cells were isolated from control or chronically infected inbred hamster spleens as described. For co-cultures, CD4^+^ T cells from control or chronically infected hamsters were added to wells with uninfected or infected macrophages. For transwell assays, the macrophages were cultured in the bottom chamber, and purified CD4^+^ T cells were cultured in a maximum of 100 μL culture medium in top transwell inserts with 0.4 μm pore polycarbonate membrane (Corning). Co-cultures and transwell assays were set up with 5x10^5^ T cells and 1x10^5^ macrophages (ratio of 5:1). Cells were cultured for 1 and 48 hours at 37°C in 5% CO~2~. At each time point, culture medium was aspirated and cells were lysed for RNA isolation as described. For co-culture experiments, a *L*. *donovani* strain transfected with an episomal vector containing the Luciferase reporter gene was used \[[@pone.0169496.ref025]\]. To determine the parasite burden, the macrophages were lysed using Promega luciferase assay kit to determine parasite luciferase activity on the FluorStar Model 403 \[[@pone.0169496.ref018], [@pone.0169496.ref025]\]. The PD-1 pathway was blocked in *ex vivo* spleen cell cultures using mouse anti-PDL-2 antibody (B7-DC, R&D) and compared to the isotype control. Statistical analysis {#sec013} -------------------- Comparison between two groups was performed using Student's *t* test (parametric) or Mann-Whitney test (non-parametric) depending on the normalcy of distribution. Comparison between more than 2 groups was performed using one-way ANOVA (parametric) or Kruskall-Wallis (non-parametric) with a correction for multiple comparisons. For experiments using 2 independent variables, comparisons were done using two-way ANOVA. Bonferroni's multiple comparisons test was used. *p*-values \<0.05 were considered significant. All analyses were conducted using GraphPad Prism version 5.04 for Windows or Mac (GraphPad Software, San Diego, California, USA). Results {#sec014} ======= Progressive increase in parasite burden is associated with upregulation of genes in T cell activation pathways {#sec015} -------------------------------------------------------------------------------------------------------------- We confirmed the dramatic increase in parasite burden over the course of the first 28 days in *L*. *donovani*-infected hamsters \[[@pone.0169496.ref016], [@pone.0169496.ref017]\] ([Fig 1A](#pone.0169496.g001){ref-type="fig"}). If allowed, the parasite burden would continue to increase until the animal's death at 10--12 weeks post-infection \[[@pone.0169496.ref017]\]. To determine the effect on the splenic CD4^+^ T cell population, we compared uninfected with 28-day infected hamsters by flow cytometry. The percentage of CD4^+^ T cells in the spleen increased significantly during chronic stages of the disease (*p* = 0.036) ([Fig 1B](#pone.0169496.g001){ref-type="fig"}). To gain insight into the splenic gene expression involved in T cell function in progressive disease, we used unbiased transcriptional profiling (RNA sequencing) to identify pathways enriched in VL in 28 day infected hamsters. Input of both upregulated and down-regulated genes identified pathways related to T cell activation and effector function ([S2 Table](#pone.0169496.s002){ref-type="supplementary-material"}). From a set of manually curated genes known to be involved in CD4^+^ T cell activation or effector function we also found evidence of significant T cell activation at 28 days of infection ([Table 1](#pone.0169496.t001){ref-type="table"}). Remarkably the upregulated genes showed a mix of CD4^+^ T cell activation/effector and exhaustion markers. The simultaneous increase in activation and exhaustion markers suggests dysfunctional T cell effector function and could explain the ineffective control of infection. {#pone.0169496.g001} 10.1371/journal.pone.0169496.t001 ###### Upregulated genes related to T cell activation and function [^a^](#t001fn001){ref-type="table-fn"}. {#pone.0169496.t001g} Gene Symbol Protein Name FC [^c^](#t001fn003){ref-type="table-fn"} ------------------------------------------------- ------------------------------------------------------------- ------------------------------------------- IFNG Interferon gamma 52.2 LAG3 [^b^](#t001fn002){ref-type="table-fn"} Lymphocyte activation gene 3 27.3 PDCD1LG2 [^b^](#t001fn002){ref-type="table-fn"} Programmed cell death 1 ligand 2 26.9 CCR5 C-C chemokine receptor 5 19.8 TNFRSF4 OX40, Tumor necrosis factor receptor superfamily member 4 9.7 CTLA4 [^b^](#t001fn002){ref-type="table-fn"} Cytotoxic T-lymphocyte associated protein 4 7.5 TNF Tumor necrosis factor alpha 6.8 IL12RB2 Interleukin 12 receptor subunit beta 2 4.8 CXCR3 C-X-C chemokine receptor 3 3.6 PDCD1 [^b^](#t001fn002){ref-type="table-fn"} Programmed cell death 1 3.3 TNFRSF9 CD137, Tumor necrosis factor receptor superfamily member 9 3.1 TBX21 Tbet, T-box transcription factor 21 3.0 PTPRC CD45 antigen, protein tyrosine phosphatase receptor type c 2.7 TNFRSF14 CD270, Tumor necrosis factor receptor superfamily member 14 2.6 CD274 [^b^](#t001fn002){ref-type="table-fn"} Programmed cell death 1 ligand 1 2.6 CD44 CD44 antigen 2.4 IL2RB Interleukin 2 receptor subunit beta 2.0 CD4 Cluster of differentiation 4 1.6 ^**a**^ Manually curated list; False Discovery rate \<0.01 ^**b**^ Inhibitory receptors ^**c**^ Fold-change Mixed splenic cytokine profile in response to chronic VL {#sec016} -------------------------------------------------------- To further characterize the functional capacity of splenic T cells, we first determined mRNA expression of Th1, Th2, and regulatory CD4^+^ T cell (Treg) markers in the spleens of *L*. *donovani*-infected hamsters over the course of early VL. We found mRNA expression of the Th1 transcription factor, T-bet, and the Th1 cytokine, IFNγ, to increase over the course of infection ([Fig 2A](#pone.0169496.g002){ref-type="fig"}). However, Th1 markers expression was preceded (at 14 days post-infection) by expression of the master regulator transcription factor for Th2 cells, GATA3, and the associated Th2 cytokine, IL-4 ([Fig 2B](#pone.0169496.g002){ref-type="fig"}). In addition to the upregulation of Th1 and Th2 markers, mRNA of the transcription factor that regulates the differentiation of Tregs, Foxp3, was transiently increased at 14 days post-infection ([Fig 2C](#pone.0169496.g002){ref-type="fig"}). This suggests a transient regulatory CD4^+^ T cell response early in infection before Th1 CD4^+^ T cells significantly increase. The regulatory cytokine produced by Treg, Th2 and some Th1 cells, IL-10, showed significant increase in the spleen at 21 and 28 days post-infection ([Fig 2C](#pone.0169496.g002){ref-type="fig"}). Although cells other than CD4^+^ T cells can produce these cytokines, these data prompted us to examine their expression in purified CD4^+^ T cells during the course of infection. {#pone.0169496.g002} Splenic CD4^+^ T cells display a mixed Th1 and Th2 profile during chronic VL {#sec017} ---------------------------------------------------------------------------- We determined the phenotype of CD4^+^ T cells that accumulate in the spleen during chronic VL by investigating mRNA expression in isolated splenic CD4^+^ T cells. CD4^+^ splenocytes were enriched using magnetic particles conjugated with anti-CD4 antibody. This yielded a \>90% pure CD3^+^CD4^+^ population that was used for real time RT-PCR analysis ([Fig 3A](#pone.0169496.g003){ref-type="fig"}). In this purified CD4^+^ T cell population we found a significant increase in mRNA and protein expression of the transcription factor, T-bet ([Fig 3B](#pone.0169496.g003){ref-type="fig"}). The mRNAs for the type 1 cytokine IFNγ and chemokine receptors associated with Th1 cells, CCR5 and CXCR3, were also significantly increased ([Fig 3C](#pone.0169496.g003){ref-type="fig"}). The Th2 transcription factor, GATA3, and Th2 cytokine, IL-4, were also significantly increased in splenic CD4^+^ T cells in VL, but the chemokine receptor commonly expressed on Th2 and Treg cells, CCR4, was not increased ([Fig 3D](#pone.0169496.g003){ref-type="fig"}). Consistent with what was found in spleen tissue at 28 days post-infection ([Fig 2C](#pone.0169496.g002){ref-type="fig"}), the transcription factor Foxp3, showed no increase in mRNA or protein in splenic CD4^+^ T cells at this time point ([Fig 3E](#pone.0169496.g003){ref-type="fig"}). IL-10 and IL-21, which can be produced by multiple T cell subsets and are thought to have a disease-promoting effect in active VL \[[@pone.0169496.ref026]\], were significantly upregulated in splenic CD4^+^ T cells in infected animals ([Fig 3E](#pone.0169496.g003){ref-type="fig"}). Collectively, our data indicate that splenic CD4^+^ T cells exhibit markers of both Th1 and Th2 development during chronic VL. Whether this Th1/Th2 profile is indicative of a double positive T cell phenotype or there are two separate T cell populations accumulating simultaneously could not be determined because antibodies against these markers in hamsters are not available. {#pone.0169496.g003} Expression of chemokine ligands and receptors is increased in spleen tissue, splenic macrophages and CD4^+^ T cells in VL {#sec018} ------------------------------------------------------------------------------------------------------------------------- Because of the accumulation of CD4^+^ T cells in the spleen during VL, we investigated the expression of chemokine ligands and their receptors known to selectively recruit Th1, Th2, or Treg CD4^+^ cells in the spleens of hamsters with VL. We found mRNAs of the type 1 chemokine ligands and their receptors (CCL4, CCL5, CCR5, CXCL9, CXCL10, CXCL11 and CXCR3) were increased at 21--28 days post-infection compared to uninfected controls ([Fig 4](#pone.0169496.g004){ref-type="fig"}). Expression of the type 2 chemokine ligands, CCL17 and CCL22 increased as early as 7 days post-infection and remained upregulated up to 21 days of infection ([Fig 4](#pone.0169496.g004){ref-type="fig"}), but somewhat surprisingly their receptor, CCR4, was not ([Fig 4](#pone.0169496.g004){ref-type="fig"}). We did not measure expression of CCR8, which is a secondary receptor for CCL17 and CCL22. {#pone.0169496.g004} In order to determine the cellular source of the chemokine ligands, we next examined their expression in splenic macrophages isolated from chronically infected hamster spleens. The chemokine ligands that bind the Th1-associated chemokine receptors CXCR3 (CXCL9, CXCL10, CXCL11) and CCR5 (CCL4 and CCL5) were significantly upregulated in splenic macrophages during VL ([Fig 5](#pone.0169496.g005){ref-type="fig"}). The Th2-attracting chemokines, CCL17 and CCL22, showed a non-significant trend of increased expression. The increase in mixed chemokine ligands in this isolated cell population is consistent with the data from total spleen tissues ([Fig 4](#pone.0169496.g004){ref-type="fig"}). Together, these data suggests splenic macrophages are a significant source of T cell-attracting chemokines and may account for the accumulation of type 1 and type 2 CD4^+^ T cell subsets in the spleen during chronic VL. {#pone.0169496.g005} Splenic CD4^+^ T cells from chronically infected animals modestly reduce intracellular parasite burden *in vitro* {#sec019} ----------------------------------------------------------------------------------------------------------------- Macrophages stimulated with IFNγ have a classically activated (M1) phenotype, characterized by production of reactive oxygen and nitrogen species \[[@pone.0169496.ref027]--[@pone.0169496.ref029]\]. Nitric oxide (NO) generated by inducible nitric oxide synthase (iNOS) is the primary effector of intracellular parasite killing. The relentlessly progressive disease in humans and hamsters in the face of Th1 cell accumulation and IFNγ production suggests that macrophages are not being effectively activated to kill the parasite \[[@pone.0169496.ref030], [@pone.0169496.ref031]\]. To investigate this, we first tested the effector function of hamster macrophages in an *in vitro* activation system. Bone marrow-derived macrophages were first primed with IFNγ for an hour before LPS stimulation and *L*. *donovani* infection. The primed-activated macrophages showed a dramatic increase in the uptake of parasites compared to unstimulated control macrophages after 1 hour ([Fig 6A](#pone.0169496.g006){ref-type="fig"}). After 48 hours, the IFNγ- and LPS-stimulated macrophages showed a significant (12-fold) reduction in parasite burden ([Fig 6A](#pone.0169496.g006){ref-type="fig"}) that was accompanied by increased expression of iNOS and Arg1 ([Fig 6A](#pone.0169496.g006){ref-type="fig"}). In contrast, there was no decrease in parasite load in the unactivated macrophages. These data suggest that in a controlled *in vitro* system, hamster macrophages are fully capable of upregulation of iNOS and intracellular parasite killing. We next evaluated the capacity of splenic CD4^+^ T cells, isolated from uninfected controls and hamsters with VL, to activate *in vitro* infected macrophages to kill intracellular parasites in a co-culture assay. To avoid any possible detection of parasite DNA carried with the CD4^+^ T cells isolated from infected hamsters, we used a luciferase-transfected *L*. *donovani* strain for the *in vitro* macrophage infections and measured parasite burden by luciferase activity. After 48 hours, the parasite burden was modestly (1.5-fold) but significantly decreased in the infected macrophages co-cultured with the CD4^+^ T cells isolated from hamsters with VL compared to those from uninfected hamsters ([Fig 6B](#pone.0169496.g006){ref-type="fig"}). This was accompanied by a significant increase in iNOS expression ([Fig 6C](#pone.0169496.g006){ref-type="fig"}), although its upregulation was substantially less than that observed in IFNγ/LPS-stimulated macrophages ([Fig 6A](#pone.0169496.g006){ref-type="fig"}). Notably, there was also an increase in Arg1 expression, a marker of alternative (M2) macrophage activation, when the infected macrophages were co-cultured with CD4^+^ T cells from hamsters with VL ([Fig 6C](#pone.0169496.g006){ref-type="fig"}). Collectively, these data indicate that the mixed phenotype of splenic CD4^+^ T cells in VL leads to expression of markers of both M1 (iNOS) and M2 (Arg1) activation that marginally reduces intracellular parasite burden. {#pone.0169496.g006} Increased splenic expression of T cell exhaustion markers in VL {#sec020} --------------------------------------------------------------- Cellular immune function is ineffective in abating the relentless disease progression in VL, and CD4^+^ T cells co-cultured with *in vitro* infected macrophages uncovered only modest T cell effector activity ([Fig 6C](#pone.0169496.g006){ref-type="fig"}). Our transcriptional profiling data in spleen tissue identified the upregulation of a number of receptors/ligands that function to inhibit T cell responses and are associated with T cell exhaustion \[[@pone.0169496.ref032]\] ([Table 1](#pone.0169496.t001){ref-type="table"}). We therefore investigated markers of T cell exhaustion in splenic CD4^+^ T cells during chronic VL. We found that the inhibitory receptors PD-1 and CTLA-4, and the ligands for PD-1 (PD-L1 and PD-L2), increased in spleen tissues as the disease progressed ([Fig 7A](#pone.0169496.g007){ref-type="fig"}). PD-1 and CTLA-4 were also significantly increased in isolated splenic CD4^+^ T cells from 28-day infected animals ([Fig 7B](#pone.0169496.g007){ref-type="fig"}). Splenic macrophages isolated from 28-day infected hamsters showed an insignificant increase in PD-L1 (*p* = 0.3429), and a trend toward increased PD-L2 expression (*p* = 0.0571) ([Fig 7C](#pone.0169496.g007){ref-type="fig"}). Collectively, these data suggest that inhibitory signaling between infected macrophages and CD4^+^ T cells may occur in the spleen during chronic VL. {#pone.0169496.g007} Blockade of the PD-1 pathway leads to enhanced parasite control {#sec021} --------------------------------------------------------------- To investigate the functional significance of the increased inhibitory molecules on macrophages and CD4^+^ T cells in the model of progressive VL, we blocked the PD-1 pathway with anti-PD-L2 antibody in *ex vivo* spleen cell cultures (containing both macrophages and T cells) from hamsters with VL. PD-L2 was targeted because of its increased expression relative to PD-L1 in the spleen and splenic macrophages. We found the parasite burden was significantly reduced after 48 hours incubation with anti-PD-L2 antibody compared to the isotype control ([Fig 8](#pone.0169496.g008){ref-type="fig"}). This was associated with a striking decrease in Arg1 mRNA expression, but no change in iNOS or IFNγ expression and a slight increase in IL-10. This suggests that blockade of the PD-1/PD-L2 interaction diminishes T cell-induced, disease-promoting arginase expression by macrophages and that the reduced parasite burden is not mediated by reduced IL-10. Together, these findings suggest activation of the PD-1 signaling pathway in the spleen plays a pathological role in progressive VL in the hamster model, and that this can be reversed by blockade of this inhibitory pathway. {#pone.0169496.g008} Discussion {#sec022} ========== Individuals with subclinical *L*. *donovani* infection demonstrate robust CD4^+^ T cell responses with IFNγ-mediated macrophage activation and parasite control \[[@pone.0169496.ref033], [@pone.0169496.ref034]\]. However, during active VL cultures of purified peripheral blood mononuclear cells show T cell unresponsiveness to *Leishmania* antigens \[[@pone.0169496.ref006], [@pone.0169496.ref007]\]. These studies may not be fully representative of T cell function because parasite-responsive cells are more likely to be found at the sites of visceral infection (e.g. spleen). Using an experimental model that closely mimics active human disease, we found an increase in accumulation of CD4^+^ T cells with a mixed Th1/Th2 cytokine/chemokine profile in the spleen during chronic VL. This corroborates studies of cytokine expression in human VL and in other models of non-healing leishmaniasis \[[@pone.0169496.ref011], [@pone.0169496.ref035]\]. The expression of T-bet and IFNγ by splenic CD4^+^ T cells in our VL model adds to the growing evidence of a strong type-1 (IFNγ) splenic immune response during active VL. But the question remains: why are macrophages in this Th1 environment unable to control *L*. *donovani* infection? Splenic CD4^+^ T cells isolated from hamsters with VL (28 days post-infection) showed marginal capacity to inhibit intracellular parasite growth in co-cultured *in vitro* infected macrophages. This suggests that co-expression of regulatory cytokines (IL-4, IL-10 and IL-21) and/or expression of inhibitory receptors in the CD4^+^ T cell population contributes to suboptimal macrophage activating capacity of the splenic CD4^+^ T cells. Splenic CD4^+^ T cell expression of IL-4, IL-10 and IL-21 was significantly increased in hamsters with VL. IL-4 has a major role in the susceptibility of mice to experimental *L*. *major* infection (reviewed in \[[@pone.0169496.ref036]\]), either through inhibiting macrophage generation of effector molecules \[[@pone.0169496.ref037]\] and/or polarization of macrophages to produce arginase 1 \[[@pone.0169496.ref018], [@pone.0169496.ref038]\]. However, in mice that have a non-progressive *L*. *donovani* infection, IL-4 does not contribute to host susceptibility \[[@pone.0169496.ref039], [@pone.0169496.ref040]\]. While IL-4 is increased in the plasma or serum \[[@pone.0169496.ref011], [@pone.0169496.ref012], [@pone.0169496.ref041]\] and spleens \[[@pone.0169496.ref011]\] of patients with active VL, its role in pathogenesis of human VL has not been defined. Our previous finding that pathologic parasite-induced arginase is amplified by IL-4 in experimental VL suggests that IL-4-producing CD4^+^ T cells may contribute to impaired control of infection. IL-10 has been more clearly demonstrated to have a role in VL pathogenesis \[[@pone.0169496.ref014]\]. Neutralization of IL-10 in spleen cell explant cultures from subjects with VL led to reduced parasite load \[[@pone.0169496.ref042]\]. The identity of cells that produce IL-10 in human VL is controversial. IL-10 producing CD4^+^ T cells has been characterized in the VL mouse model as Foxp3^−^ cells \[[@pone.0169496.ref043], [@pone.0169496.ref044]\]. In humans, Nylen, et al, found the source of splenic IL-10 to be a CD4^+^Foxp3^−^ cell population \[[@pone.0169496.ref011]\]. Contrary to this, others found CD25^+^Foxp3^+^ Treg cells accumulated in the spleen in response to *Leishmania* antigen and produced IL-10 \[[@pone.0169496.ref045], [@pone.0169496.ref046]\]. We did not find sustained expansion of a CD4^+^Foxp3^+^ regulatory cell population, despite prominent expression of the regulatory cytokine IL-10. CD4^+^ T cells that produce both IFNγ and IL-10 have been described in human VL \[[@pone.0169496.ref011]\] and in chronic infection with *Toxoplasma gondii* and *Mycobacterium tuberculosis* \[[@pone.0169496.ref047], [@pone.0169496.ref048]\]. Distinction of individual cytokine-producing CD4^+^ T cells is not possible in the hamster model because of the lack of antibody reagents for multicolor flow cytometry. Splenic CD4^+^ T cells also expressed increased IL-21 in our model. Co-expression of splenic IL-10 and IL-21, and IL-21-mediated induction of IL-10, was shown in patients with VL \[[@pone.0169496.ref026]\]. Additionally, IL-21 could promote infection via M2 polarization of macrophages \[[@pone.0169496.ref049]\]. We found increased mRNA expression of PD-1, CTLA-4, PD-L1 and PD-L2 in chronically infected hamster spleen tissues, and increased expression of the inhibitory receptors CTLA-4 and PD-1 in the splenic CD4^+^ T cell population. Blockade of PD-L2 in an *ex vivo* spleen cell explant culture from hamsters with VL effected a reduction in parasite burden. This suggests a pathogenic role for the PD-1 pathway in this model of progressive VL. PD-1 and CTLA-4 are markers of T cell exhaustion, which multiple pathogens utilize to their benefit to establish chronic infection (reviewed in \[[@pone.0169496.ref032]\]). Their expression has been shown to increase soon after T cell activation is initiated. With chronic stimulation of T cells, the regulatory functions of CTLA-4 and PD-1 control overly aggressive T cell responses that could be damaging to the host. These inhibitory receptors lead to the gradual loss of activation and expansion of T cells, decreased cytokine production, and at the extreme, clonal deletion of the population \[[@pone.0169496.ref032]\]. T cell exhaustion has typically been demonstrated for CD8^+^ T cells as a result of long-term antigen stimulation in chronic viral infections \[[@pone.0169496.ref050]\], but CD4^+^ T cells and B cells \[[@pone.0169496.ref051]\] can also express PD-1 and CTLA-4. An increase in exhaustion markers was demonstrated in several models of VL \[[@pone.0169496.ref052]--[@pone.0169496.ref054]\]. CD8^+^ T cell exhaustion markers were described in patients with active VL, however, blockade of CTLA-4 and PD-1 did not recover CD8^+^ T cell responses to soluble *Leishmania* antigen or increase parasite killing in *ex vivo* cultures of spleen cells \[[@pone.0169496.ref053]\]. Mice infected with *L*. *donovani* showed evidence of progressive CD8^+^ T cell dysfunction, and *in vivo* blockade of PD-L1 improved CD8^+^ T cell survival and reduced the splenic parasite burden \[[@pone.0169496.ref054]\]. Increased expression of exhaustion markers in CD4^+^ T cells from dogs with chronic VL was found, and blockade of PD-L1 rescued *in vitro* CD4^+^ T cell proliferation and IFNγ production \[[@pone.0169496.ref052]\]. Most studies have focused on the effects of inhibitory receptors on T cell effector function, however, few studies have investigated their effects on the antigen presenting cell or phagocyte. Recently it was shown that engagement of the inhibitory receptors on macrophages and dendritic cells delivers a suppressive signal from the T cell to the myeloid cell \[[@pone.0169496.ref055]--[@pone.0169496.ref057]\]. Thus, the interaction between PD-1 and PD-L1/2 is likely to have a bi-directional effect. In support of this, we provide the evidence that blockade of PD-1/PD-L2 in spleen cell cultures from hamsters with VL led to decreased expression of arginase 1, which we demonstrated previously to promote macrophage susceptibility and VL progression \[[@pone.0169496.ref018], [@pone.0169496.ref023]\]. This is yet another subversive mechanism that renders macrophages unresponsive to antimicrobial activation \[[@pone.0169496.ref011], [@pone.0169496.ref058]--[@pone.0169496.ref060]\]. In summary, we describe for the first time the splenic CD4^+^ T cell phenotype in an experimental model of chronic progressive VL. We demonstrate that CD4^+^ T cells in hamster VL have a mixed cytokine/chemokine profile and marginal capacity to induce parasite killing in *in vitro* infected macrophages. Splenic CD4^+^ T cells also expressed regulatory cytokines, such as IL-10 and IL-21, and inhibitory receptors typically associated with T cell exhaustion. Remarkably, blockade of the PD-1/PD-L2 pathway in *ex vivo* spleen cell explant cultures inhibited host arginase expression and enhanced parasite control. Whether the protective effect mediated by PD-L2 blockade is through a direct action on macrophages or an indirect action through T cell signals is unknown. The fact that there was no difference in IFNγ expression suggests that the killing effect is not mediated by an increase in this Th1 cytokine, but through other mechanisms. Further investigation into the role of inhibitory receptors/ligands on the interaction between CD4^+^ T cells and macrophages is warranted. Supporting Information {#sec023} ====================== ###### Hamster primer sequences designed for real time RT-PCR. (DOCX) ###### Click here for additional data file. ###### Pathway analysis. (DOCX) ###### Click here for additional data file. This work was supported by institutional funding from the University of Texas Medical Branch. AAMC was supported by Award Number T32AI007526 from the National Institute of Allergy and Infectious Diseases. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institute of Allergy and Infectious Diseases or the National Institutes of Health. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: **Conceptualization:** AAMC PCM EYO LS.**Data curation:** AAMC FK.**Formal analysis:** AAMC PCM EYO FK.**Funding acquisition:** PCM.**Investigation:** AAMC EYO OAS FK BLT.**Methodology:** AAMC PCM EYO OAS FK HS BLT.**Project administration:** PCM BLT.**Resources:** BLT FK HS.**Supervision:** PCM.**Validation:** AAMC EYO.**Visualization:** AAMC EYO PCM.**Writing -- original draft:** AAMC PCM.**Writing -- review & editing:** AAMC EYO BLT PCM LS.

Introduction ============ Although Mollusca is, after Arthropoda, the second-most species-rich animal phylum on land (at least in terms of described species), their species numbers are only moderately high compared with the arthropods ([@B5386235], [@B5386249]). Even in hyperdiverse forested regions in the humid tropics, such as northern Borneo, the numbers of known species are in the hundreds, not thousands and trends of discovery appear to be levelling off towards the true number of species. This means that, with relatively little fieldwork effort, it is possible to obtain a more or less complete inventory of the terrestrial malacofauna of a certain area. We are using field courses to several biodiversity hotspots in the world for such a purpose and we use the term \'taxon expeditions\' for this ([@B5386259]). Taxon expeditions are a new concept in which a group of taxonomic experts and lay people work together in a hybrid work form of field course and biodiversity discovery expedition to discover unknown species from a given area. On our annual taxon expedition to the Ulu Temburong lowland rainforest in Brunei Darussalam, surveys of terrestrial snails have so far yielded over 25 species. The expected diversity for this type of forest would be around 85 species ([@B5386270], [@B5386249]), but already our inventories are turning up previously unknown species. In this paper, we describe a new species from the large caenogastropod family Cyclophoridae. In the Southeast Asian tropics, non-Stylommatophora (i.e. Neritopsina and Caenogastropoda) comprise nearly half of the total malacofauna. These snails are, however, more sensitive to habitat disturbance than the Stylommatophora. Studies of malacofauna conservation biology on limestone hills in Sabah have previously shown ([@B5386290]) that these non-Stylommatophora groups are the first to disappear after forest degradation, presumably as a result of lower tolerance ranges for temperature and humidity. The description of the new cyclophorid that follows is the concerted effort of untrained 'citizen scientists' working together in a field lab. The specific epithet was decided upon during a voting session, in which participants could suggest and vote for scientific names. Materials and methods ===================== We collected living snails by hand, at night, amongst vegetation along the left (south) bank of the Belalong river, 50 m downstream of Kuala Belalong Field Studies Centre. Specimens were sorted to putative morphospecies and then further examined morphologically with a dissection microscope with 10× and 20× eye pieces. Photographs and video were made either with a smartphone through the eyepiece of a microscope or on a translucent white acrylic sheet with a Nikon D800e fitted with a Laowa 25 mm ultra-macro lens. Drawings were made based on the photographs. Specimens were deposited in the collection of the University of Brunei Darussalam Museum (UBDM). Taxon treatments ================ Craspedotropis gretathunbergae sp. n. ------------------------------------- EE00EC40-87D5-5635-B318-D36D42F10DA7 urn:lsid:zoobank.org:act:B6AFC625-21E5-4436-9C91-5B8B46A61E19 ### Materials a. **Type status:** Holotype. **Occurrence:** catalogNumber: 7.00141; recordedBy: Jonathan Lim and other Taxon Expeditions participants; individualCount: 1; sex: unknown; lifeStage: adult; preparations: total individual, dry, with dried-up tissue inside the shell; disposition: in collection; **Taxon:** scientificName: Craspedotropis gretathunbergae Schilthuizen et al.; kingdom: Animalia; phylum: Mollusca; class: Gastropoda; order: Caenogastropoda; family: Cyclophoridae; genus: Craspedotropis; specificEpithet: gretathunbergae; taxonRank: species; scientificNameAuthorship: Schilthuizen et al.; nomenclaturalCode: ICZN; **Location:** continent: Asia; island: Borneo; country: Brunei; stateProvince: Temburong; verbatimLocality: Ulu Temburong, 50 m downstream from Kuala Belalong Field Studies Centre, on green leaves near river bank; verbatimElevation: 50 m; verbatimCoordinateSystem: decimal degrees; decimalLatitude: 4.5478; decimalLongitude: 115.1578; **Event:** samplingProtocol: manual collecting at night; eventDate: 2019-09-28; eventTime: 20:00/22:00; habitat: primary dipterocarp forest; **Record Level:** type: PhyscalObject; institutionID: UBD; institutionCode: IBER-UBD; collectionCode: Zoology; basisOfRecord: PreservedSpecimen b. **Type status:** Paratype. **Occurrence:** catalogNumber: 7.00142; recordedBy: Jonathan Lim and other Taxon Expeditions participants; individualCount: 1; sex: unknown; lifeStage: adult; preparations: total individual, dry, with dried-up tissue inside the shell; disposition: in collection; **Taxon:** scientificName: Craspedotropis gretathunbergae Schilthuizen et al.; kingdom: Animalia; phylum: Mollusca; class: Gastropoda; order: Caenogastropoda; family: Cyclophoridae; genus: Craspedotropis; specificEpithet: gretathunbergae; taxonRank: species; scientificNameAuthorship: Schilthuizen et al.; nomenclaturalCode: ICZN; **Location:** continent: Asia; island: Borneo; country: Brunei; stateProvince: Temburong; verbatimLocality: Ulu Temburong, 50 m downstream from Kuala Belalong Field Studies Centre, on green leaves near river bank; verbatimElevation: 50 m; verbatimCoordinateSystem: decimal degrees; decimalLatitude: 4.5478; decimalLongitude: 115.1578; **Event:** samplingProtocol: manual collecting at night; eventDate: 2019-09-28; eventTime: 20:00/22:00; habitat: primary dipterocarp forest; **Record Level:** type: PhyscalObject; institutionID: UBD; institutionCode: IBER-UBD; collectionCode: Zoology; basisOfRecord: PreservedSpecimen c. **Type status:** Paratype. **Occurrence:** catalogNumber: 7.00143; recordedBy: Jonathan Lim and other Taxon Expeditions participants; individualCount: 1; sex: unknown; lifeStage: adult; preparations: 2 total individuals, in 70% ethanol; disposition: in collection; **Taxon:** scientificName: Craspedotropis gretathunbergae Schilthuizen et al.; kingdom: Animalia; phylum: Mollusca; class: Gastropoda; order: Caenogastropoda; family: Cyclophoridae; genus: Craspedotropis; specificEpithet: gretathunbergae; taxonRank: species; scientificNameAuthorship: Schilthuizen et al.; nomenclaturalCode: ICZN; **Location:** continent: Asia; island: Borneo; country: Brunei; stateProvince: Temburong; verbatimLocality: Ulu Temburong, 50 m downstream from Kuala Belalong Field Studies Centre, on green leaves near river bank; verbatimElevation: 50 m; verbatimCoordinateSystem: decimal degrees; decimalLatitude: 4.5478; decimalLongitude: 115.1578; **Event:** samplingProtocol: manual collecting at night; eventDate: 2019-09-28; eventTime: 20:00/22:00; habitat: primary dipterocarp forest; **Record Level:** type: PhyscalObject; institutionID: UBD; institutionCode: IBER-UBD; collectionCode: Zoology; basisOfRecord: PreservedSpecimen d. **Type status:** Paratype. **Occurrence:** catalogNumber: 7.00144; recordedBy: Jonathan Lim and other Taxon Expeditions participants; individualCount: 1; sex: unknown; lifeStage: adult; preparations: 3 total individuals, in 100% ethanol; disposition: in collection; **Taxon:** scientificName: Craspedotropis gretathunbergae Schilthuizen et al.; kingdom: Animalia; phylum: Mollusca; class: Gastropoda; order: Caenogastropoda; family: Cyclophoridae; genus: Craspedotropis; specificEpithet: gretathunbergae; taxonRank: species; scientificNameAuthorship: Schilthuizen et al.; nomenclaturalCode: ICZN; **Location:** continent: Asia; island: Borneo; country: Brunei; stateProvince: Temburong; verbatimLocality: Ulu Temburong, 50 m downstream from Kuala Belalong Field Studies Centre, on green leaves near river bank; verbatimElevation: 50 m; verbatimCoordinateSystem: decimal degrees; decimalLatitude: 4.5478; decimalLongitude: 115.1578; **Event:** samplingProtocol: manual collecting at night; eventDate: 2019-09-29; eventTime: 20:00/22:00; habitat: primary dipterocarp forest; **Record Level:** type: PhyscalObject; institutionID: UBD; institutionCode: IBER-UBD; collectionCode: Zoology; basisOfRecord: PreservedSpecimen ### Description **Spire (Fig. [1](#F5386363){ref-type="fig"})**: Protoconch without distinct sculpture; juncture with the teleoconch not visible. Spire high conical, consisting of 5.25 to 5.75 convex whorls. Suture depressed. Radial sculpture: growth lines and densely placed riblets (30-70 per mm on the body whorl). Spiral sculpture: 4 high, prominent, here and there somewhat crenellated radial ribs, which start after 1.5 to 2.0 whorls. The first spiral rib (from the top) is located between the periphery and the suture with the previous whorl. The second sits at the periphery. The third is located near the suture with the next whorl and is visible on the body whorl in some, but not all, individuals. The fourth is located on the basal side and only visible on the edge of the umbilicus. Between the ribs are fine spiral lines. Umbilicus wide, without additional spiral ribs. Height 2.7 - 2.9 mm, width 1.7 - 1.8 mm. Periostracum greenish-yellow, deep brown when thickened (e.g. near the aperture). **Aperture (Fig. [3](#F5471316){ref-type="fig"})** Aperture slightly wider than high, peristome slightly thickened, with angularities coinciding with the four spiral ribs, but without sinuosities; basal side horizontal. The peristome carries ca. 10 tightly packed accretions, coloured dark brown because of the thickened periostracum. Height 0.86 - 0.90 mm, width 0.92 - 0.97 mm. **Operculum (Figs [1](#F5386363){ref-type="fig"}, [3](#F5471316){ref-type="fig"})**: Operculum thin, corneous, whorl margins not raised but flattened into a smooth concave dish, of which the outer part is greenish-brown (presumably because of a thickly applied periostracum), the central part nearly clear. No calcareous matter visible. **Body (Fig. [2](#F5386367){ref-type="fig"})**: Body pale, tentacles dark grey, buccal mass visible as a pink-orange globule. (See also Suppl. material [1](#S5466782){ref-type="supplementary-material"}) **Genitalia**: Not studied. ### Diagnosis Amongst the Bornean cyclophorids, *Craspedotropis gretathunbergae* n. sp. is most similar to *C. borneensis* ([@B5466712]), which, however, is somewhat less slender, has 7 - 9 spiral ribs and more broadly reflected apertural lip. In addition, the operculum of *C. borneensis* has raised whorl margins, which is not the case in *C. gretathunbergae* n. sp. Other Asian species, with which the new species may be confused, are: (i) *Cyathopoma conoideum* [@B5466742], which has a more slender shell and the spiral ribs arranged in a different pattern, with the second rib located just above the suture; (ii) *C. sivagherrianum* [@B5466752], which is smaller, has 7 spiral ribs and a nearly closed umbilicus; (iii) *C. beddomeanum* [@B5466762], which has more (7 - 8) and more prominent, spiral ribs, more globular whorls and a rounder aperture; (iv) *C. procerum* [@B5466772], which is stockier, has 9 - 12 spiral ribs and a peristome that is strongly thickened by folds. ### Etymology We name this species in honour of the young climate activist Greta Thunberg, because caenogastropod microsnails from tropical rainforests, like this new species, are very sensitive to the droughts and temperature extremes that are likely to be more frequent as climate change continues. Via mutual contacts, we have approached Ms. Thunberg and learned that she would be \'delighted\' to have this species named after her. Following Recommendation 51C of the Code ([@B5386332]), if it is desired that authorship of the name be included as part of the name, instead of listing all authors, the species can be referred to as *Craspedotropis gretathunbergae* Schilthuizen et al., 2019, provided that all authors of the name are cited in full elsewhere in the same work, either in the text or in a bibliographic reference. ### Distribution Borneo: Brunei Darussalam: Temburong District: Ulu Temburong lowland rainforest. ### Ecology In tropical mixed dipterocarp lowland rainforest. All individuals were found alive at the foot of a steep hill-slope, next to a river bank, foraging at night on the upper surfaces of green leaves of understorey plants, up to 1 m above ground level. ### Taxon discussion The generic classification of minute cyclophorids in Southeast Asia is somewhat confused. The genera *Craspedotropis* [@B5466692], *Cyathopoma* Blanford 1861 [@B5466722], *Jerdonia* Blanford 1861 ([@B5466722]; sometimes considered a subgenus of *Cyathopoma*) and *Ditropopsis* [@B5466732] appear poorly defined and may be partly overlapping or synonymous ([@B5386351], [@B5386341]). In its general shell form (a tall conical shell with spiral ribs), the present species is similar to several species of *Cyathopoma* (e.g. the South Asian *C. conoideum* [@B5466742], *C. sivagherrianum* [@B5466752], *C. beddomeanum* [@B5466762] and *C. procerum* [@B5466772]) and *Craspedotropis*, especially *C. borneensis* ([@B5466712]), known from Sarawak. Given the geographical proximity of the latter, we have tentatively placed the new species in the same genus, *Craspedotropis*. Discussion ========== All work described in this paper (fieldwork, morphological study, microphotography, taxonomic description and diagnosis) was carried out in a field centre with basic equipment and no internet access, by untrained 'citizen scientists' guided by expert scientists, on a 10-day taxon expedition. While we are aware that this way of working has its limitations in terms of the quality of the output (for example, we were unable to perform dissections or to do extensive literature searches), the benefits include rapid species discovery and on-site processing of materials. Supplementary Material ====================== 9D4C3545-4C96-5CDB-8E99-342A299E687A 10.3897/BDJ.8.e47484.suppl1 ###### Craspedotropis gretathunbergae, living, active individual (paratype, IBER-UBD 7.00142). Data type: video footage (macro) Brief description: This collection of short clips shows one of the paratype individuals actively crawling on a dead leaf. Some of the soft parts (tentacles, proboscis, foot) are visible, as well as the operculum and details of the shell. File: oo_365305.mp4 https://binary.pensoft.net/file/365305 P. Escoubas & M. Schilthuizen ###### XML Treatment for Craspedotropis gretathunbergae We gratefully acknowledge Mr. Rodzay Wahab, Dyg. Hajah Roshanizah binti Haji Rosli, Mr. Mohammad Salleh and Mr. Teddy Chua at Kuala Belalong Field Studies Centre (KBFSC) for support and staff at the Institute for Biodiversity and Environmental Research for assistance in organising the field course. This field course was carried out under permit UBD/AVC/-RI/1.21.1\[a\] from Universiti Brunei Darussalam. We appreciate the help provided by Bart Van Camp and George Monbiot in making contact with Greta Thunberg. Author contributions ==================== All authors participated in the collecting of the specimens. MS & JL prepared the description of *Craspedotropis gretathunbergae*. AE, PE, AvP prepared the photographs. MS wrote the draft manuscript. All authors read, commented on and approved the manuscript. {#F5386363} {ref-type="supplementary-material"}). Image by Pierre Escoubas.](bdj-08-e47484-g002){#F5386367} {#F5471316} [^1]: Academic editor: Kenneth Hayes

Background ========== Over the past two decades, the detrimental effects of endocrine disruptors (EDs) on wildlife and humans have become a major public health concern. Endocrine disruptors, a large group of environmental pollutants, are believed to act as agonists or antagonists of androgens and estrogens, which are key hormones involved in many physiological processes. These pollutants have been linked to male reproductive defects in humans, including an increase in the incidence of testicular cancer \[[@B1]\], and declining semen quality \[[@B2]\]. Evidence of cryptorchidism, undescended testis and hypospadias have also been demonstrated \[[@B3]\]. In addition, EDs have been linked to developmental problems in the testis and reproductive tract, including reductions in fertility and litter size, induction of cryptorchidism and testicular atrophy \[[@B4],[@B5]\]. Normal development of the male reproductive tract requires interactions between many biological factors and hormones. In particular, androgen hormones are essential to this process. However, many environmental chemicals have androgenic or anti-androgenic effects, or can mimic androgenic activities (i.e., thereby stimulating an androgen-dependent response). Adverse trends in human and animal male reproductive health, particularly with regards to the regulation of environmental factors, suggest that future generations will be at greater risk. Previous reports have suggested that male reproductive system disorders, which often originate during the fetal stage, can appear as testicular dysgenesis syndrome (TDS) after birth \[[@B6]\]. Previous studies demonstrated the possible effects of antiandrogenic- EDs \[i.e., flutamide and/or di- (2 ethylhexyl) phthalate\] on the reduction of androgen synthesis during the development of the male reproductive tract \[[@B7],[@B8]\]. These EDs appear to induce abnormalities in the formation of external genitalia, i.e., hypospadias, cryptorchidism and agenesis of the epididymis, vas deferens and prostate. In additional, the effects of these EDs were also observed with regards to AGD and nipple retention \[[@B5]\]. In humans, some of these alterations are permanent and affect testes function later in life \[[@B9]\]. Although ED-induced harmful effects on male reproduction have been demonstrated, the molecular mechanisms by which EDs disrupt testis development and affect testicular dysgenesis are not clearly understood. Di-ethylhexyl phthalate (DEHP) is widely used as a plasticizer in commercial products \[[@B10]\]. The effects of DEHP on male reproductive development have been well studied in rats \[[@B11]\]. In addition, phthalates and their metabolites can be released from such products and have been detected in the environment \[[@B12]\], posing potential health risks for humans and wildlife. Infants may be exposed to phthalates in the womb \[[@B13]\], via breastfeeding \[[@B14]\] or from medical devices in neonatal intensive care units \[[@B15]\]. Although DEHP has been reported to modulate fetal testosterone production \[[@B16]\], testicular physiology, and mammalian reproduction and fertility \[[@B17]\], the exact mechanisms by which DEHP exerts detrimental effects on body have not yet been fully elucidated. Previous studies have demonstrated the adverse effects of DEHP on the hypothalamic-pituitary-gonadal axis in neonatal female rats, as well as on ex vivo steroidogenesis in granulosa cells (GCs) and secretion of LH by gonadotropes \[[@B18]\]. Moreover, exposure to phthalates during reproductive tract development reduces the number of Sertoli cells (i.e., the major somatic cell type, which supports spermatogenesis) \[[@B19]\]. In addition, DEHP and its metabolites decrease testicular testosterone levels in rodents \[[@B20]\], suggesting potential impacts of these contaminants on Leydig cells. A recent study has indicated that a variety of steroidogenesis related genes were altered following phthalate- exposure and a down-regulation of most of genes involved in testosterone (T) biosynthetic pathways was observed, indicating the potential mechanism for decreased T synthesis induced by phthalate exposure \[[@B21]\]. Other study has reported a similar genetic response in the fetal and prepubertal testes of rats exposed to these environmental chemicals \[[@B22]\]. Flutamide (Flu) is a well-known AR antagonist that is widely used in therapies for androgen-dependent prostate cancer \[[@B23]\] Pre- and postnatal exposure of rats to Flu alters androgen-dependent reproductive development and function \[[@B24]\]. Flu exposure also increases plasma LH levels and stimulates intracellular steroidogenesis in rat testes. In addition, lower ventral prostate and seminal vesicle weights have also been reported, suggesting that Flu exerts anti-androgenic effects on androgen-targeting organs \[[@B25]\]. It has been indicated that exposure of rats to Flu caused a dysregulation in expression of hypothalamus/pituitary hormone genes and consequently this may affect gonadotropin release and induce an over-expression of testicular steroidogenic enzyme genes \[[@B26]\]. In this study, we used an immature rat model of development to examine the adverse effects of DEHP and Flu on the male reproductive system. In particular, we examined the effects of DEHP and Flu on the male reproductive tract and the production of testosterone and LH. We also assessed the histopathological changes of the testis in response to ED exposure. AGD values were used as androgen-dependent markers to assess the anti-androgenic and androgenic activities of the EDs. In addition, we examined alterations in gene expression in the testis of immature rats following DEHP or Flu treatment. Testosterone propionate (TP), an androgen agonist, was used as an indication of androgenic activity in androgen-responsive organs. Our findings will contribute to a better understanding of the detrimental effects of anti-androgenic EDs on humans and wildlife. Methods ======= Chemicals --------- Testosterone propionate (TP; \# 203-08433) and di-(2-ethlhexyl) phthalate (DEHP; \# 80032) were purchased from the Wako Chemical Company (Osaka, Japan). Flutamide (Flu; \# F-9397) and corn oil (i.e., used as a vehicle) were obtained from Sigma-Aldrich Ltd. (St Louis, MO, USA). Animals and treatment --------------------- Sprague-Dawley (SD; 32 males) immature rats were purchased from SamTaKo-Bio Korea (Chungbuk, Korea). The rats were housed in polycarbonate cages in a controlled environment, with an illumination schedule of 12 hour light/12 hour dark. Rats were fed a diet of soy-free pellets (Samyang Ltd., Korea), and water was provided *ad libitum*. All experimental procedures, including those involving animals, were approved by the ethics committee of Chungbuk National University. From postnatal days (PNDs) 21 to 35, rats were treated daily with TP (1 mg/kg body weight \[BW\]/day), DEHP (10, 100 or 500 mg/kg BW/day), Flu (1, 10 or 50 mg/kg BW/day) via oral gavage or with corn oil (5 ml/kg BW/day) as a vehicle. Dosages were adjusted according to changes in body weight. Body weights, clinical signs and abnormal behaviors were recorded daily throughout the experimental periods. All animals were euthanized by exposure to ethyl ether 24 hours after the final treatment. Blood was collected from the descending vena cava and serum was prepared for hormonal analysis. Changes in the weights of testes, epididymides, prostate and seminal vesicles were recorded. Four testes were collected from each group for total RNA isolation. Other testes were fixed in Bouin\'s solution, paraffin embedded, and sectioned at 5 μm for histopathological examination. Hormonal measurements --------------------- After collecting blood from the abdominal aorta, serum was prepared and stored at -20°C for testosterone and LH analyses. The Testosterone Enzyme Immunoassay kit (No. 900-065) was obtained from Assay Designs, Inc (Ann Arbor, USA) and the LH Detect Kit was purchased from INRA (Nouzilly, France). ELISA tests were performed according to the manufacturer\'s recommendations. The coefficients of variation (CV) of intra- and inter-assays were 4% and 14%. Total RNA preparation --------------------- Total RNA was extracted from the testes of immature male rats using Trizol (Invitrogen, Carlsbad, CA, USA), according to the manufacturer\'s recommendations. To make cDNAs from mRNAs for microarray analysis, the same quantity of each RNA sample from the treated groups or control group was pooled. The concentration and quality of each RNA sample was determined by measuring absorbances at 260 nm and 280 nm. cDNA microarray analysis of testes ---------------------------------- The oligo chip of GeneChip^®^Rat Genome 230 2.0 Array (Affymetrix, CA, USA), containing over 31,000 probe sets and representing over 28,000 well-substantiated rat genes, were used according to the supplier\'s instructions. Briefly, total RNA (i.e., 30 μg) from each testis was converted to cDNA for use for microarray analyses. The cDNA products were subsequently subjected to *in vitro*transcription using biotinylated cytidine 5\'-triphosphate and uridine-5\'-triphosphate. Fluorescent labels were incorporated during reverse-transcription of pooled poly (A)^+^RNA from the testes of animals in the DEHP, Flu or control groups, after priming with oligo (dT) primers (Ambion, Austin, TX). Four replicates per group were used for hybridization, thus all together 12 microarrays (3 groups × 4 biological replicates) were employed to evaluate gene alteration by EDs in this study. The fluorescent targets (i.e., Cy3-dCTP controls and Cy5-dCTP experimental cDNAs) were mixed, added to the microarray surface and covered with a cover slip. After overnight incubation at 65°C in a humidified environment, microchips were washed and scanned using a GeneChip Scanner 3000 (Affymatrix, Inc, CA, USA). Quantitative values for signals were calculated using GenePix Pro software, version 5.1 (Molecular Devices Co., CA, USA). Data analysis was performed using the GenePlex (Istech, Inc., Korea). Logged gene expression ratios were normalized by LOWESS (Locally weighted scatter plot smoother) regression as previously described \[[@B27]\]. The biological pathways of the genes were classified based on pathway analysis in which the classification of pathway of interesting genes was determined based on Database for Annotation Visualization and Integrated Discovery (DAVID) <http://david.abcc.ncifcrf.gov/>. The statistical significance of microarray gene expression was assessed by computing a q-value for each gene. To determine the q-value, we used a permutation procedure, and for each permutation a two-sample *t*-statistic was computed for each gene. The result was considered significant when the logarithmic gene expression ratio of four independent hybridizations was more than twofold the difference in the expression level. The accuracy of microarray analysis in this study was confirmed by real-time PCR as previously done \[[@B28]\]. Real-time PCR analysis ---------------------- The results of cDNA microarray analyses were confirmed and validated via real-time PCR. The primers sequences used in real-time PCR analysis are described in Table [1](#T1){ref-type="table"}. Samples were analyzed in a 20 μl reaction volume containing 10 μl of SYBR premix Ex Taq (TaKaRa Bio., Inc.) using a 7300 Real-Time PCR system (Applied Biosystems, Foster, CA, USA), following the manufacturer\'s recommendations. The relative expression level of each gene was normalized to that of HPRT (i.e., an internal control gene) and quantified using RQ software (Applied Biosystems). ###### Primer sequences for Real-time PCR analyses of gene expression **Transcript ID** **Gene Symbol** **Sequense (5\'-3\')** **Size** ------------------- ----------------- --------- ------------------------ ---------- NM_031558 *StAR* Forward tcaaggaatcaaggtcctg 208 Reverse tgttcagctctgatgacacc NM_017286 *Cyp11a1* Forward Atccagcttctttcccaatc 229 Reverse caggatgaggttgaacttgg NM_017265 *HSD3b* Forward Cgctgctgtcattgatgtct 299 Reverse tatgcagtgtgccaccattt NM_133529 *Cabp1* Forward tgactttgtggaactgatgg 232 Reverse gaagtccactcgtccatctc XM_216030 *Vav2* Forward cagaggagacggctgaaaac 338 Reverse gatgaggtcctccaggttga NM_145880 *Lhx1* Forward Ttctggaccgtttcctcttg 198 Reverse gaaccagatcgcttggagag NM_001014242 *Isocl* Forward acacgtctgtatccagcaga 227 Reverse Tggccttaattaggttctgg NM_017035 *Plcd1* Forward agctgccaaaggtcaataag 238 Reverse ctctggccaataaagtcgtt Histopathological examination of testis --------------------------------------- Testis tissues were fixed in Bouin\'s solution and immersed in neutral formalin solution. The fixed tissues were embedded in paraffin, sectioned at 5 μm and mounted on slides. These sections were stained with hematoxylin and eosin (H&E), and histopathological changes were examined under a light microscope. Statistical analyses -------------------- Results are presented as means ± standard deviations (SD). Absolute body weights, sexual organ weights and AGD measurements were analyzed at the time of necropsy. When significant changes were detected, Tukey\'s multiple regression test was used to compare the treatments (i.e., by comparing the control and experimental groups). Data were considered statistically significant at *p*\< 0.05. Results ======= Effects of DEHP and Flu on body weight, reproductive organ weight and anogenital distances ------------------------------------------------------------------------------------------ Exposure of immature male rats to DEHP or Flu did not cause any significant changes in body weight (Figure [1A](#F1){ref-type="fig"}). However, a high doses of DEHP (i.e., 500 mg/kg BW/day) or Flu (i.e., 50 mg/kg BW/day) significantly decreased the weights of reproductive organs (e.g. testis, prostate and seminal vehicle weights) as shown in Figure [1](#F1){ref-type="fig"}. Interestingly, the diminution of epididymis weight was detected in a smaller dose of DEHP and all doses of Flu treatment group. As expected, the high doses of DEHP and Flu significantly decreased AGD when compared with a vehicle, demonstrating the potential effects of DEHP or Flu on androgen-responsive organs (Figure [1F](#F1){ref-type="fig"}). However, any significant effect in the androgen- dependent organs in animals treated with TP was not observed even though there is a tendency of gene alterations by TP as seen Figure [1](#F1){ref-type="fig"}. These results suggest that androgens have been linked with alterations in several end points measured in the immature male onset assay, but seem likely that higher doses of TP may cause this. {#F1} Effects of DEHP and Flu on serum concentrations of testosterone and LH ---------------------------------------------------------------------- To further explore the effects of DEHP and Flu on critical stages in the development of the male reproductive system, serum concentrations of testosterone and LH were measured using an ELISA kit. As shown in Figure [2](#F2){ref-type="fig"}, a significant decrease in the levels of testosterone was observed when rats were exposed to all doses of DEHP (i.e., 10, 100, 500 mg/kg BW/day). However, no significant alteration by any of the two treatments was observed in the serum levels of LH, although these compounds showed a tendency to decrease LH levels. TP treatment as an androgen agonist increased the circulating testosterone without affecting the levels of LH. These data suggest that DEHP exerts an effect on the circulating levels of testosterone in immature male rats. {#F2} Effects of DEHP and Flu on histopathology of the testes ------------------------------------------------------- Histopathological abnormalities in the testes of immature male rats exposed to DEHP and Flu were examined by hematoxylin and eosin (H&E) staining, as described in the *Materials and Methods*. Testes are androgen-responsive tissues; thus, alterations in the morphology and histology of testes can result from the anti-androgenic effects of chemicals. As seen in Figure [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"}, testes morphology and histology were influenced by treatment with DEHP and Flu, respectively. Degeneration of Leydig cells and disorders of germ cells in the reproductive tract were noted in response to all doses of DEHP and Flu. In addition, dilatation of the tubular lumen and stratification of germ cells were observed when rats were treated with DEHP (i.e., 100 and 500 mg/kg BW/day) (Figure [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"}). Hyperplasia of Leydig and Sertoli cells appeared in rat testes at PND 36 in response to high doses of Flu (i.e., 50 mg/kg BW/day) (Figure [4](#F4){ref-type="fig"}). In addition, widespread germ cells disorder or leydig cells degeneration resulted from treatment with all dose of DEHP and Flu (Figure [3](#F3){ref-type="fig"} and [4](#F4){ref-type="fig"}). These results demonstrate the adverse effects of anti-androgenic EDs on the male reproductive tract, particularly with regards to spermatogenesis. {#F3} {#F4} Effects of DEHP and Flu on gene expression in the testes -------------------------------------------------------- Gene expression in immature rat testes in response to TP, DEHP or Flu was assessed using cDNA microarrays. In total, 37,317 of 41,016 genes were dysregulated following treatment with DEHP (i.e., 100 mg) or Flu (i.e., 10 mg), when compared with a control. Among these genes, 1,272 genes were up-regulated over 2-fold change and 1,969 genes were downregulated by more than two-fold. Specific genes were then selected as marker genes for androgenic or anti-androgenic activities of TP, DEHP, or Flu (Table [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}), respectively, including *Orc4l*, *Mgat4a*, and predicted *Scrt1*, *Tmem93*, *RGD1308066*, *Omp, RGD1561053, Cnot3, Ubxd6, RGD1561121, Olr297*and *Cilp*genes. Altered genes have been also classified on the basis of gene ontology (e.g., steroid hormone biosynthetic process, regulation of transcription, signal transduction, metabolic process, catabolic process, biosynthetic process, and integral to membrane, mitochondrion and others) as shown in Table [4](#T4){ref-type="table"}. However, the biological interactions of these proteins in response to EDs (i.e., particularly DEHP and Flu) remains unknown. Six genes (i.e., *CaBP1, Vav2, Plcd1, Lhx1, Hsd3b*and *Isoc1*) were randomly selected to validate the cDNA microarray results via real-time PCR (Figure [5](#F5){ref-type="fig"}). The results of these analyses suggest that these genes may be useful markers to screen for anti-androgenic effects of EDs in androgen-responsive tissues. In addition, two well-known marker genes (i.e., *StAR*and *Cyp11a1*) were used to screen for the anti-androgenic effects of these EDs. As expected, the expressions of steroidogenesis-related genes (i.e., *StAR*and *Cyp11a1*) were significantly affected by Flu exposure (i.e., 50 mg/kg BW/day) (Figures [5A](#F5){ref-type="fig"} and [5B](#F5){ref-type="fig"}). To confirm the correlation between microarray data and real-time PCR, we plotted the fold changes (*versus*vehicle) of microarray and real-time PCR results. There was a good correlation between the results obtained from microarray and real-time PCR (R^2^= 0.824722106, *p*\< 0.0001). Taken together, these results suggest that exposure to DEHP and Flu resulted in an alteration of gene expression in the testis of immature male rats. The distinct transcriptional response to DEHP and Flu reflect that these EDs may exert their anti-androgenic effects via different mechanisms. {#F5} ###### Up-regulated gene induced by TP, DEHP and Flu in the immature rat testes. **Transcript ID** **Gene Symbol** **Gene Name** **Fold change (compared with VE)** ------------------- ------------------------ ------------------------------------------------------------------------------ ------------------------------------ ---------- ---------- **TP UP** XM_345848 *Scrt1_predicted* scratch homolog 1, zinc finger protein (Drosophila) **5.88** 1.84 1.46 XM_213394 *Tmem93_predicted* transmembrane protein 93 (predicted) **5.07** 1.49 1.43 XM_215951 *Kcng1* potassium voltage-gated channel, subfamily G, member 1 **3.07** 1.13 0.96 NM_001008295 *Fip1l1* FIP1 like 1 (S. cerevisiae) **2.44** 1.00 1.08 NM_021669 *Ghrl* ghrelin precursor **2.37** 1.02 1.00 XM_221387 *Etv5_predicted* ets variant gene 5 (ets-related molecule) (predicted) **2.14** 1.01 1.13 XM_215635 *Apoa1bp_predicted* apolipoprotein A-I binding protein (predicted) **2.11** 1.31 1.22 NM_031802 *Gabbr2* gamma-aminobutyric acid (GABA) B receptor 2 **2.06** 1.29 0.80 NM_133529 ***Cabp1\**** calcium binding protein 1 **2.05** 1.11 0.97 XM_216030 ***Vav2\**** Vav2 oncogene (predicted) **1.77** 1.41 1.11 **DEHP UP** XM_213289 *RGD1308066_predicted* similar to KIAA1960 protein (predicted) 0.99 **3.29** 1.64 NM_012616 *Omp* olfactory marker protein 1.07 **3.08** 1.19 NM_021590 *Aipl1* aryl hydrocarbon receptor-interacting protein-like 1 1.31 **2.91** 1.22 NM_031770 *Gnb5* guanine nucleotide binding protein, beta 5 1.14 **2.65** 1.10 NM_001000980 *Olr1366* olfactory receptor 1366 1.33 **2.62** 1.16 NM_022625 *Tpc1808* tropic 1808 1.11 **2.43** 1.36 NM_001014015 *Rbm34* RNA binding motif protein 34 1.28 **2.43** 1.22 NM_199291 *Doxl2* diamine oxidase-like protein 2 1.27 **2.42** 1.01 XM_220283 *Sox8_predicted* SRY-box containing gene 8 (predicted) 1.10 **2.33** 1.18 NM_023992 *Kiss1r* KISS1 receptor 1.12 **2.31** 1.24 NM_001000365 *Olr748_predicted* olfactory receptor 748 (predicted) 1.21 **2.30** 1.26 NM_017035 ***Plcd1\**** phospholipase C, delta 1 1.20 **2.20** 1.24 XM_221627 *Brwd1_predicted* bromodomain and WD repeat domain containing 1 (predicted) 1.30 **2.19** 1.18 XM_221521 *Hoxd4_predicted* homeo box D4 (predicted) 0.67 **2.17** 0.93 XM_344627 *Calm4_predicted* calmodulin 4 (predicted) 1.08 **2.16** 0.84 XM_216225 *Ppp4r2_predicted* protein phosphatase 4, regulatory subunit 2 (predicted) 1.22 **2.15** 1.45 NM_175587 *Taar7h* trace-amine-associated receptor 7 h 1.14 **2.14** 1.11 NM_020106 *Olr414_predicted* olfactory receptor 414 (predicted) 1.35 **2.13** 0.94 NM_145880 ***Lhx1\**** LIM homeobox protein 1 1.04 **1.72** 1.20 **Flu UP** XM_575355 *RGD1561053_predicted* similar to Claudin 12 (predicted) 1.49 1.16 **3.18** XM_218187 *Cnot3_predicted* CCR4-NOT transcription complex, subunit 3 (predicted) 1.21 1.10 **2.48** NM_012992 *Npm1* nucleophosmin 1 1.31 1.13 **2.25** NM_017265 ***Hsd3b1\**** hydroxy-delta-5-steroid dehydrogenase, 3 beta- and steroid delta-isomerase 1 0.87 0.97 **2.18** XM_214673 *Cog8_predicted* component of oligomeric golgi complex 8 (predicted) 0.75 0.79 **2.07** XM_225336 *Lrrc16_predicted* leucine rich repeat containing 16 (predicted) 1.00 1.31 **2.06** XM_226843 *Rnasen* ribonuclease III, nuclear 0.93 0.69 **2.05** XM_234056 *RGD1564242_predicted* similar to KIAA1218 protein (predicted) 0.97 1.02 **2.05** NM_013175 *Sgne1* secretory granule neuroendocrine protein 1 0.83 0.81 **2.04** XM_221047 ***Polg2_predicted*** polymerase (DNA directed), gamma 2, accessory subunit (predicted) 1.14 1.05 **2.01** The cDNA microarray analysis for detection of fold changes of gene expression in the testis of immature rats following treatments with TP, DEHP or Flu. Altered gene up-regulation expression was estimated by the ratio of TP, DEHP 100 mg/kg BW/day and Flu 10 mg/kg BW/day treated vs. control as follows. \*, These altered genes were selected for verification by real- time PCR. ###### Down-regulated gene induced by TP, DEHP and Flu in the immature rat testes. ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- **Transcript ID** **Gene Symbol** **Gene Name** **Fold change (compared with VE)** ------------------- ------------------------ ----------------------------------------------------------------------------------- ------------------------------------ ---------- ---------- **TP DOWN** NM_001012225 *Mgat4a* Mannoside acetylglucosaminyltransferase\ **0.30** 0.69 0.74 4, isoenzyme A XM_214360 *Ubxd6_predicted* UBX domain containing 6 (predicted) **0.35** 0.82 0.63 XM_342534 *Nat5_predicted* N-acetyltransferase 5 (ARD1 homolog, S. cerevisiae) (predicted) **0.39** 0.77 1.03 NM_053356 *Col1a2* procollagen, type I, alpha 2 **0.42** 0.81 0.79 XM_345107 *RGD1559810_predicted* similar to hypothetical protein (predicted) **0.42** 0.83 0.77 NM_133317 *Tob1* transducer of ErbB-2.1 **0.43** 0.82 1.04 XM_341903 *Nrip3_predicted* nuclear receptor interacting protein 3 (predicted) **0.44** 1.90 1.87 XM_229139 *RGD1566225_predicted* similar to RIKEN cDNA 1700001F22 (predicted) **0.44** 1.00 0.82 NM_001013185 *Cabc1* chaperone, ABC1 activity of bc1 complex like (S. pombe) **0.45** 1.08 0.73 XM_235640 *Tmem16f_predicted* transmembrane protein 16F (predicted) **0.46** 0.87 0.68 NM_001010966 *Pigv* phosphatidylinositol glycan, class V **0.49** 2.21 1.73 NM_133318 *Khdrbs2* KH domain containing, RNA binding, signal transduction associated 2 **0.49** 0.82 0.85 **DEHP DOWN** XM_218447 *RGD1561121_predicted* similar to pleckstrin homology-like domain, family B, member 3 (predicted) 1.26 **0.12** 0.99 NM_199092 *Orc4l* origin recognition complex, subunit 4-like (S. cerevisiae) 0.70 **0.40** 0.72 NM_031753 *Alcam* activated leukocyte cell adhesion molecule 0.75 **0.46** 0.80 NM_001014242 ***Isoc1\**** isochorismatase domain containing 1 0.85 **0.46** 0.90 XM_215497 *Rad1_predicted* RAD1 homolog (S. pombe) (predicted) 0.97 **0.46** 0.86 XM_575065 *RGD1559623_predicted* similar to RIKEN cDNA 5230400J09 (predicted) 0.93 **0.48** 0.69 NM_139189 *Lmbrd1* LMBR1 domain containing 1 0.76 **0.48** 0.74 XM_236578 *RGD1562949_predicted* similar to mKIAA0259 protein (predicted) 0.91 **0.49** 1.26 XM_215487 *Mrps30_predicted* mitochondrial ribosomal protein S30 (predicted) 0.90 **0.49** 0.76 **Flu DOWN** NM_001000234 *Olr297_predicted* olfactory receptor 297 (predicted) 1.02 1.54 **0.25** XM_236348 *Cilp_predicted* cartilage intermediate layer protein, nucleotide pyrophosphohydrolase (predicted) 0.65 0.89 **0.40** XM_228462 *RGD1565441_predicted* similar to KIAA1687 protein (predicted) 0.90 1.02 **0.43** XM_219296 *Rbbp6* retinoblastoma binding protein 6 0.65 0.82 **0.45** NM_021653 *Dio1* deiodinase, iodothyronine, type I 0.97 0.97 **0.46** ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Altered gene down-regulation expression was estimated by the ratio of TP, DEHP 100 mg/kg BW/day and Flu 10 mg/kg BW/day treated vs. control as follows. \*, These altered genes were selected for verification by real- time PCR. ###### Functional Categorization of Genes Significantly Altered via microarray analysis in immature male testes following TP, DEHP and Flu exposure **Functional category** **Transcript ID** **Gene Symbol** ------------------------------------------ ------------------- ------------------------ **Steroid hormone biosynthetic process** NM_031558 *Star* NM_017286 *Cyp11a1* **Regulation of Transcription** XM_221387 *Etv5_predicted* XM_220283 *Sox8_predicted* XM_221521 *Hoxd4_predicted* NM_145880 *Lhx1* XM_218187 *Cnot3_predicted* XM_229139 *RGD1566225_predicted* XM_221627 *Brwd1_predicted* NM_133317 *Tob1* **Signal Transduction** NM_031802 *Gabbr2* XM_216030 *Vav2_predicted* NM_012616 *Omp* NM_031770 *Gnb5* NM_001000980 *Olr1366* NM_023992 *Kiss1r* NM_175587 *Taar7h* NM_020106 *Olr414_predicted* NM_012992 *Npm1* NM_031753 *Alcam* XM_344627 *Calm4_predicted* **Metabolic process** NM_021590 *Aipl1* NM_001012225 *Mgat4a* XM_342534 *Nat5_predicted* NM_001014242 *Isoc1* XM_236348 *Cilp_predicted* **Catabolic process** NM_022625 *Tpc1808* NM_017035 *Plcd1* **Biosynthetic process** NM_017265 *Hsd3b1* NM_001010966 *Pigv* NM_021653 *Dio1* **Integral to membrane** XM_213394 *Tmem93_predicted* NM_001000365 *Olr748_predicted* XM_575355 *RGD1561053_predicted* XM_214673 *Cog8_predicted* XM_214360 *Ubxd6_predicted* NM_001000234 *Olr297_predicted* NM_139189 *Lmbrd1* XM_235640 *Tmem16f_predicted* **Mitochondrion** NM_001013185 *Cabc1* XM_215487 *Mrps30_predicted* **Protein binding** XM_215635 *Apoa1bp_predicted* NM_133529 *Cabp1* NM_013175 *Sgne1* XM_221047 *Polg2_predicted* NM_053356 *Col1a2* NM_133318 *Khdrbs2* XM_228462 *RGD1565441_predicted* XM_219296 *Rbbp6* **Others** XM_215951 *Kcng1* NM_001008295 *Fip1l1* NM_021669 *Ghrl* XM_213289 *RGD1308066_predicted* NM_022625 *Tpc1808* NM_001014015 *Rbm34* NM_199291 *Doxl2* XM_216225 *Ppp4r2_predicted* XM_225336 *Lrrc16_predicted* XM_226843 *Rnasen* XM_234056 *RGD1564242_predicted* XM_345107 *RGD1559810_predicted* XM_341903 *Nrip3_predicted* XM_218447 *RGD1561121_predicted* NM_199092 *Orc4l* XM_215497 *Rad1_predicted* XM_575065 *RGD1559623_predicted* Discussion ========== There are a number of mechanisms by which EDs may alter the action of the endocrine system and interfere with the reproduction of humans and animals. Previous studies have shown that EDs mimic the actions of natural hormones and can stimulate or inhibit various enzymes required for hormone synthesis. Consequently, EDs interfere with the regulation of gene expression, disrupting the natural hormone balance and interfering with the normal action of the reproductive system. In particular, EDs may exert distinct androgenic and anti-androgenic effects on the developing male reproductive system in utero \[[@B29]\]. Although previous studies have examined the effects of these chemicals on critical stages in the development of the male reproductive system, the relationship between molecular events and detrimental effects of these EDs is not well described. To further explore these effects, we used an immature rat model of development to examine the adverse effects of DEHP and Flu on the male reproductive system. Oral treatments with DEHP (i.e., 10, 100 and 500 mg/kg BW/day) and Flu (i.e., 1, 10 and 50 mg/kg BW/day) failed to induce significant effects on body weight in a dose-dependent manner. However, high doses of DEHP and Flu significantly decreased the weights of reproductive organs, with the exception of the epididymis. Although treatment with high doses of DEHP failed to induce anti-androgenic effects on epididymis weight, these effects were observed after treatment with low doses of DEHP (i.e., 10 mg/kg BW/day). These results provide evidence of the anti-androgenic effects of DEHP in the epididymis. A previous study reported that treatment with DEHP (i.e., 500 mg/kg BW/day) failed to induce significant effects on body weight \[[@B30]\]. However, other studies have shown significant decreases in body weight in response to Flu (i.e., 100 or 150 mg/kg BW/day) \[[@B31]\], suggesting that these EDs differ markedly with regards to their detrimental affects on human and animals. Further studies are required to elucidate the molecular and biochemical mechanisms underlying the effects of anti-androgenic ED exposure in humans and animals, particularly during the critical stages of male reproductive development. It has been suggested that alterations in endogenous androgen levels or dysregulation of gene expression patterns causes abnormalities in male reproductive tract (i.e., cyptochism, hypospadia, stunted testicular growth, epididymal abnormalities, and AGD \[[@B32]-[@B34]\]. In addition, the anti-androgenic and androgenic effects of EDs can be reflected by changes in the weights of the testes, prostate and seminal vesicle. A previous study of rat offspring indicated dose-dependent reductions in the ventral and dorsolateral prostate weights in response to DEHP \[[@B35]\]. Another study reported malformation of the male reproductive tract following exposure to DEHP, manifesting as disgenesis of the epididymis, decreased sperm production, Leydig cell hyperplasia and adenomas \[[@B36]\]. A significant decrease in the weights of rat testes in response to DEHP (i.e., 250, 500, 1000 or 2000 mg/kg BW/day by gavage) has also been reported \[[@B37]\]. Another study demonstrated that the epididymal, prostate and seminal vesicle weights of immature male rats were strongly affected by the anti-androgenic effects of Flu at PND 20 \[[@B38]\]. Previous studies have proposed an androgen-dependent marker for the anti-androgenic and androgenic effects of EDs using AGD. A significant reduction in AGD is evoked by DEHP exposure during lactation \[[@B39]\] and by prenatal exposure to Flu \[[@B40]\]. Moreover, changes in AGD during early postnatal periods correlate with alterations in androgen-dependent development in adults \[[@B41]\]. In the present study, AGD values decreased significantly in response to high doses of DEHP or Flu; however, no alterations were observed after treatment with their medium and low doses. Although there is a correlation between EDs and reproductive, developmental and behavioral changes at high doses in experimental animals, further studies are required to determine if low doses may also contribute to reproductive disorders in humans and wildlife \[[@B39]\]. A recent study has indicated the property of phthalates associated with endocrine disruption and these compounds are believed to act as endocrine disruptors \[[@B42]\]. Exposure to phthalates reduces testosterone synthesis during this critical stage of development \[[@B7]\]. Phthalates exert their effects on steroidogenesis in Leydig cells via modulation of testosterone-biosynthetic enzyme activity and serum LH levels \[[@B43]\]. In contrast, different responses to Flu may result from local concentrations of androgenic and anti-androgenic compounds. Exposure to Flu may block the physiological action of testosterone at AR sites \[[@B44]\] and induce changes in circulating LH levels, due to disturbances in the negative feedback loop between the pituitary and the testis \[[@B40]\]. Although DEHP and Flu share an anti-androgenic activity, the mechanisms by which these EDs exert their effects on human body are distinct. It has indicated in many previous reports that AR plays an important role in Flu-mediated response, while phthalates, including DEHP, do not interact with AR at the physiological concentration \[[@B45]\]. Additionally, DEHP is thought to activate PPAR, leading a down-regulation of SR-B1 and PBR or effects on SF-1 to consequently regulate steroidogenesis-related genes \[[@B16]\]. To further understand the effects of DEHP and Flu during the critical stages of male reproductive development, we measured the serum concentrations of testosterone and LH. However, exposure to all doses of Flu and DEHP failed to induce a significant effect in the serum LH levels, while a significant decreased level in the serum concentrations of testosterone was observed in response to all doses of DEHP. We did not expect to observe Flu-induced effect on serum concentrations of testosterone and LH, however, the response of Flu on their serum concentrations may be explained by the lack of a negative feedback mechanism. Further experiments with larger sample sizes are required to verify these findings. It has been indicated that exposure to a high dose of DEHP caused a decrease in testosterone production and consequently reduced AGD value \[[@B46]\]. In this study, serum testosterone concentration and AGD value were reduced in response to a high dose of DEHP, suggesting that an alteration in testosterone synthesis may result from phthalate-induced dysfunctional interaction between Leydig and Sertoli cells \[[@B47]\]. Thus, the pathological changes induced by anti-androgenic effects in the testis during male reproductive tract development may alter serum testosterone levels. Testes are androgen-responsive tissues, and alterations in the morphology and histology of testes can reflect the anti-androgenicity effects of chemicals. In addition, an evaluation of histological and biological effects is very important to understand the potential risk impacts for spermatogenesis from endocrine disrupting chemicals \[[@B48]\]. In this study, histopathological changes in the testes of immature male rats exposed to DEHP and Flu were examined, and abnormalities in the cell morphology and histology of the testes were observed (i.e., abnormal testicular development or dysgenesis, Leydig cell hyperplasia, morphologically distorted tubules, differentiated Sertoli cells and abnormal germ cells), particularly in response to high doses of DEHP and Flu. A previous study indicated that in utero exposure to DEHP (1,000 mg/kg BW) caused the dilatation and atrophy of seminiferous tubules in rats. In addition, exposure to DEHP (500 mg/kg BW) may result in the abnormalities of the cell morphology in which multinucleated germ cells were observed in seminiferous cords \[[@B49]\]. Another study has indicated the age-related difference in the toxic effect(s) of this ED in the testis of rats \[[@B50]\]. A testicular atrophy and/or a decline in zinc concentration were observed when exposure of immature rats to DEHP \[[@B51]\], demonstrating the DEHP- and Flu-induced effects on male reproductive development, particularly with regards to spermatogenesis. A previous study has investigated the alteration in the gene expression following DEHP treatment in which exposure to this compound caused a dysregulation in the expression of many genes, including apoptosis-, cell proliferation-, metabolism-, cell adhesion- and immune response- related genes \[[@B52]\]. In this study, the gene expression in immature rat testes was also assessed via cDNA microarray assays. A total of 37,317 genes were dysregulated in response to DEHP (i.e., 100 mg) and Flu (i.e., 10 mg) treatment, when compared with a control. Among these genes, 1,272 were overexpressed by more than two fold, and 1,969 genes were downregulated (Table [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"}). Some of these genes were chosen for use as marker genes. We previously demonstrated that maternal exposure to DEHP and Flu modulates the expression of fetal *StAR, Cyp11a1*and *Hsd3b1*genes, wherein DEHP treatment downregulated transcription \[[@B53]\]. However, no significant effects were observed in the current study. Interestingly, treatment with high doses of Flu significantly increased the expression of these genes (i.e., by 4- to 6-fold). Several genes shown in Table [2](#T2){ref-type="table"} and [3](#T3){ref-type="table"} \[i.e., calcium binding protein 1(*CaBP1*); vav2 oncogene- predicted (*Vav2- predicted*); phospholipase C delta 1 (*Plcd1*); lim homeobox protein 1 (*Lhx1*) and isochorismatase domain containing 1 (*Isoc1*)\] were previously identified because of their direct or indirect involvement in physiological processes (e.g., lipid or steroid metabolism, sex determination, calcium transduction or cell proliferation). However, the biological interactions of these proteins in response to EDs (i.e., particularly DEHP and Flu) remains unknown. In current study, we also found that some genes were strongly up-regulated by TP treatment, whereas no differences at transcriptional levels were observed in the testes of rats exposed to DEHP and/or Flu. Furthermore, expression of other genes, including nuclear receptor interacting protein 3 (predicted) and phosphatidylinositol glycan, class V, were decreased markedly in TP treated group, while an enhancement in expression of these genes was observed following DEHP or Flu treatment. A distinct response to TP and DEHP/Flu may reflect hormone properties of tested chemicals. Additionally, a different pattern of gene expression also was found between DEHP and Flu treatment. These distinct responses to DEHP and Flu may be explained by the differential mechanism of actions of the two anti-androgenic EDs. The discrepancy was observed in the microarray results compared to the previous ones, which appears to be derived from different microarray system employed and different experimental environments among the studies. It has been reported that calcium binding protein-1 (i.e., encoded by *CaBP1*) functions as a calmodulin-like protein that modulates Ca2+ channel activities \[[@B54]\]. An interaction between Ca2+ channels and *CaBP1*may regulate the Ca2+-dependent forms of synaptic plasticity by inhibiting Ca2+ influx into neurons \[[@B55]\]. In this study, the expression of *CaBP1*was significantly up-regulated by TP exposure. Our data revealed that exposure to TP activated the predicted *Vav2*oncogene. The *Vav2*gene is involved in many biological processes, including the B- and T-cell receptor signaling pathways, leukocyte transendothelial migration, natural killer cell-mediated cytotoxicity and regulation of the actin cytoskeleton. Some previous studies have demonstrated an appearance of *CaBP1*and *Vav2*oncogene genes during spermatogenesis, suggesting their potential roles in the critical window stage of male reproductive system development \[[@B56],[@B57]\]. Phospholipase C delta 1 (*Plcd1*), a ubiquitous enzyme, helps regulate a variety of cellular processes. The *Plcd1*gene has been detected in the plasma membrane, the cytoplasm \[[@B58]\], and liver mitochondria \[[@B59]\], and the involvement of *Plcd*in cardiac function has also been reported \[[@B60]\]. In addition, the activity of *Plcd1*is affected by cytoplasmic concentrations of Ca2+ \[[@B61]\]. A previous study also demonstrated that FSH-induced Ca(2+) influx is mediated by the *Plcd1*signaling pathway in rat Sertoli cells \[[@B62]\]. Moreover, a testosterone-induced increase in intracellular Ca2+ occurs via activation of a plasma membrane receptor associated with the phospholipase C signaling pathway \[[@B63]\]. Our results revealed a significant increase in the expression of its pathway. In addition, we found that the expression of LIM homeobox gene 1 (*Lhx1*) was significantly modulated by DEHP, suggesting that this gene is a potential indicator for the effects of DEHP on androgen-responsive tissues. As a member of the large homeobox gene family, *Lhx1*plays an important role in the stabilization of the intermediate mesoderm and in the formation of urogenital ridges \[[@B64]\]. This gene is expressed in the comma- and S-shaped bodies of the metanephric mesenchyme, as well as the developing Müllerian duct \[[@B65]\]. In the male reproductive system, *Lhx1*plays an important role in early gonad development, sex-reversal and gonadal cell proliferation. Several studies have reported the involvement of *Isoc1*in progressive chronic renal failure in rats \[[@B66]\]. Sexual hormones (e.g., estradiol, testosterone, prolactin and FSH) concentrations are commonly affected by chronic renal failure and the dysfunction of sex steroids may contribute to the emergence of renal osteodystrophy \[[@B67]\]. Our results showed a 6-fold downregulation in *Isco1*gene expression after exposure to DEHP, representing an anti-androgen-like ED exposure. These results demonstrate the effects of EDs on sexual hormone abnormalities, and are likely to induce similar changes in the reproductive development of male humans and rats. Overall, our data indicated that the these changes may reflect an impact of chemical exposure on transcriptional level of gene expression and associated with histology changes of Leydig cells, Sertoli cells and/or germ cells, however the role(s) of these altered genes to explain a change in altered phenotype or morphology has unknown. It is expected for us to investigate the molecular events of these maker genes in the histopathological testis changes when EDs exposure during the critical male reproductive development stage in a further study. Taken together, our findings indicate that exposure to DEHP and Flu results in an alteration in gene expression in the testes of immature male rats. In addition, these chemicals exert distinct anti-androgenic effects on the male reproductive system. Our findings provide new insight into the molecular mechanisms underlying the detrimental impacts of anti-androgenic-EDs in androgen-responsive organs, in particular, in developing male reproductive tracts. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= TTPV and EMJ carried out the overall experiments to complete this study and drafted the manuscript. VHD and YMY performed experiments, in part, and drafted the manuscript. KCC performed real-time PCR and drafted and completed the manuscript. FHY participated in its design and coordination and performed the statistical analysis. EBJ conceived of the study, and participated in its design and coordination and helped to draft the manuscript. All authors read and approved the final manuscript. Acknowledgements ================ This work was supported by a grant (07142KFDA593) from the Korean Food & Drug Administration in 2007.

The etiology of rotator cuff tearing is unclear, although likely multifactorial. Decreased vascularity, mechanical impingement, intrinsic degeneration, trauma, and a genetic predisposition have all been postulated as causative factors associated with the development of rotator cuff tearing.^[@bibr3-2325967116642173],[@bibr7-2325967116642173],[@bibr14-2325967116642173],[@bibr15-2325967116642173],[@bibr17-2325967116642173],[@bibr20-2325967116642173]^ Rotator cuff disease has been observed in excess in patients with other tendinopathies and compression neuropathies, including lateral epicondylitis, Achilles tendonitis, carpal tunnel syndrome, and de Quervain tenosynovitis.^[@bibr22-2325967116642173],[@bibr23-2325967116642173]^ Kraushaar and Nirschl^[@bibr11-2325967116642173]^ described a syndrome in patients with rotator cuff tendonitis with a predisposition to developing tendonitis at several sites, including the lateral epicondylitis, and carpal tunnel syndrome as a result of flexor tendon tenosynovitis. Several epidemiologic studies have reported an association of various tendinopathies and compression neuropathies in occupational work injuries.^[@bibr16-2325967116642173],[@bibr18-2325967116642173]^ No prior studies have evaluated a potential genetic predisposition for the concomitant development of tendinopathies or compression neuropathies in patients with rotator cuff tearing. Several authors have investigated the clinical association of rotator cuff tendonitis and other tendinopathies or compression neuropathies. Titchener et al,^[@bibr23-2325967116642173]^ utilizing a large national health care database, determined a significant association between individuals with rotator cuff disease and Achilles tendonitis, lateral epicondylitis, and carpal tunnel syndrome. The same authors reported an association between lateral epicondylitis and rotator cuff pathology, de Quervain disease, and carpal tunnel syndrome.^[@bibr22-2325967116642173]^ Limitations of their analysis include a heterogeneous rotator cuff phenotype based on hospital coding. Histologic studies comparing tendon biopsies from patients with rotator cuff tendonitis, lateral epicondylitis, and Achilles tendonitis show remarkable similarities.^[@bibr6-2325967116642173],[@bibr9-2325967116642173]^ A genetic etiology of global tendon dysfunction could potentially explain these histologic and clinical associations. The purpose of this study was to examine evidence for association of rotator cuff tearing and other tendinopathies and compression neuropathies in rotator cuff tear cases and their close and more distant relatives to identify evidence for a genetic contribution to the disease association. Identification of a genetic predisposition for global tendon dysfunction and compression neuropathies would support the use of preventative measures, including work modification and rehabilitation in individuals at risk for the development of these diseases. We tested the hypothesis of excess risk for global tendinopathies or compression neuropathies in patients with rotator cuff tearing and their relatives using a well-established method: estimation of relative risks (RRs). Our hypothesis is that patients with rotator cuff tears are at greater risk to have associated tendinopathies and compression neuropathies as well as their family members compared with a control population without a diagnosis of rotator cuff tearing. Methods {#section1-2325967116642173} ======= The Utah Population Database (UPDB) includes genealogical data for over 2 million Utah founders and their descendants; some family records extend back 15 generations.^[@bibr19-2325967116642173]^ The original Utah genealogy data came from 3 generation family group sheets with a couple, parents of each member of the couple, and all of their children. Since the early 1970s, genealogy data have been added in the form of trios from state vital statistics data (eg, mother, father, and child from a birth certificate). This has resulted in individual genealogy data of varying quality and quantity. We identified all individuals with genealogy data whose genealogy data included both parents, all 4 grandparents, and at least 6 of their 8 great-grandparents (12 of their 14 immediate ancestors) who also had linked medical data for this analysis. The University of Utah Health Sciences Center (UUHSC) data warehouse is an electronic database including all patient-related information from inpatient hospital and outpatient clinic visits since 1994. The UUHSC data has been linked to the UPDB genealogical data to allow identification of patients with a phenotype of interest and study of the genetic relationships between affected individuals. The majority of recognized genetic relationships occur in the same generation (eg, siblings, cousins) due to the small window of view to phenotype data. Methods for the analysis of the UPDB genealogical and linked hospital data have been well reported and represent examination of a number of different phenotypes.^[@bibr2-2325967116642173],[@bibr21-2325967116642173],[@bibr25-2325967116642173]^ We estimated relative risks for tendinopathies and for compression neuropathies in rotator cuff tear cases, their spouses, and in their first-, second-, and third-degree relatives in the UPDB who also had UUHSC data to test for an association between these phenotypes. Disease Rates {#section2-2325967116642173} ------------- To estimate RRs for these disease associations, it is necessary to calculate the rates of rotator cuff tearing and other tendinopathies and compression neuropathies. We estimate these disease rates from all hospital patients who have linked genealogical data (approximately 400,000 patients). An attempt is made to record-link all patients at the University of Utah Hospital and Clinics to the genealogical data in the UPDB each year. All individuals in the UPDB with UUHSC data and at least 12 of their 14 immediate ancestors also in the UPDB were assigned to 1 of 205 cohorts. These cohorts are based on sex, 5-year birth year range, birth state (Utah or not Utah), and birth county (urban or rural). Cohort-specific rates for each phenotype were calculated using the total number of UUHSC patients in each cohort as the denominator and the number of patients with the phenotype of interest in each cohort as the numerator. Relative Risks in Relatives {#section3-2325967116642173} --------------------------- The RR for a phenotype of interest among the relatives of patients with rotator cuff tear is estimated as the ratio of the number of cases observed among the relatives to the number of cases expected among the relatives. For example, to estimate the RR of compression neuropathy among the first-degree relatives of individuals with rotator cuff tearing, we compared the number of first-degree relatives of rotator cuff tear patients who were diagnosed with compression neuropathy to the expected number of relatives diagnosed with compression neuropathy. The number of expected cases can be obtained by multiplying the population rate of compression neuropathy times the number of first-degree relatives. Because disease rates can vary by sex and birth year, we perform this calculation stratified by cohort to get a more accurate count of expected number of cases. The number of expected cases of compression neuropathy among first-degree relatives of the patients with rotator cuff tearing was therefore calculated by counting all of the first-degree relatives of the rotator cuff tear patients by cohort (without duplication), multiplying the cohort-specific number of relatives by the cohort-specific rate of compression neuropathy, and then summing over all cohorts. The RR of compression neuropathy among first-degree relatives of patients with rotator cuff tear was then estimated as the ratio of observed cases to expected cases. The significance of the test of the null hypothesis (RR = 1.0) was determined by a Fisher exact test for the 2 × 2 table. Confidence intervals for the RR were estimated as given by Agresti.^[@bibr1-2325967116642173]^ No patient identifiers were used in this study; all analysis of genetic relationships between affected individuals was nonidentifiable. The study was approved by both the University of Utah Institutional Review Board as well as the oversight body for the UPDB. Results {#section4-2325967116642173} ======= Identification of Cases {#section5-2325967116642173} ----------------------- We required that all patients have a minimum genealogy data set that included at least 12 of their 14 immediate ancestors. The number of such patients with rotator cuff tear identified with various diagnoses and procedure coding is shown in [Table 1](#table1-2325967116642173){ref-type="table"}. Patients are only counted once regardless of how many diagnostic codes for a phenotype of interest were present in their record; all patient counts in [Table 1](#table1-2325967116642173){ref-type="table"} show the count for the first code encountered in the medical record. ###### UUHSC Patients With a Diagnosis of Rotator Cuff Surgery or Tear*^a^*  ICD-9 or CPT-4 Code Definition Patients in Cohort, n --------------------- ----------------------------------------------------------------------------------------------- ----------------------- ICD-9 727.61 Complete rupture of rotator cuff, nontraumatic 112 ICD-9 840.4 Rotator cuff traumatic strain 1386 CPT-4 29827 Arthroscopy, shoulder, surgical; with rotator cuff repair 275 CPT-4 23410 Repair of ruptured musculotendinous cuff open; acute 32 CPT-4 23412 Repair of ruptured musculotendinous cuff open; chronic 45 CPT-4 23420 Reconstruction of complete shoulder (rotator) cuff avulsion, chronic (includes acromioplasty) 39 Total patients 1889 *^a^*Patients were only counted once regardless of how many rotator diagnosis codes were present in their record. All patient counts show the count for the first ICD-9 or CPT-4 code encountered in the medical record. CPT-4, Current Procedural Terminology, Fourth Revision; ICD-9, International Classification of Diseases, Ninth Revision; UUHSC, University of Utah Health Sciences Center. We used the International Classification of Diseases, Ninth Revision (ICD-9) diagnosis codes and the Current Procedural Terminology, Fourth Revision (CPT-4) procedure codes to identify all patients at the UUHSC with a diagnosis of rotator cuff tearing or rotator cuff repair surgery ([Table 1](#table1-2325967116642173){ref-type="table"}). We identified 1889 patients with rotator cuff repair surgery or tear by the presence of CPT-4 codes (29827, 23410, 23412, or 23420) or ICD-9 codes (727.61 or 840.4). The tendinopathy diagnoses and compression neuropathies analyzed were identified using ICD-9 and CPT-4 codes as shown in [Table 2](#table2-2325967116642173){ref-type="table"}. ###### Codes Included for Tendon Dysfunction and Compression Neuropathy and Phenotypes.*^a^*  ICD-9 or CPT-4 Code Definition ----------------------------------- -------------------------------------------------------------------------------------------------------------------------- Rotator cuff tendon dysfunction  ICD-9 727.61 Complete rupture of rotator cuff, nontraumatic  ICD-9 840.4 Rotator cuff traumatic strain  CPT-4 29827 Arthroscopy, shoulder, surgical; with rotator cuff repair  CPT-4 23410 Repair of ruptured musculotendinous cuff open; acute  CPT-4 23412 Repair of ruptured musculotendinous cuff open; chronic  CPT-4 23420 Reconstruction of complete shoulder (rotator) cuff avulsion, chronic (includes acromioplasty) Elbow tendon dysfunction  ICD-9 726.12 Bicipital tenosynovitis  ICD-9 726.30 Enthesopathy of elbow, unspecified  ICD-9 726.31 Enthesopathy of elbow region; medial epicondylitis  ICD-9 726.32 Enthesopathy of elbow region; lateral epicondylitis, epicondylitis not otherwise specified, golfers' elbow, tennis elbow  CPT-4 24340 Tenodesis distal biceps  CPT-4 24342 Reinsertion of distal bicep or tricep  CPT-4 24350 Fasciotomy, lateral or medial elbow (tennis elbow, epicondylitis)  CPT-4 24351 Fasciotomy, lateral or medial elbow with extensor origin detachment  CPT-4 24356 Fasciotomy, lateral or medial elbow with partial osteotomy  CPT-4 24357 Tenotomy, elbow, lateral or medial epicondylitis percutaneous  CPT-4 24358 Tenotomy, elbow, lateral or medial epicondylitis, open  CPT-4 24359 Tenotomy, elbow, lateral or medial epicondylitis, open with tendon repair Hand and wrist tendon dysfunction  ICD-9 726.4 Enthesopathy of the wrist and carpus  ICD-9 727.63 Extensor tendons of hand and wrist nontraumatic rupture  ICD-9 727.64 Flexor tendons of hand and wrist nontraumatic rupture  ICD-9 727.00 Other disorders of synovium, tendon, and bursa, synovitis and tenosynovitis, synovitis and tenosynovitis unspecified  ICD-9 727.05 Other disorders of synovium, tendon and bursa, synovitis and tenosynovitis, other tenosynovitis of hand and wrist  ICD-9 727.03 Other disorders of synovium, tendon, and bursa, synovitis and tenosynovitis, trigger finger  ICD-9 727.04 Other disorders of synovium, tendon and bursa, synovitis and tenosynovitis, radial styloid tenosynovitis  CPT-4 25000 Incision, extensor tendon sheath, wrist (de Quervain disease)  CPT-4 26055 Tendon sheath incision (for trigger finger)  CPT-4 25001 Incision flexor sheath wrist (FCU) Foot and ankle tendon dysfunction  ICD-9 727.67 Achilles tendon nontraumatic rupture  ICD-9 727.68 Other tendons of foot and ankle nontraumatic rupture  ICD-9 726.70 Enthesopathy of ankle and tarsus, unspecified  ICD-9 726.71 Achilles tendonitis  ICD-9 726.79 Other enthesopathy of the ankle and tarsus  ICD-9 727.06 Other disorders of synovium, tendons, bursa, synovitis and tenosynovitis, tenosynovitis of foot and ankle  ICD-9 726.91 Peripheral enthesopathies and allied syndromes, unspecified enthesopathy, exostosis of unspecified site, bone spur  ICD-9 726.72 Peripheral enthesopathies and allied syndromes, enthesopathy of ankle and tarsus, tibialis tendonitis  ICD-9 727.67 Other disorders of synovium, tendon, and bursa, rupture of tendon, nontraumatic Achilles tendon  CPT-4 27652 Repair Achilles tendon with graft  CPT-4 27654 Revision repair Achilles tendon  CPT-4 27650 Repair ruptured Achilles tendon  CPT-428086 Synovectomy tendon sheath of the foot flexor  CPT-428088 Synovectomy tendon sheath of the foot flexor  CPT-428120 Partial excision of calcaneus (Haglunds) Knee tendon dysfunction  ICD-9 726.64 Patellar tendonitis  ICD-9 727.65 Quadriceps tendon rupture  ICD-9 727.66 Patellar tendon rupture  CPT-4 27381 Patellar tendon repair with graft  CPT-4 27385 Quadriceps tendon repair  CPT-4 27386 Quadriceps tendon repair with graft  CPT-4 27380 Suture of infrapatellar tendon Hip tendon dysfunction  ICD-9 726.5 Enthesopathy of hip region  CPT-4 27060 Excision ischial bursa  CPT-4 27062 Excision trochanteric bursa Compression neuropathy  ICD-9 354.0 Carpal tunnel syndrome  ICD-9 354.1 Pronator teres syndrome (other lesion of median nerve)  ICD-9 354.2 Lesion of ulnar nerve  ICD-9 354.3 Lesion of radial nerve  CPT-4 64721 Neuroplasty and/or transposition; median nerve at carpal tunnel  CPT-4 29848 Endoscopic carpal tunnel release  CPT-4 64719 Neuroplasty and/or transposition; ulnar nerve at wrist  CPT-4 64718 Neuroplasty and/or transposition; ulnar nerve at elbow *^a^*CPT-4, Current Procedural Terminology, Fourth Revision; FCU, flexor carpi ulnaris; ICD-9, International Classification of Diseases, Ninth Revision. We estimated RRs for all of the tendinopathy or neuropathy diagnoses shown in [Table 2](#table2-2325967116642173){ref-type="table"} for the rotator cuff tear cases; their first-, second-, and third-degree relatives; and all of their spouses. The estimated RRs are shown in [Table 3](#table3-2325967116642173){ref-type="table"}, which includes the number of cases (or relatives or spouses, respectively), the observed number of relatives with the diagnosis, the expected number of relatives, the significance of the hypothesis test for RR = 1.0 (*P*), the estimated relative risk (RR), and the 2-tailed 95% confidence interval for the RR. ###### Estimated Relative Risks for Associated Disorders in Rotator Cuff Tear Patients; Their First-, Second-, and Third-Degree Relatives; and Their Spouses*^a^*  Associated Disorder Obs/Exp *P* RR (95% CI) -------------------------------------------------------------------- ----------- -------------- --------------------- Rotator cuff disease cases (n = 1889)  Elbow tendinopathy 262/17.1 \<1.0e^--20^ 15.31 (13.51-17.28)  Hand/wrist tendinopathy 195/20.1 \<1.0e^--20^ 9.69 (8.38-11.15)  Foot/ankle tendinopathy 282/29.9 \<1.0e^--20^ 9.45 (8.38-10.62)  Knee tendinopathy 19/1.9 2.8e^--13^ 9.97 (6.01-15.58)  Hip tendinopathy 150/14.8 \<1.0e^--20^ 10.16 (8.60-11.92)  Compression neuropathy 280/33.5 \<1.0e^--20^ 8.36 (7.41-9.40)  All tendinopathies 716/77.3 \<1.0e^--20^ 9.26 (8.59-9.96) First-degree relatives of rotator cuff disease cases (n = 6216)  Elbow tendinopathy 101/50.0 1.6e^--10^ 2.02 (1.64-2.45)  Hand/wrist tendinopathy 131/59.8 1.2e^--15^ 2.19 (1.83-2.60)  Foot/ankle tendinopathy 152/87.3 2.2e^--10^ 1.74 (1.48-2.04)  Knee tendinopathy 17/6.7 5.5e^--4^ 2.56 (1.49-4.09)  Hip tendinopathy 96/42.3 1.2e^--12^ 2.26 (1.83-2.77)  Compression neuropathy 178/97.5 1.8e^--13^ 1.83 (1.57-2.11)  All tendinopathies 438/230.2 \<1.0e^--20^ 1.90 (1.73-2.09) Second-degree relatives of rotator cuff disease cases (n = 11,002)  Elbow tendinopathy 84/63.9 .01 1.31 (1.05-1.63)  Hand/wrist tendinopathy 95/84.4 .25 1.13 (0.91-1.38)  Foot/ankle tendinopathy 140/117.2 .04 1.19 (1.03-1.41)  Knee tendinopathy 17/11.7 .14 1.45 (0.84-2.32)  Hip tendinopathy 58/56.7 .89 1.02 (0.78-1.32)  Compression neuropathy 144/129.8 .21 1.11 (0.94-1.31)  All tendinopathies 373/320.0 3.1e^--3^ 1.17 (1.05-1.29) Third-degree relatives of rotator cuff disease cases (n = 23,345)  Elbow tendinopathy 134/144.8 .38 0.93 (0.78-1.10)  Hand/wrist tendinopathy 220/195.5 .08 1.13 (0.98-1.28)  Foot/ankle tendinopathy 289/270.4 .26 1.07 (0.95-1.20)  Knee tendinopathy 16/21.9 .24 0.73 (0.42-1.19)  Hip tendinopathy 158/137.9 .09 1.15 (0.97-1.34)  Compression neuropathy 340/302.5 .03 1.12 (1.01-1.25)  All tendinopathies 774/726.8 .08 1.06 (0.99-1.14) All spouses of rotator cuff disease cases (n = 686)  Elbow tendinopathy 33/6.0 1.6e^--14^ 5.51 (3.80-7.74)  Hand/wrist tendinopathy 37/8.1 1.0e^--13^ 4.59 (3.23-6.32)  Foot/ankle tendinopathy 46/11.5 1.4e^--14^ 4.01 (2.94-5.35)  Knee tendinopathy 3/0.6 .02 5.07 (1.05-14.82)  Hip tendinopathy 33/6.3 5.8e^--14^ 5.25 (3.61-7.37)  Compression neuropathy 60/12.8 \<1.0e^--20^ 4.67 (3.56-6.01)  All tendinopathies 117/29.6 \<1.0e^--20^ 3.96 (3.27-4.75) *^a^*Obs/Exp, observed/expected number with the diagnosis; RR, relative risk. Significantly elevated risk for elbow, hand/wrist, foot/ankle, knee, and hip tendinopathies; all tendinopathies; and for compression neuropathies was observed in rotator cuff tear cases themselves (*P* \< 2.8e^--13^), in their spouses (*P* \< .02), and in their first-degree relatives (*P* \< 5.5e^--4^) ([Table 3](#table3-2325967116642173){ref-type="table"}). These results suggest shared environmental and behavioral risk factors. However, significantly elevated risks for elbow dysfunction (RR = 1.31), foot and ankle dysfunction (RR = 1.19), all tendinopathies in second-degree relatives (RR = 1.17), and significantly elevated risk for compression neuropathy in third-degree relatives (RR = 1.12) suggest the presence of a genetic contribution to these disorders. Discussion {#section6-2325967116642173} ========== Prior data support a significant genetic predisposition for individuals with rotator cuff tearing to have family members with rotator cuff tearing.^[@bibr20-2325967116642173]^ Rotator cuff tearing has been associated with other tendinopathies including lateral epicondylitis, Achilles tendonitis, and compression neuropathies.^[@bibr22-2325967116642173],[@bibr23-2325967116642173]^ It is unclear whether this association is due to environmental or genetic effects.^[@bibr22-2325967116642173],[@bibr23-2325967116642173]^ The current data strongly support an environmental component as well as a genetic component to the development of global tendinopathies and compression neuropathies in patients with rotator cuff tearing. Specifically, spouses of patients with rotator cuff tearing were at a greater risk for developing all tendinopathies and compression neuropathies, suggesting environmental influences. First- and second-degree (for the all tendinopathy group) and first-, second-, and third-degree relatives (for compression neuropathies) of patients with rotator cuff tearing were at significantly increased risk for the development of these diseases, suggesting a genetic contribution to their etiology. Overall there are limited data supporting an association between various tendinopathies. Walker-Bone et al^[@bibr24-2325967116642173]^ attempted to determine the prevalence and interrelation of musculoskeletal disorders of the upper extremity. The authors surveyed 9696 adults of working age in the United Kingdom for the presence of musculoskeletal disorders and found significant overlap of patients with shoulder disorders, lateral and medial epicondylitis, de Quervain disease, and carpal tunnel syndrome.^[@bibr24-2325967116642173]^ Roquelaure et al^[@bibr16-2325967116642173]^ identified similar associations between rotator cuff syndrome, lateral epicondylitis, and carpal tunnel syndrome in 2685 French workers. Finally, Titchener et al^[@bibr23-2325967116642173]^ performed a large case-control study using a large national health care database in the United Kingdom looking for associations of tendinopathies and carpal tunnel syndrome with rotator cuff disease. The authors reported an increased risk for the development of Achilles tendonitis (odds ratio \[OR\] = 1.78), trigger finger (OR = 1.99), lateral epicondylitis (OR = 1.71), and carpal tunnel syndrome (OR = 1.55) in patients with rotator cuff disease compared with controls.^[@bibr8-2325967116642173]^ All these studies support an association between tendon dysfunction and compression neuropathies and rotator cuff tearing, although it is unclear whether these are environmental or genetic. Our data support that likely both environmental and genetic effects contribute to the associations. The methodology utilized in the current study is similar to that previously performed to identify a genetic predisposition for other chronic diseases and cancers. The UPDB combines a computerized genealogy of the Utah pioneers and their descendants with various data, including the Utah Cancer Registry, Utah death certificates, and the University of Utah Hospital and Clinics.^[@bibr4-2325967116642173],[@bibr19-2325967116642173]^ This resource has been used to define the genetic contribution to many other phenotypes and was the basis of the identification in Utah pedigrees of *BRCA1*, *BRCA2*, and *p16*.^[@bibr5-2325967116642173],[@bibr10-2325967116642173],[@bibr13-2325967116642173]^ Using this resource and similar methodology, we explored the hypothesis of a genetic contribution to global tendinopathies and compression neuropathies in patients with rotator cuff tearing and have shown a strong genetic predisposition for tear development. Almost all tendinopathies individually showed an environmental influence, as suggested by significantly elevated relative risks for all diagnoses in spouses. Only all tendinopathies, foot and ankle, and elbow showed an increased relative risk in second-degree relatives, and compression neuropathies showed increased relative risks in third-degree relatives, suggesting a genetic predisposition. A possible reason for the lack of each individual tendinopathy group to show a genetic predisposition includes smaller numbers of patients in each tendon dysfunction group, thereby limiting the analysis. The data still strongly support both environmental and genetic influences on the development of tendon dysfunction and compression neuropathies. A limitation of this study is the use of diagnostic codes for phenotype assignment. Because diagnosis codes are dependent on the clinician, a diagnosis of rotator cuff tearing, other tendinopathy, or compression neuropathy may be less accurate from a nonorthopaedist compared with that from an orthopaedic specialist. We attempted to limit this bias by limiting identification of patients to diagnoses of specific tendinopathies and compression neuropathies, excluding all codes associated with global joint-related pain, strains, or dysfunction. The use of CPT-4 codes for rotator cuff repair surgery as well as neuroplasty or transposition improves the accuracy of the diagnoses but does not completely eliminate inaccuracy due to incorrect surgical indications. We also recognize there may be inaccurate coding using the specific codes for each tendinopathy or neuropathy, especially by a nonorthopaedist, due to misdiagnosis. We elected to include both the ICD-9 and CPT-4 codes to increase the overall number of patients analyzed, improving the ability to perform a population-based analysis. We recognize that the majority of patients were included based on a clinical diagnosis as opposed to undergoing surgery. Other limitations include censoring that may have occurred in the data resource. In this analysis, we were able to identify only patients at the University of Utah Hospital and Clinics who were coded with a diagnosis or procedure indicating the presence of rotator cuff disease, other tendon dysfunctions, or compression neuropathy and who also had at least 3 generations of genealogical data. The UUHSC serves 20% of Utah state. Additionally, individuals who were not seen at the UUHSC, who were seen before 1994, or whose data did not appropriately link were also censored. This censoring of data occurs uniformly across the data resource, to relatives of both patients and controls, and has been shown to have no bias on the overall results. Finally, there is a high concentration of patients of the Church of Jesus Christ of the Latter-Day Saints (LDS or Mormon) faith in the region of the country from which the study population is derived. Consequently, there may be cultural or behavioral factors that may influence the prevalence of rotator cuff disease in these patients. Regardless, the Utah population has been shown to be representative of the Northern European regions from which its founders came and to have low to normal inbreeding levels when compared with those in the United States overall.^[@bibr8-2325967116642173],[@bibr12-2325967116642173]^ Despite these potential limitations, our results still suggest that there is a genetic component to the development of global tendon dysfunction and compression neuropathy in patients with rotator cuff tearing, taking into consideration these potential environmental effects. Tendinopathies and compression neuropathies likely have multifactorial etiologies including mechanical and environmental influences and have been associated with rotator cuff tearing in many large population-based studies. We have shown a strong environmental influence to the development of these diseases in patients with rotator cuff tears. Identification of causal environmental factors resulting in the development of global tendinopathies and compression neuropathies in these patients will aid in the prevention of these disorders. We have additionally shown a strong genetic predisposition specifically for the development of global tendinopathies and compression neuropathies in patients with rotator cuff tears and their relatives. Future applications of these data include high-risk pedigree studies to identify the predisposition gene(s) responsible for these observations and the analysis of potential candidate genes associated with global tendon dysfunction and compression neuropathies. A better understanding of abnormal genetics for tendinopathies and compression neuropathies may allow for early identification of at-risk individuals as well as the development of potential alternative biologically directed treatment methods. Conclusion {#section7-2325967116642173} ========== The current study shows strong evidence of familial clustering of rotator cuff tearing with other tendinopathies and with compression neuropathy. Observed increased risks in spouses and first-degree relatives supports shared environmental risk factors for rotator cuff tearing, most tendinopathies, and compression neuropathies. Increased risks to third-degree relatives for compression neuropathy suggest an association of these phenotypes that may have a shared genetic etiology. One or more of the authors has declared the following potential conflict of interest or source of funding: Partial support (greater than \$10,000) for all datasets within the Utah Population Database was provided by the University of Utah Huntsman Cancer Institute.

The authors confirm that all data underlying the findings are fully available without restriction. The authors confirm that all data underlying the findings are fully available without restriction. All relevant data are available through figshare (<http://dx.doi.org/10.6084/m9.figshare.1181943>). Introduction {#s1} ============ The ability of older adults to understand both auditory-only and audiovisual speech declines with age [@pone.0111121-Sumby1]--[@pone.0111121-Gosselin1]. This decline extends to other important cognitive functions, such as memory, visuospatial abilities, and speed of information processing [@pone.0111121-Peich1]--[@pone.0111121-Hedden1]. Interestingly, performance declines with age are not uniform across multiple trials of the same task. Older adults exhibit much greater variability in performance: on some trials, older adults perform as well as younger adults, but on other trials, older adults perform much worse [@pone.0111121-Bielak1]--[@pone.0111121-Lovden1]. This type of performance decline, referred to as increased intrasubject variability, may be a particularly sensitive measure of age-related changes [@pone.0111121-Lovden2]. We hypothesized that the increased intrasubject variability with age observed in behavioral paradigms should have a neural counterpart. For example, *in vivo* electrophysiology investigations of the visual cortex of experimental animals show increased variability and decreased stimulus specificity with age during viewing of simple visual stimuli [@pone.0111121-Schmolesky1]--[@pone.0111121-Liang1]. Increased neural variability in the auditory brain stem response of healthy older adults has been related to deficits in speech in noise perception [@pone.0111121-Anderson1]. Variability on a cognitive motor task has been tied to individual differences in activation and functional connectivity of the dorsal pre-motor cortex [@pone.0111121-Stewart1]. Wide variability has also been observed in the network activation and functional connectivity of posterior default mode and frontal executive networks in a large cohort (1000BRAINS) of healthy older subjects, which may be related to differences in neuropsychological tests of cognitive and motor function [@pone.0111121-Caspers1]. Conversely, other investigations of variability in healthy aging have pointed to the possibility of *decreased* variability with age [@pone.0111121-Grady1]--[@pone.0111121-Garrett2]. This necessitates continued research to investigate to fully understand the role of response variability in healthy aging. We used rapid event-related blood-oxygen level dependent magnetic resonance imaging (BOLD fMRI) to measure brain responses to passively-presented audiovisual speech syllables. This allowed us to focus on perceptual processing differences between young and old rather than differences in higher cognitive function and performance. We predicted that repeated presentations of identical speech stimuli would evoke fMRI responses that were more variable in older adults than in younger adults. Neural responses to audiovisual speech were examined using two complementary methods. First, we used a region-of-interest (ROI) analysis focused on the three core areas of the multisensory speech perception network: auditory cortex, visual cortex, and the superior temporal sulcus (STS). Second, we used a voxel-wise analysis to search for differences between older and younger subjects outside of our predefined ROIs. Materials and Methods {#s2} ===================== 2.1. Ethics Statement {#s2a} --------------------- All subjects provided written informed consent and were compensated for their time in accordance with an experimental protocol approved by the Committee for the Protection of Human Subjects of the University of Texas Health Science Center at Houston. 2.2. Subjects and exclusion criteria {#s2b} ------------------------------------ The young adult cohort consisted of 14 subjects (20--39 years, 6 female, mean age 26.1 years). 24 older subjects were recruited for the study. Five older adult subjects were excluded (two for poor speech identification scores, three for MRI data quality concerns, see details below), leaving 19 total subjects whose data are reported here (53--70 years, 12 female, mean age 63.0 years). Vision in the older subjects was assessed using a Snellen eye chart. Each eye was tested separately. The range of acuities was 20/20 to 20/70. Hearing in the older subjects were evaluated using a modified Bekesy threshold test at 500 and 2000 Hz [@pone.0111121-Price1]. Our subjects had a range of 6.7 dB--30.9 dB for 500 Hz and 7.3--39.4 dB for 2000 Hz, within the normal range (two standard deviations from the mean) for hearing thresholds based on their age [@pone.0111121-Brant1]. However, speech abilities decline at a different rate than pure audiometric measures later in life [@pone.0111121-Divenyi1]. Therefore, we also tested identification of auditory-only and audiovisual syllables. Most subjects scored near ceiling on auditory-only syllable identification (83%−100%, average performance 93%) and audiovisual syllables. Two subjects scored poorly on auditory-only syllable identification (\<70%) and were excluded from the analysis. Cognitive function was assessed using the standardized Mini Mental State Examination (MMSE) [@pone.0111121-Folstein1]. All subjects' scores indicated no decline in cognitive function (scores ranged from 26--30, mean MMSE 28.4, scores 25 out of 30 points or greater indicate normal cognitive function). Two older subjects were excluded for large head movements during fMRI data acquisition (standard deviation of motion regressor \>3 mm). One older subject was excluded because an initial analysis of the fMRI data showed response amplitudes more than 3 standard deviations greater than the mean. 2.3. Overview of fMRI experiment and analysis {#s2c} --------------------------------------------- We used two independent methods for fMRI analysis: region-of-interest (ROI) and voxel-wise whole-brain analysis, which give complementary information about brain activity [@pone.0111121-Saxe1], [@pone.0111121-Friston1]. ROI analysis allows us to examine areas for which we have an *a priori* hypothesis and does not require that data be transformed to a brain template, thus allowing for differences in individual anatomy. Furthermore, it limits Type I errors by limiting the number of statistical tests to a handful of ROIs [@pone.0111121-Poldrack1]. However, ROI analyses are blind to effects outside of the predefined ROIs and to functional specialization within ROIs. Therefore, we also performed a voxel-wise analysis to examine activity across the entire brain. 2.4. Block-design localizer {#s2d} --------------------------- A block-design localizer was used to generate the regions-of-interest (ROIs). Each block contained 10 two-second trials, one word per trial, followed by 10 seconds of fixation baseline. Each trial contained a single word from a bank of digital video recordings of 105 single-syllable words (*e.g.* "view", "door", "make") spoken by a female native English speaker. Words were selected from the MRC Psycholinguistic Database [@pone.0111121-Wilson1]. Auditory-only words consisted of the auditory component of each video with a white visual fixation crosshairs and visual-only words consisted of only the visual component of the video recording. In older adults, the localizer scan series contained six blocks (two auditory-only, two visual-only and two audiovisual blocks in random order). Each block contained a target trial (the word "press") of the same type (auditory-only, visual-only, or audiovisual) as the other stimuli in the block; subjects were instructed to pay attention to each stimulus and press a response button only during target trials. In younger adults, ten blocks were presented (five auditory-only and five visual-only in random order) with no target trials. 2.5. fMRI responses to audiovisual speech syllables {#s2e} --------------------------------------------------- For the main experiment, stimuli were presented in two-second trials in a rapid event-related design. Each trial contained a single audiovisual syllable, consisting of McGurk (auditory "ba"+visual "ga", auditory "pa"+visual "ka"), non-McGurk incongruent (auditory "ga"+visual "ba", auditory "ka"+visual "pa"), congruent ("ba", "ga", "da", "pa", "ka" and "ta"), target (audiovisual "press" in older adults, audiovisual "ma" in younger subjects) and fixation trials (fixation crosshairs only). There was only a single exemplar of each audiovisual syllable, meaning that subjects were exposed to identical stimuli repeatedly. This allowed us to isolate the effects of neural variability. Subjects were instructed to respond with a button press only to target trials and to make no response to all other trials. Behavioral data in the scanner was not collected for two younger subjects. Subjects performed near ceiling on this task (18/19 older adults at 100% accuracy; 10/12 younger adults at 100% accuracy) suggesting a high degree of alertness (no significant difference between groups, t~29~ = 0.6, p = 0.57). The total length of each video was cropped with digital video editing software (iMovie, Apple Computer) such that each clip started and ended in a neutral, mouth-closed position. Each video stimulus varied in length from 1.7 to 1.8 seconds followed by fixation crosshairs for the remainder of the trial (the crosshairs were always presented in the same screen location as the mouth of the talker visible during other trials in order to minimize eye movements). Prior to the scan, a volume check was conducted for each subject outside the scanner without the presence of scanner noise. Sample videos from the experiment were played and the volume was adjusted so that the volume was "as loud as possible without being uncomfortable or hurting in any way". After each scan series subjects were asked if they could hear the stimuli presented and if any volume adjustments were necessary. Subjects viewed audiovisual stimuli presented in a rapid event-related design with slight variations in the number of trials, as follows: older subjects, *n* = 6∶150 audiovisual syllables, 40 target trials; older subjects, *n* = 13∶160 audiovisual syllables, 50 target trials; younger subjects, *n* = 5∶220 audiovisual syllables, 70 target trials; younger subjects, *n* = 9∶200 audiovisual syllables, 80 target trials. 2.6. MRI and fMRI analysis {#s2f} -------------------------- Two T1-weighted MP-RAGE anatomical MRI scans were collected at the beginning of each scanning session with a 3 Tesla whole-body MR scanner (Phillips Medical Systems). The two anatomical scans were aligned to each other and averaged in order to provide maximal gray-white matter contrast. These scans were then used to create a cortical surface model using FreeSurfer [@pone.0111121-Dale1], [@pone.0111121-Fischl1] for visualization in SUMA [@pone.0111121-Argall1]. For the fMRI scan series, T2\* weighed images were collected using gradient echo-planar imaging (TR = 2000 ms, TE = 30 ms, flip angle = 90°) with in-plane resolution of 2.75×2.75 mm. Auditory stimuli were presented through MRI-compatible in-ear headphones (Sensimetrics, Malden, MA) which were covered with ear muffs to reduce the amount of noise from the scanner. Visual stimuli subtending approximately 20×30 degrees of visual angle were presented on a projection screen with an LCD projector and viewed through a mirror attached to the head coil. Responses to the target trials were collected using a fiber-optic button response pad (Current Designs, Haverford, PA). Analysis of the functional scan series was conducted using Analysis of Functional NeuroImages (AFNI) [@pone.0111121-Cox1]. ### 2.6.1. fMRI analysis: response amplitude and variability {#s2f1} In order to generate whole brain maps of the amplitude and standard deviation measures at each voxel, we carried out a voxel-wise analysis using the AFNI function *3dDeconvolve,* which uses maximum-likelihood estimation in the context of the generalized linear model (GLM). TENTzero functions were used to estimate the individual hemodynamic response function (using the option *-- iresp*) and standard deviation of each response function (using the option *-- sresp*) in each voxel for each stimulus type, beginning at stimulus onset and ending 16 seconds later for single syllables and 26 seconds later for blocks of words (the response was constrained to begin and end at zero amplitude). The model functions consisted of independent, piece-wise linear impulse response functions (also known as stick functions) that independently estimated the amplitude of the hemodynamic response at each time point following stimulus presentation. This methodological point is important because it allowed us to estimate the amplitude and standard deviation from the actual response in each individual voxel, not from a fixed hemodynamic response function such as a gamma variate. The use of a fixed function could introduce a confound because of differences in hemodynamic response functions; for instance, if older people had slightly broader hemodynamic response functions, then their deviation from a fixed function would be greater, unrelated to trial-to-trial variability. For single syllables, we estimated the amplitude of the response as the mean of the response at 4 seconds and 6 seconds after stimulus onset (the peak of the hemodynamic response function). To estimate BOLD variability within each subject for single syllables, the standard deviation at the 4-second and 6-second time points of each impulse response function were averaged to produce a single value per voxel. The brain response to all audiovisual syllables (both response amplitude and variability) was similar, so they were combined for further analysis and only the average across stimulus types (excluding target trials) is reported. ### 2.6.2. Region-of-interest selection {#s2f2} Data from the whole-brain voxel-wise analysis (*2.6.1*) was first grouped using regions of interest created for each subject individually in native image space. ROIs were selected to target brain areas that are reliably active during multisensory speech perception [@pone.0111121-Nath1]. A combination of anatomical and functional criteria was used. The anatomic parcellation of the cortical surface was constructed from each individual subject\'s structural scans with FreeSurfer [@pone.0111121-Destrieux1], [@pone.0111121-Fischl2]. Functional criteria were constructed from the independent localizer runs (see section 2.4 for details), eliminating bias [@pone.0111121-Kriegeskorte1]. We considered three contrasts when constructing the three regions of interest: auditory words vs. fixation baseline, visual words vs. fixation baseline, and audiovisual words vs. fixation baseline. The STS ROI was defined by finding all voxels in the posterior half of the anatomically parcellated STS that showed a significant response (*t* \>2, *p*\<0.05) during the localizer (*t* \>2 for auditory-only word blocks *vs.* baseline and t \>2 for visual-only word blocks *vs.* baseline). For 5 out of 19 older adults, no voxels in the left STS met this criterion, so an alternative criterion was used (*t* \>2 for audiovisual word blocks *vs.* baseline). The auditory cortex ROI was defined by finding voxels in the anatomically parcellated transverse temporal gyrus, lateral superior temporal gyrus and planum temporale that were significantly active during the auditory-only blocks (*t* \>2 for auditory-only word blocks *vs.* baseline). The extrastriate visual cortex ROI was defined by finding voxels in the anatomically parcellated extrastriate lateral occipitotemporal cortex that were active during the visual-only blocks (*t* \>2 for visual-only word blocks *vs.* baseline). ### 2.6.3. Whole-brain analysis {#s2f3} For the whole-brain voxel-wise analysis, subjects' individual data were first aligned to the N27 atlas brain [@pone.0111121-Mazziotta1] using the AFNI function *auto_tlrc*. Blurring kernels of approximately 3--6 mm have been found to be the most sensitive for detecting activation clusters [@pone.0111121-Skudlarski1]. We chose a 3×3×3 mm FWHM Gaussian kernel to minimize blurring between adjacent ROIs. To conduct a voxel-wise search for any differences in response amplitude, the average response amplitude (average of the response to all non-target audiovisual speech stimuli relative to fixation baseline at the 4 and 6 second time points) was calculated in each voxel in each subject. *3dttest++* was used to perform an unpaired t-test for every voxel in standard space between the old and young adult groups. The results were mapped from the MRI volume to the cortical surface with *3dSurf2Vol* and masked with the group t-statistic (*t* \>2 for the contrast of all audiovisual syllables *vs.* baseline). After the voxel-wise t-test we preformed a clustering technique [@pone.0111121-Xiong1]. This finds only voxels that are significantly active above a particular threshold and spatially contiguous. The probability of finding two voxels above a particular threshold *and* being adjacent is much smaller than the chance of a single voxel above that threshold [@pone.0111121-Forman1]. Using the AFNI program *slow_surf_clustsim.py*, we estimated that a cluster with a size of 160 mm^2^ would have a corrected p-value of 0.045. A clusterizing filter on the surface (*SurfClust)* was applied and only regions larger than 160 mm^2^ (and *t* \>2 for the contrast of all audiovisual syllables *vs.* baseline) are reported. To conduct a voxel-wise search for differences in intersubject variability, the MATLAB function *vartestn* was used to perform a Bartlett's multiple sample test for equal variances on the response amplitudes, followed by clusterizing. To conduct a voxel-wise search for differences in intrasubject variability, an unpaired t-test between groups was performed on the standard deviation of the response at each voxel (*3dttest++*) followed by clusterizing. ### 2.6.4. Motion correction {#s2f4} Functional data for each subject was first aligned to the averaged anatomical dataset for that subject and then motion-corrected using local Pearson correlation with the AFNI script *align_epi_anat.py* [@pone.0111121-Saad1]. For each volume, an estimate of the amount of motion in each direction, relative to the reference, was produced. These estimates were used as regressors of no interest in the fMRI analysis. To capture a single value describing the amount of head motion in each subject, the standard deviation of each motion direction across time was averaged across motion directions. In addition to the standard motion correction steps described above, we also performed the "motion scrubbing" procedure developed by Power and colleagues [@pone.0111121-Power1] to further investigate if motion was a potential confound in our main finding of increased intrasubject variability in older adults. First, motion estimates at each time point were calculated in each of the six motion directions (rotational measures: roll, pitch, yaw, and displacement measures: superior, left, and posterior directions). Rotational displacement measures were converted to millimeters using the formula *d = R\*(* ***π*** */180)\*r*, where R is the rotation in degrees and r is the radius (we used r = 50 mm as prescribed by Power). To express the total amount of motion for each time point in a single value, the absolute value of the displacement in each direction was summed, where the total displacement at the *i*th data point was *D~i~ = \|d~α~\|+\|d~β~\|+\|d~γ~\|+\|d~x~\|+\|d~y~\|+\|d~z~\|*. Then the framewise displacement for the *i*th time point was calculated as *FD~i~ = D~(i-1)~ - D~i~* to express instantaneous head motion. The scrubbing threshold was half of the smallest voxel dimension, as recommended by Power (EPI volumes were collected using a 2.75 mm isotropic voxel, therefore we used a threshold = 1.375 mm). The GLM analysis was then completed a second time, excluding the data points with a framewise displacement exceeding this threshold. Results {#s3} ======= 3.1 Responses in the ROIs to audiovisual speech {#s3a} ----------------------------------------------- Our initial analysis focused on three ROIs implicated as critical nodes in the network for multisensory speech perception: the left superior temporal sulcus (STS), the left auditory cortex, and the left extrastriate visual cortex. ### 3.1.1. Variability of the hemodynamic response in ROIs within subjects {#s3a1} The left STS showed a robust hemodynamic response to audiovisual syllables that was similar in amplitude in individual older and younger subjects. An important behavioral difference in many behavioral paradigms between older and younger subjects is the variability across trials *within* individual subjects (intrasubject variability). To search for a neural correlate of this phenomenon, we calculated the hemodynamic response function in each subject in every voxel and then measured the standard deviation across trials at each time point in the hemodynamic response within each subject (plotting them as error bars around the mean at each time point). As shown in [Figure 1A and 1B](#pone-0111121-g001){ref-type="fig"}, while the response amplitudes were similar for individual older and younger subjects, the variability at each time point of the response was much larger in the older subject. {#pone-0111121-g001} To quantify this difference, we averaged the standard deviation from the peak of the response (4 and 6 seconds after stimulus onset) to produce a single number for intrasubject variability for each ROI for each subject ([Figure 1C--E](#pone-0111121-g001){ref-type="fig"}). For each ROI, there was much greater within-subject standard deviation in older subjects (STS: 0.14% in older adults *vs.* 0.09% in younger adults, t~31~ = 4.2, p = 2×10^−4^; auditory cortex: 0.18% *vs.* 0.11%, t~31~ = 5.9, p = 10^−6^; visual cortex: 0.18% *vs.* 0.08%, t~31~ = 7.0, p = 10^−8^). ### 3.1.2. Mean and standard deviation of the hemodynamic response in ROIs across subjects {#s3a2} An unpaired t-test with percent signal change in the left STS as the dependent measure revealed slightly greater amplitude of mean response in younger adults (0.19% in younger adults *vs.* 0.12% in older adults, t~31~ = 2.1, p = 0.048). There were no significant differences in auditory cortex (0.26% *vs.* 0.24%, t~31~ = 0.5, p = 0.63) or visual cortex (0.16% *vs.* 0.10%, t~31~ = 1.5, p = 0.15). The standard deviation of the response across subjects was also similar between groups ([Figure 2](#pone-0111121-g002){ref-type="fig"}; left STS: SD of 0.08% for younger adults *vs.* 0.12% for older adults, Bartlett's multiple sample test for equal variances χ^2^ ~1~ = 2.5, p = 0.12; left auditory cortex: 0.14% *vs.* 0.12%, χ^2^ ~1~ = 0.7, p = 0.41; left visual cortex: 0.08% *vs.* 0.12%, χ^2^ ~1~ = 2.3, p = 0.13). {#pone-0111121-g002} 3.2. Whole-brain analysis {#s3b} ------------------------- Our first set of analyses was limited to our three *a priori* ROIs created using block-design localizers. To overcome this limitation, and to prevent any biases introduced by slight differences in the localizers between old and young subjects, our second set of analyses examined the entire brain. First, we selected all voxels that showed a significant positive response (*t* \>2 for all audiovisual syllables *vs.* baseline) to audiovisual syllables across old and young subjects ([Table 1](#pone-0111121-t001){ref-type="table"} and [Figure 3A](#pone-0111121-g003){ref-type="fig"}). {#pone-0111121-g003} 10.1371/journal.pone.0111121.t001 ###### Regions of activation in response to audiovisual speech in both older and younger adults. {#pone-0111121-t001-1} Label Area (mm^2^) Peak t-value --------------------------------------------------------------------------------------- -------------- -------------- R fusiform gyrus, inferior occipital gyrus,middle occipital gyrus *(26,* −*71,* −*6)* 4593 6.2 R fusiform gyrus, inferior occipital gyrus,middle occipital gyrus *(55,* −*21, 2)* 3054 4.3 R superior temporal sulcus and gyrus *(55,* −*21, 2)* 2361 7.4 L superior temporal gyrus *(*−*46,* −*33, 16)* 2315 5.8 L middle occipital gyrus *(*−*23,* −*93, 5)* 560 4.9 R supramarginal gyrus and subcentral gyrus *(49,* −*6, 44)* 219 3.9 Total 13102 Regions are ranked by area (only clusters greater than 160 mm^2^ are reported) and Talairach coordinates following anatomical label in *(x, y, z)* format are the weighted center of mass of the cluster. ### 3.2.1. Variability of the whole-brain hemodynamic response within subjects {#s3b1} We calculated the variability of the response *within* each subject at each voxel, and then compared the two groups. Many brain regions showed greater intrasubject variability in older adults ([Table 2](#pone-0111121-t002){ref-type="table"} and [Figure 3B](#pone-0111121-g003){ref-type="fig"}), including left auditory cortex and bilateral extrastriate visual cortex (total area on the cortical surface = 7026 mm^2^; peak t-statistic = 5.8, p = 2×10^−6^). There were no regions where intrasubject variability was greater in younger adults. 10.1371/journal.pone.0111121.t002 ###### Regions with greater intrasubject variability in older adults compared to younger adults. {#pone-0111121-t002-2} Label Area (mm^2^) Peak t-value ------------------------------------------------------------- -------------- -------------- R superior temporal gyrus *(54,* −*19, 1)* 2341 5.0 L middle and inferior occipital gyri *(*−*38,* −*69,* −*3)* 1943 5.8 L planum temporale *(*−*47,* −*29, 13)* 1697 4.9 R inferior occipital gyrus *(37,* −*61,* −*13)* 882 4.2 R superior temporal sulcus *(45,* −*56, 5)* 163 5.3 Total 7026 Regions are ranked by area (only clusters greater than 160 mm^2^ are reported) and Talairach coordinates following anatomical label in *(x, y, z)* format are the weighted center of mass of the cluster. ### 3.2.2. Mean and standard deviation of the whole-brain hemodynamic response across subjects {#s3b2} We compared the amplitude and standard deviation of the response across subjects. Younger adults had greater amplitude of response in right and left visual cortex and right superior temporal sulcus ([Table 3](#pone-0111121-t003){ref-type="table"} and [Figure 3C](#pone-0111121-g003){ref-type="fig"}; total area = 1000 mm^2^; peak t-statistic = 4.3, p = 2×10^−5^; there were no regions where response amplitude was greater in older subjects). Older adults had greater standard deviation of response in bilateral visual cortex and right superior temporal cortex ([Table 4](#pone-0111121-t004){ref-type="table"} and [Figure 3D](#pone-0111121-g003){ref-type="fig"}; total area = 4367 mm^2^; peak Bartlett's χ^2^ ~1~ = 29, p = 9×10^−8^; no regions with greater standard deviation in younger adults). 10.1371/journal.pone.0111121.t003 ###### Regions with greater response amplitude in younger adults compared to older adults. {#pone-0111121-t003-3} Label Area (mm^2^) Peak t-value --------------------------------------------- -------------- -------------- R occipital pole *(13,* −*94, 5)* 302 3.3 R superior temporal sulcus *(47,* −*36, 4)* 273 3.8 L occipital pole *(*−*17,* −*93, 5)* 223 4.3 L subcentral sulcus *(*−*53,* −*21, 13)* 202 3.6 Total 1000 Regions are ranked by area (only clusters greater than 160 mm^2^ are reported) and Talairach coordinates following anatomical label in *(x, y, z)* format are the weighted center of mass of the cluster. 10.1371/journal.pone.0111121.t004 ###### Regions with greater intersubject variability in older adults compared to younger adults. {#pone-0111121-t004-4} Label Area (mm^2^) Peak χ^2^ ---------------------------------------------------- -------------- ----------- L inferior occipital gyrus *(*−*30,* −*75,* −*12)* 1819 29 R inferior occipital gyrus *(24,* −*81,* −*8)* 1552 27 R superior temporal gyrus *(62,* −*8, 2)* 438 15 R transverse temporal gyrus *(44,* −*24, 10)* 200 7.8 R fusiform gyrus *(34,* −*49,* −*19)* 194 9.7 L superior frontal gyrus *(*−*1, 0, 56)* 164 27 Total 4367 Regions are ranked by area (only clusters greater than 160 mm^2^ are reported) and Talairach coordinates following anatomical label in *(x, y, z)* format are the weighted center of mass of the cluster. 3.3. Potential confound: differences in head movements between younger and older adults {#s3c} --------------------------------------------------------------------------------------- Our initial analysis used standard techniques for estimation and correction of head motion in the functional data, and our GLM included motion estimates as regressors of no interest. Estimated head motion was small in both groups but greater in older than younger adults (0.48 mm *vs*. 0.32 mm, p = 0.004). The finding of increased BOLD signal variability in older adults remained unchanged after applying a "motion scrubbing" procedure [@pone.0111121-Power1]: STS: 0.13% *vs.* 0.09%, t~31~ = 3.4, p = 0.002; auditory cortex: 0.18% *vs.* 0.10%, t~31~ = 5.3, p = 8×10^−6^; visual cortex: 0.17% *vs.* 0.09%, t~31~ = 5.5, p = 1.6×10^−5^. There was no correlation in either group between amount of head motion and intrasubject variability (older adults: r = 0.04, p = 0.88; younger adults: r = 0.16, p = 0.59). An ANCOVA with age group as one factor, head motion as a covariate, and standard deviation of the fMRI response as the dependent variable and revealed no interaction between age group and head motion in any ROI (p\>0.68), and the finding of increased intrasubject variability in older adults remained significant (STS: F~1,29~ = 13, p = 0.001; auditory cortex: F~1,29~ = 25, p = 3×10^−5^; visual cortex: F~1,29~ = 34, p = 2×10^−6^). A whole-brain ANCOVA that included the amount of head motion in each subject as a covariate gave results nearly identical to the analysis without the head motion covariate. Discarding the six older adults with the greatest amount of head motion rendered group differences in head movements insignificant (0.39 mm *vs*. 0.32 mm, p = 0.10), but left the main finding of increased intrasubject variability intact (left STS: 0.14% *vs.* 0.09%, t~25~ = 3. 7, p = 0.001; left auditory cortex: 0.18% *vs.* 0.11%, t~25~ = 5.4, p = 10^−4^; visual cortex: 0.17% *vs.* 0.08%, t~25~ = 6.7, p = 5×10^−7^). Discussion {#s4} ========== We compared brain responses to repeated presentations of identical audiovisual speech in healthy older and younger adults using fMRI. The most important finding was greater intrasubject variability in the older adults: across multiple presentations of identical stimuli, older adults had greater variability in their brain responses than younger adults. This was true across all of the brain areas that responded to audiovisual speech and was confirmed with two independent types of analysis (ROI and whole-brain). We also observed two less robust effects: older adults had smaller mean response amplitudes than younger adults, and the older adult group had a greater standard deviation of the response amplitude (intersubject variability) than younger adults. What is the relationship between variability and perception/processing? {#s4a} ----------------------------------------------------------------------- Across a variety of behavioral tasks, older adults have worse performance and increased intrasubject variability compared with young adults [@pone.0111121-Lovden1], [@pone.0111121-Murphy1]--[@pone.0111121-West1]. Older adults with mild dementia show more intrasubject variability than healthy age-matched controls [@pone.0111121-MacDonald2] and healthy older adults with more trial-to-trial variability showed greater cognitive declines over time [@pone.0111121-Lovden2]. While a link between increased neural (BOLD fMRI) variability and increased behavioral variability is sensible on its face, the precise link between the two is a matter of speculation. With increased neural variability the distribution of responses in a population of neurons to a given stimulus would become wider, which in turn would make it harder for the brain to differentiate between the possible stimuli that evoked the response [@pone.0111121-Li1]. Consistent with this idea, decreased stimulus specificity has been observed in single cell recordings of older cats and non-human primates [@pone.0111121-Schmolesky1]--[@pone.0111121-Liang1]. Older adults are particularly impaired in perceiving speech if it is embedded in artificially generated noise [@pone.0111121-Dubno1], [@pone.0111121-Gosselin1], [@pone.0111121-Stephan1], [@pone.0111121-Jones1]. The increased noise in the stimulus could exacerbate the effects of increased neural variability, which could be considered "neural noise". We did not test older subjects using noisy audiovisual speech, the type of speech on which they are most impaired. Therefore, we could not directly compare the increased neural variability we observed in older subjects with the decreased performance of recognizing speech in noise that they are known to have. Future studies using noisy stimuli would be expected to produce poorer performance and reveal differences between subjects correlated with BOLD variability. Neural variability at early stages of cortical sensory processing might be compounded by neural variability at decision layers higher in the cortical hierarchy. Speech perception involves categorical judgments about the identity of each syllable. Neuronal variability could impair these decisions, an effect that may be even more important than added sensory noise [@pone.0111121-Beck1]. Neuronal variability may also differentially affect the ability to make both fine and coarse discriminations. Low levels of neuronal variability favor fine discriminations performed at locations in stimulus space in which neuronal selectivity changes rapidly, while high levels of neuronal variability favor coarse discriminations performed at locations in stimulus space where neuronal responses are maximal [@pone.0111121-Butts1]. Therefore, increased neuronal variability with aging might impair fine discrimination while leaving coarse discriminations relatively intact. Within individual trials, variability in neural responses can allow a population of neurons to encode multiple stimulus attributes. For instance, the mean of the population response may encode the stimulus estimate, while the variability of the population response encodes the uncertainty [@pone.0111121-Ma1], [@pone.0111121-Knill1]. Previous fMRI studies showing greater inter/intrasubject variability in older adults {#s4b} ------------------------------------------------------------------------------------ There have been a number of fMRI studies comparing young and older adults in other tasks, and the preponderance of these studies have reported more variability in older adults, as we observed in our dataset. D'Esposito et al. [@pone.0111121-DEsposito1] measured BOLD responses in motor cortex to a bilateral button press cued by the appearance of a briefly presented white circle. Greater intrasubject variability in older adults compared to younger adults was observed; there were no differences in response amplitude and only a small increase in intersubject variability. Huettel et al. [@pone.0111121-Huettel1] measured responses in visual cortex to checkerboard stimuli with no behavioral task. Greater intrasubject and intersubject variability in older adults was observed, with no difference in amplitude of response. Samanez-Larkin et al. [@pone.0111121-SamanezLarkin1] found that healthy older adults exhibited suboptimal decision-making on a financial investment task compared to younger adults and also exhibited greater temporal variability in the nucleus accumbens. In contrast to these studies that reported more variability in older adults, Garrett et al. [@pone.0111121-Garrett2] reported *less* neural variability in older adults using a variety of complex cognitive tasks. One possible explanation for this result is that the analysis of Garrett et al. used a measure derived from multivariate voxel pattern analysis (MVPA) to measure variability across all brain voxels. MVPA analyses and traditional univariate analyses such as ours may give conflicting or even contradictory results. The explanation for these discrepancies is a matter of debate [@pone.0111121-Jimura1]. Potential Confounds {#s4c} ------------------- A potential confound in BOLD fMRI studies of older populations is vascular changes with age [@pone.0111121-Fang1]. However, studies that directly measure neuronal activity also find age-related increases in variability. Anderson et al. [@pone.0111121-Anderson2] presented auditory syllables to healthy older adults and measured the auditory brain stem response, an electrophysiological measure of neuronal activity that is not influenced by the vasculature, and found greater variability in older adults. In a study of non-human primates, single neuron responses in V1 and MT of older monkeys had greater variability than in younger monkeys [@pone.0111121-Yang1]. These findings suggest that the variability differences in our study have a neuronal component in addition to any possible vascular sources. Intergroup differences in fMRI studies may also be driven by differences in head movements, especially in studies of resting state functional connectivity [@pone.0111121-Power1], [@pone.0111121-VanDijk1]. In a task-based study such as ours, averaging the response to multiple stimuli reduces movement effects, since head movements and stimulus presentation are independent. We used five different methods to account for differences in head motion, and found no effect on our main results. Consistent with these analyses, two previous studies (Huettel et al. [@pone.0111121-Huettel1] and D'Esposito et al. [@pone.0111121-DEsposito1]) did not find a correlation between head motion and BOLD signal variability. Conclusion {#s5} ========== Our most robust finding was of greater intrasubject variability in older adults. This finding was true in both the ROI and whole-brain analysis. Among the physical changes to the aging brain, a decrease in myelination has been observed [@pone.0111121-Lu1], [@pone.0111121-Kerchner1]. These decreases in white matter integrity could lead to increases in neuronal variability by preventing neurons from firing consistently even with the same sensory input. Better understanding the neural sources of this variability and its behavioral consequences may help in designing strategies to ameliorate declines in speech perception, one of our most important cognitive functions. We thank Vips Patel for assistance with MR data collection. [^1]: **Competing Interests:**Co-author MSB is a PLOS ONE Editorial Board member. This does not alter the authors\' adherence to PLOS ONE Editorial policies and criteria. [^2]: Conceived and designed the experiments: SHB MSB. Performed the experiments: SHB. Analyzed the data: SHB. Contributed reagents/materials/analysis tools: SHB MSB. Contributed to the writing of the manuscript: SHB MSB.

**Core tip:** Data are lacking regarding the long-term outcomes of patients with disorders of consciousness (DoC) in China. This was a two-center prospective cohort study of inpatients with prolonged DoC for up to 6 years, included 93 patients (62 vegetative state/unresponsive wakefulness syndrome and 31 minimally conscious state). The results showed that patients with severe DoC, despite having strong predictors of poor prognosis, might recover consciousness after a prolonged time of rehabilitation. INTRODUCTION ============ Recent innovations in intensive care has improved the prognosis of patients with severe brain injuries, which are usually due to traumatic brain injuries (TBI), hypoxic ischemic encephalopathy (HIE), or stroke\[[@B1]\]. Patients may progress from coma to disorders of consciousness (DoC), which include the "vegetative state" (VS) (recently termed "unresponsive wakefulness syndrome", UWS) and the "minimally conscious state" (MCS)\[[@B1]-[@B3]\]. Sometimes these clinical conditions are temporary, but they may also be the stabilized state of the acute brain event and the patients will never recover consciousness; they then enter the stage of prolonged DoC\[[@B1]-[@B3]\]. Reliable epidemiologic data on DoC and emergence from prolonged DoC are scarce. When using a proxy definition, the prevalence is estimated at 5000-42000 for VS/UWS and 112000-280000 for MCS\[[@B4]\]. In Europe, the prevalence is 0.2-6.1 cases per 100000 population\[[@B5]\]. The most important concern for DoC patients and their family is prognosis because it is central to the decisions regarding medical treatment and eventual rehabilitation therapy, and profound ethical issues are involved\[[@B6],[@B7]\]. Indeed, there is a lack of evidence in the literature regarding late improvements in DoC patients and an even more deplorable lack of accurate epidemiological data about the long-term outcomes of DoC in China\[[@B8],[@B9]\]. Lifetime care for a patient with prolonged DoC can exceed \$1000000\[[@B1]-[@B3]\]. In China, the accessibility to proper management services for DoC patients is a major issue, especially for the areas far from large cities\[[@B9]\]. It is therefore necessary to study the long-term outcomes of patients in prolonged DoC in light of many factors likely to influence crucial decisions about their care and their life. Because of this lack of knowledge on the long-term outcome of patients with DoC, we implemented a two-center prospective study that aimed to collect data about the patients with prolonged DoC admitted to two specialized units in Beijing and Nantong (China). The present real-world prospective cohort study aimed to assess the recovery rate of the inpatients with a VS/UWS to MCS and to analyze and compare the long-term outcomes of patients with prolonged DoC considered in VS/UWS or MCS 1 year after the standard follow-up in the prognostic study, then yearly afterwards up to 6 years, hoping to investigate the factors associated with a higher likelihood of transition to MCS at rehabilitation facilities in China and further our understanding of the rehabilitation potential of the most severely affected patients with DoC. We also hypothesized that there was potential for some level of recovery despite the presence of strong unfavorable prognostic markers. Here we present the preliminary results six years after the start of the study. MATERIALS AND METHODS ===================== Study design and participants ----------------------------- Between October 2012 and September 2018, we recruited all consecutive comatose patients with severe brain damage admitted to the department of rehabilitation medicine of the Affiliated Hospital of Nantong University, Nantong and Xuan Wu Hospital, Capital Medical University, Beijing, China. The study was approved by the ethics committee of Xuan Wu hospital according to the guidelines of the Helsinki declaration (1964 and 2000). Written informed consent was obtained from all patients' legal guardians for the use of their clinical data for research purposes. The inclusion criteria for patient enrolment were: (1) Native Chinese speaker; (2) Diagnosis of prolonged DoC (VS/UWS or MCS, at least 28 d) after brain injury of traumatic or non-traumatic (*e.g*., anoxic, vascular, and encephalitic) causes; (3) Normal vision and auditory function; (4) Normal cognitive interpersonal functions prior to brain injury; (5) No history of vital organ injury or failure, stroke, cancer or brain tumor, neuropathy, brain surgery, or familial psychosis; and (6) The patient and his/her family members cooperated with treatment. The exclusion criteria were: (1) Artificial therapeutic coma; or (2) History of brain injury prior to the event leading to the prolonged DoC considered for enrollment in the present study. Behavioral assessment --------------------- All patients enrolled in the study were assessed using the Glasgow coma scale (GCS)\[[@B10]\] and coma recovery scale-revised (CRS-R)\[[@B11]\] at least three times within 1 wk of admission. For patients with major medical complications, the GCS and CRS-R evaluations were performed when the medical condition had stabilized. One month after starting the study, all data were evaluated to ensure their uniformity and accuracy. Follow-up --------- In order to examine the outcomes in this study and obtain a significant prognostic value, follow-up was conducted at 1, 3, 6, and 12 mo after admission. Most patients were regularly managed in a single regional rehabilitation center or long-stay hospital dedicated to patients in prolonged DoC. Phone call follow-up was performed if the patients were discharged from or transferred to other hospitals during the 12-mo follow-up. In the case where the patient had already died, the physician had to provide the last cognitive and medical state of the patient. The investigators determining outcomes were blinded to test results. All patients received rehabilitation wherever possible and appropriate. The rehabilitation program was based on passive joint movements and placing the patients into an upright sitting position using a tilt table. Definition and measures of outcome ---------------------------------- The endpoint of follow-up was recovery of full consciousness or death. The state of consciousness and emergence from MCS (EMCS) was assessed by the CRS-R and the functional disability according to the Glasgow outcome scale (GOS) outcome\[[@B12]\]. The changes in the primary clinical outcome were evaluated at 12 mo compared with baseline. Patients who progressed to EMCS or regained full consciousness \[including the "locked-in syndrome" (LIS)\] were classified as "awareness" (good outcome); those whose clinical diagnosis remained in VS/UWS or MCS (including MCS+/-) were classified as "unawareness" (poor outcome). Patients who recovered awareness and then died because of new etiologic events and those who died remaining unawareness were included in this category. Statistical analysis -------------------- Statistical analyses were conducted using SPSS 22.0 (IBM, Armonk, NY, United States). Descriptive results are summarized as proportions (percentages), means ± SD, or medians (interquartile ranges). Categorical variables were tested using the *χ*^2^ test or Fisher exact test. Continuous variables were compared with the Student's *t* test or the Mann-Whitney U test. The cumulative awareness rates were calculated using the Kaplan-Meier method, and the differences between groups were assessed by the log-rank test. *P* values \< 0.05 were considered statistically significant. RESULTS ======= Characteristics of the subjects at study entry ---------------------------------------------- The study population was composed of 93 patients (62 VS/UWS and 31 MCS, 43 from Beijing and 50 from Nantong). Median follow-up was 20 mo (interquartile range, 12-37 mo). Demographic data of the patients are shown in [Supplementary Table 1](Supplementary Table 1). Over the 6-year period, for the 93 patients initial GCS score was ≤ 8 at onset. The post-injury interval range was 28-634 d. The CRS-R total score range was 0-17. There were 69 males (74.2%) and 24 females (25.8%). Mean age was 49.8 ± 16.9 years (range, 7-85 years). Ten had HIE (10.8%), 35 had TBI (37.6%), and the remaining 48 had stroke (51.6%). There were 62 (66.7%) patients in VS/UWS (age range: 7-80 years; time since injury range: 28-496 d; CRS-R score range: 0-8) and 31 in MCS (age range: 9-85 years; time since injury range: 28-634 d; CRS-R score range: 6-17). All of the patients were put on a specialized program for slow-to-recover brain injury. The two diagnostic groups had no significant differences in age, duration, or etiology, but there was a significant difference in sex (*P* = 0.04) (Table [1](#T1){ref-type="table"}). ###### Characteristics of the patients **Characteristic** **All (*n* = 93)** **Group** ***P* value** ------------------------------------------------- -------------------- --------------- --------------- ---------- Age, yr, mean ± SD 49.8 ± 16.9 49.7 ± 16.4 50.1 ± 18.0 0.897 Maximum 85 80 85 Minimum 7 7 9 Sex, male, *n* (%) 69 (74.2) 42 (67.7) 27 (87.1) 0.044 Post injury days, median (Q~25~, Q~75~) 60 (35, 98) 59 (37, 95.5) 61 (32, 126) 0.446 Maximum 634 496 634 Minimum 28 28 28 Etiology, *n* (%) Traumatic 35 (37.6) 25 (40.3) 10 (32.3) 0.368 Vascular 48 (51.6) 29 (46.8) 19 (61.3) Anoxic 10 (10.8) 8 (2.9) 2 (6.5) GCS score at admission, median (Q~25~, Q~75~) 8 (5.5, 9) 6 (4, 8) 10 (9, 11) \< 0.001 CRS-R score at admission, median (Q~25~, Q~75~) 5 (2, 9) 3 (1, 5) 10.5 (9, 13) \< 0.001 Outcome, *n* (%) VS/UWS 35 (37.6) 35 (56.5) 0 (0.0) \< 0.001 MCS- 10 (10.8) 6 (9.7) 4 (12.9) MCS+ 1 (1.1) 1 (1.6) 0 (0.0) EMCS 4 (4.3) 2 (3.2) 2 (6.5) CS 29 (31.2) 11 (17.7) 18 (58.1) LIS 7 (7.5) 1 (1.6) 6 (19.4) Death 7 (7.5) 6 (9.7) 1 (3.2) GOS outcome, *n* (%) Death (1) 7 (7.5) 6 (9.7) 1 (3.2) \< 0.001 VS/UWS (2) 35 (37.6) 35 (56.5) 0 (0.0) Severe disability (3) 24 (25.8) 11 (17.7) 13 (41.9) Moderate disability (4) 16 (17.2) 5 (8.1) 11 (35.5) Good (5) 11 (11.8) 5 (8.1) 6 (19.4) VS/UWS: Vegetative state/unresponsive wakefulness syndrome; MCS: Minimally conscious state; EMCS: Emergence from minimally conscious state; LIS: Locked-in syndrome; CS: Conscious state; SD: Standard deviation; CRS-R: Coma recovery scale-revised; GCS: Glasgow coma scale; GOS: Glasgow outcome scale. Clinical process of consciousness and functioning recovery ---------------------------------------------------------- Six years after study start, 88/93 patients (61 in VS/UWS and 27 in MCS) had available 12-mo follow-up data; five patients (one in VS/UWS and four with MCS) had not reached the 12-mo time point. We present the preliminary results of the patients with 12 mo of follow-up at the end of the 6-year observation period. At that time point, of the 93 patients, 33 transitioned to an EMCS or full consciousness, eight had an LIS, and there were 35 patients remaining in a VS/UWS and 11 in an MCS. Seven (including one LIS) patients (7.5%) died within 12 mo of injury: Two from infectious diseases, one from respiratory complications, one from heart failure, one from intracranial hemorrhage relapse, and two from an unrecorded/unknown cause. It should be noted that no patient became vegetative during the follow-up period. The 62 patients who were in a VS/UWS had a mean age of 49.7 years (range: 7-80) and a median number of days post-injury of 59 (range: 28-496). The median CRS-R total score at admission was 3 (range: 0-8). Of the 62 patients in the VS/UWS group at 1 mo, two patients were corrected as LIS (one died at 12-mo follow-up), and 16 transitioned to MCS (13 in MCS- and three in MCS+), while 12 patients recovered consciousness (including two in EMCS) (19.4%), three died, and one was in MCS- at 12 mo of follow-up. Of the remaining 44 patients at the end of follow-up, only two recovered full consciousness (one recovered full consciousness at 24 mo), five transitioned to MCS-, and two died, while 35 were still in a VS/UWS (56.5%). At that time, of the 31 patients with MCS (22 in MCS- and nine in MCS+) at admission, except six who showed an LIS and one who died, MCS+ in all other five patients and MCS- in 15 patients transitioned to EMCS or full conscious (64.5%), and MCS- in the other four patients did not change (12.9%). The Kaplan-Meier plot of time from injury to awareness is shown in Figure [1](#F1){ref-type="fig"}. {#F1} Factors associated with improved responsiveness ----------------------------------------------- Table [2](#T2){ref-type="table"} presents the factors associated with improved responsiveness. Compared with the unresponsive group at 12 mo, the responsive group had a higher proportion of males (87.8% *vs* 63.5%, *P* = 0.008), a shorter time from injury (median, 40 d *vs* 65 d, *P* = 0.006), higher frequency of vascular etiology and lower frequency of traumatic etiology (*P* = 0.007), higher GCS score at admission (median, 9 *vs* 6, *P* \< 0.001), higher CRS-R score at admission (median, 9 *vs* 2.5, *P* \< 0.001), higher CRS-R scores at 1 mo (median, 14 *vs* 5, *P* \< 0.001) and 3 mo (median, 20 *vs* 6, *P* \< 0.001), lower frequency of VS/UWS diagnosis (36.6% *vs* 90.4%, *P* \< 0.001), and more favorable GOS outcome (*P* \< 0.001). The functional outcomes of the 41 awareness patients were: 13 with severe disabilities (31.7%), 16 with moderate disabilities (39.0%), 11 with good outcome (26.8%), and one death (2.4%). ###### Characteristics of patients with or without improved responsiveness **Characteristic** **Outcome** ***P* value** --------------------------------------------------- -------------------- ------------------- ---------- Age, yr, mean ± SD 49.4 ± 16.2 50.3 ± 17.8 0.797 Maximum 79 85 Minimum 7 13 Sex, male, *n* (%) 33 (63.5) 36 (87.8) 0.008 Days post injury, median (Q~25~, Q~75~) 65.5 (38.5, 136.5) 40.0 (30.0, 87.0) 0.006 Maximum 496 634 Minimum 29 28 Etiology, *n* (%) Traumatic 23 (44.2) 12 (29.3) 0.007 Vascular 20 (38.5) 28 (68.3) Anoxic 9 (17.3) 1 (2.4) GCS score at admission, median (Q~25~, Q~75~) 6 (4, 8) 9 (7.5, 10) \< 0.001 CRS-R score at admission, median (Q~25~, Q~75~) 2.5 (1, 5) 9 (6, 12) \< 0.001 CRS-R score change at 1 mo, median (Q~25~, Q~75~) 5 (3, 7) 14 (10, 19) \< 0.001 CRS-R score change at 3 mo, median (Q~25~, Q~75~) 6 (4, 8) 20 (17, 22) \< 0.001 Diagnosis, *n* (%) VS/UWS 47 (90.4) 15 (36.6) \< 0.001 MCS- 5 (9.6) 17 (41.5) MCS+ 0 9 (22.0) GOS outcome, *n* (%) Death (1) 6 (11.5) 1 (2.4) \< 0.001 VS/UWS (2) 35 (67.3) 0 Severe disability (3) 11 (21.2) 13 (31.7) Moderate disability (4) 0 16 (39.0) Good (5) 0 11 (26.8) VS/UWS: Vegetative state/unresponsive wakefulness syndrome; MCS: Minimally conscious state; EMCS: Emergence from minimally conscious state; LIS: Locked-in syndrome; SD: Standard deviation; CRS-R: Coma recovery scale-revised; GCS: Glasgow coma scale; GOS: Glasgow outcome scale. Table [3](#T3){ref-type="table"} presents the univariable and multivariable analyses for recovery of awareness as the outcome. In the univariable analysis, VS/UWS *vs* MCS (OR = 16.29, 95%CI: 5.32-49.93, *P* \< 0.001), etiology of stroke *vs* TBI (OR = 2.68, 95%CI: 1.09-6.62, *P* = 0.03), GCS score at admission (OR = 1.77, 95%CI: 1.37-2.28, *P* \< 0.001), and CRS-R score at admission (OR = 1.59, 95%CI: 1.32-1.91, *P* \< 0.001) were associated with recovered awareness. In the multivariable analysis, days post injury (OR = 0.99, 95%CI: 0.974-0.996, *P* = 0.006), GCS score at admission (OR = 0.31, 95%CI: 0.12-0.82, *P* = 0.02), and CRS-R score at admission (OR = 4.82, 95%CI: 1.94-11.95, *P* = 0.001) were independently associated with recovered awareness. ###### Logistic regression analysis with recovery of awareness/unawareness as the outcome **Univariable analysis** **Multivariable analysis** -------------------------- -------------------------- ---------------------------- ----------------------- ------- Age 1.003 (0.979, 1.028) 0.794 1.004 (0.962, 1.048) 0.860 Sex 0.241 (0.081, 0.719) 0.011 0.949 (0.196, 4.600) 0.948 Days post injury 0.997 (0.992, 1.001) 0.172 0.985 (0.974, 0.996) 0.006 Group (VS/UWS *vs* MCS) 16.293 (5.317, 49.925) \< 0.001 0.083 (0.004, 1.701) 0.106 Etiology Stroke *vs* TBI 2.683 (1.087, 6.623) 0.032 1.824 (0.410, 8.126) 0.430 HIE *vs* TBI 0.213 (0.024, 1.885) 0.165 0.006 (0.000, 4.798) 0.134 GCS score at admission 1.766 (1.370, 2.277) \< 0.001 0.312 (0.119, 0.823) 0.019 CRS-R score at admission 1.590 (1.322, 1.913) \< 0.001 4.822 (1.946, 11.949) 0.001 CI: Confidence interval; OR: Odds ratio; TBI: Traumatic brain injury; HIE: Hypoxic ischemic encephalopathy. DISCUSSION ========== The results suggest that patients with severe DoC, without distinction for age, etiology, duration, or extent of DoC, may recover consciousness, despite having strong predictors of poor prognosis. The present preliminary analysis provides information on consecutive patients with severe brain injury and DoC. The number of patients with VS/UWS was higher than that of patients with MCS, and the frequency of vascular etiology was higher than that of traumatic and anoxic etiologies. These findings are supported by an Italian multicenter rehabilitation registry\[[@B13]\], but not by a German multicenter rehabilitation registry\[[@B14]\], where the frequency of anoxic etiology was higher than that of the vascular or traumatic etiology, and time since injury was shorter (*i.e*., mean 28 d). Therefore, the discrepancies among registries could likely be related to the differences in the eligibility criteria. The first aim of this present longitudinal study was to present the process of consciousness recovery on clinical evolution after 12-mo of follow-up. Patients with severe brain injury, irrespective of the cause, often develop VS/UWS after an initial coma and subsequently develop MCS, but not always reaching full consciousness or self-awareness. Upon study entry, the patients with VS/UWS and those with MCS showed different clinical scale scores, which is consistent with the diagnoses of the patients, but the two groups had similar age, time post-injury, and etiology, which are different from the findings of previous studies\[[@B15],[@B16]\]. So far, 26/31 MCS patients, but only 16/62 VS/UWS patients, in our cohort recovered consciousness or corrected to an LIS, and thus the clinical outcomes were significantly different between the two diagnostic groups. This finding is supported by previous studies of patients with MCS showing a better short-\[[@B17]-[@B19]\] and long-term outcome (\> 12 mo post-injury)\[[@B17],[@B20]\]. Nevertheless, the patient populations remain heterogeneous among the studies and within the studies in terms of time post-injury, frequency of traumatic or non-traumatic etiology, level of consciousness, and sample size. The secondary aim of this longitudinal study was to examine the factors affecting prognosis. Anoxic DoC has a worse prognosis than vascular or traumatic DoC, and longer VS/UWS durations are associated with a lower likelihood of regaining awareness. Unfortunately, only one patient with HIE emerged from MCS, while all the other patients with HIE remained VS/UWS or MCS- after up to 4-year follow-up. Thus, our observations are in line with the generally very poor functional prognosis of HIE. It has been suggested that VS/UWS be considered permanent when lasting \> 12 mo after TBI or 3 mo after the original anoxic/vascular condition, but late recovery of consciousness has been reported\[[@B21],[@B22]\]. In any case, emergence can occur months or even years after DoC, and long time to emergence always leads to severe or extremely severe functional disability and poor functioning. Moreover, our study included DoC patients who entered the rehabilitation centers and were followed for 12 mo. Therefore, compared with previous studies, the patients' duration of onset was longer, which was more likely to reflect the late prognosis of DoC. Lammi et al\[[@B23]\] found that recovery after a prolonged MCS is highly variable among patients. MCS duration does not seem to influence the psychosocial condition or functional recovery of the patients. Patients in VS/UWS with a transition to MCS within 8 wk of DoC onset are likely to recover to higher functional levels, with many of them able to go home and resume daily activities\[[@B24]\], supporting the present study. Since our cohort had enrolled more MCS patients with vascular etiology, some of whom were corrected to LIS, more patients with vascular etiology recovered from DoC. It has to be stressed that the data presented here are preliminary, but they appear to confirm that patients with anoxic DoC have worse outcomes than patients with traumatic or vascular DoC, as supported by the literature\[[@B17],[@B20],[@B24]-[@B26]\]. In the present study, although the preliminary data cannot provide definitive diagnostic and prognostic information, they strongly indicate that patients in prolonged DoC should not be pooled for prognostic purposes at admission. First, in all 93 patients, we screened out eight cases of LIS, who should not be classified as prolonged DOC strictly. Second, the analysis showed that the potential for unfavorable outcome was significantly greater in VS/UWS than in MCS. Thirty-one patients in the MCS group had a median duration of 61 d, slightly longer than those reported in previous studies. However, the results suggested that six patients were corrected to an LIS, and that 80% of the remaining 25 patients converted to EMCS within 6 mo, while 18 patients regained full consciousness. Of the 62 patients with VS/UWS, two were corrected to an LIS, 13 entered to EMCS, 11 of whom recovered complete consciousness, and 10 of these 11 patients had been converted to MCS at 1-mo follow-up. Another patient achieved EMCS only one year after follow-up, and eventually recovered full consciousness at 2-year follow-up. We found that, contrary to VS/UWS, the permanence of MCS cannot be established 1 year after coma onset. Between 3 and 6 mo after coma onset, most patients in MCS emerged from this condition. At this stage, patients in MCS seem to have an important potential for recovery given that, at 1-year follow-up, approximately 50% of them recovered at least partial independence for locomotion and activities of daily living. After 6 mo, that potential seems to drop given that, among four patients who remained in MCS for more than 6 mo, none regained full consciousness. After 6 mo in the MCS state, the prognosis becomes even worse, which is the same as in VS/UWS. Among 39 patients who remained in VS/UWS for 3 mo of follow-up, only one regained full ambulation and 34 remained in this condition at the endpoint. To our knowledge, only one study has reported the outcome at 1 year after patients were considered in MCS after 6 mo of follow-up and this outcome was poor because none of the four patients could live with family; all needed total assistance for activities of daily living and had to be institutionalized\[[@B20]\]. Unsurprisingly, our data report very poor functional outcomes, because 47 patients with DoC remained in the same condition or died during follow-up. Nevertheless, EMCS after 1 year was not exceptional and also occurred mainly during the second year of coma. Long MCS duration is associated with a lower probability of recovery, in a similar way to that for VS/UWS. Altogether, these observations tend to confirm the difference in outcomes already suggested between MCS and VS/UWS. This difference lends further support to the need for a clear distinction between MCS and VS/UWS and, thus, for better ways to differentiate the two conditions. Although the design of our study precluded a detailed description of the clinical and functional status of patients who emerged from MCS, most patients remained severely disabled on the motor and cognitive points of view. Our findings emphasize the clinical importance to follow patients with DoC. VS/UWS can be considered permanent in TBI patients after 12 mo and non-TBI patients after 3 mo. This view has been challenged. Avesani et al\[[@B27]\] described two people diagnosed with VS/UWS who, at respectively 6 and 12 mo after their original trauma, had achieved a moderate level of functional independence following a significant motor and cognitive recovery after 5 years and suggested that it is important to conduct regular follow-ups to better evaluate changes and, if it is necessary, to re-adjust the rehabilitation accordingly. Also, our results have shown that patients admitted to rehabilitation in an unresponsive state can show considerable recovery even after a prolonged time. Although slow regeneration of axons in patients with brain injury could be an intriguing hypothesis as a biological mechanism of delayed recovery, no neurological interpretation of late recovery from VS/UWS has been advanced and early predictors might not apply\[[@B8]\]. Our previous study has supported the use of ERP in clinical practice to predict the likelihood of recovery from DoC\[[@B28]\]. An accurate initial diagnosis of patients with DoC is critical for predicting outcome. Misdiagnosis may lead to a worse prognosis for patients, which may restrict their access to recovery. The adoption of homogeneous assessment procedures will provide valuable and reliable data for investigating clinical issues regarding the diagnosis and prognosis of DoC, as well as the effectiveness of treatment strategies for long-term clinical progression. At the same time, a decision to conduct or withhold specialized neurorehabilitation in traumatic or non-traumatic DoC survivors should be considered comprehensively. A strength of this study is that it included all patients with brain injury admitted to two specialized centers. Of course, enrolling patients at rehabilitation facilities will bias the data and show more positive outcomes than what could be observed in non-specialized centers. The present study only included a small number of patients from only two study centers, limiting generalizability. The third limitation of this study stems from the inclusion of a disproportionate number of patients in VS/UWS as a result of stroke (compared to other published studies). Finally, the present study examined prolonged DoC and the long-term complications were not taken into consideration. In conclusion, we present a novel prospective real-world cohort study on the 12-mo outcome of patients with a severe condition at two specialized units in Beijing and Nantong with different etiologies. The results suggest that patients with severe DoC, despite having strong predictors of poor prognosis, might recover consciousness after a prolonged time of rehabilitation. An accurate initial diagnosis of patients with DoC is critical for predicting outcome and a long-term regular follow-up is also important. This preliminary study indicates that establishing a rehabilitation-based registry for patients with severe DoC after brain injury is feasible and probably relevant to improve patient management. ARTICLE HIGHLIGHTS ================== Research background ------------------- Recent innovations in intensive care have improved the prognosis of patients with severe brain injuries and brought more patients with disorders of consciousness (DoC). The most important concern for DoC patients and their family is prognosis because it is central to the decisions regarding medical treatment and eventual rehabilitation therapy, and profound ethical issues are involved. In China, the accessibility to proper management services for DoC patients is a major issue. Research motivation ------------------- Data are lacking regarding the long-term outcomes of DoC patients in China. It is necessary to study the long-term outcomes of patients in prolonged DoC in light of many factors likely to influence crucial decisions about their care and their life. Research objectives ------------------- The study aimed to assess the recovery rate of the inpatients with prolonged DoC after the standard follow-up, hoping to investigate the factors associated with a higher likelihood to recover consciousness at rehabilitation facilities in China and further our understanding of the rehabilitation potential of the most severely affected patients with DoC. Research methods ---------------- This was a two-center prospective cohort study of inpatients with vegetative state (VS)/unresponsive wakefulness syndrome (UWS). The study outcomes were the recovery from VS/UWS to minimally conscious state (MCS) and the long-term status of patients with prolonged DoC considered in VS/UWS or MCS for up to 6 years. The patients were evaluated using the Glasgow coma scale (GCS), coma recovery scale-revised, and Glasgow outcome scale. The endpoint of follow-up was recovery of full consciousness or death. The changes in the primary clinical outcome improvement in clinical diagnosis were evaluated at 12 mo compared with baseline. Research results ---------------- The study population included 93 patients (62 VS/UWS and 31 MCS). At the endpoint, 33 transitioned to an emergence from MCS or full consciousness, eight had a locked-in syndrome, and there were 35 patients remaining in a VS/UWS and 11 in an MCS. Compared with the unresponsive group, the responsive group had a higher proportion of males, shorter time from injury, higher frequency of vascular etiology, higher Glasgow coma scale score and coma recovery scale-revised score at admission, lower frequency of VS/UWS, and more favorable Glasgow outcome scale outcome. Research conclusions -------------------- Patients with severe DoC, despite having strong predictors of poor prognosis, might recover consciousness after a prolonged time of rehabilitation. An accurate initial diagnosis of patients with DoC is critical for predicting outcome and a long-term regular follow-up is also important. Research perspectives --------------------- This preliminary study indicates that establishing a rehabilitation-based registry for patients with severe DoC after brain injury is feasible and probably relevant to improve patient management. Institutional review board statement: The study was approved by the ethics committee of Xuan Wu hospital according to the guidelines of the Helsinki declaration (1964 and 2000). Informed consent statement: All study participants, or their legal guardian, provided written consent prior to study enrollment. Conflict-of-interest statement: The authors of this manuscript have no conflicts of interest to disclose. Data sharing statement: There is no additional data available. STROBE statement: The authors have read the STROBE Statement, and the manuscript was prepared and revised according to the STROBE Statement. Manuscript source: Unsolicited manuscript Peer-review started: December 30, 2019 First decision: April 1, 2020 Article in press: May 20, 2020 Specialty type: Medicine, research and experimental Country/Territory of origin: China Peer-review report's scientific quality classification Grade A (Excellent): 0 Grade B (Very good): 0 Grade C (Good): C Grade D (Fair): 0 Grade E (Poor): 0 P-Reviewer: Lanza G S-Editor: Dou Y L-Editor: Wang TQ E-Editor: Liu JH [^1]: Author contributions: Chen WG participated in study design, drafted the manuscript, was involved in data collection, and assisted with data analysis; Li R, Zhang Y, and Hao JH were involved in data collection and assisted with data analysis; Du JB and Guo AS were involved with data collection; Song WQ participated in design and supervision of the study; all authors read and approved the final manuscript. Supported by the National Natural Science Foundation of China, No. 81371194 and No. 81873723. Corresponding author: Wei-Qun Song, MD, Doctor, Department of Rehabilitation Medicine, Xuan Wu Hospital, Capital Medical University, No 45, Changchun Street, Xicheng District, Beijing 100053, China. <songwq66@126.com>

I[NTRODUCTION]{.smallcaps} {#sec1-1} ========================== The population of India of 1.247 billion accounts for a significant portion of the total population of the world, and around 68% of Indians live in rural areas.\[[@ref1]\] A number of studies and reviews of studies\[[@ref2][@ref3][@ref4][@ref5]\] report that the prevalence of neurological disorders is much higher in rural areas. At the same time, of the 1200 neurologists who are members of the Indian Academy of Neurology, most work in metropolitan cities, and rural parts of the country and interiors remain without specialists, leading to a wide gap between demand and supply of neurological health services in the country.\[[@ref6]\] One of the first steps to start bridging this gap is to obtain relevant and representative neuroepidemiological data from different rural parts of the country which provides insights into the nature of concerns of our heterogeneous and vast population. Furthermore, 8.61% of the population of India (around 104.28 million) comprise individuals belonging to various tribes according to the 2011 census.\[[@ref1]\] These tribal communities face a number of diverse health concerns compounded by the difficult terrains that they live in and the ecological challenges that they face. Factors such as illiteracy, poverty, isolation, poor access to health care, and their own particular traditions, customs, and habits may further exacerbate their concerns.\[[@ref7]\] There is a need for more neuroepidemiological studies in India targeting specific tribal communities. Systematic, community-based neuroepidemiological studies in India are limited. The prevalence rates of specific disorders have been described in studies and reviews of existing research.\[[@ref3][@ref4][@ref8][@ref9][@ref10][@ref11][@ref12][@ref13]\] Relatively fewer studies have described the prevalence rates of a range of neurological conditions;\[[@ref2][@ref14][@ref15][@ref16]\] however, most of the ones reviewed and published have been conducted in the north, east, and south parts of India,\[[@ref2][@ref4][@ref5][@ref8][@ref11][@ref14][@ref15][@ref16]\] with the western and northwestern region relatively under represented (except for a study in Gadchiroli in the Western Ghats\[[@ref13]\]). Barring some hospital studies and clinical profile-based studies in the state of Gujarat,\[[@ref17][@ref18]\] to the best of the authors\' knowledge, community-based studies of the prevalence of neurological disorders from this state are limited. This cross-sectional epidemiological study was aimed at obtaining data about the prevalence of a wide range of neurological disorders, in a rural population cluster with a predominantly tribal population, in the state of Gujarat. M[ETHODOLOGY]{.smallcaps} {#sec1-2} ========================= Study area and population {#sec2-1} ------------------------- Kaparada is one of the six talukas in the district of Valsad in the state of Gujarat. The Valsad region is located in South Gujarat between north latitudes 20°10' and 20° 45' and east latitudes 72° 48' and 73° 43', at 12 m above sea level. It is surrounded by Navsari district in the north, Nashik and Thane districts of Maharashtra in the east, and the Arabian Sea to the south and west. The geographical terrain is hilly, and there are many forests, and it is among the regions receiving the heaviest rainfall. Due to this, it is difficult to access the region during the monsoon season, making access to health care and other amenities a challenge. Three villages in this Kaparada taluka were surveyed for this study, namely, Moti Vahiyal, Arnai, and Chavshala. According to the 2011 census, these villages cover an area of 573 hectares, 855.6 hectares, and 1381.1 hectares, respectively. These villages have 568, 468, and 508 households, respectively, and a population of 2918, 2716, and 3110. Sex ratios indicate almost equal male: female representation. Around 99% of the individuals in these villages hail from Scheduled Tribes. Literacy rates at the time of the census were in the 50% for Moti Vahiyal and Arnai and around 30% for Chavshala.\[[@ref1]\] The dominant language spoken is Gujarati and predominant religion followed is Hinduism. Study design {#sec2-2} ------------ This study was part of a larger project to assess neurological health-care needs, provide services, and improve access to services, in some villages in Kaparada. A cross-sectional community-based prevalence study was conducted, and a version of Gourie-Devi\'s adaptation of Schoenberg\'s two-stage methodology was used to obtain data.\[[@ref19]\] The survey questionnaire {#sec2-3} ------------------------ The survey questionnaire was aimed at screening for a wide range of neurological disorders in both children and adults. It consisted of three sections. Section 1 noted demographic details such as age, sex, education, occupation, and religion, for all members of the household. Section 2 contained questions about symptoms of common neurological conditions in adults, specifically epilepsy, movement disorders, headaches, and stroke. Section 3 contained symptoms of neurological disorders in children, particularly cerebral palsy and developmental disorders. The questionnaire was developed by a team of neurologists and neurophysiotherapists, and it covered the most common neurological symptoms. The questionnaire was translated in Gujarati by language experts. Stage 1 {#sec2-4} ------- A preliminary survey was conducted across the three villages to identify the number of households and participants. Community health workers and teachers were identified to conduct the survey. A 2-day training on the survey was conducted for the health workers and refresher trainings followed at regular intervals. These volunteers conducted the door-to-door survey, attempting to cover every identified household in the three villages. Information was obtained about every member of each household. A team of neurophysiotherapists scrutinized the filled-in questionnaires for positive cases. Any individual who had reported yes to any single symptom was considered to have been "screened in" and was eligible for stage 2. Stage 2 {#sec2-5} ------- In stage 2, a team of six neurologists conducted a medical camp in these three villages, where they assessed the above individuals using history and clinical evaluation. The prevalence of neurological disorders was calculated based on the diagnoses provided by this team of doctors. Age- and sex-specific prevalence rates were calculated for eight broad categories of neurological disorders, namely epilepsy, stroke and stroke residue, movement disorders, upper motor neuron diseases, lower motor neuron diseases, headache, vertigo, and mental and behavioral disorders. R[ESULTS]{.smallcaps} {#sec1-3} ===================== Population characteristics {#sec2-6} -------------------------- The trained volunteers identified 8537 individuals from 1464 households across the three villages. The nonresponse rate at stage 1 was 3.33%, and a total of 8253 individuals were screened. Nonavailability at the household and unwillingness were the two primary reasons for dropouts at this stage. Of 8353 individuals, data of 36 individuals were discarded as the forms were missing or incomplete (0.44% data loss), and a total population of 8217 was considered in this study (which is 96.25% of the total population identified). Of these 8217 individuals, 615 (7.48%) individuals reported at least 1 symptom and were consequently identified for further evaluation at stage 2. Dropout rate at stage 2 was 56.91%. Of the 265 individuals who attended the medical camp and were examined by the team of neurologists, 207 (78%) individuals were diagnosed with a neurological disorder. Six (2.26%) of the ones examined were given more than one neurological diagnoses. Age and sex distribution of the 8217 samples is presented in [Table 1](#T1){ref-type="table"}. The sample included 50.27% men and 49.73% women. Distribution of men and women across the different age groups was fairly comparable. Nearly 40.57% of the individuals were below 19 years of age and 7.96% were above 60 years of age. ###### Demographic details about the population sample  Nearly 99.63% of the individuals reported following Hinduism and 0.37% were found to follow Islam. 44.52% of the individuals were illiterate, 13.85% had completed some primary education, 22.36% had completed some secondary education, and 15.01% and 0.1% had completed their Secondary School Certificate or Higher Secondary Certificate, and/or some professional qualification, respectively. Data for 4.16% of the individuals were either not available or not applicable (very young children). For the adults (aged above 18 years), the chief occupation was farming with 89.19% engaged in this. 1.79% of the individuals were student. 4.21% of the individuals were unskilled laborers and 1.69% were either skilled laborers or engaged in professional work, while data for 3.12% were not available. The average size of the household was 5--6 members, with the smallest households having 1 member and the largest having 15 members. 47.61% of the households could be characterized as small (5 or less members), 46.65% were medium sized (6--9 members), and 5.74% were large (10 or more members). Prevalence {#sec2-7} ---------- The crude prevalence rates of the major disorder categories are reported in [Table 2](#T2){ref-type="table"}. The crude prevalence of neurological disorders overall was found to be 2592.19/100,000. Lower motor neuron diseases were found to be the most frequent; this category included peripheral neuropathy and muscle diseases. Headache was the second most frequent. Some individual disorder crude prevalence rates were found to be 35.51/100,000 for Parkinson\'s disease, 73.04/100,000 for essential tremors, and 24.24/100,000 for Parkinson\'s plus syndromes. The prevalence was 24.34/100,000 for peripheral neuropathy and for intellectual disability and 60.86/100,000 for cerebral palsy. The neurologists also found a prevalence of 36.51/100,000 for leprosy sequelae -- the cases identified had been previously treated for leprosy and showed associated deformities. ###### Crude prevalence rates of neurological disorders  Age-specific prevalence rates and sex-specific prevalence rates are reported in Tables [3](#T3){ref-type="table"} and [4](#T4){ref-type="table"}, respectively. For prevalence of neurological disorders overall, a generally increasing trend with a plateau between the ages of 40 and 59 years was seen, with the highest prevalence for those aged above 70 years. The exception was the 10--19 years\' age group that had the lowest prevalence rate. A female preponderance was seen in the diagnoses, with a more than two times higher rate of diagnoses among females as compared to males. ###### Age-specific prevalence rates out of 100,000  ###### Sex-specific prevalence rates out of 100,000  Marked differences were seen in age- and sex-specific prevalence rates for different disorders. The highest prevalence rates for epilepsy were seen for the 0--9 years and 70 years and above age groups. 47.61% of those diagnosed with epilepsy were below the age of 10 years. 100% of those diagnosed with movement disorders were above 40 years of age and 80% were above 60 years, with the highest prevalence rates for the 60--69 years\' age group. 100% of individuals diagnosed with stroke were above 50 years, and the prevalence rates were highest for those above 70 years, followed by those between 50 and 59 years of age. Increasing trends were seen for vertigo, upper motor neuron diseases, and lower motor neuron diseases, and for headache and mental and behavioral disorders, the pattern was varied. The most common childhood disorders were mental and behavioral disorders and epilepsy. The prevalence rates were higher for males than for females for epilepsy, stroke and upper motor neuron diseases, lower for males than females for movement disorders, vertigo, headache and lower motor neuron diseases, and roughly equal for both sexes for mental and behavioral disorders. D[ISCUSSION]{.smallcaps} {#sec1-4} ======================== This study is one of the very few epidemiological studies conducted in the western and northwestern part of India for a tribal population. It demonstrates that it is possible to systematically assess prevalence even in such remote locations using the standard two-stage methodology, and also illustrates the challenges in doing so. Overall prevalence of neurological disorders {#sec2-8} -------------------------------------------- In a systematic review of neuroepidemiological studies in India from 1982 to 1995 conducted by Gourie-Devi *et al*.,\[[@ref19]\] rates of neurological disorders in the population were reported as ranging between 967 and 4070 with an average of 2394/100,000. In subsequent studies, rates of 3126/100,000 in the total population and 4070/100,000 in rural areas were reported from Bengaluru in 2004,\[[@ref2]\] and the rate of 5168/100,000 was reported from Kashmir in 2016.\[[@ref15]\] The current study found rates of 2592.19/100,000, which is close to the average mentioned,\[[@ref19]\] but much lower than the rates in other areas. These differences could also be due to differences in the categorization of neurological disorders and the kinds of neurological disorders included across different studies, and it could also suggest geographical differences. The age- and sex-specific prevalence trends in this study mirror those of previous studies, with rates increasing with age, and female preponderance. The rates of neurological disorders among children below 9 years in this study (1193.32/100,000) is much higher than in a previous study in Kashmir that reported 706/100,000.\[[@ref20]\] Epilepsy {#sec2-9} -------- Reviews of epidemiological studies on epilepsy from different parts of the country have found very diverse rates, with reported ranges being 559--1000/100,000\[[@ref11]\] and 300--1190/100,000,\[[@ref3]\] and these trends are reported to be similar to trends seen in high-income countries where the range is from 270 to 1240/100,000.\[[@ref3]\] The rate of epilepsy in this study was 255.6/100,000 falls below these ranges. The two peaks in prevalence rates in childhood and after 70 years of age found in this study are similar to the trends found previously in a number of Indian community-based epidemiological studies, for example, Mani *et al*. in 1998 and Banerjee *et al*. in 2010.\[[@ref3]\] A male preponderance is noted in most community-based epidemiological studies\[[@ref3][@ref11]\] and in a clinical-based study from the northwestern region of Chandigarh (geographically the closest to Gujarat compared to locations of other studies),\[[@ref21]\] and similar trends are seen here. Stroke {#sec2-10} ------ The rates of stroke described in several reviews and studies are as varied as 44--898\[[@ref9][@ref10][@ref12][@ref13][@ref14][@ref15][@ref16][@ref19][@ref22][@ref23][@ref24][@ref25]\] The rate reported in this study of 109.53 falls in this broad range. It is much lower than those in another tribal population of Gadchiroli in the Western Ghats (388.43).\[[@ref13]\] Most but not all studies have reported a male preponderance\[[@ref12][@ref13][@ref14]\] as was found in the current study, and the latter was also found in two clinical case studies from Gujarat itself.\[[@ref17][@ref18]\] Highest prevalence of stroke in the above 70 years of age group is similar to other studies.\[[@ref14]\] Previous studies have reported a fair percentage of cases of young-onset stroke,\[[@ref10]\] but no such cases were found in this study. Types of strokes were not identified in this study. However, similar to the studies based out of the Gujarat Medical College,\[[@ref17][@ref18]\] hemiplegia was the most common clinical presentation. Movement disorders {#sec2-11} ------------------ The prevalence of movement disorders was found to be 133.9/100,000. The crude prevalence rate of Parkinson\'s disease was found to be 35.5/100,000, which is in keeping with rates of 6--53 reported in review\[[@ref19]\] and in other rural studies.\[[@ref2][@ref14][@ref15]\] The rates of essential tremors are markedly different in different studies, and in some, they have been reported to vary from 8 to 395.\[[@ref19]\] Following global trends,\[[@ref26][@ref27][@ref28]\] most of the cases were reported in individuals above 60 years of age. Global trends suggest a slight male preponderance in most countries, except for Asia, where the rates are more equal across sexes; in the current study, a slight female preponderance was seen for seen for Parkinson\'s disease, but a high one was seen for movement disorders in general. There was a 24.24/100,000 prevalence of Parkinson\'s plus syndromes. The rates of Parkinson\'s plus syndromes have not previously been reported in the Indian epidemiological surveys reviewed by the authors. Other disorders {#sec2-12} --------------- The rates of headache were lower than other studies in Bengaluru, Kashmir, and West Bengal.\[[@ref2][@ref15][@ref16]\] The rates of vertigo were markedly different from those reported in another study conducted in West Bengal.\[[@ref16]\] The rate of cerebral palsy were lower than those seen in Kashmir.\[[@ref15]\] The high rates of lower motor neuron diseases could be related to the heterogeneous nature of the classification. In sum, it appears that the prevalence rates found in this tribal population are either on the lower end or much lower than available figures, especially for figures previously found in rural populations. However, apart from Parkinson\'s disease, the age and sex distribution of diagnoses show patterns similar to those in other studies in India. Limitations {#sec2-13} ----------- This study was done in a small population with limited resources. All the diagnoses were based purely on the clinical history and examination. This may have limited diagnostic accuracy pending further imaging and test results. 78% of the individuals who were screened and who attended the medical camp in stage 2 actually received a diagnosis of a neurological disorder. Around 21% of the patients identified with neurological conditions were referred for further imaging to clarify the diagnoses. The high coverage of the population (96.25%) in stage 1 is noteworthy. However, the dropout rate of 56.91% at stage 2 is a major concern. Possible reasons for the high dropout rates could have been that the individuals forgot the dates of the camp, were at work, or could not leave the house, on the specific dates that the team of the neurologists conducted the medical camps, or because their reported concerns were perceived by them to be very vague and diffuse, and the individuals felt that these were not worth the visit to a doctor. The team took several steps to attempt to reduce dropout rates. The health volunteers visited every single household the day before the camp to remind individuals of the camp date; they also visited each household on the day of the camp to remind the individuals again and on occasion, personally accompanied individuals to the camp venue. The camp was held at a prominent place that was well known and easily accessible in the community. The neurologists themselves conducted home visits to the houses of those individuals with mobility restrictions. The camp was conducted in batches held at different times including evening hours, to make it easier to accommodate individuals\' work hours. A follow-up visit by the neurologists was organized in 3 months for those who missed the first camp. Notwithstanding these efforts, the dropout rate persisted. The remoteness of the area and the near-impossibility in accessing it during the monsoons because of its geographic location made further follow-up visits by the doctors a significant challenge. This highlights the difficulties in finding prevalence estimates in such locations. The questionnaire used for identifying patients had been vetted by a team of senior neurologists. It was previously piloted in an informal settlement in Mumbai, where the volunteers were observed by neurologists and physiotherapists while they conducted the assessment to check for accuracy in administration. On its subsequent use in Kaparada, the high diagnoses rate (78%) among those who were screened as positive (out of those who attended the camp) testifies to the utility of the questionnaire as a screen for common symptoms of neurological disorders. However, difficulties faced by the team with respect to camp attendance and remoteness of the location making regular doctor visits difficult made it difficult to gather the necessary requisite data to ascertain the sensitivity and specificity of the instrument. Notwithstanding the challenges, the highlight of the study was the snapshot captured of the rates of neurological conditions in a region that is not otherwise studied, among a population that is often ignored in research studies. Kalkonde *et al*.\[[@ref13]\] highlighted that there were few epidemiological studies in rural areas and that no epidemiological study in rural areas had been conducted in 20 years (before theirs in 2016). They outlined the need for more current epidemiological data, given the transitioning nature of the Indian rural units. More robust studies in ethnically diverse groups in varied parts of the country are highly recommended. Another highlight of the study was that it combined research with service. All patients were provided medications and other treatments free of cost, and they were directed to other relevant intervention professionals. C[ONCLUSIONS]{.smallcaps} {#sec1-5} ========================= The team further intends to conduct a detailed analysis of the cost of treatments, access to delivery systems, treatment compliance and other healthcare variables, in this region, and combine this data with the prevalence data, to design and execute effective health-care models for this region, in an attempt to bridge some of the glaring treatment gap for neurological conditions in India. Financial support and sponsorship {#sec2-14} --------------------------------- The study was supported by Bombay Hospital. Conflicts of interest {#sec2-15} --------------------- There are no conflicts of interest.

1. Introduction {#sec1-ijms-20-06017} =============== Glioblastoma (GBM) is the most common and most aggressive form of primary brain tumour in adults, carrying a dismal prognosis. Current standard of care for newly diagnosed GBM remains complex consisting of maximal surgical resection, localized radiotherapy followed by adjuvant temozolomide (TMZ) \[[@B1-ijms-20-06017]\]. In the last decade, cellular heterogeneity of GBM has been extensively studied, as evidenced by the numerous single-cell RNA-seq publications describing the intra-tumoral and inter-tumoral variability of cell types and histological compartments in GBM \[[@B2-ijms-20-06017],[@B3-ijms-20-06017],[@B4-ijms-20-06017],[@B5-ijms-20-06017]\]. The GBM tumour microenvironment (TME) is characterized by complex networks of cancer cells and stromal cells such as astrocytes, microglia, and endothelial cells and an extensive release of soluble factors \[[@B6-ijms-20-06017]\], which interact with the tumour and generate permissive conditions for tumour growth \[[@B7-ijms-20-06017]\]. Recent studies \[[@B8-ijms-20-06017],[@B9-ijms-20-06017],[@B10-ijms-20-06017],[@B11-ijms-20-06017],[@B12-ijms-20-06017],[@B13-ijms-20-06017],[@B14-ijms-20-06017]\] have explored the field of cell-to-cell interaction and the potential exchange of cellular content between non-neoplastic and tumour cells but only limited evidence has been described in GBM \[[@B15-ijms-20-06017],[@B16-ijms-20-06017]\]. Astrocytes, the most numerous non-neuronal cells in the central nervous system (CNS), comprise approximately 50% of the human brain volume \[[@B17-ijms-20-06017]\]. They dynamically respond to changes in the brain environment. While altered astrocytic function is well recognized as a primary contributory factor to a number of neurological diseases, their physiological function is to maintain homeostasis of the brain microenvironment \[[@B17-ijms-20-06017],[@B18-ijms-20-06017]\]. For example, astrocytic cells are directly implicated in the structure/organization of the brain-blood barrier (BBB) and modulating cerebral blood flow \[[@B18-ijms-20-06017]\]. Astrocytes also play important roles in neuronal function coordinating synapse formation during development \[[@B19-ijms-20-06017]\], they also protect neurons by removing metabolic waste products \[[@B20-ijms-20-06017]\], and detect and integrate signals of neuronal damage and inflammation to regulate the neuro inflammatory response \[[@B21-ijms-20-06017]\]. In addition to the physiological roles that astrocytes play in the healthy CNS, during pathological conditions, such as trauma, ischemia, or stroke, astrocytes become activated, upregulating glial fibrillary acidic protein (GFAP) expression and showing a characteristic polarization with the formation of long processes \[[@B22-ijms-20-06017]\]. In GBM, astrocytes directly interact with tumour cells and communicate via gap junctions, leading to increased intracellular calcium and resistance to treatments \[[@B23-ijms-20-06017]\]. Indirect or paracrine communication with surrounding cells often occurs via secretion of exosomes known to carry miRNAs to target key tumour suppressor genes in tumour cells and surrounding microenvironment \[[@B24-ijms-20-06017]\]. Interestingly, Lin J et al. identified subpopulations of astrocytes in the adult brain and in gliomas that respectively contribute to synaptogenesis and tumour pathophysiology, providing a blueprint for understanding diverse astrocyte contributions to neurological disease \[[@B25-ijms-20-06017]\]. Recently the chemo protective role of astrocytes has been investigated in a co-culture with human breast cancer cells, lung cancer cells, or melanoma cells \[[@B26-ijms-20-06017],[@B27-ijms-20-06017]\], establishing the role of reactive astrocytes in brain metastases \[[@B28-ijms-20-06017]\]. Although several publications have investigated the mechanism of chemo resistance of GBM proposing different hypotheses, e.g., communication by growth factor release and paracrine signals, extracellular vehicles (EVs), membrane protrusions, and gap junctions (GJs) \[[@B29-ijms-20-06017]\], the mechanisms by which astrocytes promote tumour progression and drug resistance require further investigation. The latest discoveries in cancer biology have indicated new mechanisms for cell-to-cell communications, namely tunneling nanotubes (TNTs). TNT structure was initially characterized in Rustom's lab in 2004 as long, thin, filamentous membrane extensions acting as delivery tubes between two cells \[[@B30-ijms-20-06017]\]. Subsequently, these have been physically described in numerous cell types \[[@B31-ijms-20-06017],[@B32-ijms-20-06017],[@B33-ijms-20-06017],[@B34-ijms-20-06017]\], including rat astrocytes and C6 glioma cells \[[@B9-ijms-20-06017]\], PC12 cells \[[@B35-ijms-20-06017]\], and murine neuronal cells \[[@B36-ijms-20-06017]\]. Formation of TNTs has been demonstrated, both in vivo and in vitro, to facilitate direct intercellular communication between malignant and normal cells \[[@B33-ijms-20-06017]\] suggesting that TNTs may promote cancer cell resistance, recurrence and metastasis \[[@B33-ijms-20-06017],[@B37-ijms-20-06017]\]. However, the mechanisms of de novo formation of TNT and maintenance in refractory cancer are still unexplored. In addition, whether there exist truly cancer-specific components of TNTs remains to be determined in order to provide selective targets for cancer therapy. Previous studies have shown that preferential mitochondrial transfer through TNTs from stromal cells into cancer cells modulates chemo resistance \[[@B37-ijms-20-06017],[@B38-ijms-20-06017]\]. With regard to astrocytes, despite being the most abundant glial cells in the brain, and therefore highly abundant within the GBM microenvironment, their impact on the tumour microenvironment is yet to be fully understood. Based on our considerable experience in developing 3D human cell in vitro models in the context of GBM invasion \[[@B39-ijms-20-06017]\], BBB-mediated drug delivery \[[@B40-ijms-20-06017]\], brain metastasis \[[@B41-ijms-20-06017],[@B42-ijms-20-06017]\], and drug screening in relevant in vitro all human models \[[@B43-ijms-20-06017]\], we established 2D and 3D in vitro models in a hyaluronic acid-gelatin based hydrogel mimicking TME to investigate the impact of astrocytes within the GBM microenvironment. By exploring the significance of astrocytes in the TME, we demonstrated that astrocytes physically interact with GBM cells reducing drug sensitivity to TMZ, vincristine (VCR), and clomipramine (CLM) by developing TNT connections. We also found that trafficking of mitochondria through TNTs between astrocytes and GBM cells provides insights into the mechanism underlying the capacity of astrocytes to promote GBM aggressiveness and resistance to cytotoxics in 2D and 3D in vitro co-culture models, and that this mitochondria transfer could be implicated in apoptotic rescue in GBM cells. 2. Results {#sec2-ijms-20-06017} ========== 2.1. Cell Trace Dyes Label GBM and Astrocytes Tracing Cell Proliferation {#sec2dot1-ijms-20-06017} ------------------------------------------------------------------------ In order to establish co-culture models to mimic the TME and investigate the influence of non-neoplastic astrocytes on the behavior of GBM cells, we optimized cell labelling using Cell Trace dyes in order to distinguish both cell populations as per previous publication \[[@B43-ijms-20-06017]\]. Two GBM cell lines (U-87 MG, UP-007) were stained using Cell Trace carboxyfluorescein succinimidyl ester (CFSE), ([Figure 1](#ijms-20-06017-f001){ref-type="fig"}A,B; [Figure S1](#app1-ijms-20-06017){ref-type="app"}) and the percentage of CFSE-positive cells remained \>95% during 7 days of culture demonstrating low toxicity even using a high concentration of dyes (10 µM). Along with GBM cells labelled with Cell Trace CFSE, the human astrocyte cell line (UP-010) was labelled with Cell Trace Far Red to obtain a reliable labelling that allowed us to track both GBM cells and astrocytes. Using Cell Trace Far Red (1 μM), astrocytic cells retained Far Red-positive during 9 days in culture and remained viable when compared to unstained cells ([Figure S2](#app1-ijms-20-06017){ref-type="app"}). 2.2. Astrocytes Influence GBM Proliferation and Invasion {#sec2dot2-ijms-20-06017} -------------------------------------------------------- To directly evaluate the effect of astrocytes on GBM behavior, we established co-cultures of CFSE-positive GBM cells and Far Red-positive astrocytes at ratios of 90:10, 80:20, and 50:50. The purpose of the use of different GBM to astrocyte ratios was based on the evidence that astrocytes comprise approximately 50% of the cells in the brain \[[@B17-ijms-20-06017],[@B44-ijms-20-06017]\]. As a control, GBM and astrocytes were grown in a monolayer. In all the co-culture ratios of GBM to astrocytes used, both cells revealed the ability to grow together establishing direct contact and, as displayed in [Figure 1](#ijms-20-06017-f001){ref-type="fig"}A,B both cells remained viable (viability \>95%) in co-culture. To study the effects of co-culture of astrocytes on GBM proliferation, the fluorescence intensity of each Cell Trace (CFSE or Far Red) was used as an indirect measure of proliferation following flow cytometry analysis ([Figure 1](#ijms-20-06017-f001){ref-type="fig"}C,D). In the two GBM cell lines, the presence of astrocytes in culture resulted in a slight increase in GBM cell proliferation, in UP-007 while in the U87-MG the presence of astrocytes did not have any effect on GBM proliferation. Interestingly, an increase in the proliferation of UP-010 when in co-culture with GBM was found ([Figure 1](#ijms-20-06017-f001){ref-type="fig"}C,D). The effect of astrocytes on GBM motility was also assessed using a wound-healing scratch assay ([Figure 2](#ijms-20-06017-f002){ref-type="fig"}). Initially, we performed a scratch-wound assay on the two different GBM cells at confluency either alone or in a contact co-culture with astrocytes ([Figure 2](#ijms-20-06017-f002){ref-type="fig"}A,B). As depicted in [Figure 1](#ijms-20-06017-f001){ref-type="fig"}, U-87MG and UP-007 when in a co-culture with astrocytes showed no significant difference (*p* \> 0.05) in terms of rate of closure of the gap compared to a monoculture of GBM. However, it was possible to observe a positive trend in rate of closure by increasing the ratio of the astrocyte population. 2.3. Astrocytes Modulate Drug Sensitivity in a GBM Co-Culture Model {#sec2dot3-ijms-20-06017} ------------------------------------------------------------------- Astrocytes play a crucial role in GBM malignancy \[[@B44-ijms-20-06017],[@B45-ijms-20-06017]\], however few studies have investigated their function in GBM chemo resistance to TMZ, VCR, or CLM. To delineate an effective concentration for each drug, dose-response was carried out at a range of concentrations, and the IC~50~ value was obtained ([Supplementary, Table S1](#app1-ijms-20-06017){ref-type="app"}). To assess the influence of astrocytes on GBM drug-sensitivity we established a co-culture system of GBM and astrocytes at different ratios (90:10, 80:20, and 50:50) to mimic TME. As shown in [Table S1](#app1-ijms-20-06017){ref-type="app"}, in all the cell lines, treatment with TMZ (200 to 1000 μM) resulted in a reduction of cell viability up to \~75--80% ([Supplementary, Table S1](#app1-ijms-20-06017){ref-type="app"}), and with the range of concentrations tested, IC~50~ value was not obtained. However, the IC~50~ values at \< 30 μM for each cell line were calculated with CLM and VCR ([Table S1](#app1-ijms-20-06017){ref-type="app"}). Based on these results, a concentration within the IC~50~ range of CLM (20 μM) and VCR (2 μM) was used in GBM and astrocyte co-cultures. For TMZ, the concentration of 400 μM was chosen for the purpose \[[@B46-ijms-20-06017]\]. When treated with TMZ, astrocyte cells had no statistically significant effect on the sensitivity of GBM cells, and a significant increase proliferation of UP-007 was observed with astrocytes at all ratios ([Figure 3](#ijms-20-06017-f003){ref-type="fig"}A). CLM- and VCR-treated GBM cells in co-culture with astrocytes were more responsive to cytotoxics even in the presence of a low percentage of astrocytes (10--20%) ([Figure 3](#ijms-20-06017-f003){ref-type="fig"}C--F, *p* \< 0.05). These data indicate that astrocytes could be involved in human glioma cell sensitivity to TMZ, VCR, or CLM treatment, suggesting that the glial cells play fundamental roles in the tumour microenvironment with particular reference to astrocytes here in GBM drug resistance. 2.4. Interactions Between Astrocytes and GBM Cells {#sec2dot4-ijms-20-06017} -------------------------------------------------- Previous studies have demonstrated that astrocytes interact with glioma cells triggering an astrocyte phenotypic modification by the upregulation of proteolytic enzymes and strong expression of GFAP, in particular within the tumour \[[@B47-ijms-20-06017],[@B48-ijms-20-06017]\]. In the present study, astrocyte activation studies were conducted. An in vitro GBM and astrocyte co-culture model was established at ratio 50:50 GBM-astrocytes. As shown in [Figure 4](#ijms-20-06017-f004){ref-type="fig"}A, in the bottom panel, UP-007 GBM cells were infiltrated and surrounded by GFAP-positive activated astrocytes (red fluorescence) which exhibited morphological changes towards a multipolar star shape compared to astrocytes in monoculture which showed a classical fibroblastic form. The GFAP fluorescence intensity was also increased by \~57% ([Figure 4](#ijms-20-06017-f004){ref-type="fig"}B, *p* \< 0.05) especially in co-culture where the high expression was detected in both elongated processes and perikarya of astrocytes ([Figure 4](#ijms-20-06017-f004){ref-type="fig"}A). 2.5. Protective Role of Astrocytes Depends on the Physical Contact with GBM {#sec2dot5-ijms-20-06017} --------------------------------------------------------------------------- To determine whether protection by astrocytes from therapeutic drug effect requires direct contact or paracrine signals secreted by astrocytes, we established co-culture experiments using a semi-permeable Transwell membrane, which ensures physical separation of the GBM cells from the astrocytes while allowing the transmission of secreted factors, and direct contact co-cultures. As shown in [Figure 5](#ijms-20-06017-f005){ref-type="fig"}, TMZ induced-apoptosis in GBM cells in the direct contact co-culture group was approximately 50% (UP-007/UP-010: 4.59 ± 0.13 vs. UP-010/UP-007 Tw 8.67 ± 0.89). When compared with transwell co-culture, this protection was also observed when VCR was used (UP-007/UP-010: 17.41 ± 0.007 vs. UP-010/UP-007 Transwell: 24.41 ± 2.92). When challenged with CLM the apoptosis index ([Figure 5](#ijms-20-06017-f005){ref-type="fig"}) was higher even in direct contact (UP-007/UP-010: 50.5 ± 0.28 vs. UP-010/UP-007 Transwell: 62.4 ±1.83) ([Figure S3](#app1-ijms-20-06017){ref-type="app"}). 2.6. 3D Co-Culture Model Reflects the Protective Role of Astrocytes on GBM {#sec2dot6-ijms-20-06017} -------------------------------------------------------------------------- The results obtained above with the 2D co-cultures of GBM and astrocytes showed the protective effect of astrocytes on both GBM cell lines in human serum-supplemented media. Although these results demonstrate the relevance of non-neoplastic cells in in vitro models of GBM, the 3D environment is a key parameter that must be considered for pre-clinical model studies. Thus, we established a 3D in vitro co-culture mixing GBM and astrocytes using hyaluronic acid-gelatin based hydrogel aiming to mimic the extracellular matrix (ECM) of the tumour. In terms of cell viability, a Cell Titer 96^®^ Cell Proliferation assay displayed that cells, either in mono- or co-culture, growing in 3D remained viable when compared to the 2D cultures ([Figure 6](#ijms-20-06017-f006){ref-type="fig"}A). In addition, cells growing in the 3D system were stained for PI to check the number of apoptotic cells ([Figure 6](#ijms-20-06017-f006){ref-type="fig"}B). The viability of GBM cells in the 3D hydrogel was not altered by the presence of astrocytes ([Figure 6](#ijms-20-06017-f006){ref-type="fig"}B). However, in terms of cell proliferation, GBM cells in the 3D system mixed with astrocytes demonstrated a greater proliferation as the Cell Trace signal diffuses through generations ([Figure 6](#ijms-20-06017-f006){ref-type="fig"}C). Thus, these results reflect the data obtained in the 2D co-cultures, confirming that astrocytes promote proliferation of GBM. In 2D cultures, an increase in proliferation in GBM UP-007 cells was observed, while in the 3D culture in hyaluronic acid, a greater increase was obtained with growth (50:50) in co-culture, suggesting that hyaluronic acid-based ECM further supports the role of astrocytes in GBM proliferation. Subsequently, the 3D co-culture model was tested for drug-screening using TMZ, CLM, and VCR in order to establish a comparison with the 2D co-culture model ([Figure 7](#ijms-20-06017-f007){ref-type="fig"}). Cell proliferation was assessed by Cell Titer 96^®^ Cell Proliferation assay of mono- and co-cultures of GBM and astrocytes at a ratio of 50:50 in 2D and 3D co-culture models when treated with TMZ, CLM, or VCR. In the 3D co-culture model, CLM and VCR caused a decrease of cell viability of GBM in monoculture, however a co-culture with astrocytes resulted in the survival of \~100% of the population in response to the CLM and VCR ([Figure 7](#ijms-20-06017-f007){ref-type="fig"}A2,A3). Moreover, GBM and astrocytes co-cultured in the hydrogel exhibited a 1.1- and 1.2-fold increase in cell viability compared to the monoculture of GBM in 3D when treated with CLM and VCR, respectively (*p* \< 0.001). The addition of TMZ in a 3D system resulted in a statistically significant difference in the mono- and co-cultures ([Figure 7](#ijms-20-06017-f007){ref-type="fig"}A1, *p* \> 0.05), however a slight increase in viability was observed when comparing both conditions in 2D and 3D. To further understand the response of GBM, cell fluorescence intensity of CFSE-positive UP-007 cells treated with TMZ, CLM, and VCR was quantified using intensity fluorescence decrease as a measure of cell proliferation ([Figure 7](#ijms-20-06017-f007){ref-type="fig"}B1--B3). In terms of cell proliferation, the dose of TMZ, CLM, or VCR did not have a statistically significant effect on the mono culture of UP-007 in 3D (*p* \> 0.05). Nevertheless, when comparing the treated mono- and co-cultures in 3D, a significant reduction of the cell fluorescence was observed with all the cytotoxics indicating an increase in cell proliferation, confirming the supportive role of astrocytes to GBM sensitivity. 2.7. Both Homo-Cellular and Hetero-Cellular Extension Occurs Between "Reactive" Astrocytes and GBM Cells {#sec2dot7-ijms-20-06017} -------------------------------------------------------------------------------------------------------- To show the presence of this physical contact and the potential connections between astrocytes (UP-010) and GBM (UP-007) cells under co-culture conditions, we established co-cultures of 50:50 GBM and astrocytes, labelling UP-007 cells with Cell Trace fluorescent dye CFSE, and co-culturing them with UP-010 cells. Long (\~10 to 200 µm) thin, tubular, membranous-based structures connecting "reactive" astrocytes (UP-010) and GBM (UP-007) as well as GBM to GBM cells were observed in 2D ([Figure 8](#ijms-20-06017-f008){ref-type="fig"}D). These processes were able to span long distances from 10 μm to 200 μm ([Figure 8](#ijms-20-06017-f008){ref-type="fig"}D) between astrocyte and GBM cells. These nano tunneling tubules between the co-cultured cells have an F-actin filaments component as shown in [Figure 8](#ijms-20-06017-f008){ref-type="fig"}D by the staining of membrane with Phalloidin WGA AF549 on confocal microscopy and that the membranous-based structures were observed after 48 h more in "reactive "astrocytes where the number of projections were increased ([Figure 8](#ijms-20-06017-f008){ref-type="fig"}C, *p* \< 0.01, *n* = 4) compared to monoculture astrocytes. This suggested the TME induced the formation of these long extensions between astrocytic cells and GBM cells, enhancing intercellular communication and cytoplasmic content exchange. The nano tunneling formation and mitochondrial transfer was also observed in 3D in vitro co-culture ([Figure 9](#ijms-20-06017-f009){ref-type="fig"}). 2.8. TNTs Mediate Transfer of Cytoplasmic Components Between Astrocytes and GBM Cells in 2D and 3D Co-Culture Models {#sec2dot8-ijms-20-06017} -------------------------------------------------------------------------------------------------------------------- We established both 2D and 3D co-culture experiments to investigate a possible exchange of cytosolic content through TNTs. First, GBM were labelled with CFSE to distinguish them from astrocytes in the co culture system. Prior to co culture experiments, we also labelled the astrocytes with the mitochondria-specific dye Mito Tracker Orange to observe mitochondrial transfer between astrocytes and GBM cells. After 2 days of co-culture, before quantifying the mitochondrial transfer by flow cytometry we checked nano tunneling formation and Mito Tracker staining by confocal microscopy ([Figure 10](#ijms-20-06017-f010){ref-type="fig"}A). As shown in [Figure 10](#ijms-20-06017-f010){ref-type="fig"}B,C, by increasing the astrocyte population in a co-culture with GBM, a greater mitochondrial uptake from astrocytes to GBM was obtained with the highest mitochondrial transfer at a ratio of 50:50 astrocytes to GBM ([Figure 10](#ijms-20-06017-f010){ref-type="fig"}B,C). To further corroborate these results and exclude any passive diffusion of the Mito Tracker in the culture media, we stained mitochondria using citrate synthase as a specific marker for mitochondria detection ([Figure 9](#ijms-20-06017-f009){ref-type="fig"}). Here, it is clear the movement of mitochondria along the TNTs from astrocytes to GBM as well as GBM to GBM ([Figure 11](#ijms-20-06017-f011){ref-type="fig"}A,B). The nano tunneling formation and mitochondrial transfer was also observed in 3D in vitro co culture ([Figure 9](#ijms-20-06017-f009){ref-type="fig"}A,B). 3. Discussion {#sec3-ijms-20-06017} ============= Astrocytes have been widely reported in a variety of experimental models to modulate GBM progression and drug resistance \[[@B47-ijms-20-06017],[@B48-ijms-20-06017],[@B49-ijms-20-06017],[@B50-ijms-20-06017],[@B51-ijms-20-06017],[@B52-ijms-20-06017]\]. However, the exact mechanism of how astrocytes facilitate GBM invasion, proliferation, and drug resistance is not yet fully understood. In order to comprehend the role of the astrocyte within the GBM microenvironment, we established a 2D co-culture model using different ratios of GBM cells to astrocytes (90:10, 80:20, and 50:50) with human serum supplementation. Establishing human and animal cells in vitro requires adequate culture conditions and culture media. In the last decade, alternative methods to avoid the use of fetal bovine serum (FBS) in cell and tissue culture media has becoming a major goal in terms of the 3R principles and in terms of good manufacturing practice (GMP), principally to guarantee safe and animal product-free conditions for biomedical tissue engineering and stem cell therapy. In our laboratories, human serum supplementation has been shown to modulate both protein and gene expression in human biopsy-derived GBM cells as well as to influence the cell shape and proliferation \[[@B53-ijms-20-06017]\]. Human serum has also been used to expand human fibroblasts and adipose tissue-derived stem cells (ASC), showing both expansion and differentiation status of cells was superior to media supplemented with FBS \[[@B54-ijms-20-06017]\]. To move forward with the development of pre-clinical models for drug testing studies, we established a 3D in vitro co-culture model using hyaluronic acid-gelatin based hydrogel supplemented with human serum to better simulate human in vivo conditions. For methods to detect different cell populations cultured together in co-culture systems, specific antigens can be employed \[[@B55-ijms-20-06017],[@B56-ijms-20-06017]\] or genetically modified to express fluorescent proteins \[[@B57-ijms-20-06017]\] but fluorescent dyes can also be utilized for medium/long term cell monitoring \[[@B58-ijms-20-06017]\]. GBM is a highly heterogeneous tumour consisting of mutant cells that share many antigens such as GFAP with neural stem cells (NSCs), NSC-derived astrocytes and oligodendrocyte precursor cells (OPCs) \[[@B59-ijms-20-06017]\] making it difficult to distinguish from astrocytic cells by the use of antibodies. In respect of this limitation and as demonstrated in our laboratory \[[@B43-ijms-20-06017]\], without using any genetic manipulation to identify single cells, we optimized cell labelling using two amine-reactive dyes, Cell Trace CFSE and Far Red, for their flexibility, compatibility with primary cell lines and not antigen specificity. Fluorescent dye labelling was also chosen as the appropriate approach for cell tracking in order to preserve genetic heterogeneity within GBM cells. Moreover, the ability of cell trace to diffuse into daughter cells at each cell division \[[@B60-ijms-20-06017]\], allowed us to track cell proliferation. Our results showed that CFSE (5 μM) stains the GBM cells (\~99% of CFSE-positive GBM up to 7 days in culture) and is not toxic for GBM cells. Similarly, Cell Trace Far Red (1 μM) was suitable for labelling astrocytic cells maintaining them viable when compared to unstained controls. Human astrocytes, the most abundant cells in the glioma microenvironment, are key players in the healthy brain but also play dynamic roles in pathological conditions \[[@B61-ijms-20-06017]\].Very few functional data on gap-junction (GJ) changes and connexin 43 (in co-culture models of rat astrocytes and GBM cells) have been reported concerning the putative role of astrocytes in supporting GBM proliferation and invasion \[[@B61-ijms-20-06017],[@B62-ijms-20-06017]\]. Although astrocytes promote tumour progression, their influence in GBM drug response has not been investigated using all human conditions in in vitro models, particularly under human serum supplementation. We initially explored the influence of astrocytes on GBM cell proliferation and invasion using 2D direct contact co-cultures. We then translated this in a 3D in vitro model using hyaluronic acid-gelatin based hydrogels. As shown above increasing of astrocytes proliferation does not affect either GBM viability or proliferation of GBM. Zhang et al. have analyzed tumor-associated astrocytes in GBM patients and discovered similarities to astrocytes from fetal brains marked by increased proliferation \[[@B63-ijms-20-06017]\]. Taken together, these data confirm previous results about the supportive role of astrocytes in GBM aggressiveness \[[@B9-ijms-20-06017],[@B64-ijms-20-06017]\]. To comprehend the impact that astrocytes may have on GBM drug response we challenged the co-culture using TMZ, CLM, and VCR. TMZ, known as the "gold" standard option with which to treat GBM patients, acts by adding methyl groups to purine bases of DNA causing DNA damage \[[@B1-ijms-20-06017]\], and VCR, a vinca-alkaloid that that disrupts microtubules, thus inhibiting mitosis \[[@B65-ijms-20-06017]\], has been extensively used to treat many cancers including brain tumors \[[@B66-ijms-20-06017]\]. CLM is a tricyclic antidepressant drug that has been shown to have selective cytotoxicity against glioma cells inducing mitochondrially mediated apoptosis \[[@B67-ijms-20-06017],[@B68-ijms-20-06017],[@B69-ijms-20-06017]\] and autophagy \[[@B70-ijms-20-06017],[@B71-ijms-20-06017]\]. CLM was included as, not only has it been reported to act on the GBM cells through a completely different, mitochondrial, mechanism \[[@B71-ijms-20-06017],[@B72-ijms-20-06017]\] but because a sophisticated 3D human in vitro drug testing model may provide a non-live animal pre-clinical testing means to 'fast-track' re-purposed drugs like this to clinic for GBM treatment. Indeed, CLM has already shown some evidence of good tolerance and efficacy in small/anecdotal patient cohorts with 80.8% of patients from a cohort of 27 showing good partial response both clinically and radiologically \[[@B73-ijms-20-06017],[@B74-ijms-20-06017],[@B75-ijms-20-06017]\]. Thus, the three selected cytotoxic agents act through distinct mechanisms allowing us to understand the role of astrocytes in the response of GBM to the cellular mechanisms initiated by these agents. Based on the results above, when treated with cytotoxic agents (TMZ, CLM, and VCR), GBM cells showed an increase in proliferation when challenged with TMZ and VCR even in a low ratio of astrocytes and a slight reduction in proliferation when treated with CLM. Little is known about the role of tumour-associated astrocytes and their function in drug response in brain tumors. Only a handful of studies have revealed that astrocytes induce drug resistance and reduce the radio sensitivity of GBM stem cells \[[@B52-ijms-20-06017]\]. Yang et al. using human eGFP/luciferase tagged GBM cell lines and immortalized astrocytes proposed a cell-specific bioluminescent assay for screening clinical relevant drugs \[[@B52-ijms-20-06017]\]. Following these studies, Chen et al. demonstrated that the presence of astrocytes mediates a significantly higher cell survival after TMZ treatment in U251, C6, and A172 GBM cell lines and doxorubicin treatment in certain cell lines (U251 and LN-18) \[[@B64-ijms-20-06017]\]. Later, using a co-culture of human astrocytes and GBM cells they confirmed the protective mechanisms of astrocytes including a reduction of glioma cell apoptosis induced by TMZ and VCR which was essentially mediated by the distribution of Ca^2+^ through Connexin 43 gap junctions \[[@B23-ijms-20-06017]\]. Although different co-culture models have been proposed to study the interactions between astrocytes and GBM cells \[[@B47-ijms-20-06017],[@B48-ijms-20-06017],[@B49-ijms-20-06017],[@B50-ijms-20-06017],[@B51-ijms-20-06017],[@B52-ijms-20-06017]\] most of them used rat astrocytes or selected clones of GBM fluorescent-tagged cells without considering genetic heterogeneity of a GBM cell populations. Here we employed cell labelling using fluorescence ammine CFSE and Far Red as to distinguish the two different cells population in co-culture without selecting a single tagged clone and this provided a suitable method to assess the cell proliferation without using metabolic assay test in a 3D in vitro model. Cell-to-cell communication and TNTs have recently been discovered in the context of exchange of cytoplasmic content via extracellular vesicles (EVs) and organelles between cells, which alter biological functions \[[@B9-ijms-20-06017],[@B31-ijms-20-06017],[@B34-ijms-20-06017],[@B76-ijms-20-06017],[@B77-ijms-20-06017],[@B78-ijms-20-06017],[@B79-ijms-20-06017],[@B80-ijms-20-06017],[@B81-ijms-20-06017]\]. Emerging evidence points to the notions that non neoplastic cells can communicate with a variety of damaged target tumour cells via TNT structures to repair and rescue damaged cells by transferring mitochondria and promoting aggressiveness and drug resistance \[[@B82-ijms-20-06017],[@B83-ijms-20-06017]\]. Osswald et al. reported that astrocytoma cells extend ultra-long membrane protrusions, which they term "tumor microtubes" (TMs) as routes for brain invasion, proliferation, and interconnection over long distances \[[@B15-ijms-20-06017]\]. Subsequently experiments have shown that 1p/19q non-co deleted patients' gliomas are rich in long and intercellular TMs, while 1p/19q co deleted ones are not and pointed to a potential role for tumour microtubes in the resistance to therapies \[[@B16-ijms-20-06017]\]. Others have demonstrated that the formation of TNTs occurs in response to micro environmental changes in ovarian cancer and other forms of invasive refractory cancers \[[@B82-ijms-20-06017],[@B83-ijms-20-06017],[@B84-ijms-20-06017]\]. Here, we hypothesized from our data that physical contact between astrocytes and GBM cells and TNT formation are crucial in mediation of chemo protection and the latter provides a means to transfer cellular contents or energy between cells when are under stress. As shown in co-culture experiments "reactive" astrocytes surrounding GBM cells develop and increase TNT formation toward glioblastoma cells and these cells may use TNTs to transfer undamaged mitochondria, carrying useful substance or energy to GBM cells when challenged with cytotoxic agents. Reactive astrocytes are commonly present in brain injury \[[@B85-ijms-20-06017],[@B86-ijms-20-06017]\] and surrounding brain tumors \[[@B87-ijms-20-06017]\]. In stressed conditions and in response to external stimuli they typically show morphological changes moving towards a hypertrophic phenotype as well as the upregulation of the cytoskeletal intermediate filament protein GFAP \[[@B18-ijms-20-06017],[@B45-ijms-20-06017],[@B47-ijms-20-06017],[@B48-ijms-20-06017]\]. It has also been reported that stressed conditions induce TNT formation in hippocampal rat astrocytes and neurons when treated with H~2~O~2~ or by serum depletion \[[@B88-ijms-20-06017]\] and might facilitate cytoplasmic content transfer and therefore alter the proliferation potential of glioma cells \[[@B9-ijms-20-06017]\]. Therefore, our data support the significance of astrocytes in the TME and the role of TNTs formation in GBM proliferation and drug resistance. Understanding the molecular mechanisms of drug resistance in refractory cancer such as GBM is more crucial than ever in order to achieve effective cancer therapy. Drug resistance mechanisms include drug transporters, DNA damage repair (DDR) and genomic instability, inhibition of apoptosis and metabolic adaptation \[[@B89-ijms-20-06017]\]. Mitochondria play fundamental roles in cancer cells, as they constitute a center for several molecular mechanisms such as apoptosis and metabolic reprogramming \[[@B89-ijms-20-06017]\]. It has been demonstrated that tumor cells, compared to normal cells, display numerous mitochondrial dysfunctions related to energy metabolism, transmembrane potential increase, and elevated ROS generation \[[@B89-ijms-20-06017],[@B90-ijms-20-06017]\]. Moreover, it has been demonstrated that mt-DNA is an important target of therapeutic drugs that interact with DNA \[[@B91-ijms-20-06017],[@B92-ijms-20-06017]\]. Using fluorescent probes to track mitochondria, we observed that extensive mitochondrial transfer occurred from astrocytes to GBM cells and occurred through TNT-like structures ([Figure 10](#ijms-20-06017-f010){ref-type="fig"}, A). This data was confirmed by flow cytometry experiments showing that where there was an increased astrocyte population after 48 h in co-culture growth there was more mitochondrial exchange between astrocytes and GBM ([Figure 10](#ijms-20-06017-f010){ref-type="fig"}B,C). Most interestingly in 3D microenvironment, the ECM component acted as a supportive substrate to increase the number of TNTs ([Figure 9](#ijms-20-06017-f009){ref-type="fig"}A) as already shown in previous experiments where collagen-based matrix enhanced cell migration and process formation \[[@B93-ijms-20-06017]\]. The presence of these structures was also confirmed by the co-localization of mitochondria within the F-actin based TNTs under both 2D ([Figure 11](#ijms-20-06017-f011){ref-type="fig"}A,B) and 3D conditions ([Figure 9](#ijms-20-06017-f009){ref-type="fig"}B). This evidence could explain the role of astrocytes within the GBM microenvironment specifically in drug response. Moreover, tumour-associated glial cells (TAGs) have been shown to promote GBM growth in a xenograft model \[[@B94-ijms-20-06017]\]. Herein, we showed that glial cells, in particular astrocytes in a stressed tumour microenvironment, within both 2D and 3D in vitro co-culture models could form TNTs with GBM cells, thus reducing GBM-induced apoptosis and promoting cell proliferation and migration capacities of GBM via TNT mediated mitochondrial transfer. According to previous reports, TNTs can also transfer cytoplasmic material and organelles between two distant cells, and our study indicated that the "rescue" effect of astrocytes on glioblastoma was, at least, partially related to mitochondrial transfer via TNT-like structures. 4. Materials and Methods {#sec4-ijms-20-06017} ======================== 4.1. Ethics Statement {#sec4dot1-ijms-20-06017} --------------------- A biopsy from a glioma patient was resected at King's College Hospital, London, under Ethics permission LREC00-173/11/SC/0048 (11 April 2011) in accordance with a National Research Ethics Service and the study was approved by ethics committees for the University of Portsmouth and King's College Hospital (UP-007). The astrocyte cell line (UP-010) was obtained similarly from an epilepsy tissue sample under the same Ethics Permission. 4.2. Cell Culture {#sec4dot2-ijms-20-06017} ----------------- Two human GBM cell lines were used in this study (U-87 MG, UP-007). U-87 MG cells were obtained from European Collection of Cell Culture, UP-007 cells were established in house from the GBM biopsy resected at King's College Hospital. UP-010 human astrocytes cell lines were established from an epilepsy tissue biopsy. GBM and astrocytes were cultured in Dulbecco's modified Eagle's medium (DMEM, Gibco, Fisher Scientific, Loughborough, UK) supplemented with 10% (*v*/*v*) of human serum (Sigma-Aldrich, Dorset, UK). All cells were maintained at 37 °C in a humidified incubator with 5% CO2. Media was changed every 2 to 3 days, and sub culturing was carried out by incubating cells with TrypLE Express Enzyme (Gibco, Fisher Scientific, Loughborough, UK) for 3 min. 4.3. Cell Labelling {#sec4dot3-ijms-20-06017} ------------------- Cell labelling optimization screening was performed with all GBM cells and astrocytes using the CellTrace carboxyfluorescein succinimidyl ester (CFSE) and a Cell Trace Far Red (Invitrogen, Fisher Scientific, Loughborough, UK), respectively. For the labelling, cells were incubated with TrypLE Express for 3 min, centrifuged, and resuspended in 1 mL of HBSS (with Ca^2+^ and Mg^2+^). GBM cells were incubated with Cell Trace CFSE (5--10 μM) and UP-010 cells incubated with Cell Trace Far Red (1 μM) in HBSS for 20 min at 37 °C. Subsequently, to remove the unconjugated Cell Trace, cell suspensions were incubated with DMEM supplemented with 5% (*v*/*v*) human serum for 5 min at room temperature. Cells were centrifuged, resuspended in complete DMEM media, and analyzed in a Counter II FL Automated Cell Counter to assess the number of cells and viability. ### 4.3.1. Cell Labelling Efficacy {#sec4dot3dot1-ijms-20-06017} To evaluate the efficacy of the labelling, CFSE^+^ GBM cells and Far Red+ UP-010 cells were seeded at 20,000 cells/cm^2^ in 24-well plates in DMEM supplemented with 10% (*v*/*v*) human serum and allowed to grow for up to 9 days. On days 1, 3, 5, 7, and 9, cells were analyzed by flow cytometry to quantify the remaining percentage of labelled cells. CFSE-positive GBM cells (U-87 MG, UP-007) were detected using the blue 488 laser and the 530/30 channel (FL1). Far Red-positive UP-010 cells were excited with the red laser 635 and detected with a 661/16 channel (FL4). Non-stained cells were included as a control. Cells were analyzed in FACS Calibur flow cytometer (BD Biosciences, Wokingham, UK) collecting at least 10,000 events. FlowJo software (BD Biosciences, Wokingham, UK) was used for analysis of the data. ### 4.3.2. Cell Viability {#sec4dot3dot2-ijms-20-06017} The effect of the Cell Trace on the metabolic activity of GBM cells (CFSE) and UP-010 (Far Red) was evaluated by the Cell Titer 96^®^ Cell Proliferation Assay (MTS, Promega, Southampton, UK). CFSE-positive GBM (U-87 MG, UP-007) or Far Red-positive UP-010 cells were seeded at 20,000 cells/cm^2^ in a 96-well plate in DMEM supplemented with 10% (*v*/*v*) human serum and allowed to grow up to 9 days. At specific time-points, the MTS solution was added (20 μL) to each well, and the cells were incubated at 37 °C for 2 h. Absorbance was measured at λ 492 nm using a spectrophotometric microplate reader (POLARstar Omega, BMG Labtech, Ortenberg, Germany). Non-stained cells were used as a control in each experiment. 4.4. 2D Co-Cultures: Cell Viability and Proliferation {#sec4dot4-ijms-20-06017} ----------------------------------------------------- Co-cultures of GBM (U-87 MG, UP-007) and astrocytes (UP-010) were established by seeding both cells in a 24-well plate at a final density of 20,000 cells/cm^2^. CFSE-positive GBM and Far Red+ UP-010 cells were cultured either alone (40,000 cells) or cultured in co-culture at the following ratio of GBM to UP-010 cells: 90:10 (36,000:4000), 80:20 (32,000:8000), or 50:50 (20,000:20,000). Cells were allowed to grow in close contact for 3 days. ### 4.4.1. Cell Viability {#sec4dot4dot1-ijms-20-06017} To evaluate the viability of the cells in co-culture (UP-007 and UP-010), cells were incubated with TrypLE Express Enzyme for \~3 min, centrifuged, and suspended in HBSS. Propidium iodide (Sigma-Aldrich, Dorset, UK) was added at a final concentration of 50 μg mL^−1^ and incubated for 5 min protected from the light. Cells were washed with HBSS and analysed in the FACS Calibur™ flow cytometer, as previously described. Propidium iodide positive cells were detected using the blue 488 laser and the 630/30 channel (FL3). FlowJo software (BD Biosciences, Wokingham, UK) was used for analysis of the data. ### 4.4.2. Cell Proliferation {#sec4dot4dot2-ijms-20-06017} Cell proliferation was quantified by calculating the inverse of the fluorescence intensity of CFSE-positive GBMs and Far Red-positive UP-010. Both Cell Trace molecules readily diffuse into cells and bind covalently to intracellular amines within proteins resulting in stable, well-retained fluorescent staining. The fluorescence intensity of the Cell Trace diminished with an increase of cell division as the reagent is passed through generations, resulting in an inverse correlation between fluorescence intensity and proliferation. Thus, cells were analysed by flow cytometry, as previously described, and using FlowJo software fluorescence intensity for each channel (FL1 or FL4) was obtained. Proliferation was calculated in relation to a control of cells growing in monoculture according to equation below: where A represents cells growing in a co-culture (90:10, 80:20, and 50:50 ratios of GBM to UP-010) and B is the control of Cells growing in the monoculture either GBM or UP-010. Cell proliferation is expressed in function of the cells growing in a monoculture. ### 4.4.3. Cell Migration {#sec4dot4dot3-ijms-20-06017} Cell migration was evaluated using a wound-healing assay. Cells, either in a mono- or co-culture, were seeded at 100,000 cells/cm^2^ in a 24-well plate and allowed to grow to a confluent monolayer over 24 h. Afterward, a linear scratch was done using a pipette tip. Following injury, wound closure was observed using EVOS FL Auto 2 Cell Imaging System (Fisher Scientific, Loughborough, UK) equipped with a cell imaging chamber under a humidified atmosphere (37 °C, 5% CO~2~). Images were acquired every 30 min over a period of 24 h. Image analysis was performed using *Image J* (U.S. NIH, Bethesda, USA), and rate of closure (μm/h) was calculated by plotting distance of the wound gap *versus* time at different time points (0, 4, 8, 12, and 18 h). 4.5. 2D Co-Cultures: Drug Response {#sec4dot5-ijms-20-06017} ---------------------------------- GBM (U-87 MG, UP-007) or astrocytes (UP-010) were seeded at 20,000 cells/cm^2^ in a 96-well plate and allowed to grow for 24 h. Subsequently, cells were treated with a cytotoxic: temozolomide (TMZ, Sigma-Aldrich, Dorset, UK), clomipramine (CLM, Sigma-Aldrich, Dorset, UK), or vincristine (VCR, Tocris, Bristol, UK) to calculate an IC~50~ value for each cell line. TMZ was dissolved in dimethyl sulfoxide at 1 mM, and working solutions were prepared from 200 to 1000 μM in DMEM. CLM was dissolved at 50 mM in water, and solutions were prepared at concentrations ranging 20--100 μM in DMEM. VCR sulphate was dissolved in water at 1 mM and working solutions were used from 2 to 10 μM in DMEM. After drug treatment, cells were incubated for 2 days, and Cell Titer 96^®^ Cell Proliferation Assay, as previously described, measured viability. IC~50~ values were calculated using Graph Pad Prism 7.03 software (Graph Pad, San Diego, USA). To assess the effect of astrocytes in GBM drug response, co-cultures in 2D were established as previously described, and cells were treated with TMZ (400 μM), CLM (20 μM), or VCR (2 μM). Following 48 h of drug treatment, cells were analyzed by flow cytometry using a FACS Calibur™ collecting 10,000 events. The FlowJo software was used for analysis of the data, and cell proliferation was calculated using Equation 1. Untreated cells were used as a negative control and unlabeled cells were used as a control for the flow cytometry analysis. 4.6. Immunofluorescence Assay {#sec4dot6-ijms-20-06017} ----------------------------- UP-010 astrocytes and UP-007 CFSE+ GBM cells were co-cultured (50:50) in 24-well culture plates for 48 h. Cells were fixed in 4% (w/v) paraformaldehyde permeabilized with 0.5% (w/v) Triton X-100 and blocked with 1% (w/v) non-fat milk powder. Then, the cultures were incubated with primary antibodies against GFAP (Dako, rabbit monoclonal, 1:250) overnight at 4 °C, and a secondary antibody conjugated with Alexa Fluo-647 fluorescent dyes (1:1000, goat anti-rabbit IgG, Abbkine) was applied for 1 h at 37 °C. After staining with DAPI, the cells were imaged using Imaging was carried out using a LSM719 confocal laser-scanning microscope (Zeiss, Oberkochen, Germany). Mean fluorescence intensity was calculated using *Image J* Software (BD Biosciences, Wokingham, UK). 4.7. Co-Culture Chemo-Sensitivity Experiments {#sec4dot7-ijms-20-06017} --------------------------------------------- To determine whether astrocytes could protect glioma cells from apoptosis induced by chemotherapeutic drugs, we used an astrocyte--GBM co-culture system in vitro. UP-007 cells were stained with CFSE to be distinguished from the UP-010. GBM and UP-010 astrocyte (50:50) were cultured alone or co-cultured in 6-well plates. After incubating for 24 h, the cultures were treated with TMZ (400 μM), VCR (2 µM) or CLM (20 µM). 48 h later, the cells were harvested and stained with Alexa Fluor Annexin V-647 (Invitrogen) and propidium iodide (PI) to determine the level of apoptosis of CFSE^+^ GBM cells by flow cytometry. To detect whether such sensitivity to chemotherapeutic drugs is due to direct contact between astrocytes and GBM cells or paracrine signals via astrocytes, we used a transwell co-culture system. UP-010 were seeded on top of transwell chamber with 0.4 µm-pore-sized membrane and co-cultured with UP-007 cells plated at the bottom of a 6-well plate. After adding chemotherapeutic drugs for 48 h, the apoptotic fraction was determined by flow cytometry as described below. ### Apoptosis Determined by Flow Cytometry The culture medium containing floating cells and the attached cells was collected from each well. Apoptosis was analyzed by flow cytometry according to the manufacturer's instructions. Apoptotic UP-007 cells were defined as Annexin V-647-positive cells. The Annexin V-647-positive and PI-negative area represents early apoptosis, Annexin V/PI-positive area represents end-stage apoptosis and death. In co-culture systems, UP-007 cell apoptosis was detected by gating on CFSE^+^ events to exclude UP-010. Apoptotic index was calculated according to Prieto et al. \[[@B95-ijms-20-06017]\]. 4.8. Immunofluorescence Detection and Confocal Imaging of TNTs and Mitochondrial Transfer {#sec4dot8-ijms-20-06017} ----------------------------------------------------------------------------------------- ### 4.8.1. Membrane Staining and TNT Formation Processes {#sec4dot8dot1-ijms-20-06017} Cells were labelled with 1 mg/mL Alexa Fluor^®^ 594-conjugated wheat germ agglutinin (WGA, Invitrogen) at 5 μg/mL for 15 min at 37 °C in the dark. WGA is a fluorescent lectins probe for detecting glycol-conjugates, which selectively binds to N-acetylglucosamine and N-acetylneuraminic acid residues of cell membranes. ### 4.8.2. TNT Counts {#sec4dot8dot2-ijms-20-06017} Monoculture or Co-cultures (50:50 ratio) of astrocytes and glioblastoma cells CFSE-positive were grown for 3 days and after stained with WGA AF594 to specifically mark membrane projection in both cells. TNTs extensions were counted in 10 randomly selected fields using a 20× objective and averaged. ### 4.8.3. Cytoskeletal Staining {#sec4dot8dot3-ijms-20-06017} To determine the composition of TNTs, astrocytes and GBM cells were seeded onto 24--well glass bottom dishes 15.000 cells/cm2 at a 50:50 ratios and growth for 48 h, subsequently the co-culture system was treated with 4% (w/v) PFA. The cells were then permeabilized with 0.1% (w/v) Triton X-100 in PBS for one minute at room temperature and washed three times with PBS. 1% (w/v) non-fat milk powder was used as a blocking agent and applied for 30 min. The sample as then stained with Alexa Fluor 546 Phalloidin (Life Technologies) and DAPI (Life Technologies) for 45 min, washed three times and stored in fluorescent mounting medium (Dako North America, Carpinteria, CA, USA). ### 4.8.4. Mitochondrial Transfer Between Astrocyte Cells and Glioblastoma Cells via TNTs {#sec4dot8dot4-ijms-20-06017} To trace intercellular exchange of mitochondria, cells were labelled separately with Mito Tracker^®^ probe. Briefly, cells were resuspended in prewarmed (37 °C) staining solution containing the MitoTracker^®^ probe (25 nM) for 30 min under appropriate growth conditions. Cells were plated in mono or co-culture according to the experiment and allowed to grow for 3 days. To avoid potential not TNT related mechanisms of Mito Tracker^®^ transfer e.g., via uptake of cell debris from dead cells containing fluorescent particles, a medium change immediately after attachment of the living cells to the culture dish was performed. Mito Tracker^®^ passively diffuses across the plasma membrane of live cells and accumulates in active mitochondria. After staining, cells were washed three times in PBS and suspended in fresh pre-warmed medium. All imaging was performed using a Zeiss LSM719 confocal laser-scanning microscope (Carl Zeiss, Oberkochen, Germany). Post-acquisition image analysis was performed with ZEN 2012 ×32 blue software (Carl Zeiss). Images were taken with an oil immersion 40×/1.4NA objective (Carl Zeiss). ### 4.8.5. Analysis of Mitochondrial Transfer by Flow Cytometry {#sec4dot8dot5-ijms-20-06017} To visualize Mito Tracker transfer via TNTs, we seeded GBM CFSE^+^ cells at different ratios (90:10, 80:20, and 50:50) with astrocytes stained with Mito Tracker Orange 25 nM (Invitrogen). Cells were allowed 48h to grow prior to analyses. After co-culture, cells were trypsinised, resuspended, and washed with PBS. Fluorescence (FL) was quantified on a FACS Calibour (BD Biosciences). Data were processed with FlowJo Software (BD Biosciences, Wokingham, UK). Glioblastoma CFSE fluorescence cells was acquired with (FL-1 channel) 488 nm blue laser and 510/50 nm emission, while Mito Tracker Orange astrocytes was acquired with a FL-2 Channel 640 nm red laser and 670/14 nm emission. Dot plot graph display co cultured cells detected with different channels. ### 4.8.6. Citrate Synthase Immunofluorescence {#sec4dot8dot6-ijms-20-06017} To detect the mitochondria along Nano tunneling structures, astrocytes (UP-010) and CFSE-positive glioblastoma (UP-007) cells were co-cultured at ratio 50:50 in 24-well culture plates for 48 h. Cells were labelled with 1 mg/mL Alexa Fluor^®^ 594-conjugated wheat germ agglutinin (WGA) (Invitrogen, Fisher Scientific, Loughborough, UK)) at 5 μg/mL for 15 min at 37 °C in the dark and permeabilized with 0.2% (w/v) Triton X-100. The cultures were then incubated with primary antibodies against Citrate Synthase (GeneTex, rabbit polyclonal, 1:2000) overnight at 4 °C, and a secondary antibody conjugated with Alexa Fluo-488 fluorescent dyes (1:1000, goat anti-rabbit IgG, Jackson Immuno-research) was applied for 1 h at 37 °C. All imaging was performed using a Zeiss LSM719 confocal laser-scanning microscope (Carl Zeiss, Oberkochen, Germany). Post-acquisition image analysis was performed with ZEN 2012 ×32 blue software (Carl Zeiss). Images were taken with an oil immersed 40×/1.4NA objective (Carl Zeiss). ### 4.8.7. Mitochondrial and Nano Tunneling Detection in 3D Co-Culture Model {#sec4dot8dot7-ijms-20-06017} To visualize mitochondria and Nano tunneling within 3D in vitro co-culture model cells were grown co-culture, in a 3D HyStem™-HP hyaluronic acid-based hydrogel (Sigma-Aldrich, Dorset, UK). HyStem-HP hydrogel is formed when the crosslinking agent Extralink™ (polyethylene glycol diacrylate Mw 3400 g/mol) is added to a mixture of HyStem-HP (thiol-modified hyaluronan) and Gelin-S™ (thiol-modified gelatin). Hydrogel was prepared according to the manufacturers' instructions. GBM (UP-007) and astrocytes were grown in a co-culture at the ratio of 50:50 in the hyaluronic acid-based hydrogel. Briefly, UP-007 and UP-010 cells were labelled with Cell Trace CFSE and Mito Tracker, respectively, and encapsulated within the liquid hydrogel precursor solution at 1 × 10^6^ cells mL^−1^. Hydrogels were allowed to polymerize for 20 min, and then complete DMEM media was added to each well. Cells in 3D were allowed to grow for 3 days. Before collection of pictures, the 3D model was stained with WGA AF594, following the protocol. All imaging was performed using a Zeiss LSM719 confocal laser-scanning microscope (Carl Zeiss, Oberkochen, Germany). Post-acquisition image analysis was performed with ZEN 2012 ×32 blue software (Carl Zeiss). 4.9. Statistical Analysis {#sec4dot9-ijms-20-06017} ------------------------- The results are expressed as mean ± standard error of the mean (SEM). One-Way ANOVA obtained comparisons between groups using the Dunnett's post-hoc test in comparison to a control (GraphPad Prism 7.03, San Diego, USA). A significance level of *p* \< 0.05 was considered statistically significant. In each experiment, *n* = 3 replicates were carried out for each condition and the data displayed represents the mean of three independent experiments. 5. Conclusions {#sec5-ijms-20-06017} ============== In this paper, we highlighted the importance of having a human astrocyte component as well as the tumour cells grown in human serum-supplemented media to develop a highly sensitive and reproducible method to test relevant clinical drugs in this novel high-throughput, 3D in vitro, all human model. We also proposed the feasibility of using fluorescence amine staining CFSE and Far Red to distinguish two different cells in co-culture avoiding the use of genetic manipulation and clonal selection. This procedure allowed us to monitor cells using a sophisticated approach i.e., flow cytometry and confocal microscopy for pharmacological screening in a high throughput, 3D all human in vitro model and allowed us to explain the possible mechanism behind GBM drug resistance. Moreover, we showed a mechanism for drug resistance by way of TNTs delivering non neoplastic cell mitochondria. In particular, in the presence of non-neoplastic astrocytes within a hyaluronic acid-rich human serum-supplemented 3D model, GBM cells show an increase in proliferation when exposed to a therapeutically relevant dose of both TMZ or VCR while when treated with CLM there was a slight reduction in proliferation. Supplementary Materials can be found at <https://www.mdpi.com/1422-0067/20/23/6017/s1>. ###### Click here for additional data file. P.C.; D.M.L. and G.J.P. designed the research; P.C. and D.M.L. performed the research and data analyses; P.C. and G.J.P. wrote the paper. The study was kindly funded by the Jake McCarthy Foundation. The Brain Tumour Research Centre at the University of Portsmouth is also core-funded by the charity, Brain Tumour Research and receives funding for its 3D human in vitro modelling programme from Animal Free Research UK. The authors declare no conflicts of interest. GBM Glioblastoma FBS Fetal bovine serum eGFP enhanced green fluorescent protein 2D Two dimensional 3D Three dimensional WGA Wheat germ agglutinin VCR Vincristine CLM Clomipramine TMZ Temozolomide TNT Nano tunneling TW Transwell CFSE Carboxyfluorescein succinimidyl ester ECM Extracellular matrix component TME Tumour microenvironment {#ijms-20-06017-f001} {#ijms-20-06017-f002} {#ijms-20-06017-f003} {#ijms-20-06017-f004} {#ijms-20-06017-f005} {#ijms-20-06017-f006} {#ijms-20-06017-f007} {#ijms-20-06017-f008} {#ijms-20-06017-f009} {#ijms-20-06017-f010} {#ijms-20-06017-f011}

Fadoo Z, Merchant Q, Rehman KA. *Journal of Blood Medicine*. 2013;4:65--73. On page 67 the legend of [Figure 1](#f1-jbm-4-087){ref-type="fig"} incorrectly states "Reprinted from Song JW, Choi JR, Song KS, Rhee JH. Plasma factor XIII activity in patients with disseminated intravascular coagulation. *Yonsei Med J*. 2006;47(2):196--200." The correct statement is "Reprinted from Muszbek L, Bagoly Z, Cairo A, Peyvandi F. Novel aspects of factor XIII deficiency. *Curr Opin Hematol*. 2011;18(5):366--372." On page 69 the legend of [Table 1](#t1-jbm-4-087){ref-type="table"} states "Reprinted from Song JW, Choi JR, Song KS, Rhee JH. Plasma factor XIII activity in patients with disseminated intravascular coagulation. *Yonsei Med J*. 2006;47(2):196--200." The correct statement is "Reprinted from Kohler HP, Ichinose A, Seitz R, Ariens RA, Muszbek L; Factor XIII and fibrinogen SSC subcommittee of the ISTH. Diagnosis and classification of factor XIII deficiencies. *J Thromb Haemost*. 2011;9(7):1404--1406." {#f1-jbm-4-087} ###### Laboratory Diagnosis/Classification of factor XIII deficiency Deficiency Plasma FXIII activity Plasma FXIII-A~2~B~2~ antigen Plasma FXIII-A antigen Plasma FXIII-B antigen Platelet FXIII activity Platelet FXIII-A antigen ----------------------------- ----------------------- ------------------------------- ------------------------ ------------------------ ------------------------- -------------------------- Inherited FXIII-A deficiency  Type I ↓↓↓ ↓↓↓ ↓↓↓ \>30% ↓↓↓ ↓↓↓  Type II ↓↓↓ ↓-N ↓-N \>30% ↓↓↓ ↓-N FXIII-B deficiency ↓↓ ↓↓↓ ↓↓ ↓↓↓ N N Autoantibody against FXIII Anti-FXIII-A  Neutralizing ↓↓↓ ↓-N ↓-N \>30% N N  Non-neutralizing ↓↓↓ ↓↓↓ ↓↓↓ \>30% N N Anti-FXIII-B ↓↓↓ ↓↓↓ ↓↓↓ ↓↓↓ N N Other acquired deficiencies ↓ ↓ ↓ ↓-N NA NA **Notes:** ↓↓↓, highly decreased activity/concentration usually below 3%; ↓↓, considerably decreased activity/concentration, usually 5%--10%; ↓, slightly decreased activity, usually 20%--70%.

Sir, Nalidixic acid as the first quinolone was isolated as a by-product of the synthesis of chloroquine. It has been available for the treatment of urinary tract infections for many years.\[[@ref1]\] We present here the report of a 13-year-old girl who was referred to the toxicologic emergency department of Baharloo Hospital by a general physician from his office because of seizure and decrease in consciousness. Her seizure was tonic clonic and did not repeat more than one time during hospitalization. There was no history of previous disease. She was diagnosed as diabetic ketoacidosis (DKA) based on her laboratory test results on admission to the hospital, and thus the treatment was started. She was completely cured without any signs and symptoms of diabetes and DKA, 24 h after admission. Her drug history showed nalidixic acid overdosage. Her first complaints were agitation, jerking, and spasm in her upper limbs, along with low consciousness that was diagnosed as seizure by a general physician and cured by administration of diazepam in his office (her seizure might be due to nalidixic acid, but the other etiologic factors such as severe acidosis and electrolyte disturbances might not be ruled out). Her initial vital signs and lab tests on hospital admission are shown in [Table 1](#T1){ref-type="table"}. Neurological examination was intact except mild agitation and confusion. Fever was not detected. In urine analysis, ketones were large and glucose was 1+. Urine culture was sterile. Arterial blood gas (ABG) was reported as severe acidosis. Chest X-ray was normal. By taking history and performing lab tests, the other etiologies of metabolic acidosis, like renal failure, alcohol drinking, starvation and consumption of other toxins, were ruled out. ###### Initial vital signs and lab tests on hospital admission  She was diagnosed as DKA and treatment was given with insulin (10 U regular status + 5 U/h), normal saline (2000 mL for 3 h), and potassium (10 mL KCl 15% in 1 L of serum). After continuation of treatment, her blood sugar finally reached 111 mg/dL and ABG became normal \[[Table 2](#T2){ref-type="table"}\]. She had a family history of diabetes mellitus as her brother was a diabetic, but there was no history of atopic or epilepsy. Ten empty packets of nalidixic acid (ingested with a suicidal intention) were found near her bed, which was confirmed by history taking after the patient showed clinical improvement. Twenty-four hours after treatment, the patient was completely cured without any symptoms and signs of diabetes and DKA. ###### Arterial blood gas and arterial blood glucose data and treatment  There are reports of glycosuria\[[@ref2]--[@ref4]\] and hyperglycemia, convulsions, in overdosage of nalidixic acidor even therapeutic dose of nalidixic acid.\[[@ref5]\] There are also reports of metabolic acidosis in overdosage\[[@ref6]--[@ref9]\] or therapeutic dose of nalidixic acid.\[[@ref1][@ref10]\] To our knowledge, till date there has been no report of DKA following overdosage of nalidixic acid. DKA may be the initial symptom complex that leads to a diagnosis of type 1 DM, but more frequently it occurs in individuals with established diabetes. Nausea and vomiting are often prominent, and their presence in an individual with diabetes warrants laboratory evaluation for DKA\[[@ref11]\] Drugs can also be a precipitating factor. There are reports of hyperglycemia, convulsions,\[[@ref1]\] and glycosuria in overdosage of nalidixic acid.\[[@ref2]--[@ref4]\] As far as our patient is concerned, she had a positive family history of diabetes mellitus and this presentation may be due to the window period of diabetes mellitus. So, there is a need to follow-up the patient as she could develop overt diabetes mellitus. However, more precautions need to be taken while prescribing nalidixic acid for high-risk patients for diabetes mellitus. As this communication is the first report of its kind, it should encourage aggressive pharmacovigilance in future.

Epilepsy -- a disorder of the brain and the mind ================================================ Epilepsy is a neurologic disorder that is characterized by having two or more unprovoked recurrent seizures due to abnormal brain activity.[@b1-ndt-13-585]--[@b3-ndt-13-585] This complex disease consists of many syndromes, diagnostic criteria, and treatment strategies.[@b4-ndt-13-585]--[@b6-ndt-13-585] Sixty million people are affected by epilepsy worldwide.[@b7-ndt-13-585] In the US, epilepsy affects \~2.3 million and costs the health care system ≥15.5 billion either directly, through medically related costs, or indirectly, through loss of productivity.[@b8-ndt-13-585] Also, \~30% of epileptic patients have temporal lobe epilepsy (TLE) which is characterized by complex partial seizures, limbic degeneration specifically in the hippocampus, and comorbid psychiatric disorders.[@b9-ndt-13-585]--[@b11-ndt-13-585] There are two main classifications of seizures as stated by the International League Against Epilepsy, namely, generalized seizures and focal seizures. Generalized seizures are typically bilateral, rapidly occurring, and originate at a single point. Subtypes of generalized seizures include: tonic--clonic, absence, myoclonic, clonic, tonic, and atonic. Focal seizures originate in one hemisphere and can be either localized in that one area or distributed to other areas.[@b12-ndt-13-585] There are three main classifications of epilepsy diagnosis: genetic, structural/metabolic, and unknown cause. Genetic epilepsy is caused by a direct genetic defect that causes seizures as one of the primary symptoms. Structural/metabolic epilepsy is caused by a distinct structural lesion or a metabolic disease that is the root cause of seizures. Examples of structural/metabolic type include traumas such as strokes or infections that lead to seizures. Unknown cause epilepsy is not meant to be similar to idiopathic. Instead, it means that there could be a genetic component or it may be due to another disease, but the underlying cause of the seizures is unknown at the current time.[@b12-ndt-13-585] Epilepsy also presents with cognitive and behavioral deficits that are typically seen in patients with classic neuropsychiatric disorders. These deficits include: hallucination, change in affect, delusion, apathy, changes in cognition, and delirium, to name a few.[@b13-ndt-13-585],[@b14-ndt-13-585] People with epilepsy also share comorbidities with neuropsychiatric disorders such as anxiety, depression, and obsessive compulsive disorder.[@b15-ndt-13-585],[@b16-ndt-13-585] Epilepsy has been known for millennia. From the time before Hippocrates, epilepsy and its symptoms have been attributed to extra-worldly causes, as seen in its depiction in famous paintings and novels.[@b17-ndt-13-585],[@b18-ndt-13-585] The current treatment options for epilepsy are pharmacologic drugs (ie, antiepileptic drugs \[AEDs\]) and surgery. AEDs are the primary method of treatment, and function by reducing the overall firing of the brain by inhibiting sodium and other positive ion channels or by activating gamma-Aminobutyric acid (GABA)ergic channels.[@b19-ndt-13-585] Surgical resection of the temporal lobe is an option when AED treatment fails, and involves partial removal of the seizure-generating brain tissue in order to manage symptoms.[@b20-ndt-13-585] Most treatments focus on controlling the neurological aspect of epilepsy, but fail to address the neuropsychiatric component. AEDs do help with the neuropsychiatric components indirectly by attenuating neural firing and essentially shutting off pathways that are present.[@b14-ndt-13-585] Deep brain stimulation is another treatment option that indirectly treats the neuropsychiatric component of epilepsy by altering the firing of the limbic areas.[@b21-ndt-13-585] Additional research is necessary to find a treatment option that ameliorates both the neurological and neuropsychiatric components of epilepsy. Stem cell therapy has emerged as an experimental treatment for a number of neurological and neuropsychiatric disorders. For epilepsy, the goal of stem cell therapy is to enhance stem cell differentiation and growth toward the GABAergic lineage in order to rewire the GABA-deficient neural circuitry seen in epilepsy.[@b9-ndt-13-585] This therapy has been reported for treating other neurological disorders, such as Parkinson's disease and multiple sclerosis, and neuropsychiatric diseases, such as depression and schizophrenia.[@b22-ndt-13-585] These encouraging results may be extrapolated to include treatment of epilepsy which has both neurological and neuropsychiatric components.[@b23-ndt-13-585],[@b24-ndt-13-585] In this article, we present in detail the current and emerging treatments of epilepsy, noting the potential of stem cell transplantation, as well as highlighting the gap in knowledge relating to the neglected neuropsychiatric aspect of the disease that will likely direct the treatment indications for epilepsy in future. Epilepsy: mind over matter ========================== Defining epilepsy as a neurological disorder allows one to understand that it has a physiological basis in the human brain, characterized by a very real and measureable hyperactive neurological etiology. Because of the ethical dilemmas associated with studying brain tissue in human epileptics, many preclinical models of epilepsy have contributed to our understanding of epilepsy by allowing us to study the corresponding changes in neurological tissue and other variations underlying the mechanism of epilepsy.[@b25-ndt-13-585] There are a variety of models for studying epilepsy, and the specific model used depends on the question being addressed by the researcher. Because of the limitations in studying epilepsy in human subjects, good animal models are necessary for uncovering the underlying pathology of the disease.[@b25-ndt-13-585] Well-defined preclinical chemical, electrical stimulation, genetic, developmental, and trauma models for epilepsy research have been documented.[@b25-ndt-13-585] Epilepsy as a neuropsychiatric disorder ======================================= Because of its neurological basis and associated psychopathological, cognitive, and linguistic problems, one can just as easily characterize epilepsy as a neuropsychiatric disorder.[@b4-ndt-13-585] What is the use in defining epilepsy in this way? Enumerating the vast array of historical, cognitive, behavioral, psychological, and social issues surrounding epilepsy will allow us a more accurate understanding of these individuals, and can thus potentiate our therapeutic strategies for epilepsy by identifying unmet clinical needs. Widening our scope beyond the biological basis of disease allows us to take into account the obstacles faced by epileptics, which gives an opportunity to improve treatment and provide better quality of life. Studying the behavioral and cognitive abnormalities that parallel the underlying pathology gives us greater insight into the inner workings of the human mind. Epilepsy in history and literature ================================== Epilepsy was described as early as 2000 BCE in Babylonian texts.[@b18-ndt-13-585],[@b26-ndt-13-585],[@b27-ndt-13-585] The word "epilepsy" is derived from the Greek word "epilambanein", which is loosely translated as "to seize or take hold of".[@b17-ndt-13-585],[@b18-ndt-13-585],[@b26-ndt-13-585]--[@b29-ndt-13-585] Popular belief at this time held that seizures were of supernatural origin, induced by the Gods.[@b17-ndt-13-585],[@b18-ndt-13-585] It is interesting to note, however, that as early as \~400 BCE, Hippocrates had branded epilepsy as a disease of the body and had denied that seizures were divinely provoked.[@b17-ndt-13-585],[@b18-ndt-13-585],[@b26-ndt-13-585],[@b27-ndt-13-585] Galen of Pergamon further characterized epilepsy (\~130--200 AD) and went so far as to say that seizures were a phenomenon of overactivation or irritation of the brain.[@b17-ndt-13-585],[@b18-ndt-13-585],[@b27-ndt-13-585] Despite this portrayal of epilepsy as a quantifiable, observable defect in the bodies of those afflicted, the view that epilepsy has supernatural origin has persisted and is even depicted in films today.[@b30-ndt-13-585] These beliefs have led to stigmatization and fear of epileptics, and a social environment in which epileptics feel outcast.[@b26-ndt-13-585] Historically, there have been many implications and allusions toward epilepsy. Prominent historical figures such as Joan of Arc may have had seizures.[@b18-ndt-13-585],[@b31-ndt-13-585] According to scholars, Joan of Arc's supposed epileptiform seizures gave her divine purpose and the strength to lead France in an effort to free them from English dominion.[@b31-ndt-13-585] Though her actual participation in battle is questionable, her role in recovering France from the English dominion in the Lancastrian phase of the 100-year war is undeniable, as many of the noblemen leading troops apparently listened to her advice, believing it to be of divine origin.[@b32-ndt-13-585] In "La Divina Comedia", Dante Alighieri quite accurately describes the paroxysmal change in behavior typical of an epileptic, and the postictal confusion that often immediately follows a seizure, in his depiction of a thief's punishment in the seventh circle of hell.[@b18-ndt-13-585],[@b33-ndt-13-585] Though it is unclear if Dante is aware of epilepsy as a medical disorder, the thief Vanni Fucci's seizure-like episode is precipitated by the bite of a serpent, perpetuating the supernatural explanation of epilepsy.[@b33-ndt-13-585] Epilepsy and criminality ------------------------ The stigmatization of epilepsy through religious texts and literature may have pervaded the early medical literature. Some people like Cesare Lombroso thought that epilepsy was clearly linked with criminality, in that people with epilepsy were much more likely to participate in criminal acts when compared to people without epilepsy.[@b34-ndt-13-585] As late as the 1980s, the feasibility of an epileptic performing a complex aggressive act during an ictal state was studied, as the "epilepsy defense" was being used as a plea to insanity for some people who had committed violent crimes.[@b35-ndt-13-585] More recent studies have found no such correlation between epilepsy and criminality. Though a link between aggression and the postictal state was shown in a preclinical murine model, there is no evidence to show that epileptics are prone to criminal behavior when compared to individuals without epilepsy.[@b14-ndt-13-585],[@b17-ndt-13-585] Poor understanding of a disorder and stigmatization can lead to bias, which may explain why Lombroso was convinced of the link between epilepsy and criminality. Epilepsy and psychology ----------------------- Though overarching links between a disorder like epilepsy and a social construct like criminality may not be present, there is a high comorbidity between epilepsy and psychiatric, behavioral disorders and impairments in social cognition.[@b36-ndt-13-585]--[@b39-ndt-13-585] Studies that look at the link between epilepsy and psychiatric disorders have shown a correlation between peri-ictal and interictal psychiatric symptoms, stating that people who display psychiatric disturbances due to the seizure can develop or display the same symptoms between seizures.[@b17-ndt-13-585] Mood and memory are two common deficits in epileptics, with many of these individuals reporting depressed mood and diminished memory capacity.[@b14-ndt-13-585],[@b17-ndt-13-585],[@b18-ndt-13-585] Psychological disorders such as obsessive compulsive disorder, depression, and bipolar disorder have also been correlated with epilepsy.[@b14-ndt-13-585] Sleep disturbances are also common, both due to nocturnal seizures and AEDs, and are the target of some holistic approaches to treatment of epilepsy.4,14,17,18,26--28,33,40 Since quality sleep has profound implications on our memory, mood, and cognitive functioning, one can conclude that impaired sleep due to epilepsy would negatively affect memory and cognitive functioning.[@b40-ndt-13-585] Furthermore, sleep deprivation has been known to selectively impair the consolidation of positive and neutral memories, while not significantly affecting consolidation of negative memories; this could explain the negative mood experienced by many with epilepsy.[@b40-ndt-13-585] The developmental implications of epilepsy on the brain are well documented.[@b17-ndt-13-585],[@b41-ndt-13-585]--[@b43-ndt-13-585] Epilepsy can interfere with the normal development of neural networks in children, and thus, early diagnosis and treatment is suspected to have an impact on limiting comorbid psychological issues associated with neural network developmental impairment due to epilepsy.[@b17-ndt-13-585],[@b41-ndt-13-585]--[@b43-ndt-13-585] These developmental impairments may lead to the array of cognitive impairments seen in epileptics and represent an unmet clinical need in epilepsy patients.[@b4-ndt-13-585],[@b14-ndt-13-585] The effect of epilepsy on the aging mind is not well studied.[@b17-ndt-13-585] Epilepsy and art ---------------- Studying art created by individuals with epilepsy gives us insight into the peri-ictal and interictal experiences and the psychological consequences of living with epilepsy.[@b14-ndt-13-585],[@b17-ndt-13-585],[@b18-ndt-13-585],[@b44-ndt-13-585] Epileptics often describe feelings of loneliness and isolation from the mainstream society, and gaps in time due to their seizures.[@b44-ndt-13-585] This knowledge allows us to better treat individuals with epilepsy. Though there is no correlation between epilepsy and enhanced creativity in music and poetry, the impaired creativity due to epilepsy or epilepsy treatment provides an opportunity to better understand human creativity as a function of intact and properly functioning cerebral circuitry.[@b28-ndt-13-585],[@b29-ndt-13-585],[@b45-ndt-13-585] Impairment of the creative process in epileptics is thought to be a result of the impaired neural circuitry.[@b28-ndt-13-585],[@b45-ndt-13-585] Epilepsy and neuropsychology ---------------------------- Examining epilepsy under the lens of neuropsychology allows us to better understand the disease. This understanding allows us to tailor treatment to epileptic individuals, allowing them to strive for equanimity, blend in a milieu in which epilepsy is stigmatized and feared for its supposed supernatural origin, and thrive despite impairment. Studying the historical, cognitive, behavioral, psychological, and social issues surrounding epilepsy allows us to improve treatment for epileptics and provides profound insight into the human mind. Epilepsy has been found to be comorbid with many different psychiatric disorders.[@b29-ndt-13-585],[@b46-ndt-13-585],[@b47-ndt-13-585] It has also been demonstrated that patients with depression are more likely to develop epileptic seizures compared to the general population, indicating a bidirectional relationship between epilepsy and psychiatric disorders.[@b48-ndt-13-585],[@b49-ndt-13-585] Patients with epilepsy are more likely to have mood disorders, like depression, with prevalence rates from 25% to 30%, followed by anxiety, attention disorders, psychotic, and personality disorders.[@b46-ndt-13-585],[@b50-ndt-13-585]--[@b53-ndt-13-585] Patients also have a higher suicide ideation of 30% and a rate of 11.5%, which is ten times more than that of the general population (1.5%).[@b52-ndt-13-585],[@b53-ndt-13-585] Patients with epilepsy who also have psychiatric disorders exhibit different clinical characteristics and are classified as "atypical". The classification of the psychiatric episodes is determined by their onset of seizures. Symptoms are put into four distinct categories: preictal (up to 2 days before the seizure), ictal (during the seizure), interictal (independent of the seizure), and postictal (following the seizure within 5 days).[@b46-ndt-13-585] These psychiatric episodes vary from patient to patient, and the severity depends on a multitude of factors.[@b48-ndt-13-585] Due to the comorbidity of psychiatric disorders and epilepsy, it is important for clinicians to notice the signs of mood disorders in order to treat epileptic patients. Some of the key things to look for during a psychiatric exam are: anhedonia, irritability, low self-esteem, decreased concentration, suicidal ideation, lethargy, and fatigue.[@b54-ndt-13-585] Anxiety disorders are the second most prevalent form of psychiatric disorder in epileptic patients. Many patients experience anxiety as part of the seizure or aura.[@b55-ndt-13-585] Postictal anxiety episodes appear in 45% of patients with epilepsy and are typically associated with other mood-related disorders.[@b56-ndt-13-585] Attention deficit hyperactivity disorder is found in 20%--30% of pediatric epileptic patients. This disorder tends to persist through childhood into adulthood. Pediatric epileptic patients typically present with the classic signs of attention deficit disorders: impulsivity, irritability, and poor tolerance to challenging scenarios.[@b46-ndt-13-585],[@b56-ndt-13-585] Patients with epilepsy can also exhibit psychotic disorders with variable symptoms. Epileptic patients present less severe signs of psychosis compared to nonepileptic patients, thus making diagnosis of a psychotic disorder accompanying epilepsy more difficult.[@b57-ndt-13-585] Incorrect diagnosis and treatment of the psychiatric components of epilepsy impact not only the patient, but also the entire health care system. Delayed diagnosis leads to an increased burden on the individual patient, their family, and society.[@b58-ndt-13-585],[@b59-ndt-13-585] Research has demonstrated that psychiatric comorbidity has a negative impact on the efficacy of surgery and the pharmaceutical treatment for epilepsy.[@b60-ndt-13-585],[@b61-ndt-13-585] Animal models and the biology of epilepsy ----------------------------------------- The biological bases of epilepsy and psychiatric disorders need to be studied further. However, there are a few key similarities between the two types of disorders that highlight their connection. In animal epilepsy models, serotonin and nor-epinephrine have critical roles in the pathophysiology. The lack of these neurotransmitters has been shown to increase the severity of seizures.[@b62-ndt-13-585]--[@b64-ndt-13-585] Conversely, an increase in these neurotransmitters, facilitated through reuptake inhibitors, has been correlated with a decrease in severity of seizures.[@b62-ndt-13-585]--[@b64-ndt-13-585] Changes in neurotransmitter levels are thought to be similarly important in psychiatric disorders. Researchers have used positron emission tomography scans to examine serotonin receptor (5-HT1A) activity. Both epilepsy patients and mood disorder patients exhibited a decrease in 5-HT1A binding activity in the temporal lobes near the raphe nuclei.[@b65-ndt-13-585]--[@b68-ndt-13-585] Additional research is still needed to understand the complex mechanisms behind epilepsy and psychiatric disorders. Treatments: the old and the new =============================== Pharmacologic treatments of epilepsy ------------------------------------ Current treatments of epileptic seizures are primarily controlled by readily available AEDs.[@b69-ndt-13-585] One or multiple AEDs are generally used to treat symptoms in the majority of the epilepsy patient population.[@b70-ndt-13-585] AEDs use various mechanisms of action to treat the hallmark symptoms of epileptic seizures. The ultimate goal of drug treatment is to decrease the electrical activity of the brain. This inhibition of neuronal firing is accomplished by blocking sodium channels, calcium channels, and glutamate-mediated responses.[@b19-ndt-13-585] Other AEDs provide neuronal inhibition by promotion of GABAergic release or increased potassium channel conductivity.[@b19-ndt-13-585] Unfortunately, a noteworthy proportion (nearly 30%) of the epileptic patient population exhibits or develops pharmacoresistance to AEDs.[@b69-ndt-13-585] Pharmacoresistance of current AEDs is thought to be multifactorial and due to a variety of genetic and nongenetic factors.[@b71-ndt-13-585] Although the exact mechanisms of pharmacoresistance in AEDs is not fully understood, the *ABCB1* gene is thought to play a role.[@b71-ndt-13-585] P-glycoprotein, a key transporter involved in drug efflux, is thought to be encoded by the *ABCB1* gene.[@b71-ndt-13-585] It is suggested that variability within the *ABCB1* gene may play a role in preventing AEDs from reaching their target, thus resulting in pharmacoresistance to AEDs in patients with polymorphisms of this gene.[@b71-ndt-13-585] Also, recent studies have found that certain variants of apolipoprotein E are linked with susceptibility toward certain types of epilepsy and also play a role in the pharmacoresistance against AEDs.[@b72-ndt-13-585] Although the development of new AEDs has been proven to cause fewer drug interactions and side effects, they still do not reduce the frequency of drug-resistant epilepsy.[@b70-ndt-13-585] When polytherapy using multiple AEDs fails to reduce epileptic symptoms, alternative and surgical treatment options should be discussed following the diagnosis of drug-resistant epilepsy.[@b73-ndt-13-585] Surgical and alternative treatments of epilepsy ----------------------------------------------- Alternative treatment options for epilepsy include surgical resection of the epileptic brain area, ketogenic diet, and neurostimulation devices such as deep brain stimulators and vagus nerve stimulators.[@b73-ndt-13-585] It is becoming common practice to provide neurosurgical consultation to drug-resistant epilepsy patients after failure of \~2 AED regimens.[@b20-ndt-13-585] Earlier surgical consultation results in overall increased quality of life and decreased medical cost for patients.[@b20-ndt-13-585] The temporal lobe is the most common area affected by drug-resistant epilepsy and resection of this brain area has been proven to be an effective treatment method.[@b20-ndt-13-585] Effective temporal lobe resection results in seizure control with minimal neurological deficits by removing the mesial temporal structures while maximizing conservation of the neocortex.[@b20-ndt-13-585] Deep brain stimulation has proven effective in significantly reducing the seizure frequency as well as intractable aggressive behavior that is sometimes associated with drug-resistant epilepsy.[@b74-ndt-13-585] Other neuromodulation-based treatments of drug-resistant epilepsy include vagus nerve stimulation that aims to reduce seizure frequency.[@b75-ndt-13-585] Unfortunately, vagus nerve stimulation generally does not provide complete suppression of seizures, but is still effective in reducing the seizure frequency and increasing the response to initial seizure activity.[@b75-ndt-13-585] Other alternative treatment options for drug-resistant epilepsy include a ketogenic diet. Although the mechanism of action is not well understood, some studies demonstrate a reduction in seizures in patients who adhere to these dietary restrictions.[@b76-ndt-13-585] Treatments targeting neuropsychiatric symptoms of epilepsy ---------------------------------------------------------- Patients with epilepsy are particularly vulnerable to neuropsychiatric symptoms including anxiety and depression.[@b77-ndt-13-585] Although seizure is the most recognized comorbidity of epilepsy, anxiety and depression are rarely addressed in the differential diagnosis despite being the most prevalent psychiatric comorbidities of epilepsy.[@b78-ndt-13-585] Research on anxiety and depression in epilepsy patients lacks exact clarity and longitudinal perspective; however, these neuropsychiatric symptoms are thought to be derived from either psychosocial or neurological causes.[@b79-ndt-13-585] Although disorders such as anxiety and depression are often neglected in epilepsy patients, studies have demonstrated the effectiveness of specific, single-item screening methods for these disorders.[@b80-ndt-13-585] There are many traditional and nontraditional treatment options for these neuropsychiatric symptoms. Traditional treatments of these symptoms often include antidepressants, specifically selective serotonin reuptake inhibitors and serotonin--noradrenaline reuptake inhibitors.[@b79-ndt-13-585] Studies have found that antidepressants reduce the frequency of seizure episodes; however, the association of antidepressants with the underlying disease itself is yet to be investigated.[@b79-ndt-13-585] Conducting research on this association between antidepressants and epilepsy would be difficult in human clinical trials due to numerous factors such as prolonged follow-up and multifactorial assessment of outcomes.[@b79-ndt-13-585] The use of nontraditional treatments for neuropsychiatric symptoms of epilepsy is steadily increasing. A randomized controlled trial found benefits of mindfulness-based therapy for patients with drug-resistant epilepsy.[@b81-ndt-13-585] Patients experienced better outcomes in quality of life, mood, and seizure frequency, when compared to social support alone.[@b81-ndt-13-585] Another nontraditional treatment is cannabidiol-based treatments for drug-resistant epilepsy. There is preclinical evidence suggesting that cannabidiol has antiseizure effects and could potentially help mediate neuropsychiatric symptoms of epilepsy.[@b82-ndt-13-585] Despite the legal challenges in this area of research, there seems to be growing support in advancing cannabidiol-based treatments.[@b82-ndt-13-585] There is a lack of research on the association between epileptic drugs and the neuropsychiatric symptoms of the disease. It is difficult to assess the outcomes in animal models because both chronic epilepsy models and chronic antidepressant treatment are required to simulate the disease and treatment course in humans.[@b79-ndt-13-585] Recent studies have shown that the pilocarpine model is an appropriate method to recreate the neurobehavioral disruptions seen in TLE in humans.[@b83-ndt-13-585] Further investigations using the pilocarpine model with chronic antidepressant treatment may be able to replicate the disease course accurately in humans. Stem cell therapy and epilepsy ============================== Preclinical and clinical studies -------------------------------- Therapeutic use of stem cells has been reported in multiple neurological disease models such as multiple sclerosis, stroke, and Parkinson's disease.[@b22-ndt-13-585] Stem cell transplantation therapy has been increasingly explored as a potential treatment option in epilepsy.[@b9-ndt-13-585] Preclinical and clinical studies of this approach have utilized several cell types including hippocampal precursor cells, neural stem cells (NSCs), GABAergic precursor cells, and systemic administration of bone marrow-derived mononuclear cells and mesenchymal cells.[@b9-ndt-13-585] When NSCs were first injected into a pilocarpine rat model, it resulted in a significant decrease in spontaneous motor seizures.[@b84-ndt-13-585] Although each cell type utilizes a different mechanism of action, they all strive to reduce seizure activity in the epileptic area of brain tissue. Bilateral hippocampal precursor cell grafting has been shown to effectively decrease the frequency of seizures in animal models.[@b85-ndt-13-585] This therapy strives to repair the disrupted circuitry within the epileptic area of the brain by reducing aberrant mossy fiber sprouting, while also activating the GABAergic interneurons.[@b9-ndt-13-585] The increase of GABAergic interneurons translates into more inhibitory control, thereby providing the desired effects in the epileptic brain areas.[@b85-ndt-13-585] NSC transplantation is thought to utilize a multifaceted approach by creating new GABAergic interneurons and also new astrocytes.[@b9-ndt-13-585] The underlying significance of new astrocytes is the expression of glial-derived neurotrophic factor (GDNF) that has anticonvulsant properties.[@b86-ndt-13-585] There are evident therapeutic effects of NSC transplantation, including reduction of seizure frequency and duration, despite no changes occurring in cognitive function.[@b86-ndt-13-585] The loss of GABAergic neurons produces a lack of inhibitory control on the epileptic brain area.[@b9-ndt-13-585] Thus, GABAergic cell therapy focuses on replacing the loss of GABAergic neurons, thus increasing the inhibitory synaptic control of the affected area.[@b9-ndt-13-585] Studies have shown the anticonvulsant effects of directly grafting engineered GABA-producing cells.[@b9-ndt-13-585] Unfortunately, many of these beneficial anticonvulsant effects have been shown to be temporary due to poor graft survival.[@b9-ndt-13-585] Other studies have explored these effects when using the precursor cells derived from different areas such as the lateral ganglionic eminence and the medial ganglionic eminence (MGE).[@b9-ndt-13-585] Embryonic progenitor cells from the lateral ganglionic eminence treated with fibroblast growth factor-2 and a caspase inhibitor have been found to provide lasting inhibition and reduction in the frequency of spontaneous recurrent seizures (SRS).[@b9-ndt-13-585] In addition, MGE progenitor cell grafting has been explored due to the cells' characteristic ability to migrate to surrounding regions from the initial graft site.[@b9-ndt-13-585] It is suggested that there is a connection between the pyramidal neurons and the newly formed interneurons which produces the inhibitory effect.[@b87-ndt-13-585] Further investigation in clinical GABAergic cell therapy must explore the long-term effects of grafting, results in drug-resistant epilepsy, and usefulness in reducing the cognitive and mood impairments associated with epilepsy.[@b9-ndt-13-585] Autologous stem cell transplantation for epilepsy ------------------------------------------------- The debate continues on what type of transplantation, autologous or allogeneic, to use when treating diseases. These stem cells can be induced to differentiate into a specific type of specialized cells before the actual transplantation procedure.[@b88-ndt-13-585],[@b89-ndt-13-585] Both autologous and allogeneic transplantation have strengths and weaknesses related to efficacy and safety. In allogeneic transplantation, the donor is not the host, but a very close human leukocyte antigen (HLA) match, like the host's immediate family member.[@b90-ndt-13-585] The necessity for a close HLA match creates problems with availability and compliance.[@b90-ndt-13-585] In autologous transplantation, the donor and the host are the same individual, thus ensuring an exact HLA match. These autologous cells can be differentiated into many cell types.[@b89-ndt-13-585] The primary weaknesses of an autologous transplantation are the limited quantity of stem cells that can be extracted and the significant length of time required for proliferation.[@b91-ndt-13-585] Autologous transplantation offers some advantages in regard to patient ethics, prognosis, and safety. Since autologous transplants originate from the same individual as the recipient, there is no need to use embryonic or fetal tissues, thereby bypassing some of the more controversial ethical concerns of stem cell transplantation. Autologous transplantation has a reduced risk of tumorigenesis and graft rejection, compared to allogeneic transplantation.[@b92-ndt-13-585]--[@b95-ndt-13-585] As a result, most allogeneic transplantation requires the administration of immunosuppressants to attenuate the immune response of the graft and of the host, as seen in graft-versus-host disease (GVHD). Autologous transplantation does not have the risk for GVHD since the donor and host are the same. The prevalence of GVHD is increasing due to the rise in the number of transplantations.[@b96-ndt-13-585] GVHD has three requirements: the graft should have immunologically competent cells, the host and donor antigens must be different, and the host must be incapable of attacking the transplanted cells.[@b97-ndt-13-585] The mechanism of action is described as a response from the T cells that are in the graft against the host tissue.[@b97-ndt-13-585] This immune response leads to multiple organ system damage and possible rejection of the graft.[@b90-ndt-13-585] A systematic review found that autologous transplantation did not increase the risk of mortality or the incidence of tumorigenesis.[@b98-ndt-13-585] Studies have also found that autologous transplants from older patients are less likely to cause tumors.[@b93-ndt-13-585],[@b99-ndt-13-585] The reduced risks associated with autologous transplantation make it a prime candidate for expedited review by the US Food and Drug Administration which is intended to evaluate a novel therapy that satisfies an unmet medical need for a serious, debilitating medical condition. In this case, autologous stem cell transplantation offers certain advantages over current treatments for epilepsy, and thus would be well suited for expedited review.[@b100-ndt-13-585] There are clinical trials in Phase I/II for autologous transplantation of NSCs for Parkinson's disease.[@b101-ndt-13-585] Epilepsy is the next neurological disorder that can be treated with stem cell therapy. Stem cells from the subventricular zone (SVZ) and other areas along the dentate gyrus can be isolated at present.[@b102-ndt-13-585] These cells can be isolated by a temporal lobectomy or endoscopic resection.[@b103-ndt-13-585],[@b104-ndt-13-585] Autologous transplantation of NSC has emerged as an area of interest due to the observed neurogenesis in the hippocampus and the parahippocampal gyrus which are involved in TLE.[@b105-ndt-13-585],[@b106-ndt-13-585] The intended purposes of stem cell transplantation are to provide growth factors and trophic factors, while facilitating neurogenesis with a special focus on differentiation into the inhibitory GABAergic neurons whose loss has been implicated in epilepsy.[@b107-ndt-13-585] The ability of NSCs to differentiate into GABAergic interneurons has been observed in the dentate gyrus of mice with TLE.[@b108-ndt-13-585] The NSCs used in transplantation can be obtained from the SVZ, which is one area of persistent NSC presence, and then expanded in culture before transplantation. Other sources of NSCs include the embryonic MGE.[@b86-ndt-13-585] Neurogenesis presents an issue of trade-off in the TLE model. Neurogenesis in the acute phase after insult is abnormal and deleterious, whereas neurogenesis in the chronic phase is critical to counteract the neurodegeneration that is associated with epilepsy.[@b109-ndt-13-585],[@b110-ndt-13-585] High temporal specificity is, therefore, as important an element as spatial specificity in any future epilepsy therapy based on stem cells. Induced pluripotent stem cells (iPSCs) are stem cells derived from adult cells that may offer the advantages of autologous transplantation. The groundbreaking induction of somatic cells into pluripotent stem cells was first achieved in 2006 by Takahashi and Yamanaka by introducing certain reprogramming genes found to drive stem cell self-renewal, such as Oct-4 and SOX2, into fibroblasts.[@b111-ndt-13-585] The ability to generate iPSCs provides the promise of autologous cell replacement and gene therapy to reprogram cells. Viral vectors to deliver the reprogrammed transgenes may involve the use of permanently integrating viruses.[@b111-ndt-13-585],[@b112-ndt-13-585] Whereas genetic stability is achieved with permanently introducing transgenes, alternative approaches employing nonintegrating viruses such as adenoviruses and Epstein--Barr virus have been explored.[@b112-ndt-13-585] Another use of iPSCs is generation of a population of neural cells that release specific neurotransmitters (eg, GABA) and integrate them into a specific location such as the dentate gyrus.[@b113-ndt-13-585] Such targeted introduction of neurotransmitter-producing neural cells could be a powerful autologous therapy for epilepsy in the future. In parallel, iPSCs may serve as a research tool to generate patient-derived neural cells for mechanistic studies and drug testing,[@b114-ndt-13-585],[@b115-ndt-13-585] which will facilitate human modeling of epilepsy allowing a better understanding of the disease and its treatment. Our TLE model uses intraventricular injection of kainic acid (KA) to induce SRS in virtually all rats within 3--4 months, while other groups use pilocarpine.[@b116-ndt-13-585] KA was used to create a unilateral hippocampal injury model of TLE through the loss of CA1 and CA3 pyramidal neurons, similar to the hippocampal sclerosis seen in TLE.[@b117-ndt-13-585] NSCs obtained from the adult SVZ were implanted into the hippocampus of rats that were receiving KA. A statistically significant decrease in abnormal electroencephalography spiking was observed 2 weeks after NSC transplantation.[@b118-ndt-13-585] At 5 weeks posttransplantation, grafts of NSCs were detected at CA3 of the hippocampus and in the subgranular zone into which they had migrated. Most of the transplanted NSCs were found to express glial fibrillary acidic protein, indicating that they were astrocytes, but some NSCs also expressed neuronal nuclei (NeuN), indicating that they had differentiated into mature neurons. The transplantation of MGE-derived NSCs saw a 43% decrease in the frequency of SRS at 3 months postgrafting. NSC-derived cells were found in the CA3 and CA1 fields of the hippocampus with an approximate yield of 28% of the injected cells.[@b86-ndt-13-585] Transplanted NSCs have been observed to have a survival rate of nearly 30% at 3 months after grafting, with \~91% of those surviving NSCs differentiating into GABAergic neurons.[@b119-ndt-13-585] The differentiated GABAergic interneurons were found to express markers such as neuropeptide Y, which has been shown to inhibit glutamatergic excitation in the epileptic dentate gyrus.[@b120-ndt-13-585] NSCs have also demonstrated secretion of stem cell factor with concomitant increase in c-Kit, the ligand for stem cell factor.[@b121-ndt-13-585] Such stem cell factors likely constitute the mechanisms for the observed reduction in epileptic activity after transplantation of NSCs. Immunohistochemical staining revealed increased pre servation of GABAergic inhibitory neurons and a reduction in abnormal mossy fiber growth, which has been correlated with the increased excitability of the hippocampus in epilepsy.[@b116-ndt-13-585],[@b122-ndt-13-585] Our laboratory has also reported that infusion of erythropoietin significantly increased the survival of transplanted NSCs while suppressing aberrant neurogenesis in the dentate gyrus.[@b118-ndt-13-585],[@b123-ndt-13-585] The results of decreased abnormal neuronal activity, increased secretion of neurotrophic factors, and decreased abnormal mossy fiber growth suggest a neuroprotective effect against seizures from the transplanted NSCs. Grafts of NSCs have also been observed to increase the numbers of GABAergic neurons and GDNF-secreting cells in the hippocampus.[@b86-ndt-13-585] GDNF levels were increased in the hippocampus after seizures induced by KA and the levels correlated with anticonvulsant effects.[@b124-ndt-13-585],[@b125-ndt-13-585] Transplantation of NSCs in the pilocarpine-induced TLE model has demonstrated similar effects as in the KA-induced model. The transplantation of β-galactosidase--encoded human NSCs in rats exhibiting SRS due to administration of pilocarpine saw only 13% of the rats receiving NSCs to continue exhibiting SRS.[@b22-ndt-13-585] β-gal+ NSCs were found in the hippocampal CA1 and CA3 areas among others, while only a fraction of them were found to be mature NeuN-expressing mature neurons. Some of the β-gal+ cells demonstrated some degree of differentiation into GABAergic neurons by expressing identifying markers. Stem cell therapy for treatment of epilepsy-induced neuropsychiatric symptoms ============================================================================= Stem cell therapy presents an intriguing alternative approach to treat epilepsy that is fundamentally different from anticonvulsants and electrical stimulation which seek to manage the symptoms and not address the underlying causes of epilepsy. Stem cell therapy has the potential to provide neurotrophic factors and stimulate neurogenesis. While NSC grafts have demonstrated a reduction in SRS, further studies are needed to assess the differentiation of stem cells into GABAergic interneurons and other cell types. Establishment of a consistent method of controlled differentiation of stem cells into desired cells could allow for epilepsy to be treated with the replacement of deficient or defective neurons using the patient's own stem cells. At the same time, the ability of the transplanted stem cells to differentiate must be balanced with the danger of tumorigenesis. Uncontrolled differentiation and migration of stem cells could pose a grave risk for tumor growth and metastasis. Further research into the longevity of stem cells along with their long-term production of trophic factors would help evaluate the safety and efficacy of stem cell therapy toward treating epilepsy, which is most often a lifelong neurological disease. There have been various studies that investigated the administration of stem cells into rat and mouse models of status epilepticus and other relevant animal models of epilepsy that provide valuable insights into the clinical benefits of stem cells as a potential treatment option for those suffering with epilepsy. The administration of stem cells into the status epilepticus animals has been shown to ameliorate many pathological symptoms of the disease, attributed to neuroprotection. Stem cell transplanted epileptic animals display reduced seizures (both in frequency and duration), and improved memory and learning, with postulated neuroprotective effects characterized by decreased neuronal loss, increased neurogenesis, reduced microglia and astrocyte activation, and robust anti-inflammatory response.[@b9-ndt-13-585],[@b89-ndt-13-585],[@b126-ndt-13-585]--[@b134-ndt-13-585] Although these studies investigated the hallmark biological basis of the disease, the psychiatric component of epilepsy was not examined. To date, no definitive study has explored the effects of stem cells on psychiatric symptoms in animal models of epilepsy. Currently, neuropsychiatric disorders are considered a comorbidity of epilepsy. The use of stem cell transplantation in treating epilepsy-induced neuropsychiatric symptoms still requires further investigation to determine its safety and efficacy. It is currently hypothesized that neurogenesis in mood-affecting areas of the brain will improve depression and other neuropsychiatric disorders.[@b23-ndt-13-585] Despite the presence of optimistic theories on the effectiveness of stem cell transplantation, careful and rigorous preclinical studies are warranted to advance cell therapy for treating the neurologic and neuropsychiatric symptoms of epilepsy. Recapping the future of epilepsy ================================ Epilepsy is a debilitating disease that affects millions of people and is currently only managed and not cured. The use of AEDs and surgical resection are treatments that forgo the neuropsychiatric aspect of epilepsy, a component that is just as important as the neurological aspect. Autologous stem cell therapy is a treatment option that is gaining ground for treating neurological and psychiatric disorders. This cell-based therapy can be used for the treatment of epilepsy. Autologous stem cells reduce the risk of GVHD and can be harvested from the SVZ during surgical resection. In epileptic patients, these cells can be primed to reconstruct the dysfunctional neural circuitry, although not correcting the GABA deficiency. As in any novel treatment, caution in translating laboratory investigations into clinical applications will require careful and rigorous preclinical studies on the safety and efficacy of the therapies. Optimization of stem cell dose, route of cell delivery, and timing of intervention need to be tested in clinically relevant models of epilepsy that capture both neurologic and neuropsychiatric symptoms of the disease. **Disclosure** The authors report no conflicts of interest in this work.

INTRODUCTION {#s1} ============ Squamous cell cancer of the breast (SCCB) is a rare type of breast cancer that accounts for approximately 0.04-0.1% of all breast cancers, and less than 0.1% of all invasive breast ductal carcinomas \[[@R1]-[@R4]\]. SCCB is diagnosed by exclusion of other more common cancers. In particular, the diagnosis requires that: *(i)* the tumor origin does not arise from the overlying skin, nipple, or adenexal components, *(ii)* more than 90% of the tumor consists of squamous cells, *(iii)* there is no evidence of ductal or mesenchymal elements within the tissue sample, and *(iv)* no other sites of primary squamous cell cancer are present \[[@R1], [@R2], [@R5]-[@R7]\]. Because of the rarity of this cancer, there is currently no consensus on the treatment and prognosis of these patients. Many previous studies have shown that locoregional radiotherapy (RT) can improve cause-specific survival (CSS) and overall survival (OS) of female breast cancer patients \[[@R8]-[@R10]\], but there is limited research on the effect of RT in SCCB. Moreover, many of these previous studies were single-institution retrospective reviews with limited numbers of patients, so it is difficult to make recommendations for patients with different stages of SCCB. The ideal locoregional RT regimens for patients with different stages of SCCB are still uncertain. In this study, we analyzed the effect of postoperative RT on the survival of patients with SCCB using a population-based national registry, Surveillance, Epidemiology, and End Results (SEER). PATIENTS AND METHODS {#s2} ==================== Patients {#s2_1} -------- Data were obtained from the current SEER database, which consists of 18 population-based cancer registries of patients in the United States. SEER data are an open-access resource for cancer-based epidemiology and survival analyses. SEER\*Stat software from the National Cancer Institute (Surveillance Research Program, National Cancer Institute SEER\*Stat software, <http://www.seer.cancer.gov/seerstat>, version 8.2.1) was used to identify eligible patients. Patients with diagnoses of SCCB from 1973 to 2012 were identified. We obtained permission to access research data files with the reference number 11252-Nov2014 \[[@R11]\]. All included patients were females diagnosed with SCCB, received cancer-directed surgery, and had records on whether postoperative RT was used. Pathologic diagnosis was based on the primary site using the International Classification of Disease for Oncology, Third Edition (ICD-O-3). Use of the SEER database does not require informed consent. This study was approved by the ethics committee of the First Affiliated Hospital of Xiamen University (Xiamen) and Sun Yat-sen University Cancer Center (Guangdong). Clinicopathologic factors {#s2_2} ------------------------- The following clinical and pathologic factors were collected from the SEER database: age at diagnosis, race, grade, tumor stage, tumor size (pT), lymph node status (pN), estrogen receptor (ER) status, progesterone receptor (PR) status, human epidermal growth factor 2 (HER2) status, and use of adjuvant external beam RT. Survival, cause of death, and duration of follow-up were recorded. Statistical analysis {#s2_3} -------------------- The χ^2^ and Fisher\'s exact probability tests were used to analyze differences in the qualitative data. Univariate and multivariate Cox regression analyses were used to identify factors that were significantly associated with CSS and OS. Multivariable analyses were performed for factors that were significantly associated with CSS and OS in the univariate analyses. Calculation of survival rates were plotted by the Kaplan-Meier method, and compared using the log-rank test. All data were analyzed using SPSS statistical software, version 21.0 (IBM Corporation, Armonk, NY, USA). A *P*-value less than .05 was considered statistically significant. RESULTS {#s3} ======= Patient characteristics and survival {#s3_1} ------------------------------------ A total of 523 patients met the eligibility criteria, 167 of whom (31.9%) received post-operative RT (Table [1](#T1){ref-type="table"}). The median age was 66 years (range: 24-102 years). Among patients whose pT stage, pN stage, tumor stage, ER status, PR status, and HER2 status were known, 75.8% (294/388) had stage T2-T4 SCCB, and 73.6% (293/398) had negative lymph nodes. Stage I, II, III, and IV SCCB was present in 24.8% (102/412), 51.5% (212/412), 18.7% (77/412), and 5.0% (21/412) of patients, respectively. Of the 319, 315 and 50 patients whose ER, PR and HER2 status were available, respectively, a total of 81.2% (259/319) of patients were ER negative, 88.3% (278/315) were PR negative, and 92.0% (46/50) were HER2 negative. Patients who were older than 50 years (*P* = 0.002) and with more advanced cancer (*P* = .004) were more likely to have received postoperative RT (Table [1](#T1){ref-type="table"}). ###### Patient characteristics Characteristic n (%) Without RT (%) With RT (%) *P* ---------------- ------------ ---------------- ------------- ------- Age (years)  ≤50 95 (18.2) 52 (14.6) 43 (25.7) 0.002  \>50 428 (81.8) 304 (85.4) 124 (74.3) Race  Black 64 (12.2) 43 (12.2) 21 (12.7) 0.943  White 435 (83.2) 297 (84.1) 138 (83.1)  Other 20 (3.8) 13 (3.7) 7 (4.2)  Unknown 4 (0.8) pT stage  pT0-1 94 (18.0) 59 (23.9) 35 (24.8) 0.132  pT2 178 (34.0) 122 (49.4) 56 (39.7)  pT3 72 (13.8) 44 (17.8) 28 (19.9)  pT4 44 (8.4) 22 (8.9) 22 (15.6)  Unknown 135 (25.8) pN stage  pN0 293 (56.0) 196 (77.2) 97 (67.4) 0.152  pN1 74 (14.1) 39 (15.4) 35 (24.3)  pN2 20 (3.8) 12 (4.7) 8 (5.6)  pN3 11 (2.1) 7 (2.7) 4 (2.7)  Unknown 125 (23.9) Metastasis  M0 404 (77.2) 259 (93.5) 145 (98.0) 0.058  M1 21 (4.0) 18 (6.5) 3 (2.0)  Unknown 98 (18.8) Stage  I 102 (19.5) 69 (25.7) 33 (22.9) 0.004  II 212 (40.5) 143 (53.4) 69 (47.9)  III 77 (14.7) 38 (14.2) 39 (27.1)  IV 21 (4.0) 18 (6.7) 3 (2.1)  Unknown 111 (21.3) Grade  G1 50 (9.5) 36 (13.6) 14 (10.4) 0.212  G2 116 (22.2) 82 (31.1) 34 (25.2)  G3-4 233 (44.6) 146 (55.3) 87 (64.4)  Unknown 124 (23.7) ER status  Negative 259 (49.5) 156 (79.2) 103 (84.4) 0.245  Positive 60 (11.5) 41 (20.8) 19 (15.6)  Unknown 204 (39.0) PR status  Negative 278 (53.2) 175 (89.7) 103 (85.8) 0.295  Positive 37 (7.1) 20 (10.3) 17 (14.2)  Unknown 208 (39.7) HER2 status  Negative 46 (8.8) 26 (89.7) 20 (95.2) 0.473  Positive 4 (0.8) 3 (10.3) 1 (4.8)  Unknown 473 (90.4) G1, well; G2, moderately; G3, poorly; G4, undifferentiated; ER, estrogen receptor; PR, progesterone receptor; HER2, human epidermal growth factor 2. The median duration of follow-up was 55.0 months (range: 1-473 months). The 5-year and 10-year CSS rates were 69.7% and 65.6%, and the 5-year and 10-year OS rates were 60.1% and 46.0%, respectively (Figure [1A-1B](#F1){ref-type="fig"}). {#F1} Analysis of prognosis {#s3_2} --------------------- Univariate Cox survival analysis showed that patients who were black, had advanced pT stage, and advanced pN stage had significantly poorer CSS (Table [2](#T2){ref-type="table"}). However, no significant differences in CSS were observed for patients with and without postoperative RT. Univariate analysis also indicated that patients who were older, black, had advanced pT stage, advanced pN stage, PR negative disease and did not receive postoperative RT had significantly poorer OS. ###### Univariate analysis of cause-specific survival and overall survival Characteristic CSS OS ---------------- ------- -------------- --------- ------- ------------- --------- Age (years)  ≤50 1 1  \>50 1.453 0.941-2.244 0.092 2.953 2.009-4.234 \<0.001 Race  Black 1 1  White 0.557 0.369-0.839 0.005 0.641 0.463-0.888 0.007  Other 0.818 0.357-1.874 0.635 0.607 0.296-1.249 0.175 pT stage  pT0-1 1 1  pT2 2.181 1.153-4.125 0.016 1.415 0.945-2.118 0.092  pT3 4.721 2.419-9.212 \<0.001 2.801 1.789-4.385 \<0.001  pT4 9.167 4.638-18.120 \<0.001 4.405 2.682-7.208 \<0.001 pN stage  pN0 1 1  pN1 2.475 1.468-3.525 \<0.001 1.732 1.209-2.483 0.003  pN2 4.090 2.292-7.298 \<0.001 3.021 1.815-5.028 \<0.001  pN3 4.657 2.131-10.177 \<0.001 2.782 1.296-5.969 0.009 Grade  G1 1 1  G2 1.228 0.621-2.430 0.555 0.803 0.526-1.227 0.311  G3-4 1.468 0.780-2.762 0.235 0.746 0.504-1.105 0.144 ER status  Negative 1 1  Positive 1.228 0.749-2.015 0.415 0.984 0.645-1.504 0.942 PR status  Negative 1 1  Positive 0.642 0.311-1.326 0.231 0.510 0.276-0.943 0.032 Radiotherapy  No 1 1  Yes 0.807 0.573-1.137 0.220 0.650 0.497-0.849 0.002 CSS, cause-specific survival; OS, overall survival; HR, hazard ratio; CI, confidence interval; G1, well; G2, moderately; G3, poorly; G4, undifferentiated; ER, estrogen receptor; PR, progesterone receptor. We used multivariate Cox analysis, with adjustment for significant factors from the univariate analysis, to assess the association of different parameters with CSS and OS (Table [3](#T3){ref-type="table"}). The results show that advanced pT stage and advanced pN stage were independently associated with poorer CSS. Advanced pT stage, advanced pN stage, and no postoperative RT were independently associated with poorer OS. ###### Multivariate analyses of cause-specific survival and overall survival Characteristic CSS OS ---------------- ------- -------------- --------- ------- -------------- --------- Age (years)  ≤50 1 1  \>50 1.695 0.960-2.992 0.069 1.373 0.708-2.662 0.348 Race  Black 1 1  White 0.619 0.367-1.043 0.072 0.778 0.412-1.471 0.441  Other 0.638 0.212-1.921 0.424 1.783 0.561-5.665 0.327 pT stage  pT0-1 1 1  pT2 2.090 1.072-4.076 0.030 2.379 1.055-5.367 0.037  pT3 4.234 2.095-8.557 \<0.001 4.804 2.022-11.412 \<0.001  pT4 7.278 3.433-15.432 \<0.001 9.494 3.775-23.878 \<0.001 pN stage  pN0 1 1  pN1 1.553 0.953-2.529 0.077 2.064 1.178-3.619 0.011  pN2 3.406 1.851-6.267 \<0.001 4.518 2.330-8.759 \<0.001  pN3 2.652 1.188-5.919 0.017 3.064 1.256-7.464 0.014 PR status  Negative --- 1  Positive --- --- --- 0.482 0.215-1.083 0.077 Radiotherapy  No --- 1  Yes --- --- --- 0.439 0.266-0.723 0.001 CSS, cause-specific survival; OS, overall survival; HR, hazard ratio; CI, confidence interval; PR, progesterone receptor. The relationship of the postoperative RT and survival {#s3_3} ----------------------------------------------------- Kaplan-Meier analysis indicated that postoperative RT was significantly associated with better OS (log-rank test: *P* = .001) (Figure [2A](#F2){ref-type="fig"}). The 5- and 10-year OS rates were 66.7% and 54.5% for patients given RT, and were 57.0% and 42.0% for those not given RT. However, postoperative RT had no effect on CSS (log-rank test: *P* = .217) (Figure [2A-2B](#F2){ref-type="fig"}). {#F2} We also determined the influence of postoperative RT on survival of patients with different stages of SCCB (Figure [3](#F3){ref-type="fig"}). The results indicate that RT was significantly associated with improved CSS (log-rank test: *P* = .047) and OS (log-rank test: *P* \< .001) for patients with stage II SCCB (Figure [3A-3B](#F3){ref-type="fig"}). There were trends for improved CSS and OS for patients given RT who had stage I and stage III SCCB, but these were not statistically significant (log-rank test: *P* \>.05) (Table [4](#T4){ref-type="table"}). {#F3} ###### Cause-specific survival and overall survival by stage and radiotherapy Stage Median survival (months) 5-year 10-year *P* ------------- -------------------------- -------- --------- ------ ------ --------- Tumor stage  CSS   I --- 85.9 100 83.8 95.5 0.062   II --- 71.5 86.6 69.0 77.1 0.047   III 34 33.8 46.7 0 35.6 0.327   IV 5 9.3 0 0 0 0.689  OS   I 175 81.7 89.2 62.7 63.9 0.174   II 121 55.2 78.5 39.8 72.4 \<0.001   III 28 30.6 46.7 18.3 35.6 0.327   IV 5 7.4 0 0 0 0.772 Nodal stage  CSS   pN0 --- 75.0 84.0 73.1 76.0 0.163   pN1 --- 51.8 61.1 39.9 61.1 0.248   pN2 34 23.4 50.0 --- --- 0.463   pN3 22 28.6 33.0 --- --- 0.486  OS   pN0 123 61.4 78.2 44.9 65.9 \<0.001   pN1 52 48.1 50.2 19.5 46.0 0.182   pN2 26 20.8 50.0 --- --- 0.255   pN3 22 28.6 33.0 --- --- 0.486 CSS, cause-specific survival; OS, overall survival; RT, radiotherapy. We also examined the prognostic effect of postoperative RT on the OS of patients with different pN stages of SCCB (Figure [4](#F4){ref-type="fig"}). The results indicate that RT was associated with significantly improved OS for patients with stage pN0 cancer (log-rank test: *P* \< .001). RT also tended to increase the CSS in pN0-N3 stage patients and OS in pN1-N3 stage patients, although this was not statistically significant (log-rank test: *P* \> .05) (Table [4](#T4){ref-type="table"}). {#F4} DISCUSSION {#s4} ========== This is the largest retrospective analysis to assess the effect of postoperative RT on the survival of patients with SCCB. Our results indicated that RT improved the survival of SCCB patients, especially those with stage II cancer and no regional lymph node metastasis (pN0). Previous studies have shown that postoperative RT improves local control and survival in high-risk invasive breast cancers \[[@R8]-[@R10]\], but the present study is the first to identify a survival benefit for SCCB patients who receive postoperative RT. Previous research indicated that SCCB is less likely to undergo lymphatic spread than adenocarcinomas. In particular, only 10 to 30% of SCCB patients have lymph node infiltration at the time of surgery \[[@R1], [@R12]\]. Case reports have indicated that SCCB is associated with high expression of Ki-67, a marker of proliferation \[[@R2], [@R13], [@R14]\]. Over 85% of patients are ER- and PR-, and most are also HER2- \[[@R4], [@R12], [@R15]\]. SCCB patients with the triple-negative subtype (ER-/PR-/HER2-) seem to have poor prognoses.\[[@R16]\] In this study, most patients were lymph node negative, ER-, PR-, and HER2-. The 10-year CSS and OS rates of our patients were 65.6% and 46.0%, respectively. Our results therefore underline the aggressive nature of SCCB, and suggest that comprehensive treatment should be considered for patients with potentially poor prognoses. RT is provided as a standard of care for female breast cancer patients after breast conservation surgery, and is frequently given after mastectomy in high-risk patients \[[@R8]-[@R10], [@R17]\]. A previous SEER analysis showed that among 137 patients with SCCB from 1998 to 2001, only 35% received adjuvant RT, although the study did not analyze the effect of RT on survival \[[@R15]\]. Among 31 patients in the University of Texas M.D. Anderson Cancer Center with SCCB, 19 patients were treated with postoperative RT, and the recurrence-free survival (P = .210) and OS (*P* = .840) were not significantly different from those without RT.\[[@R15]\] Two Chinese studies examined 58 patients with SCCB, 12 of whom received postoperative RT, and reported that RT provided no significant survival benefit \[[@R18], [@R19]\]. Thus, SCCB seems to be relatively radioresistant, despite the fact that SCCs are generally considered to be radiosensitive \[[@R20]\]. In our cohort of 523 SCCB patients, only 167 patients (31.9%) received RT, even though 28.1% of the patients were lymph node-positive. Our results showed that postoperative RT was associated with improved OS, but had no effect on CSS. Thus, the present study is the first to identify a survival benefit for postoperative RT in patients with SCCB. Our results found that RT specifically improved the survival of SCCB patients with stage II cancer and those with pN0 stage. The reasons for the beneficial effect of RT in these particular groups are not obvious. Four patients in the M.D. Anderson Cancer Center series who experienced a locoregional recurrence (LRR) had either T1N0 or T2N0 tumors \[[@R15]\]. Thus, the benefit of RT for patients with early-stage SCCB could in part be attributed to the aggressive nature of SCCB. Among our patients, the median survival time for those with stage III-IV and pN2-3 cancer was very short (less than 3 years). Thus, more obvious benefit of RT for stage II patients and pN0 stage with moderate risk of recurrence was observed. Our research and the previous study were both retrospective analyses \[[@R15]\], so the criteria used to select patients for postoperative RT remain unclear, but RT may not omit in management of locally advanced SCCB at presentation due to the absence of clinical trials. Due to the limitations of the SEER database, we cannot identify risk factors for LRR of SCCB. Nayak et al. found that the 5-year LRR was 46% in 21 patients with SCCB, and the only statistically significant feature associated with LRR was the presence of a spindle cell component comprising \>10% of the tumor (*P* = .006) \[[@R3]\]. The lack of response of SCCB to postoperative RT may reflect the mixed cell type present or the palliative use of RT to treat advanced disease. SCCB is considered resistant to the standard chemotherapy regimens used for other breast cancers. Nevertheless, some studies found that adjuvant cisplatin-based regimens could be effective \[[@R6], [@R7], [@R21], [@R22]\]. In addition, 62.1% to 85.7% of SCCB patients have overexpression of epidermal growth factor receptor (EGFR) \[[@R15], [@R19], [@R23]\]. Although no study has shown that combined pharmacotherapy increases the sensitivity of SCCB to radiotherapy, studies of other cancers showed that a cisplatin-based regimen with an anti-EGFR agent may radiosensitize squamous cancer cells \[[@R24]-27\]. Future studies with large sample sizes are needed to investigate the effect of such regimens on the radiosensitivity of SCCB and the survival of SCCB patients. The current study had several limitations that must be considered. The main limitations are the nonrandomized nature of the dataset and the inherent biases that exist in all retrospective studies. Second, the SEER database does not have data on the type and dose of systemic therapy, use of endocrine treatments, lymphovascular invasion, margin status, and local or regional recurrence. This hindered our ability to directly assess the effect of specific factors or clinical circumstances on outcome. In addition, this database provides little information of why clinicians administered or did not administer postoperative RT to different patients. Although retrospective reviews do not carry the power of prospective studies, no prospective studies have yet examined the effect of postoperative RT on SCCB. In conclusion, postoperative RT was associated with improved survival of patients with SCCB, especially those with stage II cancer and pN0 stage patients. Prospective studies will be needed to confirm the results of this study and to establish the optimal protocols for use of RT in the management of SCCB. This work was supported by grants from the National Natural Science Foundation of China (No. 81402527), the Sci-Tech Office of Guangdong Province (No. 2013B021800157, 2013B021800458) and the Youth Foundation of Fujian Provincial Health and Family Planning Commission (No. 2014-2-63). **CONFLICTS OF INTEREST** There is no conflict of interest.

{#sp1 .192} {#sp2 .193} {#sp3 .194} {#sp4 .195} {#sp5 .196} {#sp6 .197} {#sp7 .198} {#sp8 .199} {#sp9 .200} {#sp10 .201} {#sp11 .202} {#sp12 .203} {#sp13 .204} {#sp14 .205} [^1]: The Morison Lecture delivered in the Hall of the Royal College of Physicians, 25th January 1928.

Introduction {#s1} ============ Systemic activation of coagulation cascade and formation of thrombi in the microvasculature contribute to organ dysfunction that characterizes sepsis [@pone.0045427-Pawlinski1], [@pone.0045427-Mitroulis1]. Several studies in experimental models indicate that septic inflammatory environment induces the expression of tissue factor (TF), a trans-membrane protein [@pone.0045427-Mitroulis1] which initiates coagulation cascade and results in thrombin generation. Circulating TF, in the form of TF-bearing microparticles or TF expressed in blood cells, is suggested to play a critical role in this process [@pone.0045427-Pawlinski1]--[@pone.0045427-Pawlinski2]. Additionally, thrombin and TF/factor VIIa complex signaling through protease activated receptor-1 (PAR-1) and PAR-2 respectively was implicated in the induction of inflammation in experimental models of sepsis, linking coagulation to inflammation [@pone.0045427-Pawlinski3], [@pone.0045427-Schouten1]. Neutrophils have a critical role in launching the first line of host defense against infection. They are recruited in vast numbers to the site of microbial invasion, where they engulf and kill pathogens into phagosomes [@pone.0045427-Borregaard1], [@pone.0045427-MayerScholl1]. Moreover, neutrophils release lytic enzymes for the elimination of extracellular microbes [@pone.0045427-Papayannopoulos1]. Recently, another aspect of neutrophil microbicidal activity has been described; the release of neutrophil extracellular traps (NETs; [@pone.0045427-Papayannopoulos1], [@pone.0045427-Brinkmann1]). NETs are extracellular chromatin structures that entrap microbes and are composed of nuclear and granule constituents of neutrophils [@pone.0045427-Papayannopoulos1], [@pone.0045427-Brinkmann1]. They are formed after phagocytosis of pathogens or treatment with inflammatory stimuli [@pone.0045427-Papayannopoulos1], [@pone.0045427-Brinkmann1] and are implicated in the pathogenesis of sepsis [@pone.0045427-Clark1]. Interestingly, it was shown that the entrapment of platelets in NETs is associated with platelet activation and aggregation [@pone.0045427-Fuchs1]. Moreover, the presence of NETs has been recently indentified in thrombi in a murine model of deep vein thrombosis [@pone.0045427-Brill1]. Experimental data further implicate neutrophil serine proteases in the inactivation of antithrombin and tissue factor pathway inhibitor which results in the activation of coagulation [@pone.0045427-Massberg1]. Several lines of evidence support a critical role for autophagy in the regulation of innate immune responses [@pone.0045427-Deretic1], [@pone.0045427-Levine1]. This process has been implicated in pathogen elimination and recognition by intracellular receptors [@pone.0045427-Deretic1], [@pone.0045427-Levine1]. The activation of autophagy in human neutrophils has been previously linked with phagocytosis and activation of Toll-like receptors [@pone.0045427-Mitroulis2]. Moreover, it is reported that neutrophils from patients with sepsis exhibit vacuolization and susceptibility toward a necrotic form of cell death, both associated with autophagy [@pone.0045427-Mihalache1]. Additionally, recent data suggest that autophagy is required in NET release [@pone.0045427-Remijsen1], [@pone.0045427-Mitroulis3]. Herein, we describe a novel role of neutrophils in the interface between inflammation and coagulation. We report for the first time that TF-bearing NETs are released from neutrophils derived from patients with gram-negative sepsis, trigger thrombin generation and subsequent PAR-1 signaling *in vitro*. Furthermore, inclusion of TF in autophagosomes was linked with its extracellular delivery in NETs, while inhibition of phosphatidylinositol 3-kinase (PI3K) signaling or endosomal acidification in neutrophils abolished NET release and TF trafficking. Materials and Methods {#s2} ===================== Human Sepsis Neutrophil and Serum Samples {#s2a} ----------------------------------------- Peripheral blood neutrophils and serum were collected from six healthy donors and eight patients with gram-negative sepsis, hospitalized in First Department of Internal Medicine or Intensive Care Unit of University Hospital of Alexandroupolis. Informed written consent has been obtained from every volunteer or their relatives. The study protocol design was in accordance with the Declaration of Helsinki and the procedures have been approved by the local ethics committee (Scientific Committee of the University Hospital of Alexandroupolis, Greece). Patient characterization was in accordance to current clinical practice guidelines [@pone.0045427-Levy1], [@pone.0045427-Levy2]. The identification of gram-negative bacteria (*Escherichia coli* in six patients, *Pseudomonas aeruginosa* and *Klebsiella pneumonia* in the other two) from blood cultures obtained at the time of blood collection was required for the final inclusion of the experimental data in the study. Serum from all patients was obtained at the time of enrollment. Serum was immediately transferred to ice and isolated by centrifugation at 4°C at 1400×g for 15 min. Serum was stored at −80°C till used. Neutrophils were isolated from heparinized blood after double-gradient density centrifugation (Histopaque; Sigma-Aldrich Co., St Louis, MO, USA) as previously described [@pone.0045427-Mitroulis2]. Neutrophil purity (\>98%) was assessed by Giemsa staining and viability (\>95%) by Trypan blue staining. Contamination with platelets was under 1%. The absolute number of neutrophils was adjusted approximately to 4×10^6^ cells/ml in RPMI. After collection, all patient samples were numbered and de-identified for patient confidentiality. Stimulation and Inhibition Studies {#s2b} ---------------------------------- Neutrophils were incubated in 5% CO~2~ at 37°C in a total volume of 500 µl of RPMI in the presence of 6% serum from healthy donor. Neutrophils were treated with either opsonized *Escherichia coli* (*E.* coli) at 1/30 neutrophil to bacteria ratio (Phagoburst, Phagotest; ORPEGEN Pharma, Heidelberg, Germany) or septic serum at a final concentration of 6% in neutrophil cultures for 3 h. This concentration was the optimal to stimulate neutrophils and avoid NET degradation. Moreover, 3 hours incubation with sepsis serum demonstrated the highest percentage of NET releasing neutrophils and thus this time point was chosen for further experiments. Cells were treated with the autophagy inhibitor 3-methyladenine (3-MA) (5 mM, Calbiochem, San Diego, CA, USA) or bafilomycin A1 (30 nM, Sigma-Aldrich Co.) 30 min after the addition of bacteria in order to allow phagocytosis to occur. Moreover, healthy neutrophils were incubated with a mixture of inflammatory mediators (TNF-α, 1 ng/ml, Sigma-Aldrich; G-CSF, 1 ng/ml, GRANOCYTE-Aventis, Antony, France; IL-1β, 10 ng/ml, Sigma-Aldrich). An IgG1 mouse anti-human TF mAb (10 µg/ml; American Diagnostica, Greenwich, CT, USA) was used for TF inhibition studies and an IgG1 anti-CD19 mAb (DAKO, Denmark) as control. PAR-1 signaling inhibition was performed using FLLRN peptide (500 µΜ, Anaspec, Fremont, CA, USA). TLR4 signaling inhibition was performed by pre-incubating sepsis serum with Polymyxin B (10 µg/ml, Santa Cruz, CA, USA) for 30 mins. All the substances used in this study were endotoxin free, as determined by a *Limulus* amebocyte assay (Sigma-Aldrich). Preparation of Apoptotic Neutrophils {#s2c} ------------------------------------ Apoptotic neutrophils were prepared as previously described [@pone.0045427-Afonso1]. After incubation of neutrophils for 1 h with sepsis serum, apoptosis was induced by ultraviolet irradiation exposure (245 nm) for 10 min. Cells remained in RPMI with control serum in 5% CO~2~ at 37°C for 2 h. Apoptotic cells indicated more than 80% Annexin V positive and propidium iodide negative staining as assessed by flow cytometry (BD Biosciences, FACScan). Microparticle Depletion {#s2d} ----------------------- Microparticles were depleted from sepsis serum as previously described [@pone.0045427-Wang2]. The collected serum containing microparticles was centrifuged at 20,000×g for 15 minutes at 4°C to pellet them. Microparticle-depleted serum (supernatant) was centrifuged again at 20,000×g for 15 minutes at 4°C in order to eliminate remaining microparticles. Depletion of microparticles was verified by flow cytometry as previously described [@pone.0045427-AbidHussein1]. Immunofluorescence {#s2e} ------------------ Sample preparation and visualization by immunofluorescence microscopy was performed as previously described [@pone.0045427-Mitroulis3]. Cells were stained using an anti-myeloperoxidase (MPO)-specific mouse monoclonal antibody (DAKO), an IgG1 mouse anti-TF mAb (American Diagnostica, Greenwich, CT, USA), a rabbit anti-LC3B polyclonal Ab (Sigma-Aldrich Co.) and a mouse anti-high mobility group box-1 (HMGB1) mAb (Abcam, UK). An IgG1 anti-CD19 mAb (DAKO) was utilized as control. A polyclonal rabbit anti-mouse Alexa fluor 488 antibody (Invitrogen, Carlsbad, CA, USA) was utilized as secondary antibody. DNA was counterstained using DAPI (Sigma-Aldrich). LC3B was stained with a goat anti-rabbit Alexa fluor 647 antibody (Invitrogen). LysoTracker (100 nM, Invitrogen) was used for the detection of acidified endosomes. Cell preparations were visualized in fluorescence microscope (Leica DM2000) or confocal microscope (Spinning Disk Andor Revolution Confocal System, Ireland) in a PLAPON 60×O/TIRFM-SP, NA 1.45 and UPLSAPO 100XO, NA 1.4 objectives (Olympus). The percentage of NET releasing cells was determined by the evaluation of 200 cells in a double-blind experimental procedure. NET Isolation {#s2f} ------------- For protein detection/quantification on NETs, NET proteins were purified using DNase treatment and precipitated as previously described [@pone.0045427-Mitroulis3], [@pone.0045427-Urban1]. NET structure isolation was performed by vigorous pippetting and sequential centrifugation, as previously described [@pone.0045427-Saffarzadeh1]. Western Blot Analysis {#s2g} --------------------- For TF detection, western blotting was performed in precipitated samples as previously described [@pone.0045427-Kambas1]. A rabbit anti-human polyclonal ab (Santa Cruz) was used for TF detection and a rabbit anti-human polyclonal ab (Trevigen, MD, USA) for GAPDH. A goat anti-rabbit HRP conjugated polyclonal ab (R&D Systems, MN, USA) was used as secondary. To evaluate the expression of TF in NETs, equal amounts of total protein from each sample were loaded onto the gel to eliminate a potential discrepancy, due to the absence of a reference protein. For p62/SQSTM1 detection a mouse anti-human monoclonal ab (Santa Cruz) was used. TAT Complex ELISA Assay {#s2h} ----------------------- Thrombin generation was studied in culture supernatants from healthy and neutrophils from patients with sepsis, treated for 3 h. Culture supernatants were collected and centrifuged for 10 min at 300×g. In the case of healthy neutrophils cultured *in vitro* in the presence of sepsis serum, culture medium was removed after 1 h of incubation, cells were washed twice with 1 ml pre-warmed RPMI and incubated for 2 h in fresh culture medium contained 6% serum from healthy donor. Considering that sepsis serum contains increased levels of thrombin, this step ensured the measurement of only NET-dependent thrombin generation. Afterwards, culture supernatants were collected and centrifuged at 300×g to remove any cells and at 16000×g to eliminate cellular debris. Isolated NETs structures were added in RPMI with 6% serum for 30 min. The concentration of thrombin-antithrombin (TAT) complex was measured according to manufacturer's instructions (Assaypro, Winfield, MO, USA). The minimum detectable dose of the assay was 300 pg/ml. Platelet Activation Studies {#s2i} --------------------------- Platelets were isolated from platelet rich plasma obtained from adult healthy donors as previously described [@pone.0045427-Savi1], were suspended in HEPES buffer (200 µl) and incubated with the respective stimuli for 30 min at 37°C. Afterwards, 100 µl of these suspensions were stained with either anti-CD62P-PE antibody (Beckman Coulter, Switzerland) or annexin V-FITC antibody (BD Biosciences). Platelets were immediately diluted in 500 µl of HEPES buffer and analyzed by flow cytometry (BD Biosciences, FACScan). Analyses were performed on 10000 gated platelets in Cell Quest program (BD Biosciences). RNA Isolation, cDNA Synthesis and Quantitative Real-time PCR {#s2j} ------------------------------------------------------------ RNA isolation, cDNA synthesis and TF qRT-PCR (primers and conditions) were performed as previously described [@pone.0045427-Kambas1]. Relative expression levels were determined using the 2^DDCT^ method. Statistical Analysis {#s2k} -------------------- Values are presented as mean ± SD. Mann-Whitney U-test or Wilcoxon test was used to assess the statistical significance of unpaired or paired samples (n≥6), respectively. For samples with n≥3 a Student's t test was used. All statistical analyses were performed with GraphPad Prism (GraphPad Software, Inc., San Diego, CA). *P* values of ≤.05 were considered significant. Results {#s3} ======= Septic Inflammatory Environment Induces NET Release in an Autophagy-dependent Manner {#s3a} ------------------------------------------------------------------------------------ The previously described implication of autophagy in NET release [@pone.0045427-Remijsen1], [@pone.0045427-Mitroulis3] prompted us to investigate whether autophagy inhibition attenuates the release of NETs from neutrophils derived from patients hospitalized with gram-negative sepsis (sepsis neutrophils). These cells released NETs after 3 h of incubation, as demonstrated by co-staining with myeloperoxidase (MPO) and DAPI ([Fig. 1A](#pone-0045427-g001){ref-type="fig"}, [Fig. S1](#pone.0045427.s001){ref-type="supplementary-material"}). Notably, treatment of sepsis neutrophils with 3 methyl-adenine (3-MA), a modulator of class III PI3Ks that inhibits autophagy, or bafilomycin A1, which impairs the formation of acidified autophagosomes, abrogated NET release ([Fig. 1A](#pone-0045427-g001){ref-type="fig"}, [Fig. S1](#pone.0045427.s001){ref-type="supplementary-material"}). Moreover, incubation of neutrophils from control subjects (control neutrophils) with serum from patients with sepsis resulted in NET release in a time-dependent manner ([Fig. 1B--C](#pone-0045427-g001){ref-type="fig"}, [Fig. S1](#pone.0045427.s001){ref-type="supplementary-material"}), which was also inhibited by 3-MA or bafilomycin A1 ([Fig. 1B](#pone-0045427-g001){ref-type="fig"}). Treatment of sepsis serum with an LPS inhibitor (Polymyxin B) did not affected the percentage of NET releasing cells ([Fig. 1B](#pone-0045427-g001){ref-type="fig"}), suggesting that the presence of LPS in serum from patients with sepsis was not critical for NET formation. {#pone-0045427-g001} TF is Localized in NETs Released by Sepsis Neutrophils {#s3b} ------------------------------------------------------ The critical role of blood-born TF in the pathogenesis of sepsis [@pone.0045427-Pawlinski1], [@pone.0045427-Mitroulis1] motivated the investigation of TF presence in NETs formed by sepsis neutrophils. TF was identified in NETs released from these cells ([Fig. 2A--B](#pone-0045427-g002){ref-type="fig"}) and control neutrophils treated with sepsis serum ([Fig. 2A--B](#pone-0045427-g002){ref-type="fig"}), as assessed by confocal microscopy and immunoblotting of NET-derived proteins. As expected, 3-MA ([Fig. 2B](#pone-0045427-g002){ref-type="fig"}) or bafilomycin A1 ([Fig. 2A](#pone-0045427-g002){ref-type="fig"}) attenuated NET formation and the consequent TF externalization. {#pone-0045427-g002} To further investigate the possible effect of 3-MA or bafilomycin A1 on intracellular TF levels, sepsis neutrophils or control neutrophils treated with sepsis serum were incubated in the presence of 3-MA or bafilomycin A1 for 1 h. This incubation period was selected to avoid the possible implication of TF externalization through NETs in the quantification of TF levels. TF expression levels in sepsis neutrophils ([Fig. 2C](#pone-0045427-g002){ref-type="fig"}) and control neutrophils treated with sepsis serum ([Fig. 2D](#pone-0045427-g002){ref-type="fig"}) were not altered in the presence of 3-MA or bafilomycin A1, suggesting that these agents do not interfere with the expression of TF in neutrophils. To address whether *de novo* TF production contributes to enhanced TF expression in neutrophils, microparticle-depleted sepsis serum was acquired by centrifugation ([Fig. S2](#pone.0045427.s002){ref-type="supplementary-material"}). Treatment with microparticle-depleted sepsis serum still induced TF expression, as observed by immunofluorescence microscopy ([Fig. 2A](#pone-0045427-g002){ref-type="fig"}), immunoblotting ([Fig. 2E](#pone-0045427-g002){ref-type="fig"}) and flow cytometry analysis ([Fig. 2F--G](#pone-0045427-g002){ref-type="fig"}). Inflammatory Mediators Involved in Sepsis Induce the Release of TF-bearing NETs {#s3c} ------------------------------------------------------------------------------- To simulate septic conditions, normal neutrophils were treated with opsonized *E. coli* for 3 h. We observed NET formation from *E. coli* phagocytosing neutrophils, while the addition of 3-MA or bafilomycin A1 30 min after the initiation of phagocytosis completely inhibited NET release ([Fig. 1D](#pone-0045427-g001){ref-type="fig"}, [Fig. S1](#pone.0045427.s001){ref-type="supplementary-material"}). We next studied the localization of TF in NETs produced after bacterial phagocytosis. Confocal microscopy ([Fig. 3A](#pone-0045427-g003){ref-type="fig"}) and immunoblotting of NET-derived proteins did not demonstrate TF presence in NETs released by *E. coli* phagocytosing neutrophils ([Fig. 3A--B](#pone-0045427-g003){ref-type="fig"}). Moreover, immunoblotting did not show elevated TF protein levels in these cell lysates ([Fig. 3C](#pone-0045427-g003){ref-type="fig"}). To further simulate septic conditions and considering the reported implication of TNF-α and IL-1β in the pathomechanism of sepsis [@pone.0045427-Casey1], control neutrophils were stimulated with TNF-α, IL-1β and G-CSF. Even though the percentage of NET-releasing cells was lower compared to *E. coli* phagocytosing neutrophils ([Fig. 1C](#pone-0045427-g001){ref-type="fig"}), we observed TF localization in NETs ([Fig. 3A--B](#pone-0045427-g003){ref-type="fig"}), while cellular TF levels were upregulated ([Fig. 3C](#pone-0045427-g003){ref-type="fig"}). To further enhance NET production, neutrophils were treated with the aforementioned factors in the presence of *E. coli* bacteria. We detected increased release of NETs decorated with TF ([Fig. 3A--B](#pone-0045427-g003){ref-type="fig"}), while cellular TF protein levels were elevated ([Fig. 3C](#pone-0045427-g003){ref-type="fig"}). Notably, *TF* mRNA levels were also found elevated in neutrophils incubated with all the aforementioned inflammatory stimuli ([Fig. 3D](#pone-0045427-g003){ref-type="fig"}), even with *E. coli* alone, suggesting a role for the cytokine mixture in the TF post-transcriptional regulation. {#pone-0045427-g003} Thrombin Generation and PAR Signaling are Triggered by TF-bearing NETs {#s3d} ---------------------------------------------------------------------- We next assessed whether the externalization of TF in NETs is involved in the activation of coagulation system by sepsis neutrophils. TAT complex levels were increased in culture supernatants from sepsis neutrophils ([Fig. 4i](#pone-0045427-g004){ref-type="fig"}) and control neutrophils treated with sepsis serum ([Fig. 4ii](#pone-0045427-g004){ref-type="fig"}) or TNF-α, IL-1β and G-CSF ([Fig. 4iii](#pone-0045427-g004){ref-type="fig"}). TF blockade with anti-TF mAb abolished the aforementioned increase of TAT complex levels ([Fig. 4i--iii](#pone-0045427-g004){ref-type="fig"}), while it had no effect on the production of NETs (data not shown). On the other hand, *E*. *coli* phagocytosis alone was not sufficient for thrombin generation, despite the formation of NETs ([Fig. 4iii](#pone-0045427-g004){ref-type="fig"}). Moreover, treatment with 3-MA or bafilomycin A1 normalized TAT levels in culture supernatants from sepsis neutrophils or control neutrophils treated with sepsis serum or the aforementioned inflammatory agents ([Fig. 4i--iii](#pone-0045427-g004){ref-type="fig"}). Additionally, incubation of control serum with isolated NET structures demonstrated similar results regarding thrombin generation. NET structures from control neutrophils treated with cytokines and *E.coli* induced TAT complex levels, while NETs from neutrophil treated only with *E.coli* did not ([Fig. 4iv](#pone-0045427-g004){ref-type="fig"}). This ability was also TF dependent, as suggested by the inhibitory effect of anti-TF neutralizing antibodies ([Fig. 4iv](#pone-0045427-g004){ref-type="fig"}). {#pone-0045427-g004} Platelet activation by thrombin was further investigated, in order to study its ability for signaling through PAR-1. Platelets treated with culture supernatant from sepsis neutrophils ([Fig. 5i](#pone-0045427-g005){ref-type="fig"}) or normal neutrophils incubated with sepsis serum ([Fig. 5ii](#pone-0045427-g005){ref-type="fig"}) or the aforementioned cytokines ([Fig. 5iii](#pone-0045427-g005){ref-type="fig"}) expressed increased levels of CD62P and annexin V, as assessed by flow cytometry. Inhibition of NET release with bafilomycin A1 abolished this effect ([Fig. 5i--iii](#pone-0045427-g005){ref-type="fig"}). TF blockade in neutrophil cultures and PAR-1 inhibition with FLLRN in culture supernatants also inhibited CD62P and annexin V expression in platelets ([Fig. 5i--iii](#pone-0045427-g005){ref-type="fig"}). Moreover, culture supernatant from *E. coli* phagocytosing neutrophils had no effect in the expression of these markers in platelets ([Fig. 5iii](#pone-0045427-g005){ref-type="fig"}). {#pone-0045427-g005} To compare NET-forming neutrophils with apoptotic ones in their ability to induce thrombin generation, control neutrophils were incubated with sepsis serum for 1 h, washed and incubated for 2 h in normal serum. To induce apoptosis, cells were irradiated after being treated with sepsis serum. TAT levels in culture supernatants from irradiated cells were significantly lower compared with those from non-irradiated cells treated with sepsis serum and were comparable to neutrophils treated with normal serum ([Fig. 4v](#pone-0045427-g004){ref-type="fig"}). Moreover, culture supernatants from apoptotic cells did not induce significant elevation of CD62P levels in platelets ([Fig. 5iv](#pone-0045427-g005){ref-type="fig"}). Localization of TF in LC3 Positive Structures {#s3e} --------------------------------------------- To further assess whether autophagy is involved in the delivery of intracellular TF to NETs, the localization of TF and the LC3B was scrutinized in sepsis neutrophils. After 1 h of incubation, we observed formation of TF aggregates, which were colocalized with LC3B-coated structures ([Fig. 6A](#pone-0045427-g006){ref-type="fig"}, [Fig. S5](#pone.0045427.s005){ref-type="supplementary-material"}), as assessed by confocal microscopy. Inhibition of autophagy by 3-MA attenuated the aggregation of LC3B and resulted in a more disperse TF staining ([Fig. 6A](#pone-0045427-g006){ref-type="fig"}). Additionally, control neutrophils were treated with sepsis serum for different time points and the localization of TF and LC3B were studied. We detected minimal TF levels after 5 min of stimulation, which were up-regulated at 30 min. TF colocalization with LC3B-positive structures was observed at 1 h of stimulation ([Fig. 6B](#pone-0045427-g006){ref-type="fig"}, [Fig. S3A](#pone.0045427.s003){ref-type="supplementary-material"}) and NETs decorated with TF were detected after 3 h ([Fig. 6B](#pone-0045427-g006){ref-type="fig"}). Of interest, TF colocalization with LC3B was observed in the cytoplasm, and partially in NETs, of NET releasing cells ([Fig. 6B](#pone-0045427-g006){ref-type="fig"}). The same phenomenon was also observed in control neutrophils simultaneously treated with inflammatory mediators and *E. coli* ([Fig. S3B](#pone.0045427.s003){ref-type="supplementary-material"}). As expected, neutrophils treated with *E. coli* alone exhibited increased formation of LC3B positive structures, while TF levels remained unaltered ([Fig. S3B](#pone.0045427.s003){ref-type="supplementary-material"}). {#pone-0045427-g006} Next, we studied whether the engulfment of proteins localized in NETs in LC3B-coated endosomes is a generalized phenomenon for NET targeting. Colocalization of high mobility group box-1 (HMGB1) with LC3B-positive structures was observed in neutrophils treated with sepsis serum following the same pattern with TF ([Fig. 7A, C](#pone-0045427-g007){ref-type="fig"}). On the other hand, we did not detect any colocalization of MPO with LC3B positive structures under these conditions ([Fig. 7B--C](#pone-0045427-g007){ref-type="fig"}), suggesting that MPO, which is located in the primary granules, does not follow the same route to reach the NETs. {#pone-0045427-g007} Moreover, we investigated whether the observed TF and HMGB1 containing autophagosomes matured to autophagolysosomes. The LC3-coated structures were stained positive with LysoTracker, suggesting the formation of acidified autophagosomes ([Fig. 8A](#pone-0045427-g008){ref-type="fig"}, [Fig. S4A](#pone.0045427.s004){ref-type="supplementary-material"}, [Fig. S5](#pone.0045427.s005){ref-type="supplementary-material"}). As expected, incubation with bafilomycin A1 attenuated the acidification of autophagosomes ([Fig. 8A](#pone-0045427-g008){ref-type="fig"}). {#pone-0045427-g008} To verify the induction of autophagic machinery in our model, p62/SQSTM1 immunoblotting was utilized. We observed degradation of p62/SQSTM1 in control neutrophils treated with sepsis serum in a time dependent manner, while inhibition of autophagolysosomal fusion by bafilomycin A1 abolished this phenomenon ([Fig. 8B](#pone-0045427-g008){ref-type="fig"}). Discussion {#s4} ========== In the present study, we provide evidence for the release of TF through NETs by neutrophils isolated from patients with sepsis. Moreover, it is demonstrated that an autophagy-based mechanism is implicated in the extracellular localization of TF in NETs. This form of TF induces *in vitro* thrombin generation and subsequent PAR-1 signaling, suggesting a possible involvement in coagulopathy and inflammatory responses that characterize sepsis. The activation of coagulation is a critical step in the continuum of events associated with sepsis. It participates *via* fibrin deposition in the prevention of microbial dissemination [@pone.0045427-Luo1], [@pone.0045427-Flick1]. However, the activation of coagulation and the formation of thrombi in the microvasculature may also lead to disseminated intravascular coagulation (DIC), resulting in multi-organ failure [@pone.0045427-Opal1], [@pone.0045427-Levi1]. Studies in both animal and human models suggest the implication of circulating TF in the progression towards DIC [@pone.0045427-Lupu1], [@pone.0045427-Shimura1]. Except from increased thrombogenicity, TF/thrombin/PARs pathway participates in the induction and maintenance of systemic inflammatory response during sepsis [@pone.0045427-Taylor1]--[@pone.0045427-Creasey1]. Previous studies implicated neutrophils in thrombogenicity. Neutrophil depletion attenuated thrombus formation in a mouse model of deep vein thrombosis [@pone.0045427-vonBrhl1] and plaque formation in atherosclerosis [@pone.0045427-Zernecke1]. Even though *de novo* TF production [@pone.0045427-Ritis1]--[@pone.0045427-Redecha1] or acquisition of microparticle-derived TF [@pone.0045427-Egorina1] is still a matter of debate, *ex vivo* and *in vivo* studies demonstrated that TF in neutrophils plays a significant role in the induction of both coagulation and inflammation [@pone.0045427-Ritis1]--[@pone.0045427-Redecha1]. However, the contribution of neutrophils in circulating TF levels, responsible for thrombotic or non-thrombotic disease manifestations, has not been completely elucidated. In this study, we describe a novel role for neutrophils in TF-dependent thrombogenicity. Even though the acquisition of microparticle-derived TF by neutrophils has been previously demonstrated [@pone.0045427-Egorina1], in this *ex vivo* model the inflammatory milieu of sepsis induced *de novo* TF production by neutrophils, as indicated by the effect of microparticle-depleted serum. Moreover, neutrophils treated with inflammatory mediators also expressed TF in mRNA and protein levels. However, *E. coli* phagocytosis in the absence of inflammatory mediators was not sufficient for TF production, despite the observed upregulation of *TF* mRNA levels. This finding is in accordance with the previously reported inability of lipopolysaccharide-treated granulocytes to produce TF [@pone.0045427-Egorina1]. The extracellular delivery of TF in NETs resulted in thrombin generation in culture supernatants and subsequent signaling through PAR-1. The key role of TF in NET-dependent activation of coagulation system was indicated by the inhibitory effect of anti-TF mAb. This was also supported by the lack of thrombin generation in culture supernatants from *E. coli* phagocytosing neutrophils, which released NETs that were not decorated with TF. Additionally, the generation of thrombin by isolated NETs demonstrates that TF localized in NETs is functional. The ability of NET-forming neutrophils to generate thrombin was further juxtaposed with that of apoptotic cells. Induction of apoptosis in neutrophils treated with sepsis serum prevented thrombin generation in culture supernatants. These findings propose that the extracellular delivery of TF through NETs plays a key role in the activation of coagulation system. The aforementioned contribution of NET generation to thrombus formation has been previously studied. It has been shown that platelet-neutrophil interaction in the microvasculature results not only in NET formation but also in the obstruction of blood flow and tissue damage [@pone.0045427-Fuchs1]. Recent studies demonstrated a role for NET formation in sepsis-associated vasculopathy, due to platelet entrapment and activation [@pone.0045427-Brill1]. Of interest, during the preparation of this article, a role for NETs in thrombus formation due to FXII activation has been demonstrated in a mouse model of deep vein thrombosis [@pone.0045427-vonBrhl1]. In this mouse model, TF was observed in NETs, which is in consistency with our study. We also observed the implication of an autophagy-related pathway in both NET release and protein trafficking on NETs. PI3K signaling and endosomal acidification inhibition abrogated NET release, suggesting the involvement of autophagy, which is in accordance with previous reports [@pone.0045427-Remijsen1], [@pone.0045427-Mitroulis3]. We also demonstrated the involvement of an autophagy-related mechanism in the delivery of TF in NETs. TF was accumulated in LC3B-coated vacuoles and then externalized and localized in NETs. These vacuoles were also identified as acidified autophagosomes by staining with LysoTracker. This pathway also participated in the delivery of HMGB1 to NETs. This nuclear protein, which is implicated in the pathogenesis of sepsis [@pone.0045427-Klune1], is translocated to the cytosol and released through NETs upon neutrophil activation [@pone.0045427-Mitroulis3], [@pone.0045427-Gardella1]. On the other hand, MPO, a granule protein, did not use this particular pathway. These findings suggest that the inclusion of certain proteins into LC3B-coated endosomes, either integral membrane proteins, like TF, or cytosolic, such as HMGB1, facilitates their delivery to NETs. Since no detectable degradation is observed inside the autophagosomes, at least for the time-scale of our experiments, a key role for an autophagy-based pathway is proposed as an unconventional trafficking pathway to NETs. However, proteins expressed in primary granules, such as MPO, could follow alternative routes to NETs. An autophagy-based pathway for TF delivery to NETs is in accordance with recent studies demonstrating autophagy-based pathways being involved in the secretion of cytosolic proteins. It has been recently reported that the release of acyl-CoA-binding protein by murine astrocytes [@pone.0045427-Loomis1] and IL-1β and HMGB1 by bone marrow derived macrophages is based on an autophagy-dependent secretory pathway through LC3B positive organelles [@pone.0045427-Dupont1]. In conclusion, our data, in conjunction with previous studies, provide novel insight into the implication of neutrophil derived TF in human sepsis thrombogenicity, describing a novel form of circulating TF. Activated neutrophils, expressing high amounts of TF, are recruited in vast numbers in the vasculature during sepsis. The release of TF-bearing NETs at the sites of inflammation may result in the localized activation of coagulation cascade, leading to microangiopathy and activation of several cell populations, including platelets and endothelial cells, through PAR signaling. Moreover, an autophagy-related pathway emerges as a critical component in neutrophil function during sepsis by regulating the translocation of certain neutrophil proteins to NETs, including TF and HMGB1. Targeting NET formation and/or protein trafficking through autophagy could emerge as a potential strategy in the treatment of coagulopathy that characterizes sepsis. Supporting Information {#s5} ====================== ###### **NET formation by septic neutrophils and control neutrophils treated with inflammatory stimuli.** NET release by untreated neutrophils from patients with sepsis (Septic PMNs) and control neutrophils treated with septic serum (Septic serum) or phagocytosing opsonized *E. coli* bacteria (E. coli) alone or in the presence of a mixture of TNF-α, IL-1β and G-CSF (Cyto/E.coli) after 3 h incubation, assessed by immunofluorescence microscopy. Inhibition of NET formation in neutrophils treated with 3-MA (3MA) or bafilomycin A1 (Bafil). Untreated neutrophils from healthy subjects were used as control. One representative out of six independent experiments is shown (DNA labeled with DAPI; blue, anti-MPO monoclonal antibody; green) (original magnification 400×). Scale bar represents 30 µM. (TIF) ###### Click here for additional data file. ###### **Microparticle depletion from sepsis serum.** FACS analysis of sepsis serum and microparticle depleted sepsis serum by centrifugation as indicated by Annexin V-FITC staining. Isotype: sepsis serum stained with mouse IgG1-FITC. One out of three independent experiments is shown. (TIF) ###### Click here for additional data file. ###### **Localization of TF or MPO in LC3B-coated structures in control neutrophils treated with inflammatory stimuli.** (**A**) Single plane analysis of TF or MPO localization in LC3B-coated structures in control neutrophils treated with septic serum for 60 mins. (DNA labeled with DAPI; blue, anti-TF monoclonal antibody or anti-MPO monoclonal antibody; green, anti-LC3B mAb; red) (original magnification 1000×). Arrows indicate LC3-coated autophagosomes. Scale bar represents 5 µM. (**B**) TF and LC3B localization in control neutrophils treated with inflammatory cytokines and/or *E. coli* at various time points. Colocalization of TF with LC3B is detected both intracellularly and in NETs only in neutrophils treated with both cytokines and *E. coli*. Neutrophils treated with *E. coli* alone do not express TF. (DNA labeled with DAPI; blue, anti-TF mAb; green, anti-LC3B mAb; red) (original magnification 1000×). One out of three independent experiments is shown. Scale bar represents 5 µM. (TIF) ###### Click here for additional data file. ###### **Localization of HMGB1 or MPO in late autophagosomes in control neutrophils treated septic serum.** Control neutrophils were treated at different time points with septic serum and the colocalization of HMGB1 (A) and MPO (B) with autophagolysosomes was assessed. Autophagolysosomes were visualized by confocal microscopy as LC3B and LysoTracker double positive structures (z stack analysis, 0.3 µm per plane). (DNA labeled with DAPI; blue, anti-LC3B mAb; white, anti-MPO or anti-HMGB1 mAB; green, LysoTracker; red) (original magnification 1000×). One out of three independent experiments is shown. Scale bar represents 5 µM. (TIF) ###### Click here for additional data file. ###### **Comparison of z-stack with single plane analysis in colocalization studies.** z-axies and Single Plane analysis of TF and LC3 in sepsis PMNs or control PMNs treated with sepsis serum. Scale bar represents 5 µM. (TIF) ###### Click here for additional data file. The authors would like to thank Dr G. Kolios and Dr E. Chatzaki for providing microscopy facilities and technical support. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: IM KR. Performed the experiments: KK IM EA AC I. Kotsianidis AG. Analyzed the data: KK AC I. Kourtzelis. Contributed reagents/materials/analysis tools: MK AG. Wrote the paper: KK IM I. Kourtzelis KR. Patient supervision and clinical samples: IM PS IP. Extensive scientific discussion: KK IM EA AG AC IP PS I. Kourtzelis MK I. Kotsianidis KR.

{#sp1 .667} {#sp2 .668}

Background {#Sec1} ========== Strengthening midwives' professional role is a cornerstone in the process of improving the quality of maternal and newborn health care services and reducing maternal and neonatal mortality \[[@CR1]\]. It is at the heart of many international strategic directions to support the achievement of Millennium Development Goals (MDGs) 4 (reduce infant mortality) and 5 (improve maternal health) \[[@CR1]--[@CR3]\]. Indeed, more than 80 % of maternal deaths, stillbirths, and neonatal deaths could be prevented by midwifery interventions targeting reproductive, maternal and newborn health \[[@CR4]\]. To tackle this challenge, calls to actions have been made in African countries to implement skill development interventions for health professionals in general and for midwives in particular \[[@CR2]\]. Role strengthening is one of the main interventions being considered to provide high-quality care through expanding human resources' skills, tasks and responsibilities beyond their traditional scope of practice \[[@CR5]\]. It has been also conceptualized as a whole systems change implicating essential actions in many systems, including the sociocultural, and educational systems as well as the system pertaining to the profession itself[1](#Fn1){ref-type="fn"} \[[@CR6]\], through which a health professional role (namely the midwife's role) evolves. However, it is well documented that a health workforce intervention is a multisystem intervention involving coordinated efforts at multiple levels and affecting different actors \[[@CR7], [@CR8]\]. Therefore, it is contingent upon the context in which it is implemented \[[@CR5], [@CR8]\] and runs the risk of not being implemented as intended, compromising the achievement of the intended goals \[[@CR9]\]. Consequently, implementation challenges must be studied in order to fine-tune the intervention and to provide timely solutions for increasing the chances of producing desired effects \[[@CR10], [@CR11]\]. Indeed, international recommendations such as those of the State of the World Midwifery report in 2011 emphasize that scientific evaluations of interventions targeting the midwifery workforce should be conducted. To date, few evaluation and implementation studies have addressed midwifery interventions and, more specifically, interventions aiming to strengthen midwives' professional role. A recent systematic review has reported that three main challenges were encountered when implementing task-shifting midwifery interventions in high, middle and low income countries and must be considered in the future to ensure a successful implementation \[[@CR7]\]. They relate to: defending the midwifery model of care; training, supervision and support; teamwork and task shifting. The focus of this review was mainly the identification of implementation factors associated with task-shifting interventions covering various forms of vertical (e.g., from doctors to midwives) and horizontal (e.g., from midwives to nurses) shifts and shifts involving many types of human resources among whom tasks can be shifted. The challenges for extending midwives' skills highlighted in this review mainly involved complex/illness-related tasks or a wider range of other tasks such as management of gestational diabetes mellitus, genetic and cervical cancer screening and abortion services. While identifying barriers and facilitators to implementing all forms of task-shifting seems useful, it does not explain specific issues, such as how to best to implement a particular type of skill development intervention, within the health workforce management strategy, to strengthen a professional role. With regards to deepening our understanding of the implementation process of interventions related to strengthening the midwifery role and, specifically, those adopting a systems approach[2](#Fn2){ref-type="fn"} adapted to the context in which the midwifery professional role evolves, the literature falls short. Further research is needed in this area to inform policy makers' decisions and to tailor interventions to the factors that have been identified in the context under investigation \[[@CR12]\]. To address this need, this paper presents findings from an implementation analysis of an action plan (AP) to strengthen midwives' professional role in Morocco with the aim of reducing the scientific knowledge gap on role strengthening interventions. The objectives of this study are to: 1) assess the extent to which the intervention was delivered; and 2) understand how the possible barriers and facilitators influenced the AP's implementation in Morocco. Intervention background {#Sec2} ----------------------- Morocco is recognized as one of the North-African countries that have made maternal health a top priority and exerted substantial efforts to achieve MDG 5. In 2008, the Ministry of Health (MOH) in Morocco launched a national action plan to reduce the maternal mortality ratio from 227 deaths per 100 000 live births to 50 per 100 000 by 2012. It included three main pillars: i) reducing barriers to access to emergency obstetric care and improving the availability of qualified personnel; ii) improving the quality of care during pregnancy and delivery; and, iii) improving governance and management \[[@CR13]\]. A central component of the first pillar, was strengthening the competencies of health professionals, including midwifery competencies in terms of enhancing: i) midwifery pre-service education through reviewing the midwifery curriculum according to the competency-based approach (CBA); and, ii) in-service training by providing a continuing education program \[[@CR14]\]. To fulfill this commitment, the MOH with the support of the United Nations Population Fund (UNFPA) collaborated with the Université de Montréal. An assessment of the state of midwifery in Morocco revealed that the profession faced substantial challenges related to education and other issues that need to be addressed. They pertained to: i) the lack of congruence of the curriculum with international standards; ii) a negative image of midwifery within the professional community and the society at large; iii) an old legal framework governing the profession resulting in an ill-defined scope of practice which is not built upon the ICM international definition and does not allow midwives to provide women-centered care; and, iv) the unfavorable practice environment \[[@CR15]\]. In sum, it showed that midwives were trained three years at the undergraduate level with an underlying philosophy focusing on illness and a medical model rather than on wellness and comprehensive health care approach. The educational system prepared them to work as \"technicians of birth\" according to a biomedical model that shapes their practice, in a sociocultural system that does not value their status \[[@CR15]\]. Following this assessment, a researcher from the Université de Montréal, developed an action plan (AP), aiming to strengthen the midwife's professional role, consisting of many activities in addition to the educational activities initially prescribed by the MOH. Such a plan would enable midwives to fulfill their role in relation to the international definition of the midwife \[[@CR16]\] in order to better meet the needs of Moroccan women and their families. It was designed in terms of the following objectives \"see Additional file [1](#MOESM1){ref-type="media"}\": i) educational objectives (e.g., developing and implementing a competency-based education program (CBE); ii) professional objectives (e.g., revitalizing the midwives association); iii) sociopolitical, legal objectives (e.g., setting-up marketing activities). The main objective of the AP was to provide qualified midwives functioning in an enabling environment in order to improve the quality of maternal and newborn health care delivery and outcomes (Hatem, M: Mission report Technical assistance for the revision of the midwifery education program in Morocco, unpublished). The conceptual framework {#Sec3} ------------------------ The framework \[[@CR17]\] builds on Hatem-Asmar et al.'s conceptual model \[[@CR6]\] that describes change in a health professional role and on Damschroder et al.'s Consolidated Framework for Implementation Research (CFIR) for implementation analysis \[[@CR18]\]. Hatem-Asmar et al. consider three systems (*sociocultural, educational, professional*), including their dimensions (*values, methods, actors, targets*), to have an interactional relationship within which a health professional role operates; these should be considered as a whole system for addressing a health professional role change. Damschroder et al.'s CFIR provides a taxonomy of constructs, that are likely to influence implementation of complex interventions, organized into five domains: 1) Intervention characteristics; 2) Outer setting; 3) Inner setting; 4) Characteristics of individuals; and 5) Process of implementation. The Hatem-Asmar model served to identify the context in which the implementation process took place, referring to the three systems targeted by the intervention. Whereas, the CFIR's \"menu of constructs\" was used as a framework to analyze and organize the findings related to the implementation process (Executing), to the influential factors at play in the three systems and within the intervention by inductively mapping themes onto the CFIR constructs. Constructs of two domains (Inner and Outer settings) were applied to the dimensions of the three systems as in our case there is no single set of Inner/Outer settings due to the complex nature of the interrelated systems under study. The characteristic of our framework is that it considers a set of dimensions in each system, arguing that barriers or facilitators identified according to the CFIR'S \"menu of constructs\" could exist in these dimensions interacting with the intervention which should be examined in order to understand the implementation process. Methods {#Sec4} ======= Study design {#Sec5} ------------ A case study design was chosen to conduct an implementation evaluation. The *case* was the midwifery profession undergoing the implementation process in Morocco and the *embedded units* of the case were the *sociocultural, educational, and professional* systems targeted by the intervention \[[@CR19]\]. Study settings and participants {#Sec6} ------------------------------- We conducted the study in settings affiliated to the MOH across two regions of Morocco involved in the implementation process. Using purposeful sampling, we selected stakeholders pertaining to the three systems under investigation. Participants were recruited in two stages: 1) we sent invitations to participate via email along with informational letters to potential key informants pertaining to the above mentioned systems using address lists provided by the training division of the MOH; posters were used to recruit midwifery students; and, 2) site visits were conducted by the first author in collaboration with the project coordinator to select key informants for focus groups. As such, key informants from educational institutes, health services, and governmental departments were approached in two regions: one located at the north of Morocco (Tétouan) and the second at Morocco\'s capital (Rabat). The reason for selecting those sites was that some activities of the AP were being actively implemented in the educational settings located in these regions. This would, consequently, allow for data collection from participants attending the training and the CBE program activities. Nevertheless, we were able to achieve variation on a range of views as midwifery educators attending the sessions came from a broad range of educational institutes (10) from across ten Moroccan regions. This process resulted in a total of two educational institutes, two health centers, two birthing homes, two maternity hospitals, the directorate of population, the training division, and two regional delegations of the MOH and the United Nations Population Fund's (UNFPA) office taking part in the study. Our final sample consisted of 107 informants (106 key Moroccan informants and one international consultant). Table [1](#Tab1){ref-type="table"} presents characteristics of the participants.Table 1Characteristics of the participants in the study (*n* = 107)Job titleNSexWork experience Mean(range)/years^a^Rural setting experienceUrban setting experienceYesYesMale n(%)Femalen (%)*N* = 83(%)*N* = 83(%)Midwife practitioner1701711.29 (1--25)1617Nurses51420 (9--28)25Obstetricians6517.66 (1--16)36Physicians21124.512Medical directors22023.512Senior nurse managers2111622Midwifery managers20226.522Midwifery representatives20233.512Midwifery educators2902915.25 (6--32)2429Midwifery students15015N/AN/AN/AAcademic directors21132.512Academic management staff43128.5 (24--37)44Medical administrative officers76119.85 (15--29)77Administrative nurse cadres20223.512Health programmer11016^a^11Women808N/AN/AN/AConsultant10129^a^N/AN/AN10721(19.63 %)86(80.37 %)66(79.52 %)100 %^a^Work experience in years; *N/A* not applicable Data collection {#Sec7} --------------- We used qualitative methods to assess how well the AP was executed and to identify barriers and facilitators to implementation success. Data were collected through: focus groups (FG) (*n* = 20); individual interviews (*n* = 11) as well as from field notes, observation of educational sessions (30 h) and documents related to the implementation process (*n* = 16). Information about the quality of the implemented training activities was gathered post-hoc from session reports. By using a 1--5 point scale (ranging from very unsatisfied to very satisfied), satisfaction was assessed regarding five dimensions (objectives, content relevance, themes and presentations, acquisition of new knowledge about the midwifery role and practice in reproductive health care, organization of the training sessions). Data was collected in June and July 2010, approximately 18 months after the AP's launch. Selection of a range of key-informants was based on either their involvement in the activities' implementation, or their role as health professionals or medical administrative officers who were exposed to the implementation efforts and were affected by the AP's implementation. All interviews were conducted by the first author, a midwife, who was supported by a local assistant for participants who refused to be recorded. We used an interview guide, based on our conceptual framework \"see Additional file [2](#MOESM2){ref-type="media"}\". The questions covered the: level of information regarding the AP and the activities run; the activities' level of implementation (e.g., the content of the ongoing activities, stakeholders' involvement); and factors believed to be facilitating or hindering the implementation. The FG size varied from two to 10 members. We made sure groups were homogeneous to avoid mixing informants of different professional and organizational hierarchy. All interviews were conducted in French (only pregnant and postpartum women who received midwifery services were interviewed in Arabic). The FG lasted 50 to 90 min, individual interviews ranged from 40 to 60 min. The subjects\' verbalizations were audio-taped in one region while a few stakeholders refused to be recorded in the other region. Health system's context, fear of being judged were among the main reasons for refusal. Detailed notes using abbreviations were written up for eight un-taped interviews which were reconstructed by the first author immediately after the interviews. Moreover, the researcher checked carefully the notes after the interview with each participant to ensure the accuracy of the recorded information. Direct observation of educational sessions took place at the validation midwifery's program and train-the-trainers sessions, to examine the interaction between the participants, the coordinator, the consultant and the head of the training division and to assess the influence on the implementation process. All documents were requested from the consultant and the coordinator and were obtained except for the continuing education program's report (Advances in Labor and Risk Management (ALARM)[3](#Fn3){ref-type="fn"} report) which was held by the Canadian instructors. Documents entailed e-mail correspondence between the international consultant, the coordinator and the training division (*n* = 10), project reports (*n* = 6) (UNFPA letters, session reports). As no other activities (e.g., profile, social marketing activities) were implemented, there were no data to be analyzed. Due to time constraints, analysis through an iterative process was not done; nevertheless, detailed notes were read throughout the data collection phase in order to gain a sense of the data prior to proceeding to further interviews. Data analysis {#Sec8} ------------- The analysis was conducted combining deductive and inductive approaches using QDA Miner software \[[@CR20]\]. We used an initial coding list based on codes corresponding to the system's four dimensions (*values, methods, actors, targets*) and the intervention characteristics suggested in our framework. We applied this coding scheme to the transcripts and conducted an inductive content analysis using constructivist thinking, allowing themes to emerge. Subsequently, clustered codes forming key themes were compared to CFIR constructs. The analysis proceeded through a longitudinal analysis of each corpus of data followed by a cross-sectional analysis of the entire data. Data reduction \[[@CR20]\] consisted of: aggregating codes into key themes and assigning them to two categories (barriers, facilitators); mapping themes onto the CIFR's constructs. In sum, the main themes were grouped under: intervention characteristics, values, methods, actors, targets of the three systems; and the constructs of the CFIR. Concerning how well the AP was executed, data were organized according to the dimensions of the construct \"Executing\": i) fidelity of implementation (adherence to the planned activities; and coverage: whether the stakeholders who should be participating in the activities actually were) \[[@CR21]\]; ii) intensity (quality) (satisfaction with the activities delivered); iii) timeliness of activity completion (extent to which tasks are completed in the planned timeframe); and, iv) degree of engagement (extent to which individuals are actively engaged in the implementation process). (See Table [2](#Tab2){ref-type="table"} for definition of codes).Table 2Codebook: definition of the codesMega-codesDefinitionValuesThe values, beliefs, principles, laws and rules of the systemMethodsThe organizational procedures used in the system and that include facilitators/barriers to implementation (e.g., communication and coordination mechanisms, availability and use of physical, human, financial resources, training, etc.)ActorsThe actors involved in the system's functioning whose characteristics (e.g. capacities, motivation, attitudes) can facilitate/hinder the implementation processTargetsThe goals of the system which may facilitate/hinder the implementation processImplementation process (Executing)How well the implementation is carried out according to planInterventionCharacteristicsCharacteristics of the intervention perceived to influence the implementation (e.g., advantage, perceived difficulty of implementation)CFIR's constructsDefinitionExecutingExtent to which the activities of the plan were delivered as intended FidelityContent adherence : adherence to the prescribed activitiesCoverage: whether the stakeholders who should be participating in the activities actually do Intensity (quality)The lived experience and the satisfaction towards the implemented activities Timeliness of task completionAdherence to timetable, compliance with deadlines Degree of engagementActive involvement of actors in the implementation processExternal policy and incentivesThe external strategies to spread interventions including: local, national policies, regulations, external mandates, guidelines and recommendations that influenced the decision to implement the interventionPatient needs and resourcesExtent to which patient needs are prioritized by the organizationStructural characteristicsStructural challenges (e.g., social architecture, centralization, maturity, size, etc.) that facilitate/hinder the implementationCultureOrganization\'s beliefs, norms, values that affect the implementationNetworks and communicationsThe nature of social networks and of formal and informal communications within an organizationReadiness for implementationOrganizational commitment to its decision to implement an intervention Available resourcesLevel of resources dedicated for implementation (e.g., money, training, education, physical space, equipment, etc.) Access to knowledge and informationAccess to information, knowledge and materials about the interventionCharacteristics of individuals/Personal attributesPersonal traits such as tolerance of ambiguity, motivation, intellectual ability, competence, capacity (e.g. leadership, involvement of leaders with the intervention to help make implementation successfulIntervention sourcePerceptions of stakeholders about whether the intervention is externally or internally developedComplexityHow complicated the intervention is in: duration, radicalness, number of steps required to implementRelative advantageStakeholders' perception of the advantage of implementing the intervention versus an alternative solution Scientific rigor {#Sec9} ---------------- Data sources were triangulated by cross-checking the different point of views of stakeholders and by seeking several data sources in order to check for consistency \[[@CR22]\]. A summary of the results was also sent back to eight participants for confirmatory purposes. Two co-authors other than the first author were involved in the codification and the matrix review to ensure accuracy of interpretations. One independent researcher provided additional input on the themes. Through discussion, consensus was achieved on the final themes. The findings were shared at the ICM Congress 2014 with a range of midwifery educators and practitioners, midwives representing the Midwives Association and the MOH, who had already participated in the study. This allowed us to obtain their feedback regarding recommendations to improve the implementation process. Ethical considerations {#Sec10} ---------------------- Ethics approval was obtained from the Research Ethics Committee of Université de Montréal, Québec, and the MOH in Morocco. Informed consent was obtained from all participants. Anonymity and data confidentiality were ensured. Results {#Sec11} ======= Seven themes facilitating the implementation and 17 themes describing the barriers across the three systems and the intervention's characteristics were drawn from the findings. Our results are consistent with the constructs of the CFIR domains (22 out of 24 key themes were captured by the CFIR constructs). One notable finding is that 55 of all Moroccan informants (106) were not knowledgeable about the AP due to communication problems influenced to some extent by the structural characteristics of the Moroccan system. They were mainly health practitioners and managers (36/55) (e.g., midwives, nurses, obstetricians, medical directors) working in clinical settings. As access to information was deficient due to a lack of a formal communication strategy, we decided to inform the participants about the AP and the ongoing activities. We aimed, through instructive information, to raise awareness of implementation issues. Interviews were re-conducted and participants were asked to highlight factors that may help or constrain the implementation according to their knowledge of the Moroccan context. Furthermore, interviewees provided their perspectives on strategies to address barriers. As barriers and the potential solutions often mirrored one another, we decided to group both under the same themes. Executing the implementation {#Sec12} ---------------------------- ### Fidelity {#Sec13} Content adherence {#Sec14} ----------------- The implementation analysis has shown a sizeable gap between planned versus delivered objectives. Emphasis has been placed on achieving only the first 3/9 objectives of the AP, related to: i) implementing the ALARM international program; ii) training the trainers on the competency-based approach (CBA); and, iii) developing and implementing a CBE program.The ALARM international program (objective1): Data analysis has shown that only three training workshops were held due to financial constraints. We were unable to assess the adherence to the planned activities and other dimensions due to lack of access to the workshop reports and to lack of well-defined activities related to this objective in the original AP.Train-the-trainers activities and developing and implementing a CBE program (objectives 2 and 3): Four sessions were held and focused on the reformulation of the program modules according to the CBA, and on training activities regarding the clinical (Humanization, sexual and reproductive rights, etc.) and academic concepts (Reflexive approach, etc.) required to prepare health professionals and educators to provide holistic care to the maternal-child population. With regards to objective 2: Data analysis has shown that the training needs of the stakeholders were not identified by a committee as it was planned. They were delineated during the curriculum review sessions following the consultant's request. Content of the training activities was delivered, however, according to what was planned regarding the clinical and academic concepts. Concerning the CBE program (objective 3): Three sessions involving midwifery educators selected by the training division of the MOH led only to the elaboration of the referential for the CBE program and not to its implementation, and this was done mainly by the international consultant. Also, activities carried out drifted in some ways from what was intended. Data analysis revealed that a small committee had been set up including only six midwifery educators from one institute instead of creating a national committee bringing together a variety of midwifery educators from different institutes as originally planned. Moreover, it was dysfunctional due to lack of coordination between the training division, the consultant and the committee members which prevented them from carrying out their activities. Coverage {#Sec15} -------- Data has shown poor coverage as regards to the intended stakeholders who were supposed to attend all the sessions set up for objectives 2 and 3. The selection process done by the training division targeted mainly midwifery educators instead of focusing on actors from the clinical settings and on medical administrative officers involved in policy-making. Restricted funds from UNFPA and the MOH resulted in a failure to include all actors in the process which was reported to compromise the future implementation of the CBA in clinical settings. Moreover, conditions for participation in the ongoing activities were unfavorable. Discontinuity as regards to participating to all implemented sessions was a concern expressed by midwifery educators. An attempt by the MOH to cover all institutions across Morocco has resulted in the selection of different participants at each session. Regular participation involving the same candidates was considered by many educators as crucial for maintaining informational continuity and successfully implementing the CBA's activities. Covering the educational institutes and the clinical settings across the 16 Moroccan regions was stated by almost all participants as crucial to reduce potential resistance in clinical settings and enhance the future adoption of the CBA. Intensity (quality) {#Sec16} ------------------- Our data covered the quality of the delivered educational sessions:The ALARM program: Many participants exposed to this program expressed their satisfaction regarding their positive experience with the training sessions. They found their training very instructive and practice-expanding. They learned to practice on anatomic models and gained a broader knowledge in dealing with obstetrical emergency cases. A medical administrative officer stated that the ALARM training provided participants with evidenced-based management of the causes of maternal mortality. Another academic director linked the training to the improvement of the midwives and physicians' performance.Train-the-trainers activities: Participants reported to be very satisfied and satisfied with regards to the training received in terms of: objectives (16/18 participants), relevance of content (17/18), themes and presentations (16/18), acquiring new knowledge about the midwife's role (17/18) and professional practice (16/18). Analysis of participants\' responses showed they were rather satisfied with the interactive teaching methods used. Dissatisfaction was reported regarding the duration of the session (14/18) and the logistical organization (13/18). However, the content level was sometimes a source of difficulty (4/18) (e.g. critical thinking). Finally, the program's presentation was frustrating given it intensiveness (6/18). Participants emphasized the need to be well prepared for implementing the CBA and to improve the logistical organization of the training sessions. ### Timeliness of activity completion {#Sec17} The findings revealed a significant discrepancy between the actual time frame and the one outlined regarding the training session and the implemented program review activities. Reviewing the program's sessions was postponed several times with a lag of several months between scheduled dates and those during which activities were carried out. Many reasons prevented the sessions from being organized on time such as authorizations from the MOH and administrative procedures related to the budget and to delays in allocating financial resources by UNFPA. Moreover, many participants thought that the training division rushed the overall activities and scheduled them over short periods of time. The large amount of content to cover during a limited period of time (eight hour training sessions over six days) was cited by many participants as a major impediment to understanding. Sessions were held during a busy fall/spring period, making it difficult for midwifery educators to attend the program's activities. Lengthening the training time sessions was recommended by many midwifery educators. ### Engagement of key players in the implementation process {#Sec18} The extent to which participants are engaged in the different phases of the project was assessed through the perceptions of the informants and the document analysis. There was a noticeable lack of a team of individuals actively engaged and sharing the responsibilities for the implementation as it was originally designed. The process was organized in silos between the consultant, the head of the training division and the project coordinator. The consultant was the one leading the implementation of the educational activities, but following the directives of the training division. Thus, it diverged significantly from what was intended with regards to creating committees. Most notably, no activities were designed to engage actors in the process and to share information about the AP and even about the CBE program. Clearly, many participants were not even aware of it; and many others were considered as passive recipients experiencing delayed information about the ongoing activities, invited to execute the training division's orders. Despite the consultant's recommendations, no efforts were made to engage key stakeholders across the clinical settings specifically in order to counteract possible resistance in the future. ### Perceived facilitators and barriers to implementation {#Sec19} Our findings have been categorized in two main categories*: facilitators* (7 themes) *and barriers* (17 themes) to the AP's implementation. Tables [3](#Tab3){ref-type="table"} and [4](#Tab4){ref-type="table"} show the complete list of themes, and representative quotations to illustrate facilitators and barriers according to the constructs of the CFIR in the three-systems; whereas Table [5](#Tab5){ref-type="table"} provides the characteristics of the AP acting as barriers and facilitators to the implementation according to the CFIR constructs.Table 3Facilitators to implementing the action plan in MoroccoMega codesSub-codesThemesIllustrative quotationsSocio-cultural SystemValuesValues of the societyPrimacy of human beings value and respect for human rights*"...a facilitator is that the Moroccan society has changed its thinking over the past 10 years, it is more aware of everything that concerns rights and duties, it is aware of everything related to gender... for example concerning the women's rights"*MethodsExternal policies and incentivesNational and international policies to strengthen midwifery and to reduce maternal death*"The action plan \[the ongoing activities\] stems from the national strategy, which facilitates a lot"*ActorsAttributes/CapacitiesStrategic leadership to strengthen midwifery and to reduce maternal mortality"*I think as regards to the support, the implementation is being taken seriously into consideration, at the strategic level, at the intermediate level and operational level. For the strategic level, this means at the Ministry of Health ... this function can be attributed to the head of the training division who holds key position* "TargetMinistry of Health and international agencies' goalsPrioritizing maternal health in Morocco*"I see that Morocco wants at all costs to follow the global directives... of the millennium goals")*Patient needs and resourcesMultidimensional demands for high quality services for the sake of women's well-being"*There is one environmental factor, it is the people, recipients who have become much more demanding and assertive towards the Ministry of Health, it is a facilitator because it incites the government, the Ministry to be vigilant and to move towards achieving what was stated in the national strategy regarding the reduction of maternal mortality"*Educational SystemValuesN/AMethodsReadiness for implementation/Available resourcesIntensified training support for midwifery teachers"*She passed by groups to ..., to adjust a vision or approach, and to provide guidance for going further in the process*"ActorsN/ATargetsN/A*N/A* not applicable to dataTable 4Barriers to implementing the action plan in MoroccoMega codesSub-codesThemesIllustrative quotationsSocio-cultural SystemValues/Law/RulesStructural characteristicsBureaucratic top-down and centralized organizational structure in Morocco*"The activities are centralized at the training division that deals with implementation, ...,we look after the execution phase, I mean we only take orders from the top and receive information regarding the activities to execute"*MethodsReadiness for implementation/Available resourcesInsufficient financial resources"... *We have 16 areas that need to be covered, few people were able to attend the sessions, not all the staff, because of budget problem*"ActorsCapacities/AttributesA truncated leadership to strengthen midwives*"I think the other activities of the plan other than the training are not for the moment the priority of the policy makers, I mean they don't have any interest or it is not yet official"*AttitudesAttitudes of resistance of physicians towards midwives*"Physicians are not going to ... provide a backup mainly as regards to the activities intended to obtain a council board and change the law, and to have a clear statute"*TargetsN/AEducational SystemValuesCulture/Philosophy of midwifery training and educationA dominant technocratic approach to midwives education and training*"The training model is primarily based on a task approach, we are not prepared for the role and trained to perform a global practice, this principle of training could be a problem, a difficulty to put in place any further intervention"*MethodsReadiness for implementation/Available resourcesLimited availability of human, physical and technical resources for implementing the training activities*"Computers made available to participants were not very helpful, without any protection against virus and software to download documents was not available. Thus, they \[midwives\] had great difficulty to have access to some references ... and make presentation in plenary sessions"*Deficient educational support for midwifery teachers*"As I have mentioned a while ago, the lack of material resources ..., the lack of human skills to support this change, we do not have till date competent midwives who are trained according to the competency-based approach to train the future midwives"*Deficient access to information regarding the action plan*"In general, when there is an action plan, few people are aware of the entire plan, it remains where it was developed at the top of the chain at the policy level .... I think, it's the system here, it does not provide any information..."*Networks and communicationsInadequate mechanisms of coordination and collaboration within and across the educational institutes*"We are all victims of the lack of coordination between our leaders and the educational institutes, we have done a crazy work during the first year ...in September-October 2009, we asked for a feed-back, and we suggested a deadline for receiving feedback, comments were not submitted to the consultant, we even called for a meeting to get feedback, we have not received a response due to the issue of coordination"*ActorsAttributes/CapacitiesDifficulties in adapting teachers skills to the competency-based approach's needs*"If we teachers found it difficult... to understand, what would it be for students who fail in the first year to speak French, are not able to reason properly, students will not be able to understand the new concepts, they will face difficulties in assimilating the various \"savoir, savoir-faire\"*Attributes of students unfavorable to educational changeTargetsN/AProfessional SystemValuesCultureWorkplace culture resistant to innovations*"The change is more difficult, it takes longer time, people\'s mentality is hard to change, they are not open to a new model of care primarily the medical community"*MethodsReadiness for implementation/Professional supportLethargy of the midwifery association towards the profession*"Their presence \[Professional Association\] is symbolic, we do not have a well-developed association, we are not part of it, in my opinion it no longer exists, no activities are done to strengthen the role of the midwife. At Rabat, there are no activities to strengthen the midwife's role"*Deficient access to information as regards to the Action Plan*"Here, at the maternity hospital, we did not receive any information from the Ministry of Health regarding the plan, and the ongoing training activities at the educational institutes"*ActorsAttributesLack of motivation of practicing midwives to embrace change*"We are not motivated, so how do you want to set up a plan, we are like a simple tool for doing tasks, even for planning, for making decisions, for implementing a project, we are not considered nor consulted"*TargetsN/A*N/A* not applicable to dataTable 5Facilitators/Barriers related to the characteristics of the action planMegacodesThemesIllustrative quotationsFacilitatorsRelative advantageMultilevel perceived benefits of the action plan*"Midwives are going to be trained to be autonomous in order to train the society to be autonomous and so to transfer this competency to young girls, to women, children and men*"BarriersComplexityComplex nature of the action plan*"A lot of guidance and support, many steps technically, organizationally, and lots of patience because it is complex and we are speaking here about many stages"*Intervention sourceExternal source of the plan*"The Ministry of Health is concerned with reviewing the program, despite that I \[The consultant\] developed a global plan"* Over all, the results show that barriers generally outnumbered the facilitators. In general, the barriers covered the three systems and the intervention whereas the facilitators belonged mainly to the sociocultural system which was found to have a great influence on the implementation process as a whole. Discussion {#Sec20} ========== Our results showed that implementing the AP is a complex process that is influenced by a variety of interacting factors pertaining to the three systems as well as to the intervention itself. The study offers insight into the actual factors affecting the ongoing educational activities; and the potential barriers to adopting and implementing the other activities, that were on standby, in the future. Our findings are relevant to the adoption and implementation stages of the change process \[[@CR23]\] since factors were reported by actual and potential users who were not exposed to the AP. Taking into account these factors across these stages is considered as relevant for tailoring implementation strategies and adapting the intervention to specific contexts \[[@CR12], [@CR24], [@CR25]\]. The extent of implementation {#Sec21} ---------------------------- We have found poor fidelity, deficiency in the timelines for the completion of activities, and lack of actors' active engagement in the process. Funding constraints and the bureaucratic and centralized organizational structure in Morocco were challenges to executing the AP and have played an important role in the reported failure. Despite that the AP was welcomed during the validation phase of the project, mid-level policymakers attempted to implement only the educational activities, giving the impression that a final decision on fully adopting the AP has not been made. Referring to Rogers \[[@CR26]\], the "Knowledge, Attitudes, Practice" gap results when decision-making leads to active rejection for adopting the innovation \"in toto\" (p.177). In our study, this gap stemmed from authority innovation-decisions \[[@CR26]\] where individuals in the various Moroccan systems have no influence in the innovation-decision. The AP's drift is mainly the result of barriers present in the sociocultural system which in turn impacted the educational and professional systems. One of the important implementation gaps that emerges is the lack of professionals' active engagement in the implementation process. Consistent with our findings, many studies have highlighted that effective implementation strategies are those that achieve an active involvement of front-line professionals in the change process \[[@CR27]--[@CR29]\]; this has many implications such as increasing knowledge about the innovation, gaining a sense of ownership \[[@CR9]\], thus, decreasing the potential for resistant attitudes and, also, creating a greater commitment on the part of the participants \[[@CR26], [@CR30], [@CR31]\]. The facilitators and barriers encountered with respect to implementation {#Sec22} ------------------------------------------------------------------------ ### The three-system level {#Sec23} We emphasize the interactions that occur mainly between a range of dominant factors across the three-systems and at the intervention's level illustrating how our framework is essential for understanding the implementation process. The three systems work in dynamic interaction that challenge the AP's adoption and implementation while recognizing the fundamental role that the sociocultural system plays in the implementation process. Misalignment of the *values, methods, actors and targets* of the sociocultural system with the *values, methods and actors* of educational and professional systems on one hand; and with the intervention, on the other hand, are likely to be the greatest impediments to the adoption and full implementation of the AP. Our study revealed key facilitators in the sociocultural system essentially related to the values prevailing in the Moroccan system (e.g., respect for human rights) that mainly drove the national and international policies to reinforce midwifery, the presence of leaders (the head of the training division) who strongly advocated for the implementation of the educational activities. Nonetheless, many barriers in the sociocultural system have counteracted the facilitators' effects and have also affected the other systems. Consequently, a lack of alignment between, on one hand the *values* promoted (e.g., primacy of human beings) and the high priority given to maternal health; and, on the other hand, the *methods* set up (e.g., insufficient financial resources), capacities (e.g., truncated leadership) and attitudes of *actors* (oppressive attitudes of doctors) to reach this target, was evident. Three important barriers in the sociocultural system found in the system's structure (e.g., centralized structure), system's methods (e.g., insufficient financial resources) and the actors' levels (e.g., truncated leadership to reinforce midwives) exerted a coercive influence on the educational and professional systems. For example, the hierarchical bureaucratic structure in Morocco, characterized by a high degree of centralized lines of control, and complex administrative procedures was a big hindrance to implementing any activity. Our findings are in agreement with Macfarlane et al. \[[@CR32]\] and Edward III \[[@CR33]\], who have also identified the bureaucratic top-down structure as barriers to effective implementation of human resources strategies and public policy. Scott and Caress \[[@CR34]\] suggest to move away from the rigid top-down approach and to involve staff in decision-making processes in order to increase ownership. This structure has contributed to the limited access to knowledge about the AP's activities among many stakeholders in the three systems. Our results do echo previous findings that emphasized the necessity of effective communication in managing a healthy change process and its benefits in overcoming resistance \[[@CR28]\]. Effectively, promoting an understanding of the innovative efforts, and enhancing ownership of the innovation were among the benefits reported \[[@CR35]\]. Under the *Actors* dimension, the negative attitudes of the medical group towards the future expansion of midwives' competencies should be taken into consideration as it has been highlighted in many studies \[[@CR36], [@CR37]\]. With regards to how the sociocultural system affected the two other systems, insufficient budget allocated by the MOH and UNFPA played an influential role on the readiness of the educational system in terms of lack of adequate resources to implement the educational activities intended. The data suggest that sustained financial support is also a major concern. Providing a receptive context and devoting sufficient resources to support effective implementation of work role redesign interventions are considered as key elements \[[@CR8]\] to bring the AP to fruition and to produce the intended results \[[@CR38]--[@CR40]\]. Moreover, financial constraints stood out as a barrier to mobilize the midwifery professional association. The spread of an innovation often rests on professional organizations which are identified as taking the lead to activate this process \[[@CR41]\]. Nevertheless, our findings showed the absence of revitalization of the midwifery Moroccan association which does not seem involved in promoting the profession. Many barriers were also identified as constraining the adoption of the forthcoming change at the *actors*, *values* and *methods* levels in the educational and professional systems. Characteristics of the actors (e.g., educators' skill deficits, lack of familiarity with the CBA's needs) and the unmotivated midwifery workforce would affect their receptiveness to change. Concerning the values level, the prevailing biomedical culture in the workplace and the technocratic philosophy of midwifery education were perceived as incompatible with the forthcoming holistic approach and as hindering the successful implementation of the CBA. This is in line with earlier studies that found that organizational culture deeply embedded in traditional organizational forms is highly resistant to the introduction of role redesign interventions \[[@CR32], [@CR40]\]. Moreover, coordination and collaboration mechanisms and lack of collaborative teamwork were perceived as barriers to the future implementation of the CBA. Tensions between the norms of midwifery practice and biomedical obstetric care underpinning the coordination challenges identified in the Moroccan context were similar to those found in Colvin et al.'s review of midwifery task-shifting programs \[[@CR7]\]. Intervention characteristics {#Sec24} ---------------------------- There was a general consensus concerning the AP's (and the CBE program) potential to improve quality and continuity of care, and to clarify the midwifery role. Colvin et al. \[[@CR7]\] found that midwifery up-skilling often entailed the same advantages. Future adopters' positive perception of the anticipated benefits of the AP might contribute to make it visible to mid-level policymakers and to be a push factor to reverse the decision of rejecting the entire AP and reconsider its adoption \[[@CR26], [@CR39]\]. However, the intervention source and the complexity of the AP perceived as being challenges to its implementation should be taken into consideration. The AP designed by the consultant did not align with the middle-level policy makers' initial mandate (strengthening the midwives' education). Differences in priorities between both parties and the centralized structure in the Moroccan system have contributed to obstructing the adoption decision of the global AP. It is suggested \[[@CR38], [@CR42]\] that in order to adopt a change, it must resonate with the current perceived needs and belief systems. Complexity of the AP, the radical change needed on many levels confirm earlier research in this area showing that innovations broad in scope may be difficult to implement \[[@CR18], [@CR38]\]. Interestingly, we observed significant alignment between the various activities not taken into consideration by the mid-level policymakers, and the contextual needs called forth by the front-line providers (e.g., involving potential users in the change process, improving readiness). These results emphasize the importance for policymakers to pay attention to the key stakeholders' needs to put in place concomitant change process at the sociocultural, and professional system levels in order to address potential adopters' interests and, ultimately, meet health services' needs. Authoritative decisions have been reported to reduce the chance of successfully implementing an intervention \[[@CR12], [@CR38]\]. We predict that the successful implementation of this intervention in the future lies in the extent to which policymakers are willing to shift from a narrow focused model of change to a focus on the whole system. Ignoring the systemic nature of the change process will not result in the desired outcomes. Recommendations {#Sec25} --------------- Our findings offer important implications for policymakers, health care planners, and researchers, to redirect future implementation efforts. They suggest important lessons for tailoring implementation strategies to the barriers \[[@CR12]\] identified in the Moroccan context. We recommend the following:Consider using further knowledge transfer strategies (e.g. interactive workshops) in addition to the brief reports that have already been sent to the MOH secretary office coordinator. This will help explain, to policy makers and key stakeholders, the conceptual foundations for changing a health professional role, to persuade them about the relevance of the suggested intervention and motivate the implementation of the entire AP. Implementation consultants need to work in partnership with policymakers to redesign the AP's activities and implementation strategies.Adopt a collaborative approach in the implementation efforts as a substitute for a top-down approach to facilitate the implementation process. An effective strategy would be to actively engage a broad range of key stakeholders pertaining to the three systems in the different steps of the change process (redesigning the AP's implementation strategy, executing, evaluating the change). Involving potential users in the innovation process has been considered a \"*condition sine qua non*\" based on several implementation theories \[[@CR12]\] to promote committed use of the intervention \[[@CR25]\]. Allocating responsibilities to a team of individuals (e.g., obstetricians, midwives, educators) would help building a strong coalition of stakeholders with complementary expertise to support the implementation process \[[@CR42]\]. This would help create, in the future, a receptive culture in clinical and educational settings, and prevent professional culture clashes. Champion teams have, in fact, been considered as more efficient and effective than lone champions \[[@CR29]\].Enhance communication mechanisms by establishing effective dissemination and information strategies (e.g., workshops, brochures) supporting open lines of communications among the stakeholders involved in the process.Improve organizational readiness (e.g., increase financial, human and material resources to implement the AP, provide more relevant training sessions).Adopt a participatory approach in the future by involving relevant decision makers and key stakeholders as research partners in the evaluative process. This will promote collaboration with participants and help feelings of ownership over the findings \[[@CR22]\]. Obtaining commitment from stakeholders to monitor the implementation's progress through regular feedback will contribute to the adoption of strategies to continuously improve implementation efforts and to promote shared learning \[[@CR18], [@CR22]\]. Strengths and limitations {#Sec26} ------------------------- We present information based on real-life experiences and representing heterogeneous key stakeholders' views pertaining to three systems which help increase the validity of our findings. Another strength of this study is the use of a conceptual framework adopting a holistic approach for analyzing the implementation of a health professional role change. Some limitations should be noted. Due to limited financial resources, information was gathered during a two month period. An ongoing assessment is thought to be better than a punctual implementation evaluation, nevertheless a documentary analysis and triangulation of data helped track the implementation process over time. Also, the factors mentioned by respondents who were not exposed to the AP, address the adoption rather than the implementation stages. A lack of well-defined activities related to the first objective of the AP and of access to ALARM reports did not allow us to assess how well this objective was executed with regards to fidelity and timeliness. Also, data regarding the quality dimension were extracted from secondary sources of information and gathered from interviews. Incorporating quantitative and qualitative measures to assess each objective may provide a better understanding of the extent of implementation. Finally, as a case study method is adopted, our results may not be generalizable across all contexts; however, they may help the reader decide to what extent they are transferable to similar settings. Conclusions {#Sec27} =========== Our findings are novel within the field of implementation research regarding interventions aiming to strengthen the midwife's role. Policymakers should pay particular attention to the influential factors in order to rectify the AP's shortcomings and to improve implementation. Further research must be undertaken to better understand why the implementation of the overall AP did not pass the decision-making agenda. Our results have potential implications for other countries attempting to implement interventions to strengthen the midwife's role and to reduce maternal mortality. Additional files {#Sec28} ================ Additional file 1:**The action plan.** (DOCX 17 kb)Additional file 2:**Interview guide.** (DOCX 23 kb) AP : Action Plan ALARM : Advances in Labour and Risk Management CBA : Competency-Based Approach CBE : Competency Based Education CFIR : Consolidated Framework for Implementation Research FG : Focus group MDG : Millennium Development Goals MM : Maternal Mortality MOH : Ministry of Health UNFPA : United Nations Population Fund The *sociocultural system* represents the larger societal system, encompassing political (governmental bodies), and social systems at large (civil society). It concerns the values of society laws and regulations. The *educational-system* represents the educational principles and methods for optimizing the preparation of health professionals to attain the training goals. The *professional system* corresponds to the characteristics pertaining to the profession. It outlines the values of its members, practical approaches used; also the organization of the profession including its relationships with other professional groups; and the goals of developing their role. Systems approach: It is defined as « a general conceptual orientation concerned with the interrelationships between parts and their relationships to a functioning whole, often understood within the context of an even greater whole » (Trochim, Cabrera, Milstein, Gallagher, & Leischow, 2006, p. 539). The ALARM international program: is a continuing education program for health professionals responsible for the delivery of emergency obstetric and newborn care that addresses the causes of maternal mortality, the sexual and reproductive rights approach (Lalonde, Beaudoin, Smith, Plourde, & Perron, 2006). **Competing interests** The authors declare that they have no competing interests. **Authors' contributions** SAM took the lead role in designing the study, collecting, analyzing the data, and writing the manuscript. MH supervised the project, participated in the study's design, validated the methodology and the results, and made many critical revisions of the manuscript. NL contributed to the analysis and the critical revision of the manuscript. All authors read and approved the final manuscript. The main author of this study received financial support (scholarships) from several sources: the *Institut de recherche en santé publique de l'Université de Montréal* (IRSPUM), the CHIR-Quebec Training Network in Perinatal Research program, and Professor Hatem to collect the data. The authors wish to thank all participants for their willingness to take part in this study. The authors acknowledge: the useful suggestions of Laura J. Damschroder with regard to applying the CFIR to the present study. SAM would like to thank Jamilé Khoury at Saint-Joseph University of Beirut for her assistance with questions regarding the data analysis.

Introduction {#section1-1178630217746997} ============ The US Safe Drinking Water Act (SDWA), passed in 1974 and amended in 1986 and 1996, authorizes the US Environmental Protection Agency (USEPA) to set health-based standards for contaminants in public drinking water systems. Public drinking water systems are defined as water systems that have 15 service connections or more or serve at least 25 people daily for 60 days or more out of the year. Under the authority of the SDWA, USEPA establishes National Primary Drinking Water Regulations, which are health-based enforceable upper limits for concentrations of selected contaminants measured in drinking water, ie, standards. USEPA currently has primary standards for more than 75 contaminants and an extensive Contaminant Candidate List (CCL), substances that are under review for potential future regulation. Using the Unregulated Contaminant Monitoring Rule (UCMR), USEPA can require public water systems (PSWs) to monitor certain contaminants not otherwise regulated to assess their occurrence and levels in drinking water.^[@bibr1-1178630217746997]^ Even with these standards and other regulations in place, several contemporary incidents have highlighted the vulnerability of tap water in the United States to chemical contamination. For example, lead continues to be a prevalent concern despite the long-held and extensive knowledge about its sources in public water supplies and toxicity.^[@bibr2-1178630217746997]^ More recently, perfluorooctanoic acid (PFOA) has been detected in tap water of numerous public drinking water systems at levels previously found to be associated with kidney and testicular cancer in similarly exposed populations.^[@bibr3-1178630217746997]^ The principal safeguards for tap water delivered by PSWs are located at centralized treatment facilities, which have limited ability to control contamination that arises in downstream distribution systems, such as lead, and to remove organic compounds such as PFOA.^[@bibr2-1178630217746997],[@bibr4-1178630217746997][@bibr5-1178630217746997]--[@bibr6-1178630217746997]^ The aging infrastructure in the United States contributes to the potential for contaminants to enter tap water downstream of drinking water treatment facilities. According to USEPA, 30% of pipes in water systems serving more than 100 000 people are between 40 and 80 years old and approximately 9% of pipes in those systems are even older.^[@bibr7-1178630217746997]^ The quality of drinking water produced by private wells is subject to many of the same threats as municipal water supplies. Given the limitations of current centralized treatment systems to protect drinking water quality, filtration of water at homes, places of work, schools, and other points of use (POUs) may be beneficial. However, the potential for POU filtration technologies to contribute to drinking water safety has yet to be characterized in the context of public health. The objective of this article is to begin to fill that knowledge gap. To that end, we (1) present a critical review of the existing data and scientific literature on the effectiveness of POU filtration devices, including their use to control lead in Flint, Michigan, and (2) examine data from the USEPA's Enforcement and Compliance History Online (ECHO) database to assess the scope of SDWA violations in the United States and the potential opportunity for POU technologies to reduce consumer exposures in those settings. Methods {#section2-1178630217746997} ======= Literature search {#section3-1178630217746997} ----------------- We conducted a search of the scientific literature to identify studies of the efficacy of POU filters in actual use. We searched Web of Science, PubMed, JSTOR, and Google Scholar databases for relevant published articles using the following search terms or variants of the term: "point of use" and "drinking water treatment" OR filtration OR purification OR device OR residential OR tap OR system OR homes OR POU. Most studies of residence-level drinking water filtration/treatment have been conducted in developing countries and have focused on microbiological contaminants. As the focus of this review is on the evaluation of POU filters available to US consumers, we used the 'NOT' function in JSTOR and Google Scholar databases to further narrow results and excluded the following search terms (Africa, developing, ceramic, Nepal, Asia, epidemiology, and hospital). We limited our results to articles published in the past 2 decades, 1996-2016. We screened the titles and/or abstracts of articles for eligibility and reviewed full texts of the relevant abstracts. References provided in eligible articles were screened to identify additional studies of interest. Studies from peer-reviewed journals were included in our review if they met all of the following: (1) POU performance evaluation study, (2) conducted in United States or Canada, (3) published in English, (4) investigated chemical contaminant removal by POU drinking water treatment devices/technologies for residential use, (5) conducted lab-scale and/or field evaluation studies of POU devices, and (6) provided performance data for contaminants tested. Data analyses {#section4-1178630217746997} ------------- While conducting the literature search, we identified a large set of relevant data published by USEPA that can be used to evaluate the efficacy of POU filtration for removal of lead and other elements from tap water.^[@bibr8-1178630217746997]^ We downloaded the data from the on-line USEPA database and carried out several evaluations of the filter effectiveness. POU faucet-mount filters approved for lead removal and other contaminants by an independent review body, NSF International ([nsf.org](http://nsf.org)), were distributed to residents in the Flint, MI, service area to treat residual Pb contamination after the city switched the water supply and introduced system treatments and upgrades. The data published by USEPA included drinking water samples collected in 238 residences, 1 church, and 34 commercial properties in Flint.^[@bibr8-1178630217746997]^ Where possible, 3 samples were collected at each home or facility: (1) with the POU filter in-line using the existing cartridge in place at the time of sampling ("used filter"), (2) with the POU filtration device removed ("no filter"), and (3) with a new replacement filter cartridge inserted into the POU filtration device ("new filter"). These samples were collected sequentially at one or more sinks in each home or facility. Samples were analyzed for Al, Cd, Ca, Cr, Cu, Fe, Pb, Mg, Mn, Ni, K, Na, and Zn using USEPA Method 200.8. Six samples were analyzed for tin but were not included in this analysis due to the very limited sample size. For our analysis, we matched the used filter and no filter data by property and house identification codes, which yielded 208 observations for 13 elements. We then generated summary statistics of element concentrations in filtered and nonfiltered water and the differences between the two. Next, we used Wilcoxon signed-rank 2-sample paired tests to determine whether differences between filtered and nonfiltered water were statistically different from 0 at .05 level of significance. We also assessed the differences between used and new replacement cartridges in the POU filters for the 161 observations that had both measures to determine whether concentrations of elements vary significantly between used and new filters. Data processing, summary statistics, and Wilcoxon tests were done using R.^[@bibr9-1178630217746997]^ Assessment of USEPA violation data {#section5-1178630217746997} ---------------------------------- We queried the USEPA ECHO database to assess the scope and type of drinking water contamination that has been reported to result in violations of the SDWA.^[@bibr10-1178630217746997]^ Using the USEPA Drinking Water Dashboard system, we obtained the total reported health-based violations for exceedances of maximum contaminant levels (MCLs), residual disinfectant levels, or treatment technique requirements for each major SDWA rule for all PWSs in the United States between 2010 and 2014. For completeness, we also report information on the total number of violations, which includes lack of monitoring and reporting by PWSs. Having summarized the first pass removal efficiency for various contaminants indicated by the literature search and data analyses described above, we cross-tabulated the chemicals identified in the SDWA violations with the results of the literature review and data analyses described above to determine the number of incidents and size of the populations at risk for which POU treatment technologies with demonstrated efficacy could be used to mitigate exposures. Results and Discussion {#section6-1178630217746997} ====================== Literature review {#section7-1178630217746997} ----------------- Our search of the literature returned 3142 papers related to POU drinking water filtration. We excluded 3110 papers based on review of titles and abstracts and reviewed the full text of 32 papers. Of those papers, 18 were outside the scope of our analysis and were not considered further: 3 studies evaluated treatment technologies other than POU devices, 5 studies evaluated POU technologies not intended for residential application, 2 studies were conducted outside of the United States or Canada, 3 studies did not report filter performance data, and 5 studies addressed only microbiological contaminants. The remaining 15 papers met the inclusion criteria and include 11 studies of POU filtration of inorganic contaminants and 4 of organic chemicals. Below, we summarize the filter types and corresponding removal efficiency for each of the contaminants reported in those papers along with relevant information on the settings and methods of the testing procedures. Detailed results from each of those publications are presented in [Table S1](http://journals.sagepub.com/doi/suppl/10.1177/1178630217746997) in Supplemental Material. Of the 11 peer-reviewed studies of inorganic contaminants, one or more of the following elements were studied: Ag, Al, As, Co, Cr, Cu, Fe, Mn, Ni, Pb, Sb, and Zn. Four of the 11 inorganic studies evaluated POU performance for Pb either alone or in conjunction with other metals^[@bibr11-1178630217746997][@bibr12-1178630217746997][@bibr13-1178630217746997]--[@bibr14-1178630217746997]^ and 2 for Cu.^[@bibr11-1178630217746997],[@bibr14-1178630217746997]^ Overall, faucet-mount or under-sink solid block activated carbon (SBAC) filters removed 80% to 99% of total Pb in 2 studies^[@bibr12-1178630217746997],[@bibr14-1178630217746997]^ and 68% to 98% of total Cu in 1 study.^[@bibr12-1178630217746997]^ The study by Deshommes et al included laboratory tests of 3 SBAC faucet-mount or under-sink filtration units as well as 1 pour-through pitcher. All filters removed more than 80% of dissolved Pb and up to 99% of particulate Pb. Notably, the pour-through filter reduced particulate Pb inconsistently with as little as 28% reduction. Similarly, the in-use study with more than 90 observations made in a penitentiary also showed the effectiveness of SBAC filters on both dissolved and particulate lead.^[@bibr14-1178630217746997]^ In addition, one study evaluated SBAC performance for Mn reduction and found SBAC systems to be effective only up to approximately 50% of filter capacity, but filters containing cation exchange resin were effective in removing Mn at 60 to \>99%.^[@bibr15-1178630217746997]^ No corresponding NSF standard is applicable for Mn reductions, so consumers concerned about Mn in drinking water would not be able to identify a filter to address this metal. Finally, 7 of the 11 POU filter studies measured reductions of As using POU filters.^[@bibr13-1178630217746997],[@bibr16-1178630217746997][@bibr17-1178630217746997][@bibr18-1178630217746997][@bibr19-1178630217746997][@bibr20-1178630217746997]--[@bibr21-1178630217746997]^ Reverse osmosis (RO) units were able to reduce on average 79 to \>99% of As.^[@bibr13-1178630217746997],[@bibr16-1178630217746997][@bibr17-1178630217746997][@bibr18-1178630217746997]--[@bibr19-1178630217746997],[@bibr21-1178630217746997]^ Increasing As influent concentrations and water hardness decreased the effectiveness of RO units on As reductions.^[@bibr16-1178630217746997],[@bibr19-1178630217746997]^ POU filter study in Flint {#section8-1178630217746997} ------------------------- The in-use effectiveness of NSF-approved faucet-mount filters for metals was tested in a study conducted by USEPA in residences and commercial locations in Flint, MI, during 2015.^[@bibr8-1178630217746997]^ [Table 1](#table1-1178630217746997){ref-type="table"} presents summary statistics of the elemental concentrations in filtered and unfiltered water matched by location in the study (n = 208). The median Pb concentration in the samples collected without a filter was 2.5 µg/L, more than 20 times the median concentration of used filters (0.11 µg/L). [Table 2](#table2-1178630217746997){ref-type="table"} presents summary statistics on the differences between unfiltered and filtered water. The average difference in Pb concentration between filtered and unfiltered water (filtered -- unfiltered) was --18.2 µg/L (SD = 49.0 µg/L) with a range of --418 to +0.7 µg/L (median = --2.3 µg/L). Elemental concentrations in the filtered water were significantly (*P* \< .001) lower for all elements except Na and K, which were significantly higher (*P* \< .001) in the filtered water than the unfiltered water. In addition, there was no statistically significant difference between used and new filters (n = 161) for any of the elements (*P* \< .0001, rejecting the alternative hypothesis that the true difference between new and used filters was not equal to 0 for all elements). [Figure 1](#fig1-1178630217746997){ref-type="fig"} illustrates the highly skewed distributions of Pb and Cu measured in the used and no filter samples, but the much lower concentrations after filtration are apparent. Even in cases where the unfiltered water was as high as 419 µg/L of Pb, the highest filtered Pb concentration was 2.9 µg/L. ###### Summary statistics of 208 pairs of water samples with and without a POU filter collected in Flint, MI, January to May 2016.  Used filter No filter ---- --------------- ----------- ------- ------- ------- ------- --------------- ------- ------- ------- ------- ------- Al 0.051 (0.075) 0.009 0.009 0.009 0.200 0.330 0.129 (0.300) 0.009 0.021 0.043 0.370 2.62 Cd 0.452 (0.730) 0.061 0.061 0.061 2.00 2.00 0.641 (1.149) 0.061 0.061 0.140 2.00 12.0 Ca 22 (8) 0.2 22 26 28 32 27 (2.7) 0.24 26 27 29 32 Cr 1.25 (1.97) 0.20 0.20 0.20 5.00 5.00 1.35 (2.13) 0.20 0.20 0.25 5.00 14.1 Cu 2.2 (9.7) 0.75 0.75 0.75 2.3 120 92.0 (177) 0.75 14.3 36.0 349 1800 Fe 0.038 (0.069) 0.016 0.016 0.016 0.096 0.880 0.508 (2.86) 0.016 0.043 0.084 1.10 38.5 Pb 0.25 (0.35) 0.11 0.11 0.11 0.50 2.90 18.5 (49.0) 0.11 0.61 2.53 90.4 418 Mg 7.86 (2.00) 0.048 7.80 8.10 9.80 11.2 7.82 (0.688) 0.048 7.69 7.87 8.4 10.1 Mn 0.004 (0.006) 0.001 0.001 0.002 0.009 0.068 0.012 (0.048) 0.001 0.002 0.005 0.033 0.655 Ni 1.44 (2.25) 0.23 0.23 0.31 6.00 11.6 2.25 (2.99) 0.23 0.48 0.85 6.34 19.0 K 5.32 (10.4) 0.21 0.99 1.36 32.7 70 0.99 (0.09) 0.04 0.97 0.99 1.11 1.22 Na 9.31 (9.98) 0.98 4.80 5.39 30.3 80 5.16 (3.96) 4.1 4.6 4.8 5.27 52 Zn 0.051 (0.075) 0.009 0.009 0.009 0.200 0.330 0.129 (0.300) 0.009 0.021 0.043 0.370 2.62 Abbreviation: POU, point-of-use. Units are in milligram per liter for all elements except Cr, Cu, Ni and Pb, which are in microgram per liter. ###### Summary statistics of differences in elemental levels for 208 pairs of water samples with and without a POU filter collected in Flint, MI, January to May 2016.  Element Mean (SD) Min Max P25 Median P95 --------- ---------------- -------- ------- -------- -------- ------- Al −0.1 (0.3) −2.6 0.3 −0.044 −0.014 0.005 Cd −0.2 (0.9) −11.4 0.32 −0.1 0 0 Ca −4.2 (8.2) −29.3 21.4 −4.0 −1.0 2.0 Cr −0.1 (0.6) −9.1 0.48 −0.1 0 0 Cu −89.8 (177) −1799 47.25 −98.1 −34.3 −3.3 Fe −0.5 (2.9) −38.4 0.9 −0.184 −0.051 0 Pb −18.2 (49.0) −418 0.7 −9.9 −2.3 0 Mg −0.04 (1.95) −8.1 3.4 −0.1 0.3 1.9 Mn −0.008 (0.048) −0.647 0.064 −0.005 −0.001 0.004 Ni −0.001 (0.002) −0.019 0.006 −0.001 0 0 K 4.3 (10.4) −0.7 69.0 0 0.4 31.7 Na 4.1 (9.9) −32.4 75.2 0 0.6 25.0 Zn −0.13 (0.51) −6.5 0.09 −0.094 −0.024 0.009 Abbreviation: POU, point-of-use. Units are in milligram per liter for all elements except Cr, Cu and Pb, which are in microgram per liter. {#fig1-1178630217746997} While NSF testing provides assessment of the potential effectiveness of POU filters, the data collected in Flint, MI, and released by USEPA provide a unique opportunity to assess in-use effectiveness of faucet-mount POU filters on Pb and other metals for hundreds of homes in that city. The data show substantial and statistically significant reductions in Pb and Cu with use of NSF-approved SBAC filters at the faucet. The efficacy of SBAC POU filtration for Pb removal observed in Flint may have meaningful implications for managing risks of Pb in drinking water more broadly. A cross-sectional epidemiological study of 306 children found a strong association between Pb in home drinking water and elevated blood Pb levels (odds ratio \[OR\] = 4.66, confidence interval \[CI\] = 2.12-10.24).^[@bibr22-1178630217746997]^ The 90th percentile concentrations of Pb in drinking water in study homes ranged from 4.51 to 10.06 µg/L, depending on the sample collection method employed. The 10th and 90th percentile blood lead levels were 0.77 and 1.31 µg/dL, respectively (geometric mean (GM) = 1.35 µg/dL, 95% CI = 1.27-1.43). In comparison, the 90th percentile concentration of Pb in filtered water sampled in Flint, MI, was 0.5 µg/L. The use of SBAC filters can reduce levels of Pb in drinking water and as a result may be able to reduce blood lead levels in those drinking water with elevated levels of Pb. There are limitations to the Flint data, including the lack of repeated measures within homes over time, the lack of simultaneous pH measurements, and other indicators of hardness. However, the results support the hypothesis that SBAC filters can be effective for reducing levels of Pb and Cu (as well as many other elements) in drinking water with moderately elevated concentrations, as was the case in Flint. The Flint filter study also demonstrates that moderately scaled panel studies can provide a model for assessing additional contaminants of concern and POU technologies of interest. Although NSF testing provides an assessment of the potential effectiveness of POU filters, the data collected in Flint, MI, provided a unique opportunity to assess in-use effectiveness of faucet-mount POU filters on Pb and other metals for hundreds of homes in that city. Organic contaminants {#section9-1178630217746997} -------------------- While some SBAC faucet-mount POU filtration devices have NSF approval for trihalomethanes (THMs), no pour-through devices do. Three of the 4 studies of organic contaminants we reviewed examined the effectiveness of POU filters in reducing levels of disinfection by-products (DBPs), including THMs, in drinking water.^[@bibr23-1178630217746997][@bibr24-1178630217746997]--[@bibr25-1178630217746997]^ Anumol et al^[@bibr26-1178630217746997]^ examined removal efficiencies for 16 individual trace organic compounds (TORCs), including pharmaceuticals, pesticides, and other organic contaminants, by 2 refrigerator SBAC POU devices and a commercially available activated carbon pitcher. This study evaluated POU performance over the life of the filters and found that all were effective in reducing levels of TORCs initially by 70 to \>90%. Refrigerator SBAC devices performed better than pitchers. Over time, filters performed better when challenged with groundwater than surface water, with high total organic carbon content with reduced flow occurring for several of the pitchers when surface water was the influent source. Levesque et al^[@bibr24-1178630217746997]^ conducted a field study of municipal tap water and found a POU filter pitcher device with activated carbon and ion exchange resin reduced THMs by approximately 90% and haloacetic acids (HAAs) by 50 to \>70%. Another study evaluated additional POU pitchers and found them to remove \>93% of trihalomethane-4 and 65 to \>98% of HAA9.^[@bibr25-1178630217746997]^ Scores of studies have shown adverse toxicological effects related to chlorination by-products,^[@bibr4-1178630217746997],[@bibr24-1178630217746997],[@bibr27-1178630217746997]^ which are commonly found in drinking water. In general, there is a fairly large variation in the reduction efficiencies reported for contaminants treated by POU technologies tested in these studies. Only a small number of studies have evaluated effectiveness of POU filters on removing DBPs, such as THMs, and there are no pour-through filters with NSF certification demonstrating effectiveness of filters on DBPs.^[@bibr28-1178630217746997]^ Some faucet-mount filters, particularly SBAC filters, are approved by NSF for reduction of THMs in tap water. Turning now to other types of organic chemical contaminants, cases of drinking water contamination related to PFOA have been reported for numerous locations in the United States,^[@bibr29-1178630217746997][@bibr30-1178630217746997]--[@bibr31-1178630217746997]^ and exposures via private wells have also been demonstrated,^[@bibr32-1178630217746997]^ highlighting the risks of contaminants in both public and private water supplies. Only very recently was an NSF certification established for PFOA reduction with only 2 products approved as of August 2017.^[@bibr28-1178630217746997]^ Prior to establishment of a PFOA/PFOS certification, USEPA and many state environmental and public health agencies have recommended use of POU filtration units for individuals living in areas known or suspected to have drinking water resources contaminated with PFOA or related compounds.^[@bibr33-1178630217746997],[@bibr34-1178630217746997]^ Studies have also identified many classes of additional unregulated drinking water contaminants, including prescription medications, pesticides, and antimicrobials, that can make their way into drinking water systems.^[@bibr35-1178630217746997]^ Health effects associated with relatively low-level exposures to mixtures of these compounds are not known, but health effects associated with contaminants of potential concern are under review by USEPA as part of its CCL program.^[@bibr1-1178630217746997]^ In addition, the frequency of detection and magnitude of concentrations of certain unregulated contaminants are monitored under the UCMR. Despite the UCMR and drinking water regulations, concerns remain about the potential impacts of unregulated compounds in drinking water. Furthermore, many of these contaminants, including medications, pesticides, and flame retardants, can be treated with NSF-approved POU filters. US EPA SDWA violation data {#section10-1178630217746997} -------------------------- [Table 3](#table3-1178630217746997){ref-type="table"} provides a summary of monitoring and reporting, as well as health-based SDWA violations, for inorganic and organic contaminant-related rules reported in the United States between 2010 and 2014. Violations related to monitoring and reporting are much more frequent than health-based violations as illustrated in the "total violations" columns. The number of health-based violations of the Lead and Copper Rule (LCR) was 290, potentially affecting 241 675 consumers. In comparison, when all violations, including lack of monitoring and reporting, are included, the total number of LCR violations increases to 6189, potentially affecting up to 14 461 735 consumers. This table also includes classes of POU treatment devices that have been approved by NSF to be effective in reducing contaminants associated with the corresponding SDWA violation. NSF-approved POU devices have the potential to mitigate risks from SDWA violations related to contaminants, such as lead, volatile organic compounds (VOCs), and DBPs, which have been shown to have hundreds, if not thousands, of violations across the United States in recent years. ###### Data on health-based violations of inorganic and organic contaminant rules from USEPA ECHO database, 2010-2014.  ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- SDW rule/category Contaminant(s) Health-based violations Total violations Type of POU to treat NSF/ANSI certification(s) ------------------- --------------------------------------------------------------------------------------------------- ------------------------- ------------------ ---------------------- --------------------------- ---------------- -------------- LCR Pb, Cu 290 241 675 6189 14 461 735 Carbon block Standard 53 Chemical SOC, VOCs, other inorganics,^[a](#table-fn6-1178630217746997){ref-type="table-fn"}^ radionuclides 473 774 390 5101 17 704 590 Carbon block\ Standard 53\ RO Standard 58 DBPs THMs, HAAs, chlorite, bromate 1442 8 837 437 8999 40 026 297 Carbon block Standard 53 Arsenic Arsenic 708 1 074 179 1685 3 550 325 RO\ Standard 58 Nanofiltration Nitrates Nitrate/Nitrite 669 454 532 9165 9 397 203 RO Standard 58 RO Standard 58 ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Abbreviations: DBP, disinfection by-product; ECHO, Enforcement and Compliance History Online; HAAs, haloacetic acids; LCR, Lead and Copper Rule; RO, reverse osmosis; SOC, synthetic organic chemicals (includes 25 individual pesticides, ethylene dibromide, polychlorinated biphenyls, benzo(a)pyrene, di(ethylhexyl)-adipate, di(ethylhexyl)-phthalate, dioxin); THMs, trihalomethanes; USEPA, US Environmental Protection Agency; VOC = volatile organic compounds. Other inorganics include asbestos, Ba, Cd, Cr, F, Hg, Se, Sb, Be, cyanide, and Tl. Some POU devices, such as carbon block filters, remove multiple categories of pollutants, including metals, VOCs, and particles. The efficacy of using POU filtration for drinking water contamination incidents has been demonstrated most recently for Pb in Flint, MI, drinking water. With the high incidence of monitoring and reporting violations and weak enforcement, POU filters with demonstrated effectiveness may be able to protect public health by treating water containing contaminants exceeding guidance values for the millions of individuals potentially affected by these violations. Although there are limitations to these data, they provide an indication of the potential scope of drinking water contamination and the magnitude of potentially affected populations served by PWSs with violations that may not be reported to the public. In addition, the small number of health-based violations for some SDWA rules indicates potential gaps in assessment and/or enforcement programs. Violations by definition occur after the discovery of a chemical contaminant in drinking water. As a result, elevated concentrations can also occur with no knowledge on the part of consumers or drinking water authorities. Perhaps even more concerning is the health risk for the more than 15 million US residents who obtain their drinking water from domestic private wells, which are not covered by federal regulations and have no centralized treatment or regular testing. These private wells are mostly located in rural areas, some of which are near septic and/or agricultural areas and can thus be vulnerable to chemical contamination due to nearby industrial sources or agricultural activities as well as PFOA. Perfluorooctanoic acid is an example where widespread contamination can go on unnoticed for extended periods of time. Point-of use filtration can be effective as a preventative measure in these situations. The utilization of POU filters can reduce exposures to these contaminants and should be considered as an added level of protection to increase the safety of drinking water from public or private sources. Conclusions {#section11-1178630217746997} =========== Despite promulgation and enforcement of SDWA regulations, concerns about drinking water quality are common in the United States, with 61% of Americans surveyed "worrying a great deal about polluted drinking water."^[@bibr36-1178630217746997]^ Although microbiological contaminants have historically been a focus of drinking water regulations, numerous studies have identified inorganic and organic chemical contaminants in drinking water, some of which enter drinking water supplies after the water has been tested at the drinking water treatment facility or are not removed by standard treatment methods. Finally, the extensive Pb contamination^[@bibr2-1178630217746997],[@bibr37-1178630217746997]^ of drinking water in Flint, MI, in 2014 and in Washington, DC^[@bibr6-1178630217746997]^ in 2005, as well as various cases of PFOA contamination across the United States, highlights the risks associated with poorly managed drinking water supplies in the United States and the potential role of POU faucet-mount filters. Many types of residential water treatment devices available to consumers have undergone laboratory testing and often certification for efficacy against chemical contaminants in tap water. Laboratory evaluations have traditionally focused on the contaminants regulated under the SDWA in the United States. In recent years, these certifications include "emerging" contaminants of concern, including certain novel human carcinogens, industrial chemicals, pesticides, pharmaceuticals, personal care products, and flame retardants and DBPs. Some POU filters, such as SBAC, can remove multiple classes of contaminants, including both inorganic, such as Pb, and organic contaminants, such as flame retardants. As previously mentioned, the SDWA requires only 2 contaminants be monitored at the tap, Pb and Cu, although others may be introduced into drinking water as it travels through the distribution system. A number of reports have highlighted that for many SDWA contaminants, there is weak enforcement of the required monitoring and reporting, as indicated by the large discrepancies between health-based violations and those that include monitoring and reporting violations. The timing of reported violations can often be much delayed from the actual health impacts in the community. There are a number of historical exposure scenarios in which a contaminant was introduced by the pipe network. This includes Pb in Flint, MI, Washington, DC, and elsewhere and tetrachloroethylene in parts of the northeastern United States. In all of these cases, the community was not aware of drinking water contamination for long periods of time, years in some cases. While relatively few published studies have evaluated the effectiveness of POU filters in use, the studies we identified suggest that POU filters can be effective in reducing levels of chemical contaminants in drinking water. Future studies could evaluate the potential impacts of POU filtration on health endpoints, such as reducing blood lead levels or PFOA serum levels, in populations previously exposed to those contaminants in drinking water after consumers switch to filtered water. We thank Jacob Pollien for assistance in data processing. **Funding:**The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Funding for this research was provided by Environmental Health & Engineering, Inc., Needham, MA, and Kaz, Inc., Southborough, MA. **Declaration of conflicting interests:**The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article. **Author Contributions:** KWB and DLM conceived and designed the experiments; KWB and BG analyzed the data; KWB, BG, and LJB wrote the first draft of the manuscript; BG and DLM contributed to the writing of the manuscript; KWB, LJB, BG, and DLM agree with manuscript results and conclusions; KWB, DLM, and LJB jointly developed the structure and arguments for the paper; KWB, LJB, and DLM made critical revisions and approved the final version. All authors reviewed and approved the final manuscript. **Disclosures and Ethics:** As a requirement of publication, author(s) have provided to the publisher signed confirmation of compliance with legal and ethical obligations including, but not limited to, the following: authorship and contributorship, conflicts of interest, privacy and confidentiality, and (where applicable) protection of human and animal research subjects. The authors have read and confirmed their agreement with the ICMJE authorship and conflict of interest criteria. The authors have also confirmed that this article is unique and not under consideration or published in any other publication and that they have permission from rights holders to reproduce any copyrighted material. Any disclosures are made in this section. The external blind peer reviewers report no conflicts of interest.

Background ========== Several factors can contribute to poor expression levels or low yields of soluble recombinant proteins produced in *E. coli*\[[@B1]\]: codon bias, lack of post-transcriptional modifications, incorrect membrane targeting, high protease activity, missing partners for complex formation, mRNA instability, limiting redox conditions or chaperone availability. A large number of modified strains is now commercially available which either exhibit a compensative expression of rare codons, co-express specific molecular chaperones, lack protease activity or provide an oxidizing cytoplasm. Furthermore, polycistronic vectors allow the co-expression of several complex subunits and fused leader peptides enable the translocation of ribosomal products to the oxidizing periplasm. However, the most common approach to improve the solubility of recombinant proteins is their fusion with highly soluble partners. For instance, NusA has been initially selected for being the most soluble protein in *E. coli*\[[@B2]\]. Several other proteins have been proposed as fusion partners \[[@B3]\] but only few of them are commonly exploited. Among them, thioredoxin (Trx) \[[@B4]\], glutathione S-transferase (GST) \[[@B5]\] and maltose binding protein (MBP) \[[@B6]\] offer the further advantage of being suitable for affinity purification, whereas DsbA and DsbC \[[@B7],[@B8]\] are enzymes that facilitate the formation of disulphide bonds. Furthermore, we showed that thermostable proteins improve the yields and enable the purification of the recombinant fusion by simple heat incubation \[[@B9]\]. Nevertheless, some shortcomings are well documented. The linker between GST and the target protein is often instable, leading to the breakage and loss of the target protein \[[@B10],[@B11]\], whereas the solubility of the fusion protein does not yet indicate that the target protein is correctly folded. Soluble aggregates of fusion proteins have been described for both MBP \[[@B12]\] and GST \[[@B13],[@B14]\]. For several applications, the fusion carriers must be removed before using the target proteins. Specific protease recognition sequences are engineered as linkers between the fusion partners and proteolytic digestion used to remove the carriers. Among commercially available proteases, TEV \[[@B15]\] and HRV 3C \[[@B16]\] are the most reliable ones because of their longer recognition site and, consequently, extremely limited non-specific cleavage. Previous experiments indicate that the optimal conditions to maximize yields, correct folding, and protease cleavage efficiency are protein specific. Therefore, a preliminary screening for the identification of the most suitable expression conditions for the production of soluble proteins is necessary before moving to large-scale production \[[@B17]\]. Approaches that allow the parallel cloning in different vectors simplify the comparison among fusion partners. In this paper we present the results obtained cloning a unique PCR product into the pETM expression vectors developed in our institute. Results and Discussion ====================== General features of the pETM vectors ------------------------------------ The pETM vectors are derived from pET (Novagene) backbones. They share some common features, the most important of which being a 6xHis tag, a protease recognition site and the conserved multiple cloning site (MCS) starting with an NcoI recognition site (Fig. [1](#F1){ref-type="fig"}). The NcoI recognition sequence has an in-frame ATG codon that can be used for the functional expression of the target protein, minimizing the number of non-native amino acids at the N-terminus. The conserved MCS ensures that the same couple of restriction sites inserted in the PCR product can be used for direct subcloning in all of the other pETM vectors. {#F1} The 6xHis tag suitable for immobilized metal affinity chromatography (IMAC) purification and a protease recognition site, with the exception of the pETM10, are present in all vectors. Both, the TEV and the 3C recognition site versions exist for each vector subclass. The vectors differ in their tag(s) and they are identified by the first number after the pETM code. In particular, 1 indicates the class of pETM vectors with only a His-tag, 2 those with His-tag plus Trx, 3 the combination His-tag and GST (a complete list of the vector maps is available -see additional file 1: \"The maps of the pETM vectors\"- and their sequences can be downloaded at our website \[[@B18]\]). The second number identifies the specific version that can differ from the others because of the protease restriction site, the relative position of the His-tag at the N or C-terminus of the fusion partner (Fig. [1B](#F1){ref-type="fig"} and [1C](#F1){ref-type="fig"}) and the presence of a leader peptide for periplasmic targeting (Table [1](#T1){ref-type="table"}). A stuffer gene has been cloned in some vectors (Figure [1A](#F1){ref-type="fig"}) to simplify the evaluation of the restriction enzyme digestion efficiency. ###### Specific features of the vectors. The vectors used in the experiments have conserved cloning sites but differ in the position and identity of the tags and in the protease recognition sequences. Both the EYFP constructs were cloned in all the indicated vectors. (*italic*) functional tag; (*italic*) purification tag;(bold) double-function tag. **pETM Vectors** **Composition of the expressed recombinant fusion proteins** ------------------ -------------------------------------------------------------- ----------------- ----------------- ----------- **10** *His-Tag* **EYFP** **11** *His-Tag* [TEV-site]{.ul} **EYFP** **14** *His-Tag* [C3-site]{.ul} **EYFP** **20** ***Trx*** *His-Tag* [TEV-site]{.ul} **EYFP** **22** ***Trx*** *His-Tag* [C3-site]{.ul} **EYFP** **30** *His-Tag* **[GST]{.ul}** [TEV-site]{.ul} **EYFP** **33** *His-Tag* **[GST]{.ul}** [C3-site]{.ul} **EYFP** **44** *His-Tag* **MBP** C3-site **EYFP** **50** ***DsbA*** *His-Tag* [TEV-site]{.ul} **EYFP** **52** ***Ll DsbA*** *His-Tag* [TEV-site]{.ul} **EYFP** **54** ***DsbA*** *His-Tag* C3-site **EYFP** **55** ***Ll DsbA*** *His-Tag* C3-site **EYFP** **60** ***NusA*** *His-Tag* [TEV-site]{.ul} **EYFP** **66** ***NusA*** *His-Tag* C3-site **EYFP** **70** **[CBP]{.ul}** [TEV-site]{.ul} **EYFP** *His-Tag* **80** ***DsbC*** *His-Tag* [TEV-site]{.ul} **EYFP** **82** ***Ll DsbC*** *His-Tag* [TEV-site]{.ul} **EYFP** Only pETM20 carries the β-lactamase gene that confers ampicillin-resistance, whereas all the other vectors are kanamycin-selectable. Comparison of the constructs ---------------------------- Two EYFPs (wild type and I48A mutant) were cloned into the pETM vectors. The specific features of the constructs used in the analysis are listed in Table [1](#T1){ref-type="table"}. The His-tag is directly fused to the target protein in the pETM10 vector whereas a TEV and a 3C recognition site is inserted between the target protein and the tag in vectors pETM11 and 14, respectively. The pETM20 and 22 vectors contain the stabilising Trx fusion tag \[[@B4]\] and differ only in the protease restriction site. Cloning into pETM30 and 33 vectors allows the expression of GST-fusion proteins that can be purified by two independent and alternative affinity methods. Also, the MBP-EYFP fusion protein obtained by cloning into the vector pETM44 can be purified by both, maltose affinity and IMAC. In the case that the correct folding of recombinant proteins involves the formation of disulfide bonds the yield can be improved by fusion with DsbA and DsbC, both in the cytoplasm \[[@B8]\] or after targeting into the oxidizing environment of the periplasm. The vectors pETM50, 52, 54, 55, 80, and 82 cover all of these possibilities. The pETM60 and 66 vectors allow the fusion of the target protein with the highly soluble bacterial NusA protein \[[@B2]\], and the pETM70 leads to the expression of fusion proteins with the calmodulin-binding protein. The yield of soluble EYFP expressed using the commercial pRSET vector (Invitrogen) was used as a reference. The crucial features for expression control (origin of replication, promoter) of pRSET and pETM are the same, but they differ in the length of the region between the His tag and target protein. Small-scale expression screening -------------------------------- Small-scale (1.5 mL) cultures contain enough material for a rapid comparison of recombinant protein yields. The bacterial lysates from wild type and mutant EYFP were analysed by SDS-PAGE and used to estimate the total expression levels of the recombinant constructs while the soluble fractions recovered after centrifugation were used to purify the recombinant proteins by IMAC using Talon resin. We found this information more reliable than the analysis of the total soluble fraction in which soluble aggregates can be highly represented \[[@B12]-[@B15],[@B19]\]. All of the vectors expressed recombinant proteins of the expected molecular mass but the amount of accumulated recombinant proteins differed among the vectors: GST fusion\>His only\>MBP fusion\>DsbA fusion\>DsbC fusion = NusA fusion\>Trx fusion. Apparently, the expression rate is dependent on the fusion partner because similar yields were obtained expressing the fusion partners alone: GST = MBP\>DsbC\>DsbA = NusA\>Trx (data not shown), although the regulative expression elements are the same for all the vectors. Codon bias, mRNA or protein stability \[[@B20]\] can further influence the level of protein accumulation. For instance, wild type and mutant EYFP expressed using the same vector often differed for the amount of accumulated soluble protein (Fig. [2A](#F2){ref-type="fig"}). The comparative screening indicated that the mutant EYFP was more soluble than the wild type when expressed from the pETM22 and 33, whilst the wild type when expressed from the pETM66. {#F2} SDS-PAGE analysis of the IMAC purified recombinant proteins from bacteria transformed with the different vectors is shown in Figure [2A](#F2){ref-type="fig"}. EYFP depends on a reducing environment to successfully complete its folding and its expression in the periplasm was used as a negative control. As expected, the recombinant proteins secreted in the periplasm expressed from pETM50, 54, and 80 were mostly not soluble. A negative effect of the fusion partners DsbA and DsbC is ruled out since the leaderless versions of the same vectors (pETM52, 55, and 82) largely accumulated in the soluble fraction. All of the remaining constructs were soluble, with the exception of the calmodulin-binding fusion (pETM70). The amount of purified proteins seemed strongly vector-dependent but it is necessary to remind that this comparison is rather qualitative because larger constructs (fusions with MBP or NusA, namely pETM44, 60, and 66) stain less than constructs with smaller fusions (Trx, GST, DsbA, and DsbC: pETM20, 22, 30, 33, 52, 55, 82) or no fusion partners (pETM10, 11, 14). A second round of screening can be envisaged in which different expression conditions can be tested in parallel for their effect on the soluble protein yields. For instance, the pETM20 vector is the only ampicillin-resistant and has a different backbone. The constructs from the pETM20 were poorly expressed and gave a low yield of soluble protein when standard growth conditions were used. The metabolization of the ampicillin could result in the loss of selectivity for the transformed bacteria and their substitution in the liquid culture with wild type cells. Furthermore, a higher level of expression leakage can lead to mutations of the T7 polymerase, with consequent decreased transcription functionality and repressed expression of the recombinant protein \[[@B21]\]. Therefore, the pETM20-EYFP vector was cultured using the degradation-resistant carbenicillin instead of ampicillin and either transformed in pLysS cells or cultured in the presence of glycerol to prevent expression leakage. As a result, the yields of the soluble proteins were dramatically increased (Fig. [2B](#F2){ref-type="fig"}). Apparently, even minor differences in the sequence of an expressed construct can significantly affect the protein solubility, as in the case of EYFP from pETM11 and pETM14 that differ only in the protease recognition site. We compared two other similar vectors to gain further information about the critical features involved in protein solubility. The EYFP construct from pETM10 accumulated at higher concentrations than that from pRSET (Fig. [2C](#F2){ref-type="fig"}, total lysates). Furthermore, the first construct was apparently soluble because concentrated by the affinity purification step while the EYFP expressed from pRSET seemed mostly insoluble (Fig. [2C](#F2){ref-type="fig"}). The presence of a long, not structured region, at the N-terminus of the recombinant protein expressed by pRSET could explain the different results because, otherwise, the two vectors share the same expression regulative features. The first set of data show that the screening step based on affinity purification and SDS-PAGE analysis is a reliable tool for the identification of soluble recombinant protein, select among constructs expressed from vectors belonging to the same subclass (for instance, pETM10 and 14 with respect to 11), the comparison of different growth conditions, and the evaluation of the degradation rate of the constructs. Large-scale purifications ------------------------- The constructs with the wild type EYFP version were purified from the pellet corresponding to 1 L culture. The supernatants obtained after lysate ultracentifugation were loaded onto CoCl~2~-activated sepharose columns and the recombinant His-tagged protein was eluted after a washing step. After desalting the purified protein was used for stability tests and its concentration calculated according to its specific absorbance at 280 nm (Table [2](#T2){ref-type="table"}). A typical purification pattern (EYFP from pETM10) is reported in Figure 2D to show that the non-specific binding is negligible after extensive washing. ###### Yields calculated after large-scale purification and stability of the wild type EYFP fusion proteins generated using the different pETM expression vectors. Three different aggregation tests were performed (see M & M) and the symbols indicate: aggregation (+), no aggregation (-), not performed (/). **Constructs** **Soluble fusion protein yield (mg/Lculture)** **Aggregation tests** -------------------- ------------------------------------------------ ----------------------- EYFP-10 11.5 \-\-- EYFP-11 1.1 -++ EYFP-14 13.8 \-\-- EYFP-20 2.7 \-\-- EYFP-22 4.9 \-\-- EYFP-30Gluthatione 16.6 \-\-- EYFP-30His 23.3 \-\-- EYFP-33His 18.7 \-\-- EYFP-44 4.2 \-\-- EYFP-50 0 / EYFP-52 26.8 \-\-- EYFP-54 0 / EYFP-55 23.1 \-\-- EYFP-60 42.2 \-\-- EYFP-66 40.8 \-\-- EYFP-70 0 / EYFP-80 0 / EYFP-82 1.4 +++ The highest yields of EYFP fusions were obtained using NusA as a stabilizing partner (constructs 60 and 66) but high yields were also obtained with fusions with both GST and DsbA (30, 33, 52 and 55) and using the vectors pETM10 and 14 (only His tag) (Table [2](#T2){ref-type="table"}). The yields of pure EYFP obtained from the pETM10, 14, 60, and 66 become comparable after removal of the carrier proteins. Trx, DsbA, and DsbC, (pETM20, 22, 52, 55, and 82) fusions yielded lower amounts of soluble protein, as also with the pETM11 vector (His tag only). The constructs expressed from pETM11 and 82 became instable after incubation at 30°C overnight or after repetitive freezing and thawing cycles (Table [2](#T2){ref-type="table"}). This was in contrast to all of the other proteins with fusions that remained monodisperse after these trials. The large-scale experiments confirmed that no soluble protein could be recovered from bacteria transformed with pETM70 (fusion to calmodulin-binding protein) and vectors expressing constructs for periplasmic secretion (pETM50, 54, and 80). In some cases the fusion constructs appeared to be partially cleaved at the level of the linker during the purification. Roughly 10% of the protein purified from bacteria expressing DsbA-EYFP fusions was represented by DsbA alone. Similarly, we recovered 15% of pure GST from the GST-EYFP fractions and 35% of DsbC as a degradation product of the DsbC-EYFP fusion (data not shown). These results confirm that observed at small-scale level (Fig. [2A](#F2){ref-type="fig"}) and indicate that the fragility at the linker position is not a specificity of the GST fusions \[[@B11]\]. Similar degradation products also occur in large scale screening experiments performed using GFP, GST, Trx, MBP as fusion partners \[[@B17]\]. What we observed with other proteins is that the degree of degradation using a specific vector varies from negligible levels to almost 100% \[[@B11]\], suggesting that is the specific interaction carrier-target protein to determine the fragility of the construct rather than the carrier itself. Our results give an indication of how much the length and composition of the expressed non-native N-terminus domain (His-tag/linker/protease recognition sequence) influence the stability of the expressed constructs. The solubility and yields of EYFP recovered from bacteria transformed with pRSET and pETM11 were low (0.4 mg/mL and Fig. [2](#F2){ref-type="fig"}; 1.1 mg/L, Table [2](#T2){ref-type="table"}) in comparison to EYFP from pETM10 and 14 (more than 10 mg/L, Table [2](#T2){ref-type="table"}). The EYFP construct expressed from pRSET has a 33 aa tail at the N-terminus, that of pETM11 has 26 extra aa while the EYFP from pETM14 has only 18 aa and the one from pETM10 has as few as 9. Furthermore, the pETM14 linker has more flexible glycine residues and less large aa than the linker in pETM11. EYFP obtained after proteolytic digestion of any of the fusion constructs was soluble and stable, indicating that the destabilizing effect was specifically due to the extra non-native amino acids at the N-terminal. A possible explanation for the poor results obtained using the cytoplasmic DsbC-EYFP construct from pETM82 would take into consideration the enzymatic activity of the fusion partner. In fact, DsbC has been shown to be capable of catalyzing the formation of disulfide bonds even in the reducing cytoplasm \[[@B8]\] whilst their formation seems incompatible with a stable structure of EYFP. We were also interested in comparing some other parameters. The position of the His-tag at the N-terminus (GST and MBP fusions, Table [1](#T1){ref-type="table"}) did not apparently improve the recovery of the recombinant fusions during the affinity purification. In fact, the yields of constructs in which the His-tag is expressed in the theoretical less accessible between the fusion partner and the EYFP -as for NusA and DsbA fusions- were comparable (Table [2](#T2){ref-type="table"}) and no more unbound recombinant protein was detected in the flow-through and wash fractions (data not shown). We also used the double tagged GST-EYFP to compare the purification efficiency of ion metal affinity for 6xHis and glutathione sepharose for GST. The yield of the fusion purified using the His-tag was slightly higher (Table [2](#T2){ref-type="table"}) but the higher affinity for GST using glutathione-sepharose resulted in less non-specific impurities. The amount of contaminants due to degradation was in the same proportions (data not shown). Specificity of the protease cleavage site ----------------------------------------- The protease recognition sequence had no significant influence on the protein solubility. In fact, the constructs expressed by the TEV and 3C versions of the same vectors (pETM20 and 22, 30 and 33, 60 and 66) gave similar yields (Table [2](#T2){ref-type="table"}). The cleavage efficiency of TEV and 3C was compared using EYFP-Trx fusions from pETM20 and 22 as substrates. Both proteases cleaved more than 95% of their substrates after incubation at 30°C for 2 h (Fig. [3A](#F3){ref-type="fig"}). Nevertheless, 3C could be used 5 times more diluted than TEV, namely 2 μg of protease for mg of substrate instead of 10 μg. Furthermore, 3C was confirmed to be more active than TEV at 4°C (Figure [3A](#F3){ref-type="fig"} and \[[@B16]\]). The smaller mass of 3C in comparison to TEV could lead to a decreased steric hindrance and allow the facilitated access to the recognition site \[[@B22]\]. Otherwise, both proteases remained active in buffers containing up to 500 mM NaCl, 250 mM imidazole, and 4 mM DTT (Fig. [3B](#F3){ref-type="fig"}). 3C was more sensitive to detergents (Triton X-100 and Brij 58) than TEV but less pH-dependent. Most of the combinations of salt, detergent and imidazole had no synergic inhibitory effect (Fig. [3B](#F3){ref-type="fig"}). In conclusion, the choice of either of the two proteases allows the selection of a cleavage buffer which is at least partially optimized for the stability of the substrate protein. {#F3} Conclusion ========== Comparison of the yields of the purified EYFP constructs from the different expression vectors indicated that fusions with NusA and GST, as well as constructs with His-tags and short linkers, resulted in the highest yields of soluble recombinant protein. Furthermore, we did not observe better affinity binding using constructs with a His-tag at the N-terminus than with constructs with an internal tag and the His-GST-EYFP was purified at similar quantitative and qualitative levels using any of the two affinity tags. The analysis of the aggregation state of the purified proteins showed that most of the constructs were both soluble and monodisperse \[[@B23]\]. The solubilizing effect of the different tags has often been reported but independent experiments are difficult to compare because they were performed using a great variety of growth and detection conditions and often with a limited number of test proteins \[[@B6],[@B24],[@B25]\]. More systematic investigations \[[@B17]\] and our experience involving the expression of hundreds of proteins using pETM vectors suggest that the best tag and growth conditions are protein specific. Therefore, the aim is not the identification an improbable ideal tag that always ensures better solubility to any fused passenger protein but to develop strategies for fast and easy screening among different vectors before starting the large-scale production. The pETM collection facilitates the sub-cloning and simplifies the result comparison. With respect to other modular systems used in high throughput proteomics and based on ligation independent cloning \[[@B26]\] and recombination \[[@B27],[@B28]\] rather than classical cloning, the pETM vectors can provide shorter extra sequences at both protein termini. The negative effect of extra amino acids and poorly structured regions on the protein solubility has been related to their length and composition \[[@B29]-[@B33]\] and confirmed by our experiments performed with the constructs expressed by pRSET, pETM10, 11, and 14. Finally, the pETM vectors offer the choice of the specific protease for removal of the tags thus widening the opportunity to select buffer conditions more compatible with the stability of the target protein (Fig. [3](#F3){ref-type="fig"}). Methods ======= Cloning into the pETM vectors ----------------------------- Two EYFP (Yellow Fluorescent Protein, AAU85108) sequences (wild type and mutant I48A) cloned into pRSET (Invitrogen) were used as a template for the PCR reactions. The PCR primers were designed to contain the NcoI restriction site (forward) and either the EcoRI or the NotI (reverse). Two series of pETM vectors \[[@B18]\] were digested with NcoI/EcoRI and NcoI/NotI, respectively, and the digested PCR products were ligated both overnight and by quick-ligation. The four parallel ligation products for each vector were used to transform DH5α competent cells and the highest efficiency was obtained using the combination NcoI/EcoRI and overnight ligation. After plasmid amplification and sequencing, the pETM expression vectors containing the sequence-verified EYFP insert were transformed in BL21 (DE3) competent cells and the colonies used to inoculate 3 mL of LB for preparing bacterial glycerol stocks and material for the small-scale screening. Bacterial culture and protein purification ------------------------------------------ The bacteria were grown at 37°C until they reached an OD~600~of 0.4, the temperature was then decreased to 20°C, the recombinant expression induced with 0.1 mM IPTG at an OD~600~of 0.6 and the pellet corresponding to 1.5 mL was recovered after overnight culture and stored at -20°C. Frozen pellets were resuspended in 350 μ L of 20 mM TrisHCl buffer, pH 8.0, (metal affinity purification) or 350 μ L of 20 mM PBS buffer, pH 7.3, (glutathione sepharose purification) containing 5 mM MgCl~2~1 mM PMFS, 1 mg/mL lysozyme, and 1 μ g/mL DNase. NaCl was added to the Tris buffer to a final concentration of 300 mM. The samples were sonicated for 5 min in a water bath and the total lysate samples were taken before centrifugation at 5,000 g for 10 min. The resulting supernatants were added to either 25 μ L of Talon beads (Clontech) equilibrated in 20 mM Tris buffer, 300 mM NaCl, or 25 μ L glutathione Sepharose 4 Fast Flow resin suspension (Amersham) equilibrated in 20 mM K-phosphate buffer, 150 mM NaCl, respectively. After incubation with shaking at room temperature for 30 min the affinity matrixes were recovered by a short centrifugation and the supernatant discarded. The glutathione sepharose resin was resuspended in PBS, washed for 30 min, recovered as above and then washed again before being resuspended in 25 μ L of SDS-PAGE loading buffer. The Talon beads were washed twice for 30 min in 20 mM Tris-HCl, pH 8.0, 500 mM NaCl, 20 mM imidazole and finally recovered in 25 μ L of SDS-PAGE loading buffer. The purified proteins were visualized by SDS PAGE. The purification protocols were modified at the following steps when large-scale purifications were performed. Pellets from 1L bacterial culture were re-suspended in 4 volumes of buffer (PBS + 5 mM MgCl~2~and 1 mM PMFS or 20 mM TrisHCl buffer, pH 8.0, 500 mM NaCl, 10 mM imidazole 5 mM MgCl~2~and 1 mM PMFS) and sonicated 5 min in a water bath. 1 mg/mL lysozyme and 1 μ g/mL DNase were added and the lysates were incubated with shaking at room temperature for 30 min. The samples were centrifuged at 90.000 × g for 35 min, the supernatants filtered and loaded onto a pre-equilibrated GSTrap or Hi-Trap chelating columns (Amersham) charged with CoCl~2~, respectively, connected to a FPLC system. The elution was performed using either 25 mM Tris HCl, pH 8.0, 10 mM reduced glutathione (glutathione affinity) or 20 mM Tris HCl, pH 8.0, 250 Mm imidazole (IMAC). The purified proteins were buffer exchanged into 50 mM PBS containing 10% glycerine using a HiTrap Desalting column (Amersham), the protein concentration was calculated after measurement of the absorbance at 280 nm, and samples were prepared for SDS-PAGE. Protein aggregation analysis and protease activities ---------------------------------------------------- Protein aggregation was measured according to Nominé et al. \[[@B34]\] using a luminescence spectrometer (SLM Aminco). The stability of the purified constructs was determined by comparing the aggregation index calculated using fresh samples (control) with: a) samples thawed after incubation at -80°C for 24 hours, b) samples frozen and thawed for 5 cycles, c) incubated at 30°C for 24 hours. Protein gels were stained using SimplyBlue Safe Stain (Invitrogen) and the relative protein content in each band quantified using the public domain NIH Image program (developed at the U.S. National Institutes of Health \[[@B35]\]). TEV and 3C proteases were used at concentrations varying from 1:100 to 1:1000 dilutions with respect to the substrate, at 4 and 30°C, and using 2 or 18 hour incubation. The different buffers are indicated in the legend to the Figure [3](#F3){ref-type="fig"}. **Additional File 1**. (PDF) \"The maps of the pETM vectors\". The maps and the MCS regions of the pETM vectors used in the experiments are reported. Acknowledgements ================ The authors wish to thank Anna Peyker for having provided the pRSET-EYFP vectors, Milena Lange for general assistance in the lab and all the people who contributed with their experience to the establishment of the EMBL protein expression facility, in particular Hans van der Zandt, Gunter Stier, Arie Geerlof, and David Drechsel.

Introduction ============ During the process of long bone measurement, locating the precise anatomic plane to perform ultrasound biometry and reducing inter- and intra-operator measurement error are critical for diagnosis. With respect to fetal ultrasound, the biometry of the fetal femur is technically simple due to the one-dimensional nature of the measurement. However, unlike the proximal extremities, the long bones of the distal areas can be difficult to measure. For example, in the distal upper and lower limbs, radius and ulna, and tibia and fibula, the paired bones are located too close together for proper differentiation. Other exacerbating factors, including constant fetal movements, the consequent variation in fetal position, degree of the expectant mother\'s obesity, quantity of the amniotic fluid, and location of the placenta further aggravate the difficulty in making proper measurements. In conventional two-dimensional (2D) ultrasound, the sonographer must manipulate the transducer accordingly to locate an appropriate plane for fitting measurements. The 2D-ultrasound is also burdensome as sonographers must meet with patients directly to conduct a lengthy examination. Furthermore, the accuracy of diagnosis and construction of images mainly depend on the degree of the sonographer\'s experience and expertise. However, the newer three-dimensional (3D) ultrasound scans volume data in a short period of time and reconstruct a 3D image, which can be cut into multiple sections for further manipulation to obtain precise plane much like the saved images of computed tomography or magnetic resonance imaging. 3D-ultrasound is also particularly useful in that the saved images allow diagnostic deliberation by the sonographer even after the examination ends and the patient is unavailable for further testing. Recently published studies have verified the utility of 3D-ultrasound in various diagnostic areas, including fetal biometry, skeletal dysplasia, facial anomaly, and fetal echocardiography \[[@B1][@B2][@B3][@B4][@B5]\]. However, 3D-ultrasound also requires the manipulation of 3D volume data to reconstruct the images, which requires time and effort. Furthermore, such manipulation efforts also depend on the expertise and experience of the sonographer. To compensate for these weaknesses, long bone automated detection system, five-dimensional 5D Long Bone (5D LB) was made available to streamline the process of reconstructing the lower limb long bone images and performing fetal biometry. The present study compared the biometric data of conventional 2D-ultrasound, manually manipulated 3D volume data, and 5D LB-treated 3D volume data to determine the feasibility of 5D LB functions in light of other available methods. Materials and methods ===================== 1. Study population ------------------- This study was performed with 39 pregnant women with singleton pregnancies at 26+0 to 32+0 weeks of gestation who visited the Department of Obstetrics at the Yonsei University Health System between November 2011 and August 2012. Subjects with fetal abnormality was suspected on ultrasound examination, those who had a medicosurgical disease, or multiple pregnancies were excluded from the study. This study was approved by the institutional review board and informed consent was obtained from all the participants. 2. Lower limb long bone length measurement ========================================== One experienced operator (operator 1, a maternal-fetal medicine clinical instructor) and one inexperienced operator (operator 2, two gynecology residents) participated in the ultrasound examination. Every 2D and 3D ultrasound examination was performed twice, once by the inexperienced operator and again by the experienced operator. Each ultrasound scanned the proximal (femur) and distal (tibia and fibula) areas of the long bones of the right lower limb. For the measurement, the Accuvix V20 Prestige (Medison Co., Seoul, Korea) was used. For 2D-ultrasound measurements, a 2-6 MHz transabdominal ultrasound transducer was used and 3D volume data were acquired using a 4-8 MHz volume transducer. For 2D-ultrasound measurement, the transducer was aligned to the long axis of the diaphysis, and the image was captured and used to measure the length. To obtain 3D volume data, the entire bone length of the femur was identified on the screen, as in the 2D-ultrasound measurement, and the long axis of the femur was placed in the direction of the x-axis in the image in which the ultrasound beam was perpendicular to the bone. The femur was included in the volume box such that the bone length occupied about 80% of the entire image, and the volume data were captured using 3D scanning ([Fig. 1A](#F1){ref-type="fig"}). These data were designated the longitudinal-90 (long-90) set. Then, the femur was rotated clockwise or counterclockwise by 45 degrees on the screen, and the volume data were again captured using 3D scanning ([Fig. 1B](#F1){ref-type="fig"}). These data were designated the longitudinal-45 (long-45) set. The distal lower extremity bones, i.e., the right tibia and fibula, were measured similarly: the entire bone length was identified on the screen, the long axis of the bones was placed in the direction of the x-axis in the image, as was femur long-90 and the volume data were captured using 3D scanning ([Fig. 1C](#F1){ref-type="fig"}). The 3D-ultrasonograph long bone length was measured offline by a maternal-fetal medicine specialist (JYK) who had not participated in the ultrasonograph measurements. The acquired volume data were reconstructed in a multiplanar mode to adjust the x, y, and z-axes of each plane. Then, an appropriate plane was located for the long bone length measurements of the right lower limb. Subsequently, the long bone length was measured using the 5D LB with the following procedures. The volume data used in the manual 3D-ultrasound measurement were displayed in an offline multiplanar mode, and the 5D LB set key was pressed on the system, wherein the system automatically analyzed the 3D volume data, reconstructed the 3D image of the long bones, and displayed the measured lengths of the long bones on the screen ([Fig. 2](#F2){ref-type="fig"}). 3. Statistical analysis ======================= Statistical analysis was performed using the PASW ver. 18.0 (SPSS Inc., Chicago, IL, USA) and MedCalc ver. 12.7 (MedCalc Software, Ostend, Belgium). The Kolmogorov-Smirnov test was used to test the normal distribution of the measurement data. The paired *t*-test was performed to analyze intra- and inter-observer variability. Agreement was analyzed by using Bland-Altman plot, Passing-Bablok regression and calculating the interclass correlation coefficient (ICC). *P*-value of \<0.05 was considered statistically significant. Results ======= To measure the length of the fetal femur, tibia, and fibula using different methods, two operators performed 2D- and 3D-ultrasound examinations with a total of 39 subjects two times each. Of the 624 total 5D LB measurements, the overall success rate was 86.2% (538/624); the overall failure rate, 8.3% (52/624); and the overall error rate, 5.4% (34/624). For the femur, the success rate of long-90 measurement was 91.0% (142/156) and that of long-45 measurement was 96.1% (150/156). As such, the long-45 success rate was slightly higher than that of the long-90. The success rate was lower for the measurements of the tibia (80.7%, 126/156) and fibula (76.9%, 120/156), in which the length of both bones was measured simultaneously, than for the measurements of the femur ([Table 1](#T1){ref-type="table"}). To examine the variation in the long bone length measurements depending on the volume scanning angles in the 3D-ultrasound, the variability and agreement were analyzed between the femur long-90 and femur long-45 measurements obtained by manual 3D-ultrasound and 5D LB. As shown in [Table 2](#T2){ref-type="table"}, neither 3D-ultrasound nor 5D LB exhibited a significant difference in measurement values between the different scanning angles. In fact, the scanning angles exhibited an ICC \>0.9, showing a high level of agreement. Consequently, long-90 data were used in all subsequent analyses. The intra- and inter-observer variability and reproducibility were analyzed with respect to the three measurement methods, and the results showed no significant differences among the methods; in fact, a rather high degree of agreement was noted ([Tables 3](#T3){ref-type="table"}, [4](#T4){ref-type="table"}). Then, the measurement values of 3D-ultrasound and 5D LB were compared with the standard 2D-ultrasound measurement values to verify the reproducibility of the two former methods. When the 3D-ultrasound and 5D LB methods were compared with the 2D-ultrasound, the measurement values did not exhibit a significant difference. The ICC was \>0.89 and the Bland-Altman plot showed an even distribution near zero ([Table 5](#T5){ref-type="table"}, [Fig. 3](#F3){ref-type="fig"}). Passing-Bablok regression showed no significant systemic or proportional differences ([Table 6](#T6){ref-type="table"}). Discussion ========== Ultrasound examination can be safely applied to women of childbearing age or pregnant women because it does not use ionizing radiation and is able to obtain real-time tomograms of any region depending on the position at which the sonographer places the probe. In addition, the newer 3D-ultrasonograph method can reconstruct the images from saved volume data to enable the sonographer to analyze the images in various sections even after the patient leaves the examination room. In other words, only a single scan is required to provide sufficient data for in-depth analysis. Even better, the recent development of an automated 3D-ultrasound further reduces the time and effort required to manually manipulate the volume data by automatically reconstructing images from the scanned volume data through a preloaded program on the apparatus. Volume imaging through 3D-ultrasound is useful since the desired image may be reconstructed by manipulating volume data acquired by volume sweeping even if the correct plane is not found at the time of examination. Benacerraf et al. \[[@B5]\] reported that volume data were acquired within 2 minutes and interpreted in 6 to 7 minutes using 3D-ultrasound in the standard fetal anatomic survey, indicating that the temporal efficiency of the 3D-ultrasound was greater than that of 2D-ultrasound, which took 19.6 minutes. They also reported that 3D-ultrasound may be useful in fetal anatomic surveillance because it allows for tomographic imaging study, like computed tomography or magnetic resonance imaging. Acquired 3D volume data can be displayed in three orthogonal planes and provide all the 2D planes for a complete anatomical evaluation of the particular organ. Therefore, 3D-ultrasound with multiplanar reconstruction or surface rendering has advantages in the evaluation of complex anatomy, such as fetal heart, central nervous system and face \[[@B6][@B7][@B8][@B9][@B10]\]. For example, a single volume sweep provides reconstruction of a midsagittal plane for evaluation of nuchal translucency and nasal bone \[[@B11][@B12]\]. From 3D fetal echocardiography, standardized planes and intracardiac structures of the fetal heart can be reconstructed from volume data in real time. Spatiotemporal image correlation enables an automated volume acquisition of the fetal heart by using a mechanical volume transducer and software. Real-time three-dimensional echocardiography is another kind of technology using a matrix array transducer which can display the fetal heart in real time. Furthermore, 3D-ultrasound provides the ability to store volume data that can be manipulated after the patient has left the examination room and also be transmitted electronically to evaluate elsewhere. Despite these advances of 3D-ultrasound, the manipulation of volume data requires a learning curve and is also operator dependent. Automated 3D imaging will allow an operator independent and standardized approach to evaluate complex anatomical evaluation and improve the efficiency by reducing the time to complete the ultrasound examination. Buoyed by the recent advancements in ultrasound technology, numerous studies have been conducted to develop a more efficient and precise automated ultrasound system for biometry. Zador et al. \[[@B13]\] performed automated measurements of the biparietal diameter, occipitofrontal diameter, and head circumference using a personal computer-based system. The results showed that the automated measurements not only were highly correlated with the measurements taken by an experienced sonographer using conventional methods but also decreased the amount of time needed to make the measurements available for analysis. Thomas et al. \[[@B14]\] conducted computerized measurements of the femur and humerus. Other studies have verified the feasibility of computerized automated fetal biometry measurement and its good intra-interobserver variability \[[@B15][@B16]\]. Ultrasound diagnosis based on automated computerized systems can be considered semi-automatic, since the operator searches for the correct plane and the computer performs the automated caliper placement through image subtraction. Thus, the ultimate goal is to create a fully automated system, in which the correct plane is also determined by the instrument. The present study is significant in that it verified the feasibility of automated long bone length measurements through 5D LB system. Our study focused on gestational ages of 26 to 32 weeks to assist less experienced operator in avoiding falsely including the triangular spur artifacts and the distal femoral epiphysis in the late trimester; 20 weeks long bone measurement is quite straight forward as only the diaphysis is seen and to facilitate multiple long bone measurements. In a fetus with suspected skeletal dysplasia, measuring all of the long bones may be cumbersome and skeletal dysplasia (or long bone shortening) usually manifests in the late 2nd to 3rd trimester. The volume sweep initiated by the automated long bone detection system only takes a few seconds to perform and the acquired volume data are available for offline manipulation. Such processes decrease the duration of the examination period, reduce the incidence of musculoskeletal stress-induced injuries common among sonographers \[[@B17][@B18]\], and construct a more precise image plane by being less affected by the movement or position of the fetus. In addition, the manual volume data manipulation required by 3D-ultrasound was replaced by an automated system that enables the manipulation to be completed in just a few seconds. In our study, volume sweep takes 2 seconds with sweep angle of 30 degrees and software execution to measurement output takes 4 seconds. Nonetheless, biometric measurement using 5D LB still requires improvement. In this study, the overall success rate of 5D LB measurements was 86.2% (538/624), the overall failure rate was 8.3% (52/624), and the overall error rate was 5.4% (34/624). Among the above measurements, the femur long-90 and femur long-45 exhibited a relatively high success rate of 91.0% (142/156) and 96.1% (150/156), respectively. However, the tibia and fibula length measurement exhibited lower success rates at 80.7% (126/156) and 76.9% (120/156), respectively. Although it is estimated that the femur long-90 has highest success rate theoretically, the femur long-45 has slightly higher than long-90. Based on 4 cases that had failed, the thigh was abutting uterine wall without space filled with amniotic fluid between the wall and the thigh, thus the software was unable to differentiate bright bone outline and high echogenic skin line. However, when the femur was at 45 degrees-angle of inclination, there was amniotic fluid pocket thus less error for the software in selecting out bone outline. Some of the reasons for 5D LB measurement failures include failure to locate the appropriate sagittal plane to construct an image of the long bone, misplacement of the caliper due to the unusually high soft tissue echo near the target long bone, unclear differentiation between the bone and soft tissue due to the low bone echo, image blurring due to fetal movements or maternal obesity, and acoustic shadowing due to the obstruction of nearby organs or the position of the two long bones. Such measurement errors have been reported since conventional 2D-ultrasound was first used. While volume data manipulation has made strides in rectifying some of the measurement errors, such measurement errors still occur in 3D-ultrasound examination \[[@B19]\]. These errors need to be eliminated, perhaps by compensation using the 5D LB system algorithm. Ultrasound is highly operator dependent thus the quality of ultrasound examination can vary accordingly. In hope to aid in standardization and reduce operator dependency, software that semi-automatically or automatically reproduce 2D image plane or measurement using 3D volume data was developed and has been promising \[[@B6][@B7][@B9][@B10][@B11][@B12]\]. But limitation lies in that with current ultrasound machine that enables us to perform 2D- and 3D-ultrasound using two separate transducers, adding 3D software may not be practical but rather cumbersome and time-consuming since we need to switch back and forth between functions to apply the 3D software. Thus, automated femur measurement software by itself may not be attractive at present setting, however, ultrasound technology is innovating towards combining 2D and 3D transducer functions \[[@B20][@B21][@B22]\] which will allow integration of multiple 3D software with realtime scanning in a most practical way. In such future perspective on where ultrasound technology is heading, 5D LB function should serve as one of gateway to automated biometry software. The present study performed a comparative analysis of the feasibility of the 5D LB program with respect to its fetal long bone length measurement capacity. The capacity of the 5D LB program was compared with that of the conventional 2D- and 3D-ultrasound. Fetal lower limb long bone measurement using 5D LB demonstrated low intra and -inter-observer variability and a high level of agreement compared with data acquired using the conventional imaging techniques. As such, the 5D LB program exhibited high feasibility with respect to its capacity to perform fetal biometry. However, further improvements on the measurement failures and errors must be made for 5D LB to replace conventional 2D-ultrasound fetal biometry. Through this process, more efficient and precise ultrasound diagnosis will be possible. This work was supported by the 3D and 4D Research Task Force, The Korean Society of Ultrasound in Obstetrics and Gynecology, Korea. **Conflict of interest:** No potential conflict of interest relevant to this article was reported. {#F1} {#F2} {#F3} ###### Success rate, error rate and fail rate of measurements of the femur (longitudinal-90 and longitudinal-45), tibia and fibula by five-dimensional Long Bone  Values are presented as n (%). ###### Comparison of femur length measurement by 3D-US and 5D LB in relation to volume sweeping angles.  3D-US, three-dimensional ultrasound; 5D LB, five-dimensional Long Bone; diff, difference between pairs of measurements; SD, standard deviation; CI, confidence interval; ICC, interclass correlation coefficient. ###### Intra-observer agreement in long bone measurement for each technique: 2D-US, 3D-US, and 5D LB  2D-US, two-dimensional ultrasound; 3D-US, three-dimensional ultrasound; 5D LB, five-dimensional Long Bone; SD, standard deviation; CI, confidence interval; diff, difference between pairs of measurements; ICC, interclass correlation coefficient. ###### Inter-obserever agreement in Long Bone measurement for each technique: 2D-US, 3D-US, and 5D LB  2D-US, two-dimensional ultrasound; 3D-US, three-dimensional ultrasound; 5D LB, five dimensional Long Bone; diff, difference between pairs of measurements; SD, standard deviation; CI, confidence interval; ICC, interclass correlation coefficient. ###### Comparison of Long Bone measurement techniques by comparing 3D-US and 5D LB measurements with the 2D-US measurements  3D-US, three-dimensional ultrasound; 5D LB, five-dimensional Long Bone; 2D-US, two-dimensional ultrasound; diff, difference between pairs of measurements; SD, standard deviation; CI, confidence interval; ICC, interclass correlation coefficient. ###### Comparison of Long Bone measurement techniques by comparing 3D-US and 5D LB measurements with the 2D-US measurements  3D-US, three-dimensional ultrasound; 5D LB, five-dimensional Long Bone; 2D-US, two-dimensional ultrasound; CI, confidence interval.

All relevant data are within the paper. Introduction {#sec001} ============ Eating disorders (EDs) are serious psychiatric disorders that predominantly affect female adolescents. In addition, EDs can cause serious medical and psychosocial morbidities. The prevalence rates for different types of EDs range from 0.3% to 1% in young females \[[@pone.0132050.ref001]\]. Anorexia nervosa (AN) is an ED characterized by distorted body image, self-induced starvation, and excessive weight loss with a pathological fear of becoming fat. Psychological, social, genetic, and cognitive deficits play a role in the onset and the maintenance of EDs. Body image distortion is a major symptom of AN. It is widely known that AN patients perceive their own body, particularly body parts (e.g. thighs, abdomen or face) to be fatter than they are.\[[@pone.0132050.ref002]\]. Neural basis of body image distortion has gradually become clear in recent years. Over the past decade, several functional neuroimaging studies have researched the possible neurofunctional correlates of body images distortion in patients with AN and other EDs. These studies have documented different neural activation patterns between patients and controls \[[@pone.0132050.ref003]--[@pone.0132050.ref006]\]. The face is an important component of self-concept, and it is the least occluded part of the body that plays crucial role in interpersonal communications. AN patients see and feel their own face as fatter than it actually is. However, few studies have attempted to investigate the self-face perception in patients with AN. Facial perception is related to the part of the temporal cortex \[[@pone.0132050.ref007]--[@pone.0132050.ref009]\] as well as body perception \[[@pone.0132050.ref010], [@pone.0132050.ref011]\]. Research on self-face perception has been conducted for more than a decade and several studies have suggested that self-face perception involves distinct brain mechanisms compared with stranger-face perception \[[@pone.0132050.ref012]--[@pone.0132050.ref015]\]. Therefore, we hypothesized that subjects with AN who present with own-body image distortion have different self-face perception compared with controls. To our knowledge, there are no published reports on the perception of self-face and stranger-face in patients with AN. Thus, the purpose of this study was to compare neuronal correlates of self-face and stranger-face (female) perception in female adolescents with AN and without AN (controls). Near infrared spectroscopy (NIRS) is a noninvasive neuroimaging methodology that monitors blood volume and oxygenation in the brain \[[@pone.0132050.ref016], [@pone.0132050.ref017]\]. The general rationale of NIRS is that it provides an index of brain function by assessing changes in the concentration of oxygenated hemoglobin (oxy-Hb), deoxygenated hemoglobin (deoxy-Hb), and total-hemoglobin (total-Hb) using measurements of the diffuse transmittance of near-infrared light at an appropriate range of wavelengths. It is widely accepted that brain activity is related to regional changes in blood flow and oxygenation \[[@pone.0132050.ref018]\]. NIRS has been used for more than a decade to examine brain activity during various cognitive tasks \[[@pone.0132050.ref019]--[@pone.0132050.ref023]\]. With regard to facial perception, NIRS has been used to measure neural activities in the temporal area. Several lines of evidence have suggested that the hemodynamic response in the temporal cortex of the brain is activated during face processing \[[@pone.0132050.ref024]--[@pone.0132050.ref031]\]. Compared with other neuroimaging techniques such as functional magnetic resonance imaging (fMRI), NIRS is completely silent, providing a non-intrusive environment for the patients. Furthermore, NIRS can be conducted in a brief period of time, and it requires less restriction of the body and the head. NIRS has been utilized for the assessment of brain activities in children with psychiatric disorders including AN, while the subjects performed cognitive tasks. Nagamitsu et al. reported that the analysis of information obtained by NIRS revealed that the prefrontal hemodynamic response during a cognitive task (word fluency task) differed between patients with AN and controls \[[@pone.0132050.ref032]\]. In the current study, we used NIRS to measure the hemodynamic responses in patients with AN and in age- intelligence quotient (IQ)-matched controls while they viewed self-face and stranger-face photographs in full color. The purpose of this study was to compare the neuronal correlates of viewing self-face and stranger-face images in female adolescents with AN and in controls using a NIRS system which was placed on the bilateral temporal regions. There were different patterns of brain activation in response to the sight of the self-face and stranger-face in female adolescents with AN and controls. Methods {#sec002} ======= 2.1 Participants {#sec003} ---------------- The current study involved 15 female adolescents who fulfilled the criteria for AN in the 4th text revision edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM-IV-TR). The AN subtype of all patients was restrictive. None of the patients had other psychiatric or neurological disorders including past history of pervasive developmental disorder. The Children\'s version of the Eating Attitude Test-26 (ChEAT-26), a screening instrument which is widely used to measure the characteristics of EDs in children \[[@pone.0132050.ref033]\] was administered to all participants. ChEAT-26 has been commonly used as an index of the symptoms of AN; a high score indicates a tendency toward EDs. The patients were recruited from the Department of Pediatrics, Center for Child Development and Psychosomatic Medicine, Dokkyo Medical University Koshigaya Hospital in Japan. They were tested either 2--4 weeks after admission or at the first visit to the clinic. All female adolescents with AN were in the acute phase, yet their general conditions were stable enough to walk and have conversation with others. Their mean age and body mass index (BMI) at the time of study were 13.8 years \[standard deviation (SD) = 1.59\] and 14.4 kg/m^2^ (SD = 2.20), respectively ([Table 1](#pone.0132050.t001){ref-type="table"}). Their severity of weight loss was clinically severe. The control group comprised 15 female volunteers, who were recruited from a local school. They had no history of psychiatric or neurological disorders, and no irregularities in eating behavior. Their mean age and BMI at the time of study were 13.1 years (SD = 1.10) and 18.7 kg/m^2^ (SD = 2.15), respectively ([Table 1](#pone.0132050.t001){ref-type="table"}). All the participants were right-handed. Written informed consent (as outlined in PLOS consent form) was obtained from the individuals in this manuscript. This study was approved by the Ethical Committee of the Dokkyo Medical University Koshigaya Hospital (hosp-k 24016). 10.1371/journal.pone.0132050.t001 ###### Clinical characteristics of all participants. {#pone.0132050.t001g} AN (n = 15) Control (n = 15) *P* ------------- ------------- ------------------ ---------- Age (years) 13.8±1.59 13.1±1.10 0.0671 BMI (kg/m2) 14.4±2.20 18.7±2.15 \<0.0001 BMI-SDS -3.1+1.80 -0.25±0.71 \<0.0001 ChEAT26 34.0±20.7 8.7±3.71 0.0008 IQ 102.3±13.8 104.3±9.32 0.4465 Data is presented as mean ± SD (Mann-Whitney *U* test) 2.2 Stimuli and design {#sec004} ---------------------- The stimuli presentation sequence consisted of baseline and test periods. For the test period, there were two stimulus conditions: self-face and stranger-face conditions. In those conditions the participants viewed self-face or stranger-face color photographs, respectively ([Fig 1](#pone.0132050.g001){ref-type="fig"}). The order of the conditions was counterbalanced across the participants. The self-face photograph was the participant's own face. The stranger-face photograph was the face of another age-matched female adolescent with AN in the case of the AN group and another age-matched control in the case of the control group. To eliminate the external features of the faces, we removed the hair from the face images so that only the internal features were visible. The sizes of the stimuli were approximately 13 × 10 deg for the faces and 3.5 × 3.5 deg for the blinking black dots. {#pone.0132050.g001} The duration of the test period was fixed at 15s. Each face image was shown 15 times in one test period. The duration of each face image was 800ms, and the 200ms inter-stimulus interval was filled by the presentation of a fixation point (a blinking black dot). Each test period was preceded by a baseline period that was fixed at 20 s. During the baseline period, the screen of the monitor was filled with a white uniform blank. The blank was presented for 800 ms, and a 200 ms inter-blank interval was filled with the same fixation point (a blinking black dot) used in the test period. The results obtained from viewing the uniform blank were used as the baseline. 2.3 Procedure {#sec005} ------------- Each participant was tested while sitting in a chair and facing a computer screen approximately 50 cm away. The participants passively watched the stimuli while their brain activity was measured, and they were allowed to watch the stimuli for as long as they were willing. The behavior of the subjects was observed by three evaluators, and the viewing time was recorded using a video system that allowed tracking the line of sight during the experiment. 2.4 Recording {#sec006} ------------- Real time cerebral-cortex imaging and measurement were conducted using the Hitachi ETG-4000 and ETG-7100 systems (Hitachi Medical, Chiba, Japan). The NIRS wavelengths used for measurement were the same for both systems (695 and 830 nm); therefore, we believe that the differences in the specifications of the two systems had no effect on our results. We recorded 24 channels simultaneously, with 12 channels for the right temporal area and 12 for the left. Both instruments generate two wavelengths of NIR (695 nm and 830 nm) and measured the time courses of the levels of oxy-Hb, deoxy-Hb, and their sum (total-Hb) with 0.1 s time resolution. The NIRS probes contained nine optical fibers (3 × 3 arrays). Of the nine fibers, five were emitters and four were detectors. The distance between the emitters and detectors was set at 3 cm. Each pair of adjacent emitting and detecting fibers defined a single measurement channel. The placement of the participants' probes was set on the bilateral temporal area centered at T5 and T6 according to the International 10--20 system \[[@pone.0132050.ref034]\] ([Fig 2](#pone.0132050.g002){ref-type="fig"}). This was the same location of probes used by Nakato et al. \[[@pone.0132050.ref024]\]. When the probes were positioned, the experimenter ensured that the fibers were attached to the participants' scalp correctly. The Hitachi ETG-4000 and 7100 systems automatically detect whether the contact adequately measures emerging photons for each channel. Channels were rejected from the analysis if adequate contact between the fibers and participants' scalp could not be achieved because of hair interference. {#pone.0132050.g002} 2.5. Data analysis {#sec007} ------------------ The NIRS data analysis was performed in a manner similar to that of Nakato et al. \[[@pone.0132050.ref024]\]. Before performing data analysis, we monitored the video recording of the participants' behavior to evaluate adequate trials for the analysis. We removed the trial from the analysis when either of the following occurred: \[[@pone.0132050.ref001]\] accumulative looking time within the trial did not reach 5 s, or \[[@pone.0132050.ref002]\] movement artifacts were detected by the analysis of sharp changes in the time series of the raw NIRS data. The raw data on oxy-Hb, deoxy-Hb, and total-Hb concentrations from each channel were digitally band-pass-filtered at 0.02--1.0 Hz to remove the noise caused by the heartbeat pulsations or any longitudinal signal drift. The mean concentration of each channel within a subject was then calculated by averaging data across the trials in a time series with a 0.1 s resolution. Based on the mean concentrations in the time series and the "baseline" values (mean concentration at 3 s before the trial), we calculated the Z-scores for oxy-Hb, deoxy-Hb, and total-Hb in the self-face and stranger-face conditions for each channel within a subject \[[@pone.0132050.ref024]--[@pone.0132050.ref030], [@pone.0132050.ref035]\]. The Z-score at each time point was used to indicate the deviation of hemodynamic response from the "baseline" during the presentation of facial images. The Z-scores were calculated separately for oxy-Hb, deoxy-Hb, and total-Hb in the self-face and stranger-face conditions. A previous NIRS study reported that the inferior region in the temporal cortex is activated more than the other regions in the temporal cortex when participants looked at facial images \[[@pone.0132050.ref024]\]. Therefore, the Z-scores obtained from the set of 6 channels within the left inferior temporal area and the set of 6 channels in the right inferior temporal area were each averaged across trials for every participant, as previously described \[[@pone.0132050.ref024]\]. Although the raw NIRS data were relative values and could not be averaged directly across participants or channels, the normalized data (i.e., Z-scores) could be averaged regardless of the unit \[[@pone.0132050.ref017], [@pone.0132050.ref036]\]. The Z-score is a reliable indicator of the concentration changes because the analysis is independent of the differential path length factor. Consistent with our previous studies using NIRS \[[@pone.0132050.ref024], [@pone.0132050.ref027]--[@pone.0132050.ref029]\], the hemodynamic response of oxy-Hb increased gradually during the test period and reached a plateau approximately 12 s after the stimuli onset ([Fig 3](#pone.0132050.g003){ref-type="fig"}). Thus, we used the mean Z-score for the 12--17 s interval of the test period for the statistical analysis. To examine the possibility that the activity resulting from the self-face or stranger-face stimuli were greater or smaller than 0, we conducted two-tailed one-sample t-tests with the mean Z-scores for oxy-Hb deoxy-Hb, and total-Hb concentrations in each condition. The statistical significance threshold was set to P \< 0.05. {#pone.0132050.g003} Results {#sec008} ======= 3.1. Clinical characteristics of participants {#sec009} --------------------------------------------- The characteristics of the study participants are listed in [Table 1](#pone.0132050.t001){ref-type="table"}. There was no significant difference in age or IQ between the two groups. The BMI and BMI-standard deviation score (BMI-SDS) values in the AN group were significantly lower compared with those in the control group (*P* \< 0.0001, Mann-Whitney *U*-test), and the ChEAT-26 \[[@pone.0132050.ref033]\] scores in the AN group were significantly higher compared with those in the control group (*P* = 0.0008, Mann-Whitney *U*-test). 3.2. Available trials for statistical analysis {#sec010} ---------------------------------------------- Data for 30 participants who looked at faces in five trials in both the self-face and stranger-face conditions were obtained. The date were averaged across trials and across participants for each condition (self-face: mean = 4.87 trials, SD = 0.51; stranger-face: mean = 4.93 trials, SD = 0.59). There was no significant difference between the numbers of available trials under each condition. 3.3. NIRS measurements {#sec011} ---------------------- At the wavelengths measured by the Hitachi ETG-4000 and ETG-7100 systems (695 and 830 nm), the estimations of oxy-Hb and total-Hb concentrations are more sensitive than those of deoxy-Hb concentrations \[[@pone.0132050.ref022], [@pone.0132050.ref037]\]. However, when compared with deoxy-Hb and total-Hb concentrations, the oxy-Hb concentration is a better indicator of changes in regional cerebral blood flow \[[@pone.0132050.ref038]\]. The time-course of the average change in oxy-Hb, deoxy-Hb, and total-Hb concentrations was measured in the participants during the presentation of the self-face and stranger-face images ([Fig 3](#pone.0132050.g003){ref-type="fig"}), and Z scores were calculated based on the differences between the values after stimulus onset and the background values ([Fig 4](#pone.0132050.g004){ref-type="fig"}). {#pone.0132050.g004} The Z-scores for oxy-Hb concentrations in the right temporal cortices of the AN group were significantly greater than 0 in both the self-face and stranger-face conditions (*P* = 0.015 and *P* = 0.032, respectively). In contrast, the Z-scores for oxy-Hb concentrations in the right temporal cortices of the control group were significantly greater than 0 in the self-face condition (*P* = 0.011), but not in the stranger-face condition (*P* = 0.35). The Z-scores for deoxy-Hb concentrations in the right temporal cortices of the AN group were significantly less than 0 in both of the self-face and stranger-face conditions (*P* = 0.039 and *P* = 0.011, respectively), and the Z-scores for deoxy-Hb concentrations in the right temporal cortices of the control group were significantly less than 0 in the stranger-face condition (*P* = 0.012), but not in the self-face condition (*P* = 0.38). The Z-scores for total-Hb concentrations in the right temporal cortices of the AN group were not significantly different from 0 in either condition. In contrast, the Z-scores for total-Hb concentrations in the right temporal cortices of the control group were significantly higher than 0 in the self-face condition (*P* = 0.0068). None of the Z scores (oxy-Hb, deoxy-Hb, or total-Hb) were significantly different from 0 in the left temporal cortices with the exception of the Z-scores for deoxy-Hb concentrations, which were significantly less than 0 in the AN group (*P* = 0.0015). Discussion {#sec012} ========== In this study, we hypothesized that patients with AN who present with own-body image distortion have different self-face perception compared with controls. To compare the neuronal correlates of viewing self-face images and stranger-face images between AN and controls, we measured hemodynamic responses while the participants viewed full-color self-face and stranger-face photographs using NIRS. NIRS is a noninvasive neuroimaging method that does not require physical restraint of the subject and has a brief examination time compared with other neuroimaging methods. Thus, it is suitable for children and patients in the acute phase of AN. The objective of this study was to use NIRS to investigate the brain activity of participants in the AN and control groups while they viewed full-color self- and stranger-face photographs. The fusiform gyrus and superior temporal sulcus, which are located in the inferior temporal lobe play important roles in face recognition \[[@pone.0132050.ref007], [@pone.0132050.ref008]\]. Therefore, the placement of the NIRS' probes was attached to the bilateral temporal regions of the participants, as in Nakato et al. \[[@pone.0132050.ref024]\]. The Z-scores obtained from the six channels within the inferior temporal areas were analyzed. NIRS monitors blood volume and oxygenation in the brain. When deoxy-Hb and total-Hb concentrations are compared, the oxy-Hb concentration reflects changes in regional cerebral blood flow more accurately \[[@pone.0132050.ref038]\]. In our study, the hemodynamic response, as indicated by oxy-Hb, increased gradually during the test period (i.e., photograph-viewing period) and was found to occur predominantly in the right hemisphere, suggesting that the brain activity of facial perception mainly occur on the right side. A recent neuroimaging study also reported that the main aspects of facial processing occur predominantly in the right hemisphere \[[@pone.0132050.ref009]\]. Sugiura et al. reported that activation selective to the self-face compared with the stranger-face was observed in the right occipito-temporo-parietal junction \[[@pone.0132050.ref012]\]. Furthermore, in recent surveys that adopted a stringent statistical threshold, the neural response specific to the self-face was detected predominantly in the right hemisphere \[[@pone.0132050.ref013], [@pone.0132050.ref014]\]. Here, the AN and control groups comprising adolescent females differed in their changes in brain activity in the right temporal cortices in response to viewing self-face and stranger-face photographs. In other words, the oxy-Hb concentrations of the control group significantly increased only in the self-face condition, whereas oxy-Hb concentrations of the AN group significantly increased in both conditions. These findings suggest that self-face and stranger-face were normally distinguished in controls, which is in line with previous studies \[[@pone.0132050.ref012]--[@pone.0132050.ref015]\], whereas self-face and stranger-face were processed in the same manner in patients with AN. Body image distortion is a key symptom of AN, and it plays a role in the onset and maintenance of AN \[[@pone.0132050.ref002]\]. The face is one of the most important a part of the body, and body (including a face) recognition is related to a part of the temporal cortex \[[@pone.0132050.ref010], [@pone.0132050.ref011]\]. Additionally the fusiform gyrus \[[@pone.0132050.ref003]--[@pone.0132050.ref005]\] and temporal gyrus \[[@pone.0132050.ref039], [@pone.0132050.ref040]\], which play important roles in facial perception, are also related to self-body image perception in AN. The reason why patients with AN could not distinguish between self-face and stranger-face in this study may be related to their body image distortion. Some studies have focused on the possible presence of cerebral functional disturbance in the development of AN. Cognitive deficits in body image, attention, learning, cognitive flexibility, memory, emotion, and facial affect have been observed in patients with AN by various types of neuropsychological examinations. However, it is still uncertain whether these cognitive deficits precede ED or appear as a consequence of malnutrition \[[@pone.0132050.ref041]--[@pone.0132050.ref043]\]. In the current study, all female adolescents with AN were in the acute phase. In the future, we intend to investigate brain activity again after recovery. In conclusion, to our knowledge, the present study is the first to assess the brain activity during self-face and stranger-face processing in adolescents with AN. We have demonstrated that patterns of brain activation in response to the sight of the self-face and stranger-face are different between female adolescents with AN and controls. The face is an important part of the body; moreover, we found hemodynamic differences in the right temporal area, which corresponds to structures known to contribute to facial perception and self-body image perception \[[@pone.0132050.ref003]--[@pone.0132050.ref005], [@pone.0132050.ref039], [@pone.0132050.ref040]\]. Thus, different patterns of brain activation in response to the sight of the self-face and stranger-face in this study may be related to body image distortion in AN. We are very grateful to the participants and their parents. We also thank Mr. M. Fujiwara (Hitachi Medical Cooperation, Japan) for his technical support regarding the NIRS measurement of adolescents. [^1]: **Competing Interests:**The authors have declared that no competing interests exist. [^2]: Conceived and designed the experiments: TI YS KS HI MK RO MKY SK RK RS. Performed the experiments: TI YS KS HI MK RO MKY SK. Analyzed the data: TI YS KS HI MK MKY SK. Contributed reagents/materials/analysis tools: MKY SK RK RS. Wrote the paper: TI YS RO MKY SK RK RS. Making the test-stimuli: TI MKY SK RK RS. Collection of participants: TI KS RO RS. Collection of clinical characteristics: TI RO. Reviewing manuscript: TI YS KS HI MK RO MKY SK RK RS.

Introduction ============ Primary osteosarcoma of the breast is a rare tumor, which is indistinguishable from conventional osteosarcoma of the bone and other extraskeletal sites using histological examination ([@b1-ol-06-03-0745]--[@b6-ol-06-03-0745]). In comparison, bone-producing spindle cell neoplasms with an epithelial origin, known as metaplastic (sarcomatoid) carcinomas, and malignant phyllodes tumors are more common ([@b7-ol-06-03-0745]). Primary breast osteosarcomas are considered to be highly aggressive tumors that are associated with early recurrence and a tendency for hematogenous, instead of lymphatic, spread, most commonly to the lungs ([@b2-ol-06-03-0745],[@b3-ol-06-03-0745],[@b6-ol-06-03-0745],[@b8-ol-06-03-0745]--[@b11-ol-06-03-0745]). The tumor subtype, size and mitotic figures may be predictors of prognosis. The present study describes a case of a chondroblastic/fibroblastic variant of this rare tumor that had a good prognosis. Case report =========== A 77-year-old Chinese woman presented to the Chengde Central Hospital Outpatient Department with complaints of a lump in the left breast that was self-detected a month prior to presentation. There was no history of nipple discharge, fever and pain. There was no history of breast trauma, prior local irradiation and surgery, nor any other tumor history. The patient denied using any hormonal therapy or a family history of breast disease. A breast examination showed a 7×7×6-cm irregular, firm mass in the lower inner quadrant of the left breast. The mass was poorly mobile and adherent to the skin and chest wall. No axillary lymphadenopathy was detected upon physical examination. Mammography showed that the mass was relatively well demarcated and partially calcified. The tumor did not invade the overlying skin and underlying chest wall. Breast carcinoma was thus indicated. The patient refused a needle biopsy and underwent a mastectomy. The mastectomy specimen contained a 7-cm, relatively well-circumscribed mass, which had a gray-white cartilaginous to firm calcified appearance in the cross-section. Microscopically, the tumor was slightly lobular and relatively well demarcated, however, the adjacent fat tissue had been invaded and the surrounding non-neoplastic breast parenchyma revealed compressed lobular units ([Fig. 1A and B](#f1-ol-06-03-0745){ref-type="fig"}). The tumor was mainly composed of cartilaginous components. The abundant cartilaginous proliferation varied from mature lacunar cartilage to poorly-differentiated areas displaying myxoid changes with no lacunar arrangement. In certain areas there was a transition from cartilaginous proliferation to fibrous cells. More than half of the tumor cells (\~60%) were spindle-like and sparse with minimal cytological atypia, which were mainly observed in the central portion, while the other cells (40%), which were mainly in the periphery, were epithelioid, atypical and dense. High-power magnification revealed low mitotic activity even in the dense area (1 mitoses/10 high power microscopic fields; [Fig. 1C and D](#f1-ol-06-03-0745){ref-type="fig"}). Neoplastic osteoid woven bone or trabeculae were observed in the central portion ([Fig. 1E and F](#f1-ol-06-03-0745){ref-type="fig"}). Hemorrhage and necrosis foci were also observed in the central portion. No lymph-vascular invasion or neural invasion was observed. There was no histological evidence of an epithelial or carcinomatous component, despite extensive sampling of the tumor. No evidence of a preexisting malignant phyllodes tumor was present. The tumor was negative for cytokeratin, as well as for the estrogen and progesterone receptors and HER2. The tumor was classified as a chondroblastic/fibroblastic variant of osteosarcoma. The 15 axillary lymph nodes studied showed no metastasis. The patient underwent a simple mastectomy without post-operative adjuvant chemotherapy or radiation therapy. At 60 months post-mastectomy, the patient was alive and well without clinical evidence of local recurrence or distant metastasis. Written informed consent was obtained from the patient for publication of this case report and all accompanying images. Discussion ========== Carcinoma is the most common malignancy of the breast and sarcomas form a minority of breast neoplasms. Primary osteosarcoma of the breast is rare and represents \<1% of all primary breast malignancies. However, the actual incidence of primary osteosarcoma is difficult to determine, as a number of the \~100 previously-reported cases are likely to have included metaplastic carcinomas, as well as osteogenic sarcomas arising in association with a biphasic tumor, such as a phyllodes tumor or carcinosarcoma ([@b7-ol-06-03-0745]). Almost every previous reference to primary osteosarcoma of the breast in the literature is in the form of single case reports. Silver and Tavassoli reported a clinicopathological analysis of 50 cases observed over a 38-year period, the largest collection of primary breast osteogenic sarcomas to date ([@b6-ol-06-03-0745]). In almost all cases, the patients were diagnosed clinically as having breast carcinoma and the final diagnosis was established by histology. The histogenesis of primary osteosarcoma of the breast remains unclear, but an origin from totipotent mesenchymal cells of the breast stroma or a transformation from a preexisting fibroadenoma or phyllodes tumor has been suggested ([@b1-ol-06-03-0745]). The presentation of breast osteosarcoma usually occurs at an advanced age, in contrast with skeletal osteosarcomas where the patients are younger. There has been a report of a breast osteosarcoma diagnosis at the age of 96 years, however, the usual mean age at presentation has been reported to be \~64 years, and cases are more frequently postmenopausal ([@b1-ol-06-03-0745],[@b6-ol-06-03-0745]). Risk factors for extraskeletal osteosarcomas have not been identified to date, although certain cases have been attributed to local irradiation, trauma or the presence of a foreign body ([@b6-ol-06-03-0745]). The majority of patients present with a mobile, often large, irregular lump, without axillary metastases. Primary osteosarcomas should be separated from malignant phyllodes tumors with malignant heterologous differentiation. Other diagnoses that should be excluded are: i) osteosarcomatous differentiation in carcinoma of the breast (metaplastic carcinoma), which are likely to be myoepithelial differentiation; ii) osteogenic sarcoma arising from the underlying ribs or sternum; and iii) metastatic osteosarcoma. The present patient was 77 years old with no history of trauma or irradiation, no tumor in other sites and absence of a biphasic image in the breast. Other diagnoses were therefore excluded, and it was concluded that the patient had primary osteosarcoma of the breast. The reverse zonation pattern is observed in half of all breast osteosarcomas (the majority of fibroblastic types) ([@b6-ol-06-03-0745]). In the zonation pattern, the bone tissue at the peripheral region is more mature than that in the central region, which is generally the phenomenon observed in myositis ossificans. In the reverse zonation pattern, the central portion of the tumor was largely composed of neoplastic osteoid woven bone or trabeculae, with sarcomatous stroma confined to the periphery of the neoplasm, imparting a zonal pattern (the reverse of that observed in myositis ossificans). A reverse zonation pattern was observed in the present case. In addition, central necrosis has been noted in 10% of all breast osteosarcomas. Generally, central necrosis is observed in large tumors, such as tumors with surface ulceration ([@b6-ol-06-03-0745]). Areas of necrosis were identified in the center of the tumor in the present case. Primary breast osteosarcomas are considered to be highly aggressive tumors associated with early recurrence and a tendency for hematogenous, instead of lymphatic, spread, most commonly to the lungs ([@b2-ol-06-03-0745],[@b3-ol-06-03-0745],[@b6-ol-06-03-0745],[@b8-ol-06-03-0745]--[@b11-ol-06-03-0745]). It has been reported that the predictive factors for outcome in this disease are tumor size, histological grade and subtype. Silver and Tavassoli ([@b6-ol-06-03-0745]) observed that mammary osteosarcomas that were on average 4.6 cm in diameter were associated with a significantly higher survival rate than larger tumors. Histological differentiation is important since fibroblastic osteosarcomas have a improved survival outcome compared with other pathological types, in breast osteosarcomas as well as in primary bone and soft tissue osteosarcomas. The histological appearance of primary osteosarcoma of the breast varies according to the cellular composition (fibroblastic, osteoblastic and osteoclastic), as well as the type and amount of the matrix (osteoid, osseous and chondroid). Abundant chondrosarcomatous components are rarely observed ([@b6-ol-06-03-0745],[@b12-ol-06-03-0745]). Silver and Tavassoli ([@b6-ol-06-03-0745]) classified primary osteosarcoma of the breast into the fibroblastic, osteoblastic and osteoclastic types. The osteoclastic type is not in the current World Health Organization (WHO) classification of osteosarcomas ([@b13-ol-06-03-0745]). Silver and Tavassoli ([@b6-ol-06-03-0745]) report that this subtype contains abundant, multi-nucleated osteoclast giant cells diffusely scattered within the sarcomatous stroma or surrounding osteoid islands. The fibroblastic type constituted the majority of cases (56%) in their study, while the osteoblastic type was the smallest subtype (16%). Patients with osteosarcoma of the fibroblastic subtype have a significantly improved 5-year survival rate (67%) than those with osteosarcoma of the osteoclastic or osteoblastic subtype (31%). Mitotic activity in osteosarcoma of the breast ranges between 5 and \>20 mitotic figures/10 high-power fields. The osteosarcoma of the breast that has a high mitotic rate is limited to the osteoblastic and osteoclastic variants. This may be the reason for the different survival rates among subtypes. The present tumor was classified into the chondroblastic/fibroblastic variant, and the mitotic activity of this tumor was low. The reasons for the lack of recurrence in the 5 years subsequent to the surgery may be related to the subtype, minimal cytological atypia, low mitotic rate and good local control due to adequate resection, even though the tumor size was 7 cm. Treatment for localized disease should include complete surgical removal of the tumor with an adequate margin. A simple mastectomy may be indicated to ensure complete excision of large tumors with cryptically infiltrative margins. Axillary lymph node dissection is not indicated in the setting of clinically-negative nodes, as axillary node involvement is rare. Whereas, for metastatic disease, chemotherapy based on the classic drugs (doxorubicine, ifosfamide, cisplatinium and methotrexate) used for osteosarcoma is the main treatment. Distinguishing metaplastic carcinoma and carcinosarcoma from osteosarcoma of the breast is important, since the former requires treatment as a primary breast cancer. The majority of reported primary breast osteosarcomas are considered highly aggressive tumors associated with early recurrence and a propensity for hematogenous spread. However, the presence of a chondroblastic/fibroblastic variant, minimal cytological atypia, a low mitotic rate and good local control due to adequate resection may result in a good prognosis. {#f1-ol-06-03-0745} [^1]: Contributed equally

Introduction {#Sec1} ============ Huntington's disease (HD) is a hereditary autosomal dominant neurodegenerative disease caused by an unstable expansion of over 35 glutamines in the amino-terminus of the huntingtin (HTT) protein \[[@CR1], [@CR2]\]. Although both wild-type and mutated HTT (mHTT) proteins are ubiquitously expressed, HD is associated with selective neuronal loss. Degeneration occurs preferentially in the striatum and cortex and, in later stages, it extends to a variety of brain regions, including the hippocampus and hypothalamus \[[@CR3], [@CR4]\]. As a result, HD symptoms include cognitive alterations, psychiatric abnormalities and motor dysfunction \[[@CR4], [@CR5]\]. To date, there are no disease-modifying treatments for HD, mostly due to the lack of full understanding of the physiological functions of HTT and the pathological cascade initiated by mHTT in the brain. However, HTT is found to be crucial for neuronal survival, vesicle transport, calcium homeostasis, and transcriptional regulation \[[@CR6]\]. Previous studies have identified the repressor element 1-silencing transcription factor/neuron-restrictive silencer factor (REST/NRSF) as a key regulator of HTT-mediated gene expression \[[@CR7]--[@CR9]\]. REST/NRSF target genes are mainly involved in neuronal development and synaptic transmission and their expression is found to be dysregulated in HD \[[@CR8]\]. Specifically, HTT sequesters REST/NRSF in the cytoplasm and prevents it from forming the nuclear co-repressor complex at the neuron-restrictive silencer element (RE1/NRSE) nuclear site, thereby facilitate gene transcription. On the other hand, mHTT is incapable of retaining REST/NRSF in the cytosol leading to the pathological entry of REST/NRSF into the nucleus and inhibition of gene transcription \[[@CR7]--[@CR9]\]. REST/NRSF expression is regulated by the canonical Wnt/β-catenin pathway \[[@CR10]--[@CR12]\]. β-catenin is usually retained at the plasma membrane in a complex with the cytoplasmic portion of cadherins \[[@CR13]--[@CR15]\]. Phosphorylation of β-catenin disrupts its interaction with cadherins leading to either β-catenin cytosolic degradation or its translocation to the nucleus where it can regulate REST/NRSF transcription \[[@CR16], [@CR17]\]. The metabotropic glutamate receptor 5 (mGluR5) is a G~αq~-coupled receptor and it has been shown to regulate cadherin/β-catenin complex assembly in the vascular endothelium \[[@CR17]\]. However, the molecular link(s) between mGluR5 and cadherin/β-catenin complex formation and whether mGluR5 can regulate REST/NRSF expression and signaling in neurons remains unclear. Additionally, we and others have shown that mGluR5 plays a key role in the progression of HD in mouse models of the disease \[[@CR18]--[@CR20]\]. However, it remains to be defined whether mGluR5-mediated regulation of N-cadherin/ β-catenin signaling and REST-dependent gene expression is altered in HD. In the present study, we show that mGluR5 regulates N-cadherin phosphorylation via Src kinase to alter N-cadherin/β-catenin assembly and REST/NRSF expression in mouse primary corticostriatal neurons. Moreover, we show that in 15 month old zQ175 mice, enhanced Src and N-cadherin phosphorylation can be corrected by chronic pharmacological inhibition of mGluR5 using the negative allosteric modulator (NAM), CTEP. This is paralleled by a reduction in REST/NRSF and consequent increase in its target gene synaptosomal nerve-associated protein 25 (SNAP-25) expression in CTEP-treated zQ175 mice. These findings are also validated in a BACHD mouse model of HD as we show that genetic ablation of mGluR5 reduces REST/NRSF and increases SNAP25 mRNA and protein expression. Thus, our results show that mGluR5 regulates REST/NRSF expression and signaling and highlight the relevance of Wnt pathway in HD progression and therapeutics. Results {#Sec2} ======= mGluR5 modulates N-cadherin phosphorylation in primary neuronal cultures {#Sec3} ------------------------------------------------------------------------ Several previous studies demonstrated the importance of mGluR5 in the pathophysiology of HD \[[@CR18]--[@CR21]\]. Moreover, mGluR5 was shown to modulate cadherin/β-catenin association in the vascular endothelium \[[@CR17]\]. Since the involvement of Wnt pathway in the development of HD is not well-known, we tested whether the N-cadherin/ β-catenin complex was modulated by mGluR5 in neuronal cells and whether this modulation was affected by mHTT. To start with, in vitro experiments were performed using primary corticostriatal neuronal cultures derived from wild-type E15 mouse embryos. Neuronal cultures were treated with either 100 μM (S)-3,5-Dihydroxyphenylglycine (DHPG), a group I mGluR agonist, or 10 μM 2-chloro-4-((2,5-dimethyl-1-(4-(trifluoromethoxy)phenyl)-1H-imidazol-4-yl)ethynyl) pyridine (CTEP), a mGluR5-specific NAM, for 10, 30, 60, 180 and 360 min. DHPG induced significant N-cadherin phosphorylation at Y860 following treatment for 60 and 180 min (Fig. [1](#Fig1){ref-type="fig"}a and b), whereas CTEP reduced N-cadherin-pY680 phosphorylation following 10, 30, 60 and 180 min of treatment compared to non-treated neurons (Fig. [1](#Fig1){ref-type="fig"}c and d). Total levels of β-catenin protein expression were not changed by either DHPG (Fig. [1](#Fig1){ref-type="fig"}a and b) or CTEP (Fig. [1](#Fig1){ref-type="fig"}c and d) treatment. These results indicated that mGluR5 could trigger N-cadherin phosphorylation without affecting β-catenin expression. Therefore, it was likely that mGluR5 activation modulated the assembly of N-cadherin/β-catenin complex, as this association was known to be tightly-regulated by phosphorylation of N-cadherin \[[@CR16], [@CR17]\]. Fig. 1mGluR5 modulates N-cadherin phosphorylation but not β-catenin expression in primary neuronal cultures. Representative western blots and quantification of fold change of N-cadherin phosphorylation at Y860 (pY860) and β-catenin expression with the corresponding loading controls in primary cultured corticostriatal neurons derived from E15 wild-type mouse embryos stimulated with either the mGluR group I agonist DHPG (100 μM) **(a** and **b)** or the mGluR5-selective NAM, CTEP (10 μM) **(c** and **d)** for 10, 30, 60, 180 and 360 mins. N-cadherin-pY860 was normalized to total N-cadherin, and β-catenin was normalized to vinculin (*n* = 4--5 for each group). Values represent mean ± SEM and are expressed as a fraction of the non-treated (NT) cultures. \* denotes *P* \< 0.05 versus NT cultures. Statistical significance was assessed by one-way ANOVA and Fisher's LSD multiple comparisons mGluR5 modulates N-cadherin/β-catenin complex through Src kinase in primary neuronal cultures {#Sec4} --------------------------------------------------------------------------------------------- We next tested the mechanism by which mGluR5 triggered N-cadherin-pY680 phosphorylation to disrupt N-cadherin/β-catenin interaction. Previously published studies highlighted the regulatory role of Src kinase in cadherin/β-catenin dissociation \[[@CR22]--[@CR25]\]. Since Src kinase was also demonstrated to be a downstream substrate activated by mGluR5 \[[@CR26]--[@CR28]\], we analyzed Src phosphorylation at Y416 in neuronal cultures treated with either 100 μM DHPG or 10 μM CTEP for 10, 30, 60, 180 and 360 min. DHPG induced Src phosphorylation following 60 and 180 min of treatment (Fig. [2](#Fig2){ref-type="fig"}a and b), whereas CTEP reduced Src phosphorylation after 10, 30 and 60 min of treatment compared to non-treated values (Fig. [2](#Fig2){ref-type="fig"}c and d). These results suggested that mGluR5 could activate Src kinase which potentially phosphorylated N-cadherin and regulated the scaffolding of N-cadherin and β-catenin complex. Fig. 2mGluR5 modulates Src phosphorylation expression in primary neuronal cultures. Representative western blots and quantification of fold change Src phosphorylation at Y416 (pY416) with the corresponding loading controls in primary cultured corticostriatal neurons from E15 wild-type mouse embryos stimulated with either the mGluR group I agonist DHPG (100 μM) (**a** and **b)** or mGluR5-selective NAM, CTEP (10 μM) (**c** and **d)** for 10, 30, 60, 180 and 360 mins. pSrc was normalized to vinculin (*n* = 4 for each group). Values represent mean ± SEM and are expressed as a fraction of the non-treated (NT) cultures. \* *P* \< 0.05 vs NT cultures. Statistical significance was assessed by one-way ANOVA and Fisher's LSD multiple comparisons To assess whether mGluR5-dependent Src phosphorylation of N-cadherin influenced the interaction of N-cadherin with β-catenin, we immunoprecipitated N-cadherin and blotted for co-immunoprecipitated β-catenin in wild-type corticostriatal neurons following 60 min exposure to either 100 μM DHPG, 10 μM CTEP or 1 μM A419259 (a Src family kinase inhibitor). We found that treatment with either CTEP or A419259 increased the interaction between N-cadherin and β-catenin in neurons, whereas a 60 min exposure to DHPG did not alter the interaction between N-cadherin and β-catenin (Fig. [3](#Fig3){ref-type="fig"}a and b). More so, addition of DHPG to A419259 did not alter the effects of A419259 on N-cadherin and β-catenin interaction. These results suggested that the inhibition of either mGluR5 or Src signaling could increase N-cadherin and β-catenin complex formation. Fig. 3Inhibition of either mGluR5 or Src activity increase the interaction between N-cadherin and β-catenin in primary neuronal cultures. Representative western blots (**a**) and quantification (**b**) for coimmunoprecipitation of N-cadherin with β-catenin **(upper panel)** in primary cultured corticostriatal neurons from E15 wild-type mouse embryos stimulated with either mGluR5-selective NAM, CTEP (10 μM), Src family kinase inhibitor A419259 (1 μM) or the mGluR group I agonist DHPG (100 μM) for 60 min. **Lower panel (a)** shows N-cadherin lysates used to normalize N-cadherin co-immunoprecipitation with β-catenin (*n* = 4 for each group). Values represent mean ± SEM and are expressed as a fraction of the non-treated (NT) cultures incubated with β-catenin antibody only. \**P* \< 0.05 versus NT cultures. Statistical significance was assessed by one-way ANOVA and Fisher's LSD multiple comparisons REST/NRSF signaling is modulated by mGluR5 in primary neuronal cultures {#Sec5} ----------------------------------------------------------------------- REST/NRSF was previously shown to be an important transcriptional repressor regulated by canonical Wnt pathway and modulates the transcription of genes involved in neurogenesis, neuronal development, and synaptic transmission \[[@CR12], [@CR29]\]. Our findings so far suggested the possibility that mGluR5 was signaling via the Wnt/β-catenin pathway. Therefore, we assessed whether either DHPG or CTEP treatment altered the expression of REST/NRSF in neuronal cultures. We found that a 30 min treatment of cultures with DHPG resulted in an increase in REST/NRSF expression (Fig. [4](#Fig4){ref-type="fig"}a and b), whereas CTEP treatment for 30, 60 and 180 mins reduced REST/NRSF expression compared to non-treated control cells (Fig. [4](#Fig4){ref-type="fig"}c and d). We then tested whether mGluR5-mediated regulation of REST/NRSF expression was paralleled by changes in the expression of SNAP-25, a protein that contains a RE1 transcriptional regulatory sequence and was previously demonstrated to be a key target gene of REST/NRSF \[[@CR30]--[@CR32]\]. We found that DHPG treatment for 10, 30, 60, 180 and 360 min significantly reduced SNAP-25 expression when compared to non-treated cultures (Fig. [4](#Fig4){ref-type="fig"}a and b). Conversely, CTEP treatment for 60 min significantly enhanced SNAP-25 expression when compared to non-treated cultures (Fig. [4](#Fig4){ref-type="fig"}c and d). Together, these results demonstrated that mGluR5 can regulate REST/NRSF expression and such regulation was paralleled by changes in the expression of its target gene SNAP-25. Fig. 4mGluR5 modulates both REST/NRSF and SNAP-25 expression in primary neuronal cultures. Representative western blots and quantification of fold change in REST/NRSF and SNAP-25 protein expression with the corresponding loading controls in primary cultured corticostriatal neurons from E15 wild-type mouse embryos stimulated with either the mGluR group I agonist DHPG (100 μM) (**a** and **b**) or mGluR5-selective NAM, CTEP (10 μM) (**c** and **d**) for 10, 30, 60, 180 and 360 mins. REST/NRSF was normalized to vinculin, and SNAP-25 was normalized to actin (*n* = 3--4 for each group). Values represent mean ± SEM and are expressed as a fraction of the non-treated (NT) cultures. \**P* \< 0.05 versus NT cultures. Statistical significance was assessed by one-way ANOVA and Fisher's LSD multiple comparisons. REST/NSRF was probed on β-catenin blots and therefore the same vinculin blot is presented for REST/NRSF (Fig. 4a and c) and β-catenin (Fig. [1](#Fig1){ref-type="fig"}a and c) Chronic mGluR5 antagonism alters REST/NRSF signaling via N-cadherin/β-catenin complex in zQ175 mice {#Sec6} --------------------------------------------------------------------------------------------------- We previously provided evidence that aberrant mGluR5 signaling played a key role in HD pathology and that both pharmacological and genetic silencing of mGluR5 could rescue motor deficits and mitigate HD pathology in zQ175 and Q111 mouse models, respectively \[[@CR18], [@CR19]\]. Therefore, we assessed whether altered REST/NRSF was one of the potential components contributing to pathological mGluR5 signaling in zQ175 mice and whether CTEP treatment could correct REST/NRSF expression and signaling. To do this, we assessed brain lysates from 15 month old male wild-type and zQ175 mice that were treated with either vehicle or CTEP (2 mg/Kg every 48 h) for 12 weeks. We found that Src and N-cadherin phosphorylation was significantly increased in brain lysates derived from vehicle-treated zQ175 mice and was reduced to values comparable to wild-type mouse lysates following chronic CTEP treatment (Fig. [5](#Fig5){ref-type="fig"}a-c). In contrast, β-catenin expression was not altered between either wild-type or zQ175 mice that were treated with either vehicle or CTEP (Fig. [5](#Fig5){ref-type="fig"}a and d). However, we found that REST/NRSF expression was reduced in CTEP-treated zQ175 mice compared to vehicle-treated mice (Fig. [5](#Fig5){ref-type="fig"}a and e) and was associated with a significant increase in SNAP-25 expression in CTEP treated zQ175 mice (Fig. [5](#Fig5){ref-type="fig"}a and f). Thus, our in vivo findings in zQ175 mouse model corroborated our in vitro data and confirmed that mGluR5 modulated REST/NRSF expression and its target genes via the Wnt pathway in HD mice. Fig. 5Chronic CTEP treatment modulates REST/NRSF signaling by N-cadherin/β-catenin complex in zQ175 mice. Representative western blots (**a**) and quantification of fold change in Src phosphorylation at Y416 (pY416) (**b**), N-cadherin phosphorylation at Y860 (pY860) (**c**), β-catenin protein expression (**d**), REST/NRSF protein expression (**e**) and SNAP-25 protein expression (**f**) with the corresponding loading controls in brain lysates from heterozygous zQ175 and wild-type (WT) mice after chronic treatment with either vehicle or CTEP (2 mg/kg) for 12 weeks. N-cadherin-pY860 was normalized to total N-cadherin, and Src-pY416, β-catenin, REST/NRSF and SNAP-25 were normalized to vinculin (*n* = 5--6 to each group). Values represent mean ± SEM and are expressed as a fraction of the vehicle-treated WT value. \* denotes *P* \< 0.05 and statistical significance was assessed by two-way ANOVA and Fisher's LSD multiple comparisons mGluR5 modulates REST/NRSF signaling in BACHD mice {#Sec7} -------------------------------------------------- To further validate our findings, we assessed REST/NRSF and SNAP-25 mRNA levels in the bacterial artificial chromosome (BAC) HD mouse (BACHD) model, which was previously shown to exhibit a progressive neurodegenerative phenotype \[[@CR33], [@CR34]\], in the absence (BACHD/mGluR5^−/−^) and presence of mGluR5 (BACHD). We analyzed hippocampal lysates from both 6 and 12 month old, mice since REST/NRSF was previously demonstrated to be highly expressed in post mitotic hippocampal neurons \[[@CR35]\]. We detected a reduction in REST/NRSF mRNA expression in BACHD/mGluR5^−/−^ mice as early as 6 months of age when compared to wild-type mice (Fig. [6](#Fig6){ref-type="fig"}a), and found that at 12 months of age both mGluR5^−/−^ and BACHD/mGluR5^−/−^ mice presented with reduced REST/NRSF mRNA expression when compared with both wild-type and BACHD mice (Fig. [6](#Fig6){ref-type="fig"}b). SNAP-25 mRNA expression was increased in mGluR5^−/−^ and BACHD/mGluR5^−/−^ mice at 6 and 12 months of age when compared with age-matched BACHD mice, whereas SNAP-25 mRNA was significantly reduced in 12 month old BACHD mice when compared to age-matched wild-type mice (Fig. [6](#Fig6){ref-type="fig"}c and d). We finally validated that the changes in mRNA expression were translated at protein level by immunoblotting for REST/NRSF and SNAP-25 proteins in the same group of mice. At 6 months of age, we detected a reduction in REST/NRSF protein in mGluR5^−/−^ and BACHD/mGluR5^−/−^ compared to BACHD mice and an increase in SNAP-25 protein in mGluR5^−/−^ and BACHD/mGluR5^−/−^ compared to wild-type mice (Fig. [7](#Fig7){ref-type="fig"}a). REST/NRSF protein levels of 12 month old mGluR5^−/−^ and BACHD/mGluR5^−/−^ mice were significantly reduced, whereas, SNAP-25 levels were significantly increased when compared to age-matched wild-type and BACHD mice (Fig. [7](#Fig7){ref-type="fig"}b). Taken together, we were able to validate in a different mouse model and using a genetic silencing approach that mGluR5 is a critical regulator of REST/NSRF signaling in HD. Fig. 6Genetic ablation of mGluR5 modulate REST/NRSF and SNAP-25 mRNA levels in BACHD mice. mRNA levels of REST/NRSF (**a** and **b**) and SNAP-25 (**c** and **d**) in hippocampus samples from wild-type (WT), mGluR5^−/−^, BACHD and BACHD/ mGluR5^−/−^ mice at 6 and 12 months of age. mRNA levels were assessed by quantitative RT-PCR, which was performed in triplicate and normalizes to actin mRNA levels (*n* = 6 to each group). Values represent mean ± SEM and are expressed as a fraction of WT. \* denotes *P* \< 0.05 and statistical significance was assessed by one-way ANOVA and Fisher's LSD multiple comparisonsFig. 7Genetic ablation of mGluR5 modulate REST/NRSF and SNAP-25 protein levels in BACHD mice. Representative western blots and quantification of fold change in protein levels of REST/NRSF and SNAP25 with the corresponding loading controls in hippocampal lysates from wild-type (WT), mGluR5^−/−^, BACHD and BACHD/ mGluR5^−/−^ mice at (**a**) 6 and (**b**) 12 months of age. REST/NRSF was normalized to vinculin and SNAP-25 was normalized to actin (*n* = 4--6 to each group). Values represent mean ± SEM and are expressed as a fraction of WT value. \* denotes *P* \< 0.05 and statistical significance was assessed by one-way ANOVA and Fisher's LSD multiple comparisons Discussion {#Sec8} ========== Emerging evidence indicates that the dysregulation of the transcriptional repressor REST/NRSF cell signaling and the consequent epigenetic remodeling represents a critical mechanism in the progression of the neurodegeneration associated with ischemia and AD \[[@CR30], [@CR31], [@CR36], [@CR37]\]. Specifically, an increase in REST/NRSF signaling is linked to neuronal death during ischemia \[[@CR36], [@CR37]\]. On the other hand, the loss of REST/NRSF expression is associated with cognitive impairment in AD \[[@CR30]\]. Studies in HD have shown that mHTT, contrary to HTT, cannot maintain the cytoplasmic localization of REST/NRSF leading to its nuclear translocation resulting in the repression of many targets genes, including brain-derived neurotrophic factor (BDNF) \[[@CR7]--[@CR9], [@CR38]--[@CR40]\]. Here, we show that mGluR5 controls REST/NRSF-mediated gene expression by inducing Src-dependent disassembly of the N-cadherin/β-catenin scaffold. We also show that pathological activation of mGluR5 in two distinct mouse models of HD is associated with aberrant REST/NRSF signaling that is mitigated by either genetic or pharmacological silencing of mGluR5. Thus, it is evident that impaired REST/NRSF signaling represents one of the mechanisms by which mGluR5 contributes to HD pathophysiology at the nuclear level. We provide in vivo and in vitro evidence that Wnt signaling downstream of mGluR5 regulates REST/NRSF expression and its target gene SNAP-25. We also show in two HD mouse models, zQ175 and BACHD, that the inhibition of mGluR5 signaling either pharmacologically or genetically can reduce REST expression and consequently enhance SNAP-25 expression. This is in line with previous work that shows that mHTT causes the pathological entry of REST/NRSF into the nucleus, leading to a transcriptional repression of its target genes such as SNAP-25 \[[@CR7]--[@CR9]\] and that both pharmacological and genetic ablation of mGluR5 reduces mHTT burden in HD mice \[[@CR18], [@CR19]\]. The pharmacological and genetic inhibition of mGluR5 is also associated with improved motor function and disease pathology in zQ175 and HdhQ111/Q111 mouse models \[[@CR18], [@CR19], [@CR41]\]. Therefore, we propose that defects in mGluR5-regulated REST/NRSF signaling contribute to the pathophysiology of HD. REST/NRSF has been implicated in the regulation of more than 2000 genes within the mammalian genome \[[@CR42]\], but only a subset of target genes responsive to REST/NRSF are associated with the widespread neuronal dysfunction in HD \[[@CR7], [@CR31]\]. SNAP25 is a t-soluble NSF attachment receptor (SNARE) presynaptic protein, which is involved in the regulation of synaptic vesicle exocytosis \[[@CR43]\]. Reduction of SNAP25 expression in brain samples from patients with a higher HD pathological grade has been reported, which is correlated with a defect in the neurotransmitter release machinery \[[@CR44]\]. A defect in the pre-synaptic release machinery may also potentially affect other processes of relevance to the pathogenesis of HD, such as BDNF release from cortical neurons that is important for the survival of the striatal medium-sized spiny neurons \[[@CR45]\]. Interestingly, BDNF is targeted by REST/NRSF and reduction in BDNF expression has been reported in both mouse models and patients of HD \[[@CR41], [@CR46], [@CR47]\]. A decreased synthesis and transport of BDNF is believed to underlie the neuronal loss in the caudate nucleus and the putamen in the dorsal striatum, and the striatum vulnerability could explain the involuntary motor dysfunction characteristic of HD \[[@CR45], [@CR46], [@CR48]\]. In line with these reports, we have previously reported that brain BDNF levels are reduced in zQ175 mice in a mGluR5-dependent manner \[[@CR41]\]. The modulation of REST/NRSF, its target gene SNAP-25 and BDNF expression by mGluR5 NAM may therefore represent a novel pharmacological tool to halt the progression of HD and potentially other neurodegenerative diseases. Targeting mGluR5 did not alter in vitro expression of β-catenin, the key factor in Wnt pathway that induces REST/NRSF expression, and we did not detect any change in β-catenin expression in either vehicle or CTEP-treated zQ175 mice. Activation of tyrosine kinases is known to drive β-catenin translocation to the nucleus and promote its binding to the TCF/LEF family of transcription factors to facilitate gene expression. Specifically, Src kinase can regulate N-cadherin/β-catenin association and, consequently, the nuclear accumulation of β-catenin \[[@CR49], [@CR50]\]. Previous work in melanoma cells showed that Src activation leads to N-cadherin phosphorylation at Y860 resulting in the subsequent uncoupling of β-catenin \[[@CR50]\]. Interestingly, Src kinase is a known downstream target of mGluR5 \[[@CR26]--[@CR28]\]. Our findings using cultured neurons show that the interaction between N-cadherin and β-catenin is modulated by mGluR5 in a Src kinase-dependent manner, since the inhibition of mGluR5 by CTEP or Src by A419259 increased the interaction between β-catenin and N-cadherin. Moreover, Src kinase is likely responsible for N-cadherin phosphorylation at Y860, as the kinetics of Src phosphorylation correspond with N-cadherin phosphorylation and its association with β-catenin. The change in DHPG-evoked N-cadherin phosphorylation required at least 60 mins of exposure to be detectable and therefore, it is likely that discerning changes in the N-cadherin and β-catenin association after DHPG exposure may require more than 60 mins. This may explain why DHPG did not reduce the co-immunoprecipitation of N-cadherin with β-catenin. It is noteworthy that DHPG is a Group I mGluR agonist and can potentially activate mGluR1 \[[@CR51]\] . However, it is evident that we detect opposite changes in N-cadherin and Src phosphorylation as well as REST and SNAP-25 expression when cultures were exposed to CTEP indicating that DHPG-induced changes in neuronal cultures are mGluR5-mediated. More so, the effects of CTEP on REST/NRSF signaling were more robust compared to DHPG that can be possibly attributed to the constitutive activity of the receptor and the inverse agonistic properties of CTEP \[[@CR52], [@CR53]\]. We have also validated our neuronal culture findings in vivo and detected a substantial increase in mGluR5-mediated Src and N-cadherin phosphorylation in brain lysates from zQ175 mice that was sensitive to treatment with CTEP. Our findings are in line with previously published work in HN33 cells where the expression of polyglutamine-expanded huntingtin is associated with ∼5-fold increase of Src phosphorylation and induces the translocation of activated Src from cytoplasm to cell membrane \[[@CR54]\]. Thus, it is possible that mHTT enhances mGluR5-dependent Src activation and triggers tyrosine-phosphorylation of its targets, such as N-cadherin. Because the antagonism of mGluR5 in zQ175 mice improves the motor phenotype and disease pathology and normalizes Src and N-cadherin phosphorylation, it is likely that aberrant Src/REST signaling is one of the mechanisms by which mGluR5 contribute to HD pathophysiology. It is worth noting that autophagy can play a role in the degradation process of REST/NRSF and hence, determines its nuclear availability \[[@CR30], [@CR31]\]. More so, mGluR5 is known to regulate autophagy \[[@CR19], [@CR55], [@CR56]\], and indeed we detected an increase in REST/NRSF expression in DHPG-treated neurons and vehicle-treated zQ175 mice as well as a reduction in REST/NRSF expression in CTEP-treated neurons and CTEP-treated zQ175 mice. Thus, it is possible that regulation of autophagy may be another mechanism by which mGluR5 modulates the expression and nuclear availability of REST/NRSF in neurons. In summary, although the contribution of the impaired Wnt canonical pathway to the pathology of HD has been reported previously \[[@CR57]--[@CR59]\], this study highlights a potential mechanistic link between mGluR5 and Wnt pathway and its contribution to HD pathology. We show that mGluR5 via Src kinase can regulate the assembly of the N-cadherin/ β-catenin complex and, as a consequence modulates the expression of REST/NRSF and of its downstream gene targets. Moreover, mHTT can alter Src activity and the expression of REST/NRSF and its target genes (Fig. [8](#Fig8){ref-type="fig"}), which can be reversed following either mGluR5 pharmacological blockade or genetic deletion. Thus, enhanced efforts should be directed towards exploiting the impact of mGluR5 on REST/NRSF-mediated gene expression, as this pathway may provide a conserved pathophysiological mechanism between HD and other neurodegenerative diseases. Fig. 8Schematic representation of the proposed model for the modulation of REST/NRSF signaling by mGluR5 and mHTT. **a** Shown is a schematic for mGluR5 signaling modulation by both mGluR5 agonist, DHPG, and the mGluR5-selective negative allosteric modulator, CTEP in the presence of HTT. mGluR5 induces Src family phosphorylation that phosphorylates N-cadherin at Y860, and this phosphorylation site in N-cadherin disrupts its interaction with β-catenin in cellular membrane. Then, β-catenin is released to cytoplasm and translocates to the nucleus, which becomes available to bind the TCF/LEF family of transcription factors to induce target gene expression, such as REST/NRSF. Under basal conditions, HTT sequesters REST/NRSF in the cytoplasm, thereby preventing it from forming the nuclear co-repressor complex at the RE1/NRSE nuclear site, allowing the transcription of REST/NRSF target gene, such as SNAP-25. SNAP-25 could affect the brain-derived neurotrophic factor (BDNF) release, supporting the survival of the striatal medium-sized spiny neurons. **b** The presence of mHTT enhances mGluR5-dependent Src activation, which culminates an increase of N-cadherin phosphorylation. Also, mHTT cannot retain REST/NRSF in the cytosol, causing the pathological entry of REST/NRSF into the nucleus, reducing SNAP-25 gene transcription and potentially BDNF release, which is correlated to neuronal death Materials and methods {#Sec9} ===================== Reagents {#Sec10} -------- CTEP was purchased from Axon Medchem and DHPG and A419259 from Tocris. Horseradish peroxidase (HRP)--conjugated anti-rabbit immunoglobulin G secondary antibody was from Bio-Rad. Neurobasal medium, N2 and B27 supplements, GlutaMAX (50 mg/ml penicillin and 50 mg/ml streptomycin), TRIzol, Nuclease-Free Water, and Power SYBR® Green PCR Master Mix were purchased from Thermo Fisher Scientific. Rabbit anti-REST (07--579) was from Merck. Rabbit anti-Src family (pY416; 2101) was from Cell Signaling Technology. Rabbit anti-N-cadherin (ab18203), phospho-N-cadherin (pY860; ab119752), β-catenin (ab6302), and SNAP25 (ab5666) were from Abcam. Reagents used for Western blotting were purchased from Bio-Rad, and all other biochemical reagents were from Sigma-Aldrich. zQ175 mice and drug administration {#Sec11} ---------------------------------- All animal experimental protocols were approved by the University of Ottawa Institutional Animal Care Committee and were in accordance with the Canadian Council of Animal Care guidelines. Animals were individually caged and housed under a constant 12-h light/dark cycle and given food and water ad libitum. Heterozygous zQ175 HD mice were obtained as a courtesy of CHDI Foundation from The Jackson Laboratory (stock \#370476) and bred to establish littermate-controlled male wild-type (WT). zQ175 knockin mice carry \~ 188 CAG repeat expansions. Groups of 12 male wild-type and zQ175 mice were aged to 12 months of age, and 6 mice from each group were treated every 48 h with either vehicle \[dimethyl sulfoxide (DMSO) in chocolate pudding\] or CTEP (2 mg/kg; dissolved in DMSO and then mixed with chocolate pudding) for 12 weeks. This drug dose was calculated weekly on the basis of weight and is consistent with the dose given to fragile X and Alzheimer's disease mice \[[@CR60], [@CR61]\]. At the end of the 12-week treatment, mice were sacrificed by exsanguination, and the brains were collected and randomized for western blot analysis. BACHD/mGluR5^−/−^ (double mutant) {#Sec12} --------------------------------- FVB/NJ (wild-type, RRID: IMSR_JAX:001800) and FVB/N-Tg (HTT\*97Q) IXwy/J (BACHD) transgenic mice \[[@CR34]\] and mGlu5R knockout B6; 129-Grm5tm1Rod/J (mGluR5^−/−^) mice were purchased from The Jackson Laboratory (Bar Harbor, USA). For the generation of the double mutants, mGluR5^−/−^ mice and BACHD mice were crossed, obtaining the F1 parental lineage. Afterwards, F1 mice were crossed to obtain littermate male mice at the ages of 6 and 12 of WT, mGluR5^−/−^, BACHD and BACHD/mGluR5^−/−^ (double mutant). Mice were housed in an animal care facility at 23 °C on a 12 h light/12 h dark cycle with food and water provided ad libitum. All mice that euthanized in this study were first anesthetized with ketamine/xylazine (80/8 mg/kg) i.p. before cervical dislocation and the brains were collected, dissected and randomized for PCR and immunoblotting analyses. All animal experimental protocols were conducted in accordance with the Universidade Federal de Minas Gerais Ethics Committee on Animal Use, CEUA, 234/2016. Neuronal primary culture preparation {#Sec13} ------------------------------------ Neuronal cultures were prepared from the corticostriatal region of WT E15 embryo brains. After dissection, corticostriatal tissue of each embryo was digested by trypsin followed by cell dissociation using a fire-polished Pasteur pipette. Cells were plated on poly-L-ornithine coated dishes in Neurobasal medium supplemented with N2 and B27 supplements, 2 mM GlutaMAX, 50 μg/ml penicillin, and 50 μg/ml streptomycin (Thermo Fisher Scientific). Cells were incubated at 37 °C and 5% CO~2~ in a humidified incubator and cultured for 12 to 15 days with medium replenishment every 4 days at the day of the experiment. Cell were starved in Hank's balanced salt solution (HBSS) for 1 h. Cells were then treated with CTEP or 100 μM DHPG for 10 min, 30 min, 60 min, 180 min and 360 min and 1 μM A419259 for 60 min at 37 °C, as indicated in the Figure legend. Following treatment, neuronal cultures were collected for immunoblotting and coimmunoprecipitation. Immunoblotting {#Sec14} -------------- Mouse brains were dissected and lysed in ice-cold triton lysis buffer \[50 mM tris (pH 8.0), 150 mM NaCl, and 1% Triton X-100\]. Neuronal primary cultures obtained from WT embryos were lysed in ice-cold RIPA buffer \[0.15 M NaCl, 0.05 M tris-HCl, pH 7.2, 0.05 M EDTA, 1% Nonidet P40, 1% Triton X-100, 0.5% sodium deoxycholate, 0.1% SDS\]. Both buffers contained protease inhibitors (1 mM AEBSF \[4-(2 aminoethyl) benzenesulfonyl fluoride hydrochloride\], leupeptin (10 μg/ml), and aprotinin (2.5 μg/ml)) and phosphatase inhibitors (10 mM NaF and 500 μM Na3VO4) and all samples were centrifuged at 15,000 rpm at 4 °C for 15 min. The supernatant was collected, and the total protein levels were quantified using Bradford protein assay (Thermo Fisher Scientific). Samples were prepared by adding 3x loading buffer containing β-mercaptoethanol to homogenates containing 30--70 μg of total proteins. Samples were then boiled for 10 min at 95 °C, resolved by electrophoresis on a 7.5% SDS--polyacrylamide gel and transferred onto nitrocellulose membranes (Bio-Rad). Membranes were blocked in tris-buffered saline (pH 7.6) containing 0.05% Tween 20 (TBST) and 5% nonfat dry milk for 2 h at room temperature and then incubated overnight at 4 °C with primary antibodies (1:1000) diluted in TBST containing 1% nonfat dry milk. Membranes were then incubated with secondary antibodies (anti- rabbit/mouse) diluted (1:5000) in TBST containing 1% nonfat dry milk for 1 h. Membranes were washed in TBST and bands were detected and quantified using a Bio-Rad chemiluminescence system. Quantitative RT-qPCR {#Sec15} -------------------- RNA from cortical samples of mice at 6 and 12 months of age was isolated using TRIzol reagent as per manufacturer's instructions (Thermo Scientific). RNA was resuspended in of nuclease-free water, and its concentration was analyzed by spectrophotometer (NanoDrop™, Thermo Scientific). cDNAs were prepared from 2 μg of total RNA extracted and RT-qPCR was performed from 10 × diluted cDNA using Power SYBR Green PCR Master Mix in the QuantStudio7 Flex real-time PCR system platform (Applied Biosystems). mRNA levels of REST/NRSF and SNAP25 were quantified using the following primers: REST (forward: 5′-CATGCTGATTAGAGGCCACA-3′; reverse: 5′GTGCGAACTCACACAGGAGA -3′); SNAP25 (forward: 5′ GCCTTCTCCATGATCCTGTC − 3′; reverse: 5′- CTTCATCCGCAGGGTAACAA-3′). Changes in gene expression were determined with the 2^−ΔΔCt^ method using actin as a housekeeping gene. Co-immunoprecipitation {#Sec16} ---------------------- Neuronal primary cultures obtained from WT embryos were lysed in ice cold triton lysis buffer \[0.5 M HEPES, 2.5 M NaCl, 0.5 M MgCl2, 0.5 M EDTA, 0.2% Triton X-100, pH 7.4\] containing protease inhibitors. Lysates were rotated for 1 h at 4 °C and centrifuged to pellet insoluble material. Precleared supernatant was incubated with anti-β-catenin antibody to immunoprecipitate N-cadherin. Following this incubation, freshly washed protein G-sepharose beads were added to lysate/antibody mixture and samples were rotated for 2 h at 4 C. Beads were washed three times with phosphate buffered saline, eluted with 3x SDS sample buffer containing β-mercaptoethanol and analyzed by immunoblotting. Statistical analysis {#Sec17} -------------------- Means ± SEM are shown for the number of independent experiments indicated in figure legends. GraphPad Prism was used to analyze data for statistical significance. Statistical significance (*p* \< 0.05) was determined by one-way or two-way analysis of variance (ANOVA) testing followed by Fisher's LSD as indicated in each figure legend. BACHD : Bacterial artificial chromosome (BAC) Huntington's disease BDNF : Brain-derived neurotrophic factor CTEP : 2-chloro-4-((2,5-dimethyl-1-(4-(trifluoromethoxy)phenyl)-1H-imidazol-4-yl)ethynyl) pyridine DHPG : (S)-3,5-Dihydroxyphenylglycine HD : Huntington's disease HTT : Huntingtin mGluR5 : Metabotropic glutamate receptor 5 mHTT : Mutant huntingtin NAM : Negative allosteric modulator REST/NRSF : Repressor element 1-silencing transcription factor/neuron-restrictive silencer factor RE1/NRSE : Neuron-restrictive silencer element SNAP-25 : Synaptosomal nerve-associated protein-25  **Publisher's Note** Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. S.S.G.F holds a Tier I Canada Research Chair in Brain and Mind. K.S.A is a Lecturer at the Department of Pharmacology & Toxicology, Faculty of Pharmacy, University of Alexandria, Egypt. Thanks to Shaunessy Hutchinson for breeding and drugging the zQ175 colony. J.M.D., K.S.A, F.M.R and S.S.G.F were responsible for the conception and design of all experiments. J.M.D. and K.S.A performed experiments and data analysis. J.M.D. and K.S.A wrote the manuscript and F.M.R and S.S.G.F edited manuscript and supervised the study. The author(s) read and approved the final manuscript. This study was supported by the Huntington's Society of Canada, Krembil Foundation and Canadian Institutes for Health Research (CIHR) grants PJT-148656, PJT-153317 and PJT-165967 to S.S.G.F, and clinician postdoctoral fellowship from the Alberta Innovates Health Solutions and CIHR to K.S.A. All data generated or analyzed during this study are included in this published article. All animal experiments and protocols were approved by the University of Ottawa animal care committee in accordance with the Canadian Council of Animal Care guidelines (CMM 2519) and by Universidade Federal de Minas Gerais Ethics Committee on Animal Use (CEUA 234/2016). Human ethics approval is not applicable. Not applicable. The authors declare that they have no competing interests.

*Porcine circovirus-*2 (PCV2), belongs to the *Circovirus* genus of the family *Circoviridae* and is associated with a number of syndromes defined as porcine circovirus associated diseases (PCVAD). The genome of PCV2, composed of DNA, which is circular and covalently closed, contains two major reading frames (ORF1 and 2) and some minor but not less important ORFs. ORF1 encodes the replicase and ORF2 encodes the capsid protein of approximately 30 kDa ([@b9]). The virus is recognized worldwide as endemic in commercial swine herds. Phylogenetically, PCV2 is divided into two major groups, PCV2a and PCV2b and the most distinct nucleotide differences between the two groups are found in the capsid protein ([@b9]). A third genotype (PCV2c) had been detected, recently, in Danish animals by Dupont ([@b8]). PCV2a and PCV2b are worldwide distributed and no association with pathogenicity has been found. In healthy herds in Sweden and Spain, PCV2b is the most prevalent genotype, whereas PCV2a more related to PMWS affected herds. However, in Switzerland, there was an association of PCV2b and the presence of disease in animals ([@b13]). In Brazil, both group were also detected and no relationship were observed between clinical signs and different genotypes of PCV2 ([@b4]). These wildlife boars were initially introduced in different regions of Brazil with commercial explorations purpose to attend an alternative for the segment of exotic flavors of the meat market. However, over time, it proved to be commercially disadvantageous and these animals began to live into the wild being a significant reservoir of the disease. The prevalence of PCV2 in wildlife boars has been reported in different countries. The seroprevalence in Belgian and Spanish herds were from 30 to 40% ([@b12]). PCV2 infection was detected in about 20% of Hungarian wild boars using polymerase chain reaction (PCR) ([@b7]). In Greece, there was a report stating that wild boars with wasting and dyspnea had died and PCV2 was detected in tissue samples and the histopathologic lesions were consistent with PMWS described in domestic pigs ([@b10]). In Brazil, a high (54.09% to 89.53%) prevalence was reported ([@b2]) indicating the importance of this virus in the wild boar population. In an effort to understand the PCV2 diversity, the purpose of this study was to phylogenetically characterize the genetic material of the virus detected from two Brazilian wildlife boars. The clinical samples were collected from two 90 days-old wild boar pigs, with weight loss (medium weight 15 Kg) and wasting. The semi-intensive wild boars farm presented a clinical history of high mortality in 70 - 90 days-old pigs (\> 50 %). The two pigs were necropsied and lymph nodes (mesenteric and inguinal) were collected. Macroscopically, an enlargment was observed of mesenteric and inguinal lymph nodes. Homogenates (20% w/v) of the samples were prepared and DNA extraction was carried out by guanidine isothiocyanate/glycogen method described by Chomkzynski ([@b6]). The detection of PCV2 was performed by PCR using primers Fa (5' ATT ACC AGC AAT CAG ACC CCG T 3')/Ra (5' CAA CCC TTC TCC TAC CAC TCC 3') that amplify a 476 bp fragment from ORF1 and ORF2 (position 807--1282 nt). For each enzymatic reaction, 5 µL of DNA was added to a PCR mixture with final concentrations of 1.25 mM MgCl2, 1 x PCR buffer, 1.25 U Taq DNA polymerase, 0.2 mM each dNTP and 50 pmol of each primers. Amplification was performed in a thermocycler with the following conditions: 95 **°**C for 5 minutes, 39 cycles with 95 °C for 1 minute, 60 °C for 1 minute and 72 °C for 1 minute and 72°C for 10 minutes ([@b4]). The samples of the two animals were positive for PCV2 and the complete PCV2 genome was sequenced from the two samples using four pairs of primers described by An ([@b1]). Amplicons with the expected molecular sizes, which were 700 bp, 630 bp, 705 bp and 700 bp for P1/P2, P3/P4, P5/P6 and P7/P8, respectively, were excised from the 1.5 % agarose gel and purified using a commercial kit (Concert, Gibco-BRL™). Bi-directional sequencing reactions were performed using BigDyeTM Terminator v3.1 (Applied Biosystems ™) and run using ABI model 377 sequencer analyzer (ABI model 377, Applied BiosystemsTM). The consensus sequence assembling was done using the PHRED/PHRAP and CAP3 (<http://bioinformatica.ucb.br/electro.html>) program with an analyses quality point of 20. The obtained sequences were aligned with sequences from GenBank using CLUSTAL X software ([@b11]). The phylogenetic analyses were done using the complete sequence or partial sequences of ORF2 depending on the availability of the sequences in GenBank was performed on the aligned data set, and a rooted tree was constructed in Mega v.2.1 using the distance-based neighbour-joining method with the PCV1 sequence as an out-group. The genome sequence of the two strain (BRA_JAV1_2008 and BRA_JAV2_2008) recovered were of 1,767 nt, which shared a nucleotide and aminoacid identity of 100 %. The phylogenetic tree analyses classified the two strain of this study in PCV2a subgroup together with PCV2 sequences of strain from wild boars from Poland, Brazil, Slovenia and Greece. The results confirmed that the virus is circulating within the Brazilian the wildlife boar herd, as showed previously by serological analysis ([@b2]) and it could be probably involved in the increase in the mortality rates observed in growing pigs in the herd. In the present case, the inadequate management, which was overpopulation, mixing pigs and failure on use of the all in/all out systems in the growing phase and in the herd was an essential factor in disease outcome. The improvement of this management reduced the mortality rates after 60 days. The detection of PCV2 in wildlife boar has been related in Europe and also in others Brazilian state ([@b2], [@b7], [@b12]). However, to confirm the association of PCV2 with the clinical signs, others tests, such as immunohistochemistry and histopathological analysis should be used. The detection of PCV2 by PCR is a indicative that the virus is circulating in the wild boar herd and that it could be one of the factors involved in the increase in the mortality rates observed in growing phase. The BRA_JAV1_2008 and BRA_JAV2_2008 sequences recovered were 1,767 bp long and lack the insertion of 11 nucleotides observed at position 42, nine bases upstream from the initiation codon of ORF1, which was also observed in sequences previously reported by Boisseson ([@b3]) and Dupont ([@b8]). The phylogenetic tree analyses classified the two strain of this study in PCV2a together with PCV2 sequences of strain from wild boars from previous reports. Unlike what was observed with the sequences from Germany, Slovenia and Poland, which are distributed in both groups (PCV2a and PCV2b), the strain of this study are grouped in PCV2a ([Figure 1](#fig1){ref-type="fig"}). This can be related to the year in which the sample was collected, since strain of PCV2b genotype were only detected in Brazil prior to 2005 ([@b4],[@b5]). Dupont ([@b8]) also reported the association of year of detection with the PCV2 genotype, since the authors demonstrated that PCV2a is detected in more recent samples when compared with PCV2b and PCV2c. There were no differences in the grouping of strain when used at total (data not shown) or partial sequence of the genome (ORF2), indicating, as observed by Olvera ([@b9]), that ORF2 is a good region for genetic analyses. {#fig1} This is the first complete sequence and characterization of PCV2 detected in wildlife boar in Brazil. Despite the small number of samples, the phylogenetic analyses demonstrated that the same PCV2 genotype is circulating in wild and domestic pigs in Brazil. However, further investigations are needed to better understand the epidemiological role of wild boars in the transmission of PCV2, in particular as a potential reservoir for the populations of domestic swine living in the same geographical area.

1. Introduction =============== Cervical cancer is the most common type of gynecological malignancy affecting women worldwide, and the third leading cause of cancer-related death among women in developing countries.^\[[@R1]\]^ In 2012, an estimated 527,600 new cervical cancer cases were identified worldwide, and 265,700 deaths were attributed to this type of cancer.^\[[@R1]\]^ Approximately, 80% of newly diagnosed cases and nearly 85% of related deaths occur in developing countries.^\[[@R1]\]^ Worse, more than 80% of new cervical cancer cases are detected at a locally advanced stage (International Federation of Gynecology and Obstetrics \[FIGO\] stage ≥Ib2), and in developing countries, \>50% of these cases involve stage III--IV disease.^\[[@R2]\]^ Since 1999, radical concurrent chemoradiotherapy (CCRT) has been recommended as the standard treatment for locally advanced cervical cancer (LACC), based on the results of 5 randomized studies.^\[[@R3]--[@R7]\]^ Radical CCRT for LACC generally includes pelvic external beam radiotherapy, concomitant chemotherapy, and intracavitary brachytherapy.^\[[@R8]\]^ Unfortunately, a lack of brachytherapy equipment restricts the use of radical CCRT in most developing countries. In Mainland China, only 31% (439/1413) of the radiation oncology centers are equipped with brachytherapy machine, with the number of well-equipped centers having risen only marginally, from 400 in 2005 to 439 in 2015.^\[[@R9]\]^ Therefore, roughly two-thirds of the patients cannot be treated conveniently with brachytherapy. Accordingly, the majority of patients must alternatively undergo preoperative chemoradiotherapy and radical surgery as a replacement for radical CCRT.^\[[@R10]\]^ Many studies have shown that for LACC, CCRT followed by radical surgery could yield encouraging results and a favorable long-term toxicity profile.^\[[@R11]--[@R13]\]^ The performance of radical surgery after CCRT has a couple of notable advantages. First, preoperative CCRT can shrink bulky tumors and therefore improve the successful resection rate^\[[@R14],[@R15]\]^ and counteract the negative effects of brachytherapy omission. Second, radical surgery after CCRT could allow clinicians to remove potentially chemoresistant and radioresistant foci,^\[[@R11],[@R15]\]^ leading to improved local control and overall survival. Although many previous studies have reported acceptable treatment-related urinary and gastrointestinal complications associated with CCRT followed by radical surgery,^\[[@R11],[@R12],[@R14]--[@R17]\]^ postoperative complications such as ureterohydronephrosis and lymphatic sequelae should not be ignored.^\[[@R12],[@R17]\]^ From our previous single-center experience ^\[[@R12]\]^ and the clinical research observations of Fanfani et al,^\[[@R13]\]^ lymphatic sequelae such as leg edema comprise the main type of adverse effect after CCRT followed by radical surgery, and the incidence of leg edema with this combined therapy is significantly greater than with radical CCRT alone. Because lower extremity edema results from damage to the lower limb lymphatic circulation, which is a risk of pelvic lymphadenectomy,^\[[@R18],[@R19]\]^ we wondered whether we could reduce the risk of this complication by avoiding or limiting the extent of pelvic lymphadenectomy, and whether such limitation would increase the rate of local recurrence or reduce overall survival. The present study therefore analyzed prognostic factors such as primary tumor size, pathological primary tumor type, squamous cell carcinoma antigen (SCC-Ag) level at diagnosis, LN size on diagnostic imaging, and pathologic response after preoperative CCRT in an attempt to identify patients in whom pelvic lymphadenectomy could be reduced in extent or omitted to decrease the incidence of leg edema without sacrificing local control and survival. 2. Methods and materials ======================== 2.1. Patient selection and data collection ------------------------------------------ Patients with clinical evidence of cervical involvement and treated with neoadjuvant CCRT and subsequent radical surgery, from January 2009 to December 2014 at our center, were retrospectively reviewed. The exclusion criteria in this study were as follows: FIGO stage Ia--Ib1 and IV disease (FIGO stage Ib2--IIIb according to gynecological examination and confirmation by at least 2 experts were enrolled), Eastern Cooperative Oncology Group (ECOG) performance status ≥3, history of other malignancies or cancer therapies, para-aortic lymph node (PALN) metastasis according to computed tomography (CT) and magnetic resonance (MR) images, and failing to complete preoperative radiotherapy and radical surgery. This study was approved by the Research Ethics Committee of the local hospital, and informed consent was obtained from all included patients. The consent forms were preserved in the patients' medical records. The pretreatment evaluation included a gynecological examination, chest radiography, cardiovascular evaluation, abdominopelvic CT, and MR imaging (except for patients who could not undergo MR because of an intrauterine coil or other metal in the body), complete blood cell count, transvaginal ultrasound, serological evaluation of liver and kidney functions, and SCC-Ag level evaluation. In brief, the cervical tumor size and LN size were determined using anteroposterior, lateral, and proximodistal tumor measurements (cm) from MR images (or transvaginal ultrasound images in the absence of MR). Patients were classified into 2 groups according to the pelvic LN size as determined on contrast-enhanced CT and MR images at diagnosis: enlarged LN short diameter \<0.8 cm, and enlarged short diameter ≥0.8 cm. 2.2. Treatment protocol ----------------------- ### 2.2.1. Preoperative chemoradiotherapy or radiotherapy alone Preoperative whole pelvic radiotherapy was delivered via a 6- or 15-MV photon beam according to a 3-dimensional conformal radiation protocol; a linear accelerator (Clinac 23EX or 600 CD or Clinac iX, Varian Medical Systems) was used for therapy. During treatment planning, all patients were immobilized on a custom vacuum mattress in the supine position and subjected to an enhanced CT simulation scan (Philips, Amsterdam, the Netherlands) with a slice thickness of 5 mm. The clinical target volume (CTV) was defined as the gross tumor, cervix, uterus, parametria, upper part of the vagina to 3 cm below the level of tumor invasion, and regional LN (external iliac, internal iliac, obturator, and presacral LN). The planning target volume (PTV) was defined as a uniform 3-dimensional expansion around the CTV, with 7-mm margins around the LNs, 10-mm margins around the vagina and parametria, and 15-mm margins around the gross cervical tumor and uterus. In patients undergoing intensity-modulated radiotherapy (IMRT), treatment was delivered via a 6-MV photon beam through 5 to 7 fields. The total planned dose for a pelvic field was 50 Gy/25 fractions. Concurrent chemotherapy (CCT) regimens comprised either cisplatin every 3 weeks (75 mg/m^2^) or weekly (40 mg/m^2^). If grade 3 or 4 toxicity necessitated a delay in chemotherapy, we re-evaluated the toxicity status after 1 week and withheld CCT until the white blood cell and platelet counts recovered to \>3000/mm^3^ and \>75,000/mm^3^, respectively. ### 2.2.2. Surgery and histopathological examination At 3 to 4 weeks after CCRT completion, patients again underwent pelvic MR or CT imaging and transvaginal ultrasound to evaluate the objective response. We, then, consulted at least 2 gynecological experts to determine eligibility of patients for radical surgery. Accordingly, all the eligible candidates would undergo type B radical hysterectomy, and pelvic lymphadenectomy via laparotomy, laparoscopy, or robotic techniques. Pathological responses to neoadjuvant therapy were evaluated based on a histopathological examination of resected specimens (e.g., uterus, vaginal cuff, parametria, pelvic LN). Pathological primary tumor responses were classified as a pathological complete response (pCR; no microscopic residual cancer cells), microscopically heterotypic cells (MHC; noncancerous but abnormal cells, including degeneration or necrosis of heterotypic cell nests), or residual carcinoma cells (RCC). Pathological responses of resected LNs were described as LN positive or LN negative. ### 2.2.3. Data collection and statistical analysis #### 2.2.3.1. Response, recurrence, and survival Local failure was defined as either pathological proof of cancer in the vaginal stump or the reappearance or enlargement of a tumor or any pelvic LN on imaging studies. Distant metastasis was confirmed using pathological, cytological, and/or radiological evidence. In this study, local or distant failure was defined according to the locations of lesions detected at the time of the first postoperative relapse. Patients were considered to have both local and distant failure when different lesions were detected synchronously or within a 1-month interval. Pelvic recurrence-free survival (PRFS), distant metastasis-free survival (DMFS), and overall survival (OS) were defined as the intervals from the end of CCRT to the time of pelvic recurrence, distant metastasis, and cancer-related death or last follow-up, respectively. 2.3. Toxicity and follow-up --------------------------- Acute and chronic toxicities, including hematologic, gastrointestinal (GI), and genitourinary (GU) toxicities, were evaluated according to the Acute Radiation Morbidity Scoring Criteria and Late Radiation Morbidity Scoring Scheme of the Radiation Therapy Oncology Group (RTOG). Lower extremity edema was graded using the Common Terminology Criteria for Adverse Events (CTCAE), version 3.0. Acute toxicity was evaluated weekly during treatment. After completing treatment, patients were followed-up after 1 month, and every 3 months thereafter during the first year. Subsequently, patients were followed up every 3 to 12 months. Toxicities were graded basing on the clinical findings described in the medical records. 2.4. Statistical analysis ------------------------- The normal distribution of the continuous data was described with mean ± standard deviation (SD), while the skewed distribution data were described with the median and interquartile range (IQR). Categorical data were presented as the number of patients and a percentage. OS, DMFS, and PRFS were calculated using the Kaplan--Meier method, and 95% confidence intervals (CIs) were estimated using the Cox proportional hazards model. Comparative analyses were performed using Chi-square or Fisher\'s exact test. SPSS, version 18.0 (IBM, NY) was used for the statistical analyses. A *P* value \< .05 was considered statistically significant. 3. Results ========== 3.1. Baseline of patient characteristics ---------------------------------------- A total of 472 (31.5%) of the 1497 LACC patients received the combined treatment during the study period, while others received the standard treatment of CCRT alone; 38 patients did not undergo subsequent surgery for various reasons and complete data were not available for 24 patients. Ultimately, 410 patients with FIGO stage Ib2--IIIb disease were observed in this study. The median age at diagnosis was 48 years, and the tumor diameter was 4.38 (± 0.92) cm. All patients underwent radical surgery, including type B hysterectomy and systematic pelvic lymphadenectomy via laparotomy, laproscopy, or robotic techniques. Squamous cell carcinoma (SCC) was the predominant pathologic subtype (391/410, 95.4%); non-SCC cases, including adenocarcinoma (AC) and adenosquamous carcinoma (ASC), accounted for only 4.6% of all cases. Patient classification according to an enlarged LN short diameter on CT and MR images yielded the following information: 26.6% (109/410) had an enlarged LN ≥0.8 cm. Regarding preoperative treatment, 18.5% (76/410) of the patients received no or incomplete CCT (Table [1](#T1){ref-type="table"}). ###### Clinical characteristics and postoperative pathological details of the study population.  3.2. Assessment of pathological responses to CCRT ------------------------------------------------- During a postoperative pathological evaluation, 38.8% of patients (159/410) were found to have achieved a pCR to preoperative CCRT, 35.8% (147/410) had MHC, and 25.4% (104/410) patients exhibited RCC. A pathological analysis of metastatic involvement in the resected pelvic LNs revealed that 92.7% of cases (380/410) were LN negative, whereas only 7.3% (30/410) were LN positive. Superficial, deep, and no stromal invasion were observed in 12.4% (51/410), 14.1% (58/410), and 73.4% (301/410) of cases, respectively. Negative and positive lymphovascular space invasion (LVSI) were reported in 98.3% (403/410) and 1.7% (7/410) of patients, respectively (Table [1](#T1){ref-type="table"}). 3.3. Survival outcomes and prognostic factors --------------------------------------------- During a median follow-up of 51.3 months (range: 4--97 months), from September 2009 to December 2016, 48 patients died, 44 developed distant metastases, and 10 experienced local failures (including vaginal stump recurrence in 9 patients and pelvic recurrence in 1 patient). The 5-year OS, DMFS, and PRFS rates were 86.7%, 88.6%, and 96.9%, respectively. The relationships of the patient clinicopathological characteristics at baseline with OS and DMFS are presented in Table [1](#T1){ref-type="table"}. A univariate analysis identified the pathological primary tumor type, LN size on imaging before CCRT, pathological responses of the primary tumor and resected LN after CCRT, and stromal invasion status as factors related to OS. Factors found to associate with DMFS included FIGO stage, tumor size, LN size on imaging before CCRT, SCC-Ag level before CCRT, pathological primary tumor type, pathological responses of the primary tumor and resected LN after CCRT, and stromal invasion status (Table [1](#T1){ref-type="table"}). A multivariate analysis identified the following prognostic factors as strongly predictive of OS: pathological type of primary tumor (non-SCC vs SCC, 5-year OS: 52.0% vs 89.3%; hazard ratio \[HR\] = 4.034, *P = *.001), LN size on CT/MR imaging before CCRT (LN ≥0.8 cm vs LN \<0.8 cm, 5-year OS: 73.9% vs 91.1%; HR = 2.792, *P = *.013), and postoperative pathologic response of the primary tumor (RCC vs pCR, 5-year OS: 75.8% vs 90.6%; HR = 2.092, *P = *.021). DMFS was found to associate significantly with the pathological primary tumor type (non-SCC vs SCC, 5-year DMFS: 55.4% vs 90.2%; hazard ratio \[HR\] = 2.587, *P = *.017), the LN size on CT/MR imaging before CCRT (LN ≥0.8 cm vs LN \<0.8 cm, 5-year DMFS: 73.3% vs 94.2%; HR = 2.325, *P = *.031), pathological response of the resected LN (LN positive vs LN negative, 5-year DMFS: 50.8% vs 91.7%; HR = 3.998, *P = *.001), and stromal invasion status (deep invasion vs no invasion, 5-year DMFS: 69.6% vs 93.5%; HR = 2.537, *P = *.025) (Table [2](#T2){ref-type="table"}). ###### Multivariate Cox proportional hazard regression model analysis for OS and DMFS.  3.4. Relationship between pretreatment LN size on imaging and pathological responses ------------------------------------------------------------------------------------ We identified a strong correlation between the postoperative pathologic response and LN size on CT/MR imaging before CCRT. Patients with a LN ≥0.8 cm had a significantly higher RCC rate when compared to those with a LN \<0.8 cm (33% vs 22.6%, *P = *.032). Patients with a LN ≥0.8 cm also had a significantly higher rate of stromal invasion when compared to those with a LN \<0.8 cm (33.9% vs 23.9%, *P = *.042), and a significant difference in deep stromal invasion was also observed (LN ≥0.8 cm vs LN \<0.8 cm, 20.2% vs 12.0%, *P = *.035). Furthermore, the frequency of postoperative pathological LN positivity differed significantly between the 2 LN size groups (LN \<0.8 cm vs LN ≥0.8 cm, 3.0% vs 19.3%, *P* \< .0001). In brief, the risk of a postoperative pathological positive LN was very low among patients with a LN \<0.8 cm on CT/MR imaging (Table [3](#T3){ref-type="table"}). ###### Comparative analysis of the outcome indices according to the LN size on imaging before CCRT.  The pretreatment SCC-Ag level is known to be a predictor of LN metastasis. Accordingly, we studied the relationship between SCC-Ag levels and LN in the 72.4% (297/410) of patients for whom precise SCC-Ag data at diagnosis were available. Of these patients, 73.4% (218/297) had a SCC-Ag level \<5 ng/mL, and 26.6% (79/297) had a SCC-Ag level ≥5 ng/mL. In the former and latter subgroups, 27.1% (59/218) and 40.5% (32/79) of patients, respectively, had a LN ≥0.8 cm on CT/MR imaging (*P = *.038), and approximately 5.0% (11/218) and 13.9% (11/79), respectively, had a pathologically positive LN (*P = *.010) (Table [4](#T4){ref-type="table"}). In summary, SCC-Ag levels may predict LN positivity. ###### Relationship between pretreatment SCC-Ag level and LN metastasis.  3.5. Identifying patients at low risk for LN metastasis according to the prognosis factors ------------------------------------------------------------------------------------------ Using the above results from an analysis of relationships among prognosis factors, we defined patients with SCC pathologic subtype, a pretreatment LN diameter \<0.8 cm on CT/MR imaging, and SCC-Ag level \<5 ng/mL as the low-risk group, while classifying the remaining patients as the high-risk group. Among our cohort, 51.5% (153/297) of patients were classified as low risk, and 48.5% (144/297) were classified as high risk. The low-risk group had very good outcomes. In contrast, patients in the high-risk group had a much worse prognosis. The corresponding 5-year OS and DMFS rates were 91.9% versus 81.1% and 95.9% versus 80.1%, respectively (*P* \< .05). The rates of pathological LN positivity in the low-risk and high-risk groups were 2.0% (3/153) and 13.2% (19/144), respectively (*P \< *.0001; Table [5](#T5){ref-type="table"}; survival curves are included in Fig. [1](#F1){ref-type="fig"}). ###### Different outcomes between the high- and low-risk groups which were classified according to the prognosis factors.  {#F1} 3.6. Toxicities --------------- The median interval between CCRT completion and surgery was 26 days (range: 14--71 days). The overall incidences of grade 3 and 4 acute hematological toxicities were 18.5% and 1.8%, respectively. The overall incidences of grade 2 and 3 acute gastrointestinal (GI) toxicities were 26.3% and 2.4%, respectively, and no grade 4 GI toxicities were observed. The incidences of grade 1 and 2 acute genitourinary (GU) toxicities were 11.2% and 1.8%, respectively, and no grade 3 or 4 acute GU toxicities were observed. Only 7.3% of patients developed chronic GI toxicities, and 11.0% of patients developed chronic GU toxicities (grade 1, 10.5%; grade 2, 0.5%). The incidence of lower extremity edema was 23.8% (grade 1, 14.6%; grade 2, 8.5%, grade 3, 0.7%). All such cases involved unilateral leg edema, and only 2 (0.7%) patients involved thrombus of a lower extremity vein (Table [6](#T6){ref-type="table"}). ###### Treatment-related toxicities (N = 410).  We used the Chi-square test to analyze factors that might contribute to the incidence of lower extremity edema. Among the 380 patients with complete data regarding LN resection, a median of 15 pelvic LNs were resected (range: 4--49). Our data showed that leg edema occurred more frequently when ≥15 pelvic LNs were resected (25.9% versus 16.9%) than that for \<15 LNs (*P = *.022). However, the incidence of lower extremity edema was not associated with the choice of surgical method (e.g., transabdominal or laparoscopic hysterectomy or da Vinci surgical system). 4. Discussion ============= Many studies have reported encouraging results and favorable long-term toxicity profiles associated with CCRT followed by radical surgery for LACC.^\[[@R11]--[@R13]\]^ The conclusions of our study are in agreement with these results, and our respective 5-year OS, DMFS, and PRFS rates of 86.7%, 88.6%, and 96.9% were not inferior to the reported rates from previous studies (OS: 57--85%, disease-free survival: 64--90%).^\[[@R20],[@R21]\]^ Furthermore, our toxicity analysis indicated acceptable levels of acute and chronic toxicities with this treatment combination. The satisfactory OS outcomes achieved with this combination of treatment modalities may be attributed to the considerable improvements in local control, as described in the introduction section. For example, our study revealed a good overall prognosis status and very low local recurrence rate (10/410; vaginal stump recurrence in 9 patients and pelvic recurrence in one patient), with distant metastasis as the main cause of death. Notably, residual tumor after radiotherapy or chemotherapy is a major risk factor for recurrence and poor OS.^\[[@R11],[@R14],[@R22]--[@R24]\]^ Our results demonstrated very encouraging pathological responses of cervical tumors after CCRT and subsequent surgery, as well as a very low positive LN rate (7.3%). Gadducci et al^\[[@R22]\]^ reported that patients who did not achieve an overall optimal response had a 2.757-fold higher risk of recurrence and 5.413-fold higher risk of death, compared with those who achieved such a response (5-year RFS: 87.4% vs 47.5%, OS: 96% vs 53.7%, *P \< *.0001). Moreover, in the present study, significantly better survival results were noted among patients who achieved a primary tumor pCR, compared to those who harbored residual carcinoma cells after CCRT. We further note that the pathological response of the resected LN was found to play an important prognostic role, consistent with previous reports by Ferrandina et al^\[[@R11]\]^ and Classe et al.^\[[@R14]\]^ Our study also observed that the pelvic LN status on imaging could be used to estimate the possibility of LN metastasis. Klerkx et al^\[[@R25]\]^ have concluded that best sensitivity and specificity rates are acquired when short axis diameter of pelvic LN on the MR image is \>8 mm, resulting in a sensitivity of 42.9% and specificity 96.6%. In one meta-analysis ^\[[@R26]\]^, the authors recommended LN diameter ≥8 mm as the best cut-off value when short axis diameter was adopted as a positive criterion in MRI examination. According to the above research, we classified the patients into 2 groups: enlarged LN short diameter \<0.8 cm, and enlarged short diameter ≥0.8 cm. Specifically, patients with a pretreatment LN size ≥0.8 cm on imaging had a positive LN rate of 19.3% and very poor prognosis (5-year OS, 56.6%), whereas those with LN \<0.8 cm had a positive LN rate of only 3.0%. In addition, a pretreatment LN size ≥0.8 cm was associated with a significantly higher likelihood of a pathological residual primary tumor and deep stromal invasion. These results suggest that the pelvic LN status on imaging could be used to predict the eventual prognosis. Ohara and colleagues^\[[@R27]\]^ noted that the CT-determined LN status might be a strong and useful predictor of cervical cancer confinement to the pelvis. In addition, despite the lack of a fixed prognostic cut-off value, SCC-Ag is considered an important predictor of overall survival, particularly with regard to distant metastasis.^\[[@R28],[@R29]\]^ According to our data, a SCC-Ag level ≥5 ng/mL was significantly associated with an increased risk of a pathological positive LN and reduced 5-year DMFS. The pathological primary tumor type is unquestionably the most important prognostic factor for a cervical cancer. Adenocarcinoma (AC) is widely considered to have a poorer survival outcome than adenosquamous carcinoma (ASC) or SCC.^\[[@R30],[@R31]\]^ In our study, the pathological primary tumor type was found to be an independent prognostic factor for OS. Notably, only 15.8% (3/19) of non-SCC patients achieved a pCR after CCRT, indicating a high level of resistance to CCRT that was accompanied by a low rate of survival (5-year OS, 52.0%). Finally, our toxicity analysis indicated acceptable levels of both acute and chronic toxicities, and we noted that only the incidence of leg edema after CCRT followed by radical surgery remained unimproved. Regarding our previous single-center experience, Wang et al^\[[@R12]\]^ reported that the rate of leg edema was significantly higher with this therapeutic combination than with radical CCRT alone (35.29% vs 4.96%, *P \< *.001). These authors also demonstrated that leg edema occurred more frequently when ≥20 pelvic LNs were dissected (55.56% vs 29.79% for \<20 LNs, *P = *.022). Additionally, Fanfani et al^\[[@R13]\]^ observed that in contrast to radical CCRT alone, vascular complications (including lymphocele and leg edema) were observed only in patients treated with CCRT followed by radical surgery (16.4%, *P \< *.001). Todo et al^\[[@R32]\]^ analyzed the relationship between the removal of circumflex iliac nodes distal to external iliac nodes (CINDEIN) and lower extremity lymphedema after systematic lymphadenectomy and concluded that the elimination of CINDEIN dissection could help to reduce the incidence of leg edema. These authors also concluded that patients with ≥31 LNs resected had a significantly higher incidence of edema than patients with ≤ 30 LNs resected (26.0% vs 16.5%, *P* = .0237). Although our study set a cut-off value of 15 resected LNs, we similarly concluded that a higher number of removed LNs correlated with a higher incidence of leg edema (25.9% versus 16.9%, *P = *.022). In addition, the number of removed LNs indirectly represents the region and extent of pelvic lymphadenectomy, which are associated with the occurrence of leg edema. Our results suggest that patients with a LN \<0.8 cm on imaging before CCRT, SCC histology, and a SCC-Ag level \<5 ng/mL at diagnosis have a low risk of LN metastasis and can expect good OS and DMFS outcomes. We therefore speculate that in these patients, pelvic lymphadenectomy could be avoided or that the extent of LN dissection could be limited to eliminate the risk of leg edema. While patients with any LN ≥0.8 cm, a SCC-Ag level ≥5 ng/mL and/or non-SCC histology must undergo standard pelvic lymphadenectomy. However, prospective randomized trials are needed to confirm the feasibility of individualized pelvic lymphadenectomy. One limitation of the present study is the short follow-up period. At the time of the analysis, the median follow-up period for patients who remained alive was 51.3 months (range 4--97 months). Moreover, we used SPSS 18.0 software to calculate the 5-year OS according to the Kaplan--Meier method. Therefore, the estimated 5-year OS in this study could be a representation of the overall survival trend. In addition, our study is a retrospective analysis of patients who were treated at a single center. However, to the best of our knowledge, this is the first and probably the largest study which demonstrated that CCRT followed by radical surgery is an effective therapeutic approach for LACC, especially, in those countries that have a shortage of brachytherapy equipment. Moreover, we identified the patient group that would benefit from the individualized pelvic lymphadenectomy, by analyzing the relationship between imaging at diagnosis and postoperative pathological LN involvement. However, prospective randomized trials are needed to compare the effectiveness of this combined treatment with the standard radical CCRT alone. 5. Conclusion ============= Preoperative CCRT and radical surgery are feasible treatment options for LACC, particularly in economically underdeveloped areas. The outcomes of CCRT followed by radical surgery in this study were encouraging, and moreover this treatment was not associated with unacceptable complications. Postoperative pathologic responses correlated strongly with the pretreatment LN size on CT/MR imaging, such that patients with a LN \<0.8 cm on diagnostic imaging, SCC histology, and a SCC-Ag level \<5 ng/mL at diagnosis had a very low risk of a postoperative pathologically positive LN. Accordingly, individualized pelvic lymphadenectomy such as omitting or limiting the extent of LN dissection might be an alternative option for the patients with a low risk of LN metastasis. Acknowledgments =============== The authors thank Xiao-Meng Wang, MB for her assistance in the work of follow-up visits. Author contributions ==================== **Conceptualization:** Chun Li Wei, Xin Li, Wei Wei Li, Mei Shi. **Data curation:** Chun Li Wei, Xin Li, Mei Shi. **Formal analysis:** Chun Li Wei. **Funding acquisition:** Chun Li Wei. **Investigation:** Xin Li, Ying Zhang, Wei Wei Li, Ping Jian Li, Na Li Zhao. **Methodology:** Xin Li, Ying Zhang, Zhi Yun Dang, Wei Wei Li, Ping Jian Li, Na Li Zhao. **Project administration:** Mei Shi. **Resources:** Chun Li Wei, Juan shu Liu, Xia Li. **Software:** Xin Li. **Supervision:** Chun Li Wei, Ying Zhang, Zhi Yun Dang, Mei Shi. **Validation:** Juan shu Liu, Xia Li, Mei Shi. **Visualization:** Xin Li, Mei Shi. **Writing -- original draft:** Xin Li. **Writing -- review & editing:** Mei Shi. Abbreviations: AC = adenocarcinoma, ASC = adenosquamous carcinoma, CCRT = concurrent chemoradiotherapy, CCT = concurrent chemotherapy, CINDEIN = circumflex iliac nodes distal to external iliac nodes, CIs = confidence intervals, CT = computed tomography, CTV = clinical target volume, DMFS = distant metastasis-free survival, FIGO = International Federation of Gynecology and Obstetrics, GI = gastrointestinal, GU = genitourinary, HR = hazard ratio, IMRT = intensity modulated radiotherapy, LACC = locally advanced cervical cancer, LN = lymph node, LVSI = lymphovascular space invasion, MHC = microscopically heterotypic cells, MR = magnetic resonance, OS = overall survival, PALN = para-aortic lymph node, pCR = pathological complete response, PRFS = pelvic recurrence-free survival, PTV = planning target volume, RCC = residual carcinoma cells, SCC = squamous cell carcinoma, SCC-Ag = squamous cell carcinoma antigen. Li-Chun Wei and Xin Li shared first co-authorship. This work is supported by National Natural Science Foundation of China (NSFC,81272346). The authors have no conflicts of interest to disclose.

The vitamin D hypothesis has received strong experimental support over the past two decades on the basis of the almost ubiquitous expression, in colon cancer cells, of vitamin D receptors (VDR) ([@bib21]; [@bib33]) and 1-*α*-hydroxylase ([@bib39]), which converts plasma 25-hydroxyvitamin D~3~ (25(OH)D) into 1,25-dihyroxycholecalciferol (1,25(OH)~2~D). Binding of VDR by 1,25(OH)~2~D leads to differentiation and apoptosis ([@bib34]; [@bib6]), and inhibition of proliferation ([@bib28]), angiogenesis ([@bib17]; [@bib9]), and metastatic potential ([@bib8]; [@bib18]). The best indicator of vitamin D status is plasma 25(OH)D level, as it reflects not only total vitamin D intake but also cholecalciferol production in the skin from type B ultraviolet (UV-B) radiation and hydroxylation of all sources of cholecalciferol in the liver. Prospective studies have shown that higher baseline plasma levels of 25(OH)D are associated with a significant reduction in the risk of colorectal cancer ([@bib11]; [@bib3]; [@bib31]; [@bib10]; [@bib36]; [@bib37]). A meta-analysis of five epidemiological studies found a 51% decrease in the risk of colorectal cancer associated with high plasma 25(OH)D levels (*P*\<0.0001) ([@bib15]). In contrast, the influence of vitamin D on survival of patients with established colorectal cancer remains uncertain. In a previous study, we found that higher pre-diagnosis plasma 25(OH)D levels were associated with a significant improvement in overall survival among 304 colorectal cancer patients ([@bib25]). However, this study was limited by the small number of patients who had available 25(OH)D plasma levels, and relied on a single measurement of 25(OH)D drawn at least 2 years before cancer diagnosis. To evaluate the impact of vitamin D status after a diagnosis of colorectal cancer in a larger number of patients, we developed a model to predict plasma 25(OH)D levels. This 25(OH)D score takes into account the combined influence of the major determinants of vitamin D status (race as a surrogate of skin pigmentation, residential state as a surrogate of UV-B radiation exposure, leisure-time physical activity as a surrogate of sunlight exposure, adiposity, and dietary and supplementary vitamin D intake). In previous analyses, the score was correlated with plasma 25(OH)D levels and was significantly associated with the risk of developing colorectal cancer ([@bib13]). To assess the influence of post-diagnosis vitamin D status on colorectal cancer survival, we calculated the post-diagnosis 25(OH)D score among 1017 patients participating in two ongoing prospective cohort studies, the Nurses\' Health Study (NHS) and the Health Professionals Follow-up Study (HPFS). Materials and methods ===================== Study population ---------------- In 1976, the NHS was established when 121 700 US female registered nurses aged 30--55 years answered a questionnaire on risk factors for cancer and cardiovascular disease ([@bib2]; [@bib4]). Every 2 years, participants receive follow-up questionnaires to update information on the potential risk factors and new disease diagnoses. Dietary information was first collected in 1980 and is updated in alternate follow-up cycles. Blood samples were provided by 32 826 participants aged 43--70 years from 1989 to 1990. In 1986, the HPFS was established when 51 529 male dentists, optometrists, osteopaths, podiatrists, pharmacists, and veterinarians aged 40--75 years responded to a questionnaire on risk factors for cancer, cardiovascular disease, and diabetes. A follow-up questionnaire is sent to participants every 2 years requesting an update on non-dietary exposures and medical history, with dietary history updated every 4 years. Blood samples were provided by 18 018 participants from 1993 to 1995. This study was approved by the Human Subjects Committee at Brigham and Women\'s Hospital and the Harvard School of Public Health in Boston, MA, USA. All participants provided informed consent for questionnaire and blood data to be used in research studies. Identification of colorectal cancer ----------------------------------- On each follow-up questionnaire, participants were asked whether they had a diagnosis of colorectal cancer during the previous 2 years. When a participant (or next of kin for decedents) reported a diagnosis of colorectal cancer, we asked permission to obtain hospital records and pathology reports, and blinded study physicians reviewed and recorded information on tumour characteristics. For non-respondents, we searched the National Death Index to discover deaths and ascertain any diagnosis of colorectal cancer that contributed to death. We estimate that 96--97% of cases were identified through these methods ([@bib12], [@bib13]). Individuals in this analysis were participants with pathologically confirmed colorectal adenocarcinoma diagnosed between 1986 (when physical activity was first assessed) and 2004. Measurement of mortality ------------------------ Participants in the study cohort were followed until death or 2006. Ascertainment of deaths included reporting by family or postal authorities, and names of persistent non-responders were searched in the National Death Index ([@bib27]). More than 98% of deaths have been identified by these methods ([@bib30]; [@bib13]). Cause of death was assigned by blinded physicians. Colorectal cancer-specific survival was defined as the time from diagnosis to death from colorectal cancer. Overall survival was defined as the time from diagnosis to death as a result of any cause. Exposure assessment ------------------- Derivation of the 25(OH)D prediction score has been described previously ([@bib13]). Briefly, linear regression was performed on 1095 men in the HPFS who were free of diagnosed cancer at the time of blood draw and who had available plasma 25(OH)D levels measured by radioimmunoassay ([@bib16]). Race, geographic region, dietary vitamin D intake, body mass index (BMI), and physical activity were identified as independent predictors of plasma 25(OH)D levels. Using the predictors\' regression coefficients ([Table 1](#tbl1){ref-type="table"}), a 25(OH)D score was calculated for each cohort member. The model was able to predict a wide range of 25(OH)D levels, from 9 to 36 ng ml^−1^. The mean actual circulating level of 25(OH)D for men in the highest decile of predicted 25(OH)D was 11 ng ml^−1^ higher (95% CI, 9--13) than that of men in the lowest decile (1 ng ml^−1^=2.496 nmol l^−1^). To validate this model, a 25(OH)D score was calculated for an independent sample of 542 men in HPFS who also had available plasma 25(OH)D measurements ([@bib13]). The actual plasma level rose across increasing deciles of predicted 25(OH)D (*P* trend\<0.001), and the difference in the mean actual 25(OH)D level between extreme deciles was 10 ng ml^−1^, similar to the difference of 11 ng ml^−1^ in the initial dataset. Finally, in a separate analysis of 47 800 men, a statistically significant inverse association was observed between predicted 25(OH)D levels and the risk of developing colorectal cancer ([@bib13]). Predicted 25(OH)D levels were calculated for our study cohort using the regression coefficients in [Table 1](#tbl1){ref-type="table"} and extrapolating them to women. Post-diagnosis 25(OH)D score was the primary exposure variable, calculated using baseline race and geographic region, and values of physical activity, BMI, and vitamin D intake reported 1--4 years after colorectal cancer diagnosis (to avoid assessment during the period of active cancer treatment). Scores were divided into quintiles, with cut-offs determined separately for each cohort. We also examined the influence of pre-diagnosis predicted 25(OH)D levels, using values of physical activity, BMI, and vitamin D intake from the questionnaire administered immediately before diagnosis. If a response was missing, one previous assessment was carried forward; else, the patient was not included. Covariates ---------- Prognostic factors known to influence cancer mortality were extracted from the medical record, including age, stage, grade of differentiation, tumour location, and year of diagnosis. In addition, BMI, physical activity, and total energy-adjusted calcium intake were taken from the questionnaire administered immediately before diagnosis. Race and season of diagnosis were also evaluated, but not found to be confounding variables, and were therefore excluded from the multivariable model. All analyses were adjusted for the cohort. Statistical analyses -------------------- Death from colorectal cancer was the primary end point, and death from any cause was a secondary end point. The primary statistical analysis used the two-tailed linear test for trend with predicted 25(OH)D level as a continuous variable, consistent with earlier studies ([@bib10]), to avoid the possibility of selecting cut points with maximal *P*-values. To facilitate the display of our results, predicted 25(OH)D levels were defined in quintiles. Cox proportional hazards models were used to calculate hazard ratios (HRs) of death, adjusted for other risk factors for cancer survival. The HRs were calculated according to quintiles of predicted 25(OH)D levels, with the lowest quintile as the reference group. The proportionality of hazards assumption for the effect of predicted 25(OH)D level was satisfied by examining it as a time-dependent covariate in the Cox model. Colorectal cancer-specific and overall mortality by tertile of predicted 25(OH)D levels were examined using Kaplan--Meier curves and the log-rank test ([@bib32]). Tertiles were used instead of quintiles for ease of graphical viewing. Subgroup analyses were performed, and adjusted HRs and 95% CIs for an increment of 10 ng ml^−1^ of predicted 25(OH)D levels for cancer-specific and overall mortality were reported. Tests of interaction between predicted 25(OH)D levels and relevant covariates were assessed by entering in the model the cross product of predicted 25(OH)D level as a continuous variable with the covariate as a continuous or binary variable. All analyses used SAS software, version 9.1 (SAS Institute, Inc., Cary, NC, USA). Results ======= Among 1017 eligible participants, there were 283 deaths, 119 of which were due to colorectal cancer. The median time of follow-up of participants who were still alive was 116 months (range 41--238 months). The median predicted 25(OH)D level was 27.18 ng ml^−1^ in NHS and 29.18 ng ml^−1^ in HPFS. This pattern is consistent with what we saw in our previous analysis of pre-diagnosis plasma 25(OH)D levels and colorectal cancer survival, in which circulating 25(OH)D concentrations were also slightly higher in men than in women: median plasma 25(OH)D level was 27.1 ng ml^−1^ in men and 23.9 and 25.7 ng ml^−1^ for two laboratory runs in women ([@bib25]). The difference in mean post-diagnosis predicted 25(OH)D levels between the highest and lowest deciles in this study cohort was 10 ng ml^−1^, which is similar to the difference in mean actual circulating 25(OH)D levels across extreme deciles of predicted 25(OH)D in the original and validation cohorts. Baseline characteristics according to quintiles of post-diagnosis predicted 25(OH)D levels are shown in [Table 2](#tbl2){ref-type="table"}. Overall, participants with higher predicted 25(OH)D levels were more likely to be of white race, have a lower BMI, report higher physical activity, have higher calcium intake, and were more likely to be diagnosed with colorectal cancer in the summer or autumn. Other prognostic characteristics did not differ significantly between quintiles. Higher post-diagnosis predicted 25(OH)D levels were associated with a significant reduction in colorectal cancer-specific and overall mortality ([Figure 1A and B](#fig1){ref-type="fig"}, respectively, and [Table 3](#tbl3){ref-type="table"}). This relationship remained largely unchanged after adjusting for other predictors of cancer survival ([Table 3](#tbl3){ref-type="table"}). Compared with patients with post-diagnosis 25(OH)D scores in the lowest quintile, those in the highest quintile had an adjusted HR of 0.50 (95% CI, 0.26--0.95; *P* trend=0.02) for cancer-specific mortality and 0.62 (95% CI, 0.42--0.93; *P* trend=0.002) for overall mortality. We considered the possibility that the 25(OH)D score may be acting as a surrogate for the causal factor, such as BMI or physical activity, which are both in the prediction equation. We therefore repeated our analyses after adjusting for BMI and physical activity. When BMI was included, the significant relationship between post-diagnosis 25(OH)D score and cancer-specific and overall mortality did not change. The adjusted HR was 0.51 (95% CI, 0.26--0.99; *P* trend=0.04) for cancer-specific mortality and 0.62 (95% CI, 0.41--0.94; *P* trend=0.005) for overall mortality, comparing extreme quintiles. Similarly, when we controlled for physical activity, we continued to observe a benefit for higher post-diagnosis predicted 25(OH)D levels, with an adjusted HR of 0.44 (95% CI, 0.22--0.87; *P* trend=0.01) for cancer-specific mortality and 0.67 (95% CI, 0.44--1.00; *P* trend=0.02) for overall mortality. When both BMI and physical activity were included in the model, the adjusted HR was 0.45 (95% CI, 0.22--0.91; *P* trend=0.02) for cancer-specific mortality and 0.66 (95% CI, 0.43--1.03; *P* trend=0.03) for overall mortality. Moreover, when both BMI and physical activity were included in our model, the remaining components of the post-diagnosis 25(OH)D score that were not 'accounted for\' were region of residence, race, and dietary and supplemental vitamin D intake. We therefore explored the impact of each of these individual variables on mortality in our study cohort, and found that patients who reported higher total vitamin D intake showed a trend towards lower risk of death (*P* trend=0.08). Compared with those in the lowest quintile, patients in the highest quintile of vitamin D intake had an adjusted HR of 0.72 (95% CI, 0.49--1.04) for overall mortality. We also adjusted for calcium intake in our models and found a persistent significant association between post-diagnosis predicted 25(OH)D levels and survival. The adjusted HR was 0.44 (95% CI, 0.23--0.87; *P* trend=0.008) for cancer-specific mortality and 0.56 (95% CI, 0.37--0.84; *P* trend=0.0005) for overall mortality, comparing extreme quintiles. Of note, the addition of race and season of diagnosis to the multivariable model also did not change the adjusted HRs for cancer-specific and overall mortality. When race was added to the model, the adjusted HR comparing extreme quintiles was 0.53 (95% CI, 0.28--1.02; *P* trend=0.04) for cancer-specific mortality and 0.65 (95% CI, 0.43--0.97; *P* trend=0.004) for overall mortality. When season of diagnosis was included, the adjusted HR was 0.49 (95% CI, 0.26--0.94; *P* trend=0.02) for cancer-specific mortality and 0.63 (95% CI, 0.42--0.93; *P* trend=0.004) for overall mortality. Given that lower levels of post-diagnosis predicted 25(OH)D could reflect the presence of occult cancer or other major illness, we excluded patients who died within 6 months of their post-diagnosis assessment. We continued to observe significant reductions in the risk of cancer-specific and overall mortality with increasing post-diagnosis 25(OH)D scores, with participants in the highest quintile having an adjusted HR of 0.50 (95% CI, 0.25--0.98; *P* trend=0.03) for cancer-specific mortality and 0.63 (95% CI, 0.42--0.95; *P* trend=0.003) for overall mortality. In a separate analysis, pre-diagnosis 25(OH)D scores were calculated for colorectal cancer patients with available information (*n*=1955). Higher pre-diagnosis 25(OH)D scores were found to be associated with a decrease in cancer-specific and overall mortality (*P* trend=0.03 and 0.01, respectively). When we adjusted for pre-diagnosis predicted 25(OH)D levels as well as other predictors of cancer survival in our model, higher post-diagnosis predicted 25(OH)D levels were still associated with a significant reduction in both cancer-specific (*P* trend=0.02) and overall (*P* trend=0.008) mortality, whereas the effect of pre-diagnosis predicted 25(OH)D level was no longer significant. We examined the influence of post-diagnosis 25(OH)D scores across the other predictors of cancer mortality ([Figure 2A and B](#fig2){ref-type="fig"}). Stratified analyses of cancer-specific and overall survival showed no significant interactions; the inverse relationship between post-diagnosis predicted 25(OH)D levels and mortality remained largely unchanged across most subgroups. Of note, there was a trend towards a greater impact of higher post-diagnosis 25(OH)D scores on overall survival in patients diagnosed in the winter or spring compared with those diagnosed in summer or autumn (*P* interaction=0.06). Discussion ========== Among patients with colorectal cancer, higher post-diagnosis predicted 25(OH)D levels were associated with a significant reduction in cancer-specific and overall mortality. This relationship was evident even after excluding patients who died within 6 months of their post-diagnosis 25(OH)D assessment, and across different subgroups of patients. Previous studies have suggested an inverse relationship between vitamin D and cancer incidence. Prospective observational studies showed that higher plasma 25(OH)D levels are associated with a significant reduction in risk of colorectal cancer ([@bib11]; [@bib3]; [@bib31]; [@bib10]; [@bib36]; [@bib37]). In a prospective, placebo-controlled trial of vitamin D and calcium supplementation in 1179 women, a statistically significant decrease of 60% in all-cancer risk (including colorectal cancer) was seen in the intervention arm (*P*\<0.03) ([@bib19]). We have previously shown that higher pre-diagnosis plasma levels of 25(OH)D are associated with a significant 48% reduction in overall mortality ([@bib25]). However, we only had a single measurement of plasma 25(OH)D levels taken at a median of 6 years before diagnosis. Furthermore, only 304 patients had plasma available for analysis. Therefore, to assess the influence of vitamin D status after cancer diagnosis in a larger population, we used a validated prediction score to estimate 25(OH)D levels after colorectal cancer diagnosis. This score is a reasonable predictor of circulating 25(OH)D levels, and is associated with cancer incidence and mortality ([@bib13]). There are several mechanisms through which vitamin D may influence survival after a diagnosis of colorectal cancer. Binding of VDR by 1,25(OH)~2~D leads to transcriptional activation and repression of target genes, resulting in differentiation and apoptosis, and inhibition of proliferation and angiogenesis (reviewed by [@bib5]). *In vitro* and *in vivo* data have shown growth inhibition, and differentiation of colon carcinoma cell lines and xenografts by administration of 1,25(OH)~2~D ([@bib7]; [@bib14]; [@bib40]; [@bib29]; [@bib17]), and rat models of colorectal cancer maintained on a 1,25(OH)~2~D diet developed fewer metastases compared with controls ([@bib8]). We found that season of diagnosis may modify the effect of predicted 25(OH)D levels on overall mortality, with participants diagnosed in the winter or spring showing a slightly greater reduction in the risk of death. Perhaps the impact of higher vitamin D levels is greater in the winter and spring, when sunlight exposure is at a minimum. For example, a patient with colorectal cancer with a high post-diagnosis 25(OH)D score, despite decreased UV-B exposure in winter and spring, may have a survival advantage over a patient with a low score in these seasons. The strengths of our study include its prospective design, use of a validated prediction score, data on many potential confounders, and excellent follow-up rate. As participants were health professionals, the accuracy of self-reported data is likely to be high. To our knowledge, this study is the first to examine colorectal cancer survival by use of a comprehensive assessment of factors that determine 25(OH)D level. The similarity of our findings for survival based on one measurement of plasma 25(OH)D level and those based on our predictor score indicates that each may provide comparable information on long-term 25(OH)D level, the presumed factor of interest. The most apparent limitation of our approach is the possibility that our 25(OH)D score is acting as a surrogate for the causal factor, such as BMI or physical activity, through alternative mechanisms. Physical activity has previously been shown to be associated with improved outcomes in colorectal cancer patients ([@bib22], [@bib23]). In contrast, the data for BMI and colorectal cancer survival is conflicting ([@bib24]). However, our results for predicted 25(OH)D levels did not change when adjusted for BMI or physical activity, and no significant interactions were observed. Moreover, when both BMI and physical activity were included in our model, examination of the remaining components of the post-diagnosis 25(OH)D score revealed that patients who reported higher total vitamin D intake showed a trend towards a lower risk of death. As dietary sources of vitamin D account for only 10% of circulating 25(OH)D levels ([@bib1]), it is not unexpected that the finding is bordered on significance. This lends further support to the ability of our score to represent vitamin D status, and to our hypothesis that higher total vitamin D status -- rather than simply adiposity or physical activity -- is driving the significant association with survival. Furthermore, there is evidence that calcium may have an independent role in colorectal cancer pathogenesis ([@bib38]; [@bib20]; [@bib26]). As the vitamin D pathway is intimately linked to calcium homoeostasis, we controlled for calcium intake in our analyses and found that our results did not change. Although the 25(OH)D score had been developed and validated extensively in our cohort of men ([@bib13]), we recognise that the score has not been validated in women. However, there are no data in the current literature to suggest that the determinants of serum 25(OH)D differ between men and women. Furthermore, when our results were stratified by gender, similar HRs were obtained for men and women for both cancer-specific and overall mortality, with no significant interactions (see [Figure 2A and B](#fig2){ref-type="fig"}). Indeed, when post-diagnostic predicted 25(OH)D level was evaluated solely in the 606 women with colorectal cancer, a significant association was found between predicted 25(OH)D level and cancer-specific (adjusted HR 0.26; 95% CI, 0.10--0.72; *P* trend=0.02) and overall mortality (adjusted HR 0.48; 95% CI, 0.27--0.88; *P* trend=0.003). Moreover, any impression in the 25(OH)D score among women in our study would bias results in that cohort tend towards the null. We also cannot completely exclude the possibility that lower levels of post-diagnosis predicted 25(OH)D reflect other occult predictors for poor prognosis. However, our findings remained unchanged after adjusting for potential risk factors of colorectal cancer mortality. To minimise bias in the post-diagnosis predicted 25(OH)D level by the presence of occult cancer or other major illness, we excluded patients who died within 6 months of their post-diagnosis vitamin D assessment, and continued to observe a positive impact of higher predicted 25(OH)D scores. In this cohort, data on treatment were limited. Approximately half of the participants had stage I or II disease, for which surgery alone is often the standard of care. In addition, although there have been changes in the chemotherapeutic treatment of colorectal cancer during the timeframe under study, we adjusted for year of diagnosis in our models. Furthermore, the fairly homogeneous socioeconomic and educational makeup of this cohort likely minimises any disparities in chemotherapy receipt ([@bib35]). The NHS and HPFS are composed exclusively of working health professionals with extensive access to health care; as such, differential access to state-of-the-art health care among participants is likely minimised. Lastly, because this was an observational rather than a randomised study, we cannot definitively attribute our results to 25(OH)D levels; further support for a causal role of higher vitamin D status on survival may require a randomised placebo-controlled trial. In conclusion, our data suggest that higher levels of predicted 25(OH)D after a diagnosis of colorectal cancer may be associated with improved survival. Additional efforts to understand the mechanisms through which the vitamin D pathway influences colorectal carcinogenesis and cancer progression are warranted. {#fig1} {#fig2} ###### Factors contributing to predictors of age-adjusted plasma 25(OH)D level from a multiple linear regression model of 1095 men in the Health Professionals Follow-Up Study ([@bib13]) **Factor** **Change in 25(OH)D (ng ml^−1^**) ------------------------------------------------------------------------------------------- ----------------------------------- Intercept 33.81     *Race*  White 0 (referent)  African American −5.31  Asian −5.49  Other −0.17     *Residence*  Northeast/Mid-Atlantic 0 (referent)  Midwest/West +1.61  South +2.56     *Quintile of leisure-time physical activity (MET-hr per week)*[a](#t1-fn2){ref-type="fn"}  5 0 (referent)  4 −1.81  3 −3.07  2 −3.59  1 −5.40     *Body mass index (kg m*^−*2*^)  \<22 0 (referent)  22--24.9 −0.40  25--29.9 −1.80  30--34.9 −2.58  ⩾35 −3.44     *Dietary vitamin D (IU per day)*  ⩾400 0 (referent)  300--399 −1.39  200--299 −1.04  100--199 −2.85  \<100 −4.16     *Supplementary vitamin D (IU per day)*[b](#t1-fn3){ref-type="fn"}  ⩾400 0 (referent)  200--399 −0.71  1--199 +0.97  \<1 −0.83     *Season of blood draw*[c](#t1-fn4){ref-type="fn"}  Autumn (September, October, and November) 0 (referent)  Summer (June, July, and August) −0.74  Spring (March, April, and May) −4.83  Winter (December, January, and February) −5.42 Abbreviations: 25(OH)D=25-hydroxyvitamin D~3~; hr=hour; IU=international units; MET=metabolic equivalents. Physical activity is used as a proxy for outdoor activities, which will tend to increase solar UV-B exposure. Not statistically significant. Season of blood draw was adjusted but not used in the predictive model because season is a strong determinant of 25(OH)D level and reflects the time of blood draw, but it is not a factor in determining long-term average between-person variation in 25(OH)D level. ###### Baseline characteristics of cohort according to quintile of predicted 25(OH)D (*n*=1017)   **Predicted 25(OH)D (ng ml^−1^)**   ------------------------------------------------------------------------------------ ------------------------------------- ------------------------- ------------------------- ------------------------- ------------------------- ------------------------------------- No. of patients 204 202 205 203 203 --- Median predicted 25(OH)D (ng ml^−1^)[a](#t2-fn2){ref-type="fn"} -- NHS (*n*=606) 23.3 (range 19.1--24.8) 25.7 (range 24.9--26.4) 27.2 (range 26.5--28.0) 28.6 (range 28.0--29.7) 31.0 (range 29.7--34.9) --- Median predicted 25(OH)D, (ng ml^−1^)[a](#t2-fn2){ref-type="fn"} -- HPFS (*n*=411) 25.8 (range 21.6--26.8) 27.8 (range 27.0--28.4) 29.2 (range 28.5--29.9) 30.8 (range 29.9--31.5) 32.8 (range 31.5--36.9) --- Mean age at diagnosis (years) 65.6 (s.e. 0.6) 66.0 (s.e. 0.5) 65.8 (s.e. 0.6) 65.7 (s.e. 0.5) 65.3 (s.e. 0.5) 0.92[b](#t2-fn3){ref-type="fn"}               *Stage* (%) 0.95[c](#t2-fn4){ref-type="fn"}  I 33 30 34 34 34    II 28 30 25 30 32    III 23 25 26 22 21    IV 4 4 2 4 3    Unknown 12 11 13 10 10                 *Grade of tumor differentiation* (%) 0.65[c](#t2-fn4){ref-type="fn"}  Well differentiated 12 14 12 11 19    Moderately differentiated 59 58 58 60 57    Poorly differentiated 11 10 12 11 11    Unknown 18 18 18 18 13                 *Location of primary tumor* (%) 0.43[c](#t2-fn4){ref-type="fn"}  Proximal colon 36 38 39 40 41    Distal colon 33 37 33 35 31    Rectum 22 22 23 19 24    Unknown 9 3 5 6 4                 *Year of diagnosis* (%) 0.32[c](#t2-fn4){ref-type="fn"}  1986--1995 47 46 52 49 55    1996--2004 53 54 48 51 45                 *Season of diagnosis*[d](#t2-fn5){ref-type="fn"} (%) 0.003[c](#t2-fn4){ref-type="fn"}  Summer 30 27 32 31 23    Autumn 20 23 20 34 26    Winter 25 22 27 19 23    Spring 25 28 21 16 28                 *Race* (%) \<0.0001[c](#t2-fn4){ref-type="fn"}  White 89 93 95 98 95    Black 5 1 0 0 0    Other 6 6 5 2 5   Mean body mass index at diagnosis (kg m^−2^) 28.7 (s.e. 0.3) 27.3 (s.e. 0.3) 26.1 (s.e. 0.3) 25.0 (s.e. 0.2) 24.4 (s.e. 0.2) \<0.0001[b](#t2-fn3){ref-type="fn"} Median physical activity at diagnosis (MET-hr per week)[e](#t2-fn6){ref-type="fn"} 5.3 (range 0--148.1) 8.4 (range 0--98.0) 12.9 (range 0--131.8) 15.4 (range 0--127.7) 23.0 (range 0--625.2) \<0.0001[f](#t2-fn7){ref-type="fn"} Median total energy-adjusted calcium intake (mg)[e](#t2-fn6){ref-type="fn"} 786 (range 248--3446) 817 (range 260--3375) 930 (range 329--3192) 936 (range 239--2935) 1010 (range 346--4496) \<0.0001[f](#t2-fn7){ref-type="fn"} Abbreviations: 25(OH)D=25-hydroxyvitamin D~3~; HPFS=Health Professionals Follow-up Study; hr=hour; IU=international units; MET=metabolic equivalents; NHS=Nurses\' Health Study; No.=number; s.e.=standard error. 1 ng ml^−1^=2.496 nmol l^−1^. *P*-value calculated using one-way analysis of variance. *P*-value calculated using *χ*^2^-test. Summer defined as June, July, and August; autumn defined as September, October, and November; winter defined as December, January, and February; spring defined as March, April, and May. Analysis restricted to those participants with available information. *P*-value calculated using Kruskal--Wallis test. ###### Age-adjusted and multivariate hazard ratios of death according to quintile of predicted 25(OH)D in the entire study cohort (*n*=1017)   **Predicted 25(OH)D**   ------------------------------------------ ----------------------- ---------- -------- ------------------- -------- ------------------- -------- ------------------- -------- ------------------- -------- *Colorectal cancer-specific mortality*  Age-adjusted[b](#t3-fn3){ref-type="fn"} 204/29 Referent 202/29 1.02 (0.61--1.70) 205/28 0.95 (0.56--1.59) 203/19 0.62 (0.35--1.10) 203/14 0.45 (0.24--0.84) 0.006  Multivariate[c](#t3-fn4){ref-type="fn"}       0.99 (0.58--1.68)   1.04 (0.61--1.78)   0.62 (0.34--1.11)   0.50 (0.26--0.95) 0.02                         *Overall mortality*  Age-adjusted[b](#t3-fn3){ref-type="fn"} 204/64 Referent 202/71 1.13 (0.80--1.58) 205/62 0.96 (0.68--1.36) 203/44 0.62 (0.42--0.91) 203/42 0.57 (0.38--0.84) 0.0004  Multivariate[c](#t3-fn4){ref-type="fn"}       1.19 (0.85--1.68)   1.05 (0.74--1.50)   0.63 (0.43--0.94)   0.62 (0.42--0.93) 0.002 Abbreviations: 25(OH)D=25-hydroxyvitamin D~3~; CI=confidence interval; HR=hazard ratio; No.=number. Calculated by using predicted 25(OH)D as a continuous variable, adjusted for cohort. HRs, 95% CIs, and *P*-values are adjusted for age at diagnosis (years). Multivariate HRs, 95% CIs, and *P*-values are adjusted for age at diagnosis (in years as a continuous variable), gender (male or female), cancer stage (I--IV or unknown), grade of tumor differentiation (well-differentiated, moderately-differentiated, poorly-differentiated, or unspecified/missing), location of primary tumor (proximal, distal, rectum, or unknown), and year of diagnosis (as a continuous variable).

Introduction {#s1} ============ In acromegaly, prolonged exposure to growth hormone (GH) and insulin-like growth factor (IGF-1) is associated with worsening comorbidities and impairment in quality of life ([@B1]). Biochemical dysregulation of the disease can reduce acromegalic life expectancy by up to 10 years ([@B1], [@B2]), but, when the disease is controlled, mortality may reach parity with the normal population ([@B2]). Despite this, biochemical control of acromegaly through currently available treatments---transsphenoidal surgery, radiotherapy and pharmacotherapy---has been insufficient to improve quality of life ([@B2]--[@B4]). In addition to clinical complications, orofacial changes and orthopedic comorbidities can also impair the quality of both functional and social life ([@B1]). Acromegalics show a preoccupation with the performance of tasks involved in executive dysfunction syndrome, which may impair the execution of everyday tasks ([@B5]), as well as jeopardize the development of effective coping strategies ([@B6]). Receptors for GH and IGF-1 are present in the central nervous system, especially in the limbic system and frontal lobe. An excess of GH and IGF-1 results in cognitive impairment may be irreversible, as in valvulopathy and arthropathy ([@B6]). Delay in diagnosis is related to social isolation, distortion of body image, and depression, with consequent psychosocial impairment ([@B1], [@B7]). The impairment in quality of life may also be related to psychiatric comorbidities, such as anxiety, cognitive impairment, and depression, more frequently than in other somatic diseases ([@B1], [@B8]). Therefore, quality of life impairment may be related to psychopathology and not only to biochemical dysregulation of the disease ([@B3]). Cognitive dysfunctions due to emotional imbalance have been observed in acromegalics, especially in executive functions and in memory ([@B5], [@B9], [@B10]). Lower activation of areas of the brain strongly associated with cognitive functions---the prefrontal and temporal cortex---has been observed as well ([@B5]). Considering the importance of psychopathology in acromegaly ([@B3]), particularly, related to the supposed irreversible effect of excess GH on mood and behavior ([@B1]), acromegalics can benefit from psychotherapeutic treatment ([@B2], [@B8], [@B9]). However, no results on the effects of psychotherapy on improvement of quality of life in acromegalics had been published until recently, despite being well-established for other chronic diseases ([@B4]). In this context, cognitive-behavioral therapy (CBT) can be applied to cases of acromegaly. Over the years, research involving CBT has contributed to clinical practice, since improvements in techniques favor its application in different clinical contexts ([@B11]). More than 300 research studies have applied CBT to various diseases, as it is an effective intervention with a low cost ([@B12]). Initially, the cognitive model, proposed by Aaron Beck in the 1960s, was based on the cognitive distortions present in depression ([@B13]). Depressive content was characterized by low self-esteem and guilt with a consequent desire for isolation ([@B14]). In CBT, thought, emotion, and behavior are intertwined ([@B14], [@B15]). The cognitive model has been adapted for various diseases. In this model, schemas are sets of beliefs associated with greater emotional imbalance ([@B13], [@B16], [@B17]). Due to facial changes and growth of their hands and feet, ([@B1]), acromegalics can present schemas of inadequacy, devaluation and failure. As with acromegalics, depressed patients overestimate their defects, failures, and problems, and tend to underestimate or ignore any favorable traits, skills, or achievements ([@B14]). The cognitive conceptualization of each case is facilitated by Socratic questioning. The therapist formulates questions to help the patient understand the effects of emotions and thoughts on behavior. Such questioning aims at developing new coping perspectives from the learning process ([@B18]). Among several clinical applications, in social phobia, for example, the patient feels threatened in social situations, thinking "*If I try to talk, I\'ll look stupid"*; as a result the patient avoids social interaction ([@B13]). In acromegaly, because of their physical appearance, the cognitive model may be similar, based on the belief: "*If I look horrible, everyone will stare and criticize me."* Uncertainty, cost and benefit assessment, and perceived risk precede the choice of a given behavior, and require some degree of flexibility ([@B19]). Despite the importance of decision making, not much research has been carried out the influence of emotional imbalance on choices ([@B20]). In the brain, activation of the cingulate cortex and the amygdala is important because of their role in emotional regulation, which is one of the main objectives of CBT ([@B21], [@B22]). In this context, the "Think healthy" technique was applied in a group setting with the aim of improving the quality of life of acromegalics ([@B23]). The technique is composed of six stages and uses the figures of gray and green avocados to represent problems and possible coping alternatives ([@B23]). One of its fundamental principles is decision making toward healthy behaviors, facilitated by the following thought: "*Despite the negative repercussions of acromegaly, what can I do that is healthier toward thinking and behavior*?" In a previously published study, we showed that CBT, with the "Think Healthy" technique, significantly improved the quality of life of acromegalics; evaluated by the SF-36 in the areas of general health (d′ = −0.264, *p* = 0.031) and mental health (d′ = −1.123, *p* = 0.012). In the dimensions of physical and emotional aspects there were no significant improvement, although four of the five patients with floor effects did improve ([@B23]). The aim of the present study is to evaluate whether the initially obtained results were still present after 9 months from the end of the treatment. It also compares whether, after that period of time, the treatment group shows any difference to the control group that did not receive the treatment at the same time. Methods {#s2} ======= This was a non-randomized longitudinal study performed on acromegalics attending the neuroendocrinology outpatient clinic of the University Hospital of Brasília in the period from September 2015 to November 2016. The following inclusion criteria were used: patients with acromegaly in follow-up at the above-mentioned clinic, males and females, aged between 18 and 75 years; intellectual capacity to follow the intervention; and agreement to participate in the study. The exclusion criterion was risk of suicide, evaluated according to the response to item 9 of the BDI ([@B24]). The meetings were held in a space that comfortably accommodated the number of participants, arranged in a circle, with the possibility of handling the material, taking notes and completing exercises. The room was well lit, cool, and had good privacy. The clinical trial was registered in a public trial registry, and the accession number is UTN---U 1111-1220-9846. The research project was approved by the Research Ethics Committee (CEP) of the Faculty of Health Sciences of the University of Brasília---UnB. A Free and Informed Consent form was read and signed by the participants of the group. Sample Composition and Evaluation Steps --------------------------------------- One week before the start of group therapy, which was considered the pre-intervention stage, the work proposal, the dates scheduled for the meetings, and the content scheduled for each session were provided. Information was collected about age, gender, time, and activity of the disease, arterial hypertension, diabetes, arthralgia, and type of acromegaly treatment---surgery, radiotherapy and/or oral or injectable medication. The disease was considered active when IGF-1 was above the normal value for age and gender, even when using cabergoline, octreotide, or pegvisomant; considered controlled when the use of these drugs kept blood concentrations of IGF-1 within the normal range; and considered cured when the concentrations of this hormone were within the normal range without the use of these drugs. Hypertensive participants were considered as those with blood pressure equal to or \>140 × 90 mmHg, or those using anti-hypertensive drugs ([@B25]). A participant was considered to have diabetes mellitus type 2 when glycemia was equal to or higher than 125 mg/dL, or 2 h post oral glucose tolerance test higher than 199 mg/dL or HbA1 levels equal to or higher 6.5%, or if he/she was using hypoglycemic drugs ([@B26]). The groups were selected according to convenience, availability, and the choice of each participant. Patients of the control group did not participate on this phase of the research due to the distance of their homes or difficulties with transportation. Despite the interest in participating, they did not manage to do so due to the need to be at the ambulatory on a weekly basis. Those who chose to participate in the therapy sessions were considered as the intervention group (IG)---aged between 42 and 69 years---and those who preferred not to participate in this phase of the experiment were considered as the control group (CG)---aged between 30 and 75 years. All participants in the two groups responded to the evaluation instruments before the treatment started. The details of all steps of the treatment were clarified. Acromegalics attended weekly, one and a half hour group sessions that applied the technique "Think healthy and feel the difference" ([@B23], [@B27]--[@B30]); for a total of 9 weeks. The IG was treated in the 3 month period from September to November 2015. In the post-intervention step, after the end of the nine sessions of application of the technique, all IG and CG participants were submitted to evaluations with the aim of analyzing the effect of CBT on the IG. The results were compared between groups. For the follow-up phase, in August 2016, 9 months after the end of the IG sessions, the IG and CG participants were once again submitted to evaluations. Three participants in the control group did not respond to the evaluation instruments. As a result, the total number of participants dropped from 23 to 20 ([Figure 1](#F1){ref-type="fig"}). {#F1} Evaluation Instruments ---------------------- For acromegaly, AcroQol is the instrument used in 80% of studies to evaluate quality of life. It consists of 22 questions, 11 of them more directly related to physical alterations. Regarding the level of cognition, seven of the questions are conditional beliefs with emotional dysregulation of moderate intensity, as for example, item 8: "I feel rejected by people because of my illness." Four of them are core beliefs with greater emotional dysregulation, for example, item 4: "I look awful in photographs." Because of the risk of intense imbalance emotion, AcroQol questions were used only in the seventh session for the participants to identify their cognitive distortions, restructure them and build healthier behaviors. Because of this, two self-administered evaluation instruments were answered by the participants of the two groups in the three stages of the present study: pre-intervention, post-intervention, and follow-up. Quality of Life Questionnaire (SF-36) ------------------------------------- The Short Form 36 Question Health Survey (SF-36) was used to evaluate quality of life ([@B31]). This is a multidimensional questionnaire, composed of 36 items, subdivided into eight sections with their respective questions: physical functional (item 03); role-physical (item 04); bodily pain (items 07 and 08); general health (items 01 and 11); vitality (options a, e, g, and i of item 09); social functioning (items 06 and 10); role-emotional (item 05), and mental health (options b, c, d, f, and h of item 09). In addition to the eight sections, item 02 is not part of the calculation of any section, being used only to evaluate how much the individual is better or worse compared to a year ago. It presents a final score of 0 (zero) to 100 (hundred), in which zero shows the worst state and 100 the best state ([@B31]). Beck Depression Inventory (BDI) ------------------------------- The Beck Depression Inventory (BDI) was used to evaluate signs of depression ([@B24]). The following levels were considered for this scale, as they are for psychiatric patients: minimal: 0 to 11; mild: 12 to 19; moderate: 20 to 35; serious: 36 to 63. A score of 18 to 19 indicates likelihood of depression ([@B24]). Intervention Protocol --------------------- The "Think healthy and feel the difference" protocol ([@B23], [@B27]--[@B30]), adapted for patients with acromegaly, was applied in a group setting and all participants received printed material. The technique is available to download as an application for mobile devices at the *Apple Store and Google Play*, under the name "*Think Healthy*," with versions in Portuguese and English. Patients allowed the researcher to record the sessions for later transcription. A small, plastic, coping card, measuring 10 × 6 cm, with an illustration of the avocados, was delivered to encourage the practice of the techniques and healthy decision making in moments of emotional imbalance. In each session, specific topics were covered. The content of the sessions was presented as follows: (1) basic concepts of CBT and practical examples of the relationship between unhealthy and healthy levels of cognition with emotion and associated behaviors; (2) emotional regulation; (3) increase self-esteem and self-confidence, and supply of a brain-shaped piggy bank (identification and valuation of personal qualities and achievements); (4) "Think healthy and feel the difference" technique---guidelines for completing the six steps adapted for the practice of physical activity; (5) the technique for approaching anger, assertiveness training, and the coping card; (6) the technique for approaching fear; (7) the technique adapted for shame about one\'s physical appearance in acromegaly; (8) clarifications on acromegaly, and (9) a summary of the topics covered in the sessions. To consolidate knowledge and to assist in the practice of cognitive and behavioral skills between the therapy sessions, a therapy notepad and block of coping cards were provided ([@B29]). Data Analysis ------------- To verify the equivalence of profiles of both the control and the intervention groups, we performed the non-parametric Mann-Whitney U test for the scalar variables ([@B32]). For the categorical variables, Chi-square or Fisher\'s tests were performed ([@B32]). To control the impact of non-randomization on the distribution in the groups, we performed covariance analysis (ANCOVA) to compare the SF-36 and BDI measurements in the post intervention period. The baseline means were used as covariates; thus, the groups were equalized by their baseline in the post-intervention comparison ([@B33], [@B34]). In order to verify whether the clinical effect was sustained 9 months after the intervention, the same procedure to compare the measures post-intervention was done; however, to it was added the measures of the post-intervention as a covariant, as well as the baseline. Moreover, the comparison between the intervention and the control groups was made. Student\'s *t*-Test, paired, was used when comparing the measures of the baseline and of the post-intervention, and whether the clinical effect was sustained. To interpret the size of the effects, Cohen\'s criterion was adopted: small (d′ \< 0.3), moderate (0.3 ≤ d′ \< 0.5), or large (d′ ≥ 0.5) ([@B35]). In all analyses, statistical significance was considered to be *p* \< 0.05. In order to determine if the sample size was adequate for the effect size, thus avoiding a type II error, *post-hoc* analysis was made using the G^\*^POWER 3.1 software, where the acceptable statistical power found was β = 0.80 ([@B36], [@B37]). Results {#s3} ======= Sixty-two acromegalics were selected from the endocrinology clinic of the University Hospital of Brasília. Thirty-nine of them did not agree to participate in the study, mainly because they could not attend the study site on a weekly basis. Thus, the sample consisted of 23 patients. There were 15 women (65.2%). Ages ranged from 30 to 75 years (mean = 52.78 ± 12.86 years, CV = 0.24, n = 23), from 42 to 69 years in the intervention group, and from 30 to 75 years in the control group. The mean time since disease diagnosis was 11.60 ± 7.14 years (CV = 0.62, n = 23). The time lag between the diagnosis and the beginning of the treatment ranged from 0.17 to 18 years (mean = 6.37 ± 4.94 years, CV = 0.78, n = 23). Regarding the stages of the disease, four (17.40%) were classified as cured, in 10 it was under control (43.50%), and in nine it was active (39.10%). The MRI assessment of the tumor showed the mean of the largest diameter was 23.36 ± 16.49 mm (CV = 0.71; n = 22). The percentage of IGF-1 levels (ULNR IGF-1) ranged from 30.30 to 384.30 (mean = 144.33 ± 85.00, CV = 0.59, n = 23). As for the treatment strategies, 20 (87%) underwent surgery, 21 (91.3%) used medication, and seven (30.40%) underwent radiotherapy. Regarding the level of depression, 11 (47.8%) presented minimal scores according to the BDI, six (26.1%) had mild scores, and six (26.1%) had moderate scores. As shown on [Table 1](#T1){ref-type="table"}, there was no significant difference between all the variables studied: age, time since diagnosis, delay in diagnosis, largest tumor diameter in MRI, IGF-1 ULNR %, gender, disease activity, underwent therapies, and symptoms of depression. Among the variables in the study, only bodily pain presented significant differences (*p* = 0.002), where the control group had a higher score (mean = 66.67 ± 24.00) compared to the intervention one (mean = 38.89 ± 12.00). ###### Comparison between the control group and the intervention group in relation to age, gender, the clinical and laboratory aspects of the disease, and the study variables in the baseline (*n* = 23). **Control (*n* = 13)** **Intervention (*n* = 10)** **p-value[^\*^](#TN1){ref-type="table-fn"}** ---------------------------------- ------------------------ ----------------------------- --------------------------------------------------- Age (years) 55.00 (50.0--62.0) 46.00 (37.8--58.3) 0.335 Time since diagnosis (years) 11.00 (4.8--19.3) 12.00 (8.75--16.5) 0.926 Delay in diagnosis (months) 48.00 (24.0--108.0) 84.00 (66.0--120.0) 0.532 Largest diameter in MRI (mm) 20.00 (17.0--27.0) 12.00 (11.3--49.5) 0.815 ULNR IGF-1 (%) 117.45 (109.8- 121.3) 118.70 (103.38--245.88) 0.664 **f(%)** **f(%)** ***p*****-value[^†^](#TN2){ref-type="table-fn"}** **Gender** Male 6 (46.15) 1 (10.00) 0.074 Female 7 (53.85) 9 (90.10) **Disease activity** Active 6 (46.15) 3 (30.00) 0.734 Controlled 5 (38.46) 5 (5.00) Cured 2 (15.38) 2 (2.00) **Underwent surgery** Yes 12 (92.31) 8 (80.00) 0.560 No 1 (7.69) 2 (20.00) **Using acromegaly medication** Yes 11 (84.62) 10 (100.00) 0.486 No 2 (15.38) 0 (0.0) **Underwent radiotherapy** Yes 3 (23.08) 4 (40.00) 0.650 No 10 (76.92) 6 (60.00) **Symptoms of depression (BDI)** Minimal 7 (53.85) 4 (40.00) 0.805 Mild 3 (23.08) 3 (30.00) Moderate 3 (23.08) 3 (30.00) **Study variables** **Mean** **Standard deviation** **Mean** **Standard deviation** ***p*****-value[^‡^](#TN3){ref-type="table-fn"}** --------------------------- ---------- ------------------------ ---------- ------------------------ --------------------------------------------------- Physical functional 70.00 26.06 58.50 26.67 0.311 Role-physical 50.00 36.80 32.50 39.18 0.284 Bodily pain 66.67 24.00 38.89 12.00 0.002 General health 53.08 15.07 46.00 21.58 0.364 Vitality 58.85 14.74 55.00 10.00 0.487 Social functioning 62.50 27.95 67.50 14.67 0.614 Role-emotional 56.41 45.92 43.33 49.81 0.521 Mental health 67.08 20.73 62.00 15.23 0.523 Beck depression inventory 12.23 9.57 13.30 9.03 0.788 Q1, 1° Quartile; Q3, 3° Quartile; Non-parametric Mann-Whitney test; Chi-squared test; *t-test for repeated measurements*. The results of the effect of the treatment on the intervention group compared to the control one (patients who did not receive treatment) are presented in [Table 2](#T2){ref-type="table"}. Concerning the SF-36 sections, Vitality was higher in the control group (mean = 66.15 ± 9.61; *p* = 0.041; β \> 0.80). However, Mental Health (mean = 76.80 ± 11.12) was higher in the intervention group when compared to the control group (mean = 68.92 ± 15.68; *p* = 0.025; β \> 0.80). There were no significant differences between both groups in all the other sections. Regarding the BDI, no significant differences were observed between the groups either. ###### Effect of the treatment on the intervention group compared to the control group (no treatment) in relation to the quality of life questionnaire SF-36 and the Beck Depression Inventory (*n* = 23). **Control--post (*****n*** **=** **13)** **Intervention--post (*****n*** **=** **10)** ***p*-value[^**§**^](#TN6){ref-type="table-fn"}** --------------------------- ------------------------------------------ ----------------------------------------------- --------------------------------------------------- ------- ------- ---------------------------------------- ------- Physical functional 66.92 26.34 −0.117 52.50 27.21 −0.223 0.481 Role-physical 59.62 47.37 0.228 65.00 42.82 0.793 0.252 Bodily pain 65.81 25.44 −0.035 50.00 17.57 0.752 0.705 General health 57.69 18.44 0.275[*^†^*](#TN5){ref-type="table-fn"} 51.50 20.15 0.264[^†^](#TN5){ref-type="table-fn"} 0.984 Vitality 66.15 9.61 0.600[*^†^*](#TN5){ref-type="table-fn"} 54.50 14.03 −0.042 0.041 Social functioning 77.88 17.04 0.684 76.25 18.11 0.534 0.571 Role-emotional 53.85 48.19 −0.155 76.67 35.31 −0.243 0.740 Mental health 68.92 15.68 0.101 76.80 11.12 1.123[^\*^](#TN4){ref-type="table-fn"} 0.025 Beck depression inventory 7.54 7.69 −0.544 12.50 8.49 −0.091 0.139 d′- d′ of Cohen; *t-test for repeated measurements p\<0.01*; *t-test for repeated measurements p \< 0.05*; *ANCOVA; Post: Intervention x Control, Covariant: Baseline (SF36 and BDI)*. When comparing the baseline to the post-intervention, there was a significant increase in the score of General Health both in the control group (D′ = 0.280; *P* = 0.038; B \> 0.80) and in the intervention one (D′ = 0.26; *P* = 0.024). Regarding the Vitality section, there was a significant increase only in the control group (D′ = 0.60; *P* = 0.022; β \> 0.80). The score for Mental Health has a significant increase (*P* = 0.008), the greatest effect found (D′ = 1.123; β \> 0.80). There were no significant differences in the other sections nor in the BDI evaluation. [Table 3](#T3){ref-type="table"} presents the adjustments measures for the ANCOVA models of the effect of the treatment in the intervention group compared to the control group, which did not receive the treatment, in the post-intervention. Concerning the SF-36, the sections were well-adjusted to the model, except for the Vitality variable (*R*^2^ = 0.176). ###### Adjustment measures for the ANCOVA models (*n* = 23) when comparing the control group (no treatment) to the intervention group in the post-intervention in relation to the quality of life questionnaire SF-36 and the Beck Depression Inventory. **Control (*n* = 13)** **Intervention (*n* = 10)** ***R*^2^** **Base level standardization** --------------------------- ------------------------ ----------------------------- ---------------------------------------- -------------------------------- Physical functional 62.88 57.75 0.647 65.00 Role-physical 53.89 72.44 0.334[^\*^](#TN7){ref-type="table-fn"} 42.39 Bodily pain 57.37 60.98 0.339[^\*^](#TN7){ref-type="table-fn"} 54.59 General health 55.00 54.94 0.639[^\*^](#TN7){ref-type="table-fn"} 50.00 Vitality 65.81 54.95 0.176 57.17 Social functioning 78.77 75.10 0.225[^\*^](#TN7){ref-type="table-fn"} 64.67 Role-emotional 50.99 80.38 0.292[^\*^](#TN7){ref-type="table-fn"} 50.72 Mental health 67.77 78.29 0.477[^\*^](#TN7){ref-type="table-fn"} 64.87 Beck depression inventory 7.76 12.21 0.311[^\*^](#TN7){ref-type="table-fn"} 12.70 R^2^, Model adjustment; *p \< 0.001*. Nine months after the end of the intervention, patients of both groups were again evaluated using SF-36 and BDI. Only the Mental Health section sustained a significant difference (*p* = 0.033) when comparing the control group (mean = 65.60 ± 15.91) to the intervention group (mean = 74.80 ± 21.25), in which case the intervention one had higher score (β \> 0.80). The patients during the 9 months before the final evaluation had no changes in their disease stage. The comparison between the post-intervention and the 9 month follow-up is presented in [Table 4](#T4){ref-type="table"}. The *t*-test for repeated measurements did not show significant difference either in the control group or the intervention group in any of the SF-36 sections or BDI. ###### Comparisons of follow up between the control and intervention groups in relation to the quality of life questionnaire SF-36 and the Beck Depression Inventory (*n* = 20). **Control--follow-up (*****n*** **=** **10)** **Intervention--follow-up (*****n*** **=** **10)** ***p*-value[^\*^](#TN8){ref-type="table-fn"}** --------------------------- ----------------------------------------------- ---------------------------------------------------- ------------------------------------------------ ------- ------- -------- ------- Physical functional 69.50 24.43 0.041 50.00 27.59 −0.091 0.364 Role-physical 58.33 43.30 0.130 50.00 45.64 0.413 0.989 Bodily pain 56.67 20.59 −0.448 51.11 24.68 0.053 0.852 General health 54.00 16.47 −0.215 47.00 20.58 −0.221 0.255 Vitality 59.50 12.12 −0.562 50.00 14.14 −0.319 0.395 Social functioning 73.75 18.11 −0.296 65.00 27.51 −0.493 0.308 Role--emotional 55.56 47.14 −0.155 53.33 50.18 −0.546 0.970 Mental health 65.60 15.91 −0.133 74.80 21.25 −0.124 0.033 Beck depression inventory 11.40 12.02 0.424 10.60 6.24 −0.258 0.465 d′- d′ of Cohen; ANCOVA, Post: Intervention x Control, Covariant: Baseline and Post (SF36 and BDI) [Table 5](#T5){ref-type="table"} presents the adjustment measures for the ANCOVA models when comparing the follow-up between the groups. Regarding the SF-36 instrument, only the Bodily Pain section (*R*^2^ = 0.156), General Health Perception (*R*^2^ = 0.141) and Mental Health (R^2^ = 0.351) were well-adjusted to the model. ###### Adjustment measures for the ANCOVA models (*n* = 20) when comparing the control group (no treatment) and the intervention one, during the follow-up, in relation to the quality of life questionnaire SF-36 and the Beck Depression Inventory. **Control (*n* = 10)** **Intervention (*n* = 10)** ***R*^2^** **Base level standardization** --------------------------- ------------------------ ----------------------------- ---------------------------------------- -------------------------------- ------- Physical Functional 64.215 55.285 0.488 63.75 60.50 Role-Physical 53.792 54.087 0.018 42.11 65.79 Bodily Pain 55.152 52.626 0.156[^\*^](#TN9){ref-type="table-fn"} 52.22 58.33 General Health 55.645 45.355 0.141[^\*^](#TN9){ref-type="table-fn"} 51.00 54.75 Vitality 57.384 52.116 0.221 57.25 60.25 Social Functioning 75.03 63.72 0.017 63.75 77.50 Role-Emotional 53.908 54.816 0.059 52.63 70.13 Mental Health 60.997 79.403 0.351[^\*^](#TN9){ref-type="table-fn"} 63.60 72.20 Beck Depression Inventory 12.53 9.47 0.149 12.50 9.80 R^2^, Model adjustment; *p \<0.01*. Discussion {#s4} ========== The results showed that the application of the "Think Healthy" technique, for 9 weeks, resulted in an improvement in the quality of life in the intervention group when compared to the control group. This confirms the previous observation that the technique is useful in psychological support of acromegalics. ([@B23]). Furthermore, the present study supplements that the effect was maintained at a 9 month follow-up, which suggests stability in the effects produced by the treatment. The strategy used here---nine sessions of group therapy---should be effective, since in groups with obsessive-compulsive disorder, a program of seven sessions was as effective as the same program conducted over 12 sessions ([@B38]). Thus, the good long-term results, described here, of the use of CBT in improving quality of life in acromegalics fills a gap in the management of these patients. This is because, the psychotherapeutic approach has been described as important in other chronic diseases, but not in acromegaly ([@B4]). Regarding the results achieved in the SF-36, in the intervention group, the mental health dimension had a significant increase, showing the greatest effect with the treatment ([Table 4](#T4){ref-type="table"}). In both groups there was an improvement in general health, which could be explained by the fact that all participants of the study where treated on the same neuroendocrinology outpatient clinic of the University Hospital of Brasília. In the control group, the difference in the vitality variable should be interpreted with caution, since it did not present a good fit for the model (*R*^2^ = 0.176) ([Table 3](#T3){ref-type="table"}). This result shows the need of having other studies as well to try and explain this finding. What could have happened is that the IG participants have learned to observe themselves while the CG participants stayed lenient to their evaluation. The other dimensions did not show significant differences, although some presented moderate and large effects, and fitted the model well. The improvement in quality of life after treatment and the fact that it was maintained after a 9 month follow-up may have been facilitated by the characteristics of CBT. The intervention stimulates visual memory, because it uses visual, verbal, and written support. Tasks based on CBT require planning and execution---executive function---and can be considered cognitive training. Socratic questioning---characteristic of the systematization of the technique in its six stages---favors the learning and active participation of patients in their treatment process ([@B18]). In terms of outcome and maintenance of the achieved improvement, the results of this research are similar to those obtained in other studies on other diseases. Improvements achieved in obsessive-compulsive symptoms were maintained in evaluations after a 1-year ([@B38]) and a 2-year follow-up ([@B39]). Positive results were also achieved and maintained in follow-ups for other disorders: anxiety in young people ([@B40]), obesity with binge eating ([@B41]), general anxiety, ([@B42]) psychiatric disorders in comorbidity ([@B43]), compulsive shopping ([@B44]), social anxiety ([@B17]), and body dysmorphic disorder in adolescents ([@B45]). The results described here, using CBT to improve quality of life, are of particular importance because it is still uncertain whether the control of the disease is associated with some gain in this parameter. For example, with biochemical control of acromegaly, a study with a longer duration of follow-up (mean follow-up 5.7 ± 0.6 years) did not show improvement in quality of life when compared to the control group ([@B46]). These authors emphasized that addressing physical and psycho-social comorbidities is as important as the biochemical control of the disease ([@B46]). This study ([@B45]) evaluated the quality of life of 106 surgically treated patients, 3 and 12 months after the surgical intervention. Regarding the SF-36, there was improvement in all dimensions, but it was not statistically significant ([@B47]). On the other hand, a study that used the SF-36 to evaluate the quality of life of 41 acromegalics before surgery, after surgery, and at a 1-year follow-up, showed a statistically significant improvement in the following dimensions: role-physical, general health, social functioning, and mental health ([@B7]). These results were similar to those described in this study for the domains of general health and mental health. The short-lasting treatment and the possibility of recurrence at the end of the treatment are of real concern; however, in the present study the results were maintained at a 9-month follow-up. Some participants reported the benefits of the group format for learning coping strategies. This may be because learning and living experiences, such as psychotherapy, can induce epigenetic changes in brain circuits, similar to the effect of medication ([@B48]). When pharmacology is combined with psychotherapy, due to therapeutic synergy, the result tends to be greater than the sum of the two treatments (1 + 1 = 3) ([@B48]). In the present study, the BDI score levels of the psychiatric patients (i.e., minimum, mild, moderate, and severe) were used because of the lack of a cutoff point for patients with acromegaly. In the pre-intervention assessment, three participants showed an indication of moderate depression, three showed mild depression, and four showed minimal depression. No participant in the present study showed an indication of severe depression. The lowest depression score possible (i.e., zero) was found in only one study participant. In that case, no improvement was possible, which is considered a floor effect. In addition, the decrease in BDI score was not significant. Even though there is no gender difference between both groups, there was a tendency, which may point to a limitation in the generalization of the study. For this reason, the authors suggest that new gender stratification studies be carried in the future. A limitation of the present study is the small case-by-case analysis; however, acromegaly is a rare disease. In addition, we used a convenience sample, in which the patients presented for the treatment, rather than a random sample. However, the psychotherapeutic approach was showed to reduce the suffering and improve the quality of life for acromegalics, with good short- and long-term results. The results achieved by this study reinforce the importance of more research on the use of cognitive-behavioral techniques in different clinical contexts. Ethics Statement {#s5} ================ The research project was approved by the Research Ethics Committee (CEP) of the Faculty of Health Sciences of the University of Brasília---UnB. A Free and Informed Consent form was read and signed by the participants of the group. Author Contributions {#s6} ==================== LK, LC, and LN analyzed the patient data. LK and LC were major contributions in writing the manuscript. LK is a cognitive-behavioral therapist by The Beck Institute and conducted the cognitive-behavioral group sessions. All authors agreed to be accountable for the content of the work and approved the submitted version of the manuscript. Conflict of Interest Statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. BDI : beck depression inventory CBT : cognitive-behavioral therapy CG : control group GH : growth hormone IG : intervention group IGF : insulin-like growth factor-1 MRI : magnetic resonance imaging SF-36 : quality of life questionnaire. [^1]: Edited by: Aleksandra Jawiarczyk-Przybyłowska, Wroclaw Medical University, Poland [^2]: Reviewed by: Odelia Cooper, Cedars-Sinai Medical Center, United States; Leandro Kasuki, Instituto Estadual do Cérebro Paulo Niemeyer, Brazil [^3]: This article was submitted to Pituitary Endocrinology, a section of the journal Frontiers in Endocrinology

The views expressed are those of the authors and not necessarily those of the NHS, the University of Bristol, the NIHR or the Department of Health. Background {#s1} ========== In 2018 there were 17 million new cases of cancer and 9.6 million deaths worldwide ([@B1]). One of its most common forms is breast cancer, a leading cause of cancer mortality worldwide ([@B2]), with over two million new cases in 2018. Dementia is also a major cause of suffering and death globally, with 9.9 million new cases estimated each year ([@B3]); 60--70% of these are diagnosed as Alzheimer\'s disease (AD) ([@B4]). AD and breast cancer, as examples of each disease spectrum, are contrasted here with respect to differences in the PI3K/Akt pathway. By comparing four specific key molecules, we hope to provide some insight into potential, differential therapeutic targeting. Although, due to the limitations of a mini-review we needed to narrow our selection, we acknowledge that additional molecules contributing to the inverse nature of these pathologies have also been reviewed previously ([@B5]). Every normal cell in the body will acquire mutations over a lifetime, which may result in cancer. It has been clear for many years that the initiating mutations and neoplastic transformation may occur decades before symptoms become present and the cancer is diagnosed. Most breast cancers are epithelial tumors that develop from cells lining ducts or lobules: carcinoma *in situ*, and are located exclusively in the breast, tending to be detected by routine physical examination or mammography. Invasive breast cancer can spread however, to most organs, with the main sites being the lungs, liver, bone and brain. There are five main subtypes of breast cancer, depending on the expression of the estrogen, progesterone and human epidermal growth factor receptor 2 (HER2) receptors which dictate treatment strategies ([@B6]). One mutational profile often observed in many cancers is hyperactivity of the PI3K/Akt signaling pathway leading to deregulated control of cell proliferation ([@B7]). Another common feature associated with cancer risk and progression is chronic inflammation, which can be initiated by triggers, such as infections, obesity and autoimmune diseases, the effects of which can be mediated by cytokines, such as tissue necrosis factor (TNF) and interleukins (IL-1 and 6) ([@B8]). As for cancer, the diagnosis of AD usually occurs long after the onset of neuropathology, often 10--20 years later, mainly because symptoms do not generally become evident until the brain has been severely compromised. Loss of short-term memory is usually the first symptom; later, cognitive failure and confusion, and finally an inability to carry out tasks required for successful daily living. Its two defining brain pathologies are the presence of amyloid plaques, comprised mainly of the toxic peptide Aβ42 (processed from the amyloid precursor protein (APP), which quickly fibrillises and deposits in the parenchyma of the brain, and hyperphosphorylated tau, which accumulates within neurones into neurofibrillary tangles (NFT). The parallel spread of these two pathologies across the brain, occurs over a long period before clinical symptoms become evident. Until recently, this has made early diagnosis and assessment of treatment effectiveness difficult. Positron emission tomography (PET) scans with ligands which register amyloid and NFT, as well as markers of neuroinflammation, are now available, helping diagnosis, clinical trial investigation and basic scientific discovery ([@B9]). Recent investigations with PET ligands in living patients suggest that symptoms are noticeable when amyloid and NFT both reach sufficiently high levels ([@B10]). The brain, separated from the peripheral immune system by the blood-brain-barrier (BBB), relies on its innate immune system for defense, this includes production of Aβ42 peptide ([@B11]) and activation of the resident macrophages, microglia, resulting in neuroinflammation, neuronal loss and ultimately death ([@B12]). Unless constantly cleared, Aβ42 forms plaques, whilst toxic, soluble oligomeric forms also contribute to neuronal death. Familial forms of AD with mutations with increased Aβ42 formation, led to the "amyloid cascade hypothesis" ([@B13]) where amyloid precipitates the full spectrum of pathology and symptoms. Although clearly still very useful, this is undergoing re-appraisal in terms of the non-familial or common sporadic form ([@B14], [@B15]). Whilst most cancers, including breast cancer, involve apparently unrestrained cell proliferation, AD involves cell loss. Neurones in the brain, are terminally differentiated post-mitotic cells, which if forced into cycle re-entry usually die ([@B16]). Cancer is associated with an increased glucose uptake by tumor cells, that is preferentially converted to lactate fermentation: a phenomenon known as the Warburg effect ([@B17]). The Warburg effect co-ordinates a number of cellular processes however, in addition to lactate fermentation, including preventing damage from reactive oxygen species (ROS), ensuring that cancer cells have a supportive microenvironment for cell proliferation ([@B18]). By contrast, AD is associated with an early reduction of glucose uptake and utilization in certain areas of the brain ([@B19], [@B20]). Due to its commonly seen insulin-resistance brain profile, AD is sometimes referred to as Type3 diabetes mellitus (T3DM) ([@B19]--[@B22]). Despite the apparently different pathologies, we investigate here aspects of insulin/IGF signaling and the PI3K/Akt pathway that may determine these differences and briefly explore underlying commonalities between the mechanisms which play a role in the two disease states. Glucose intolerance increases generally with age ([@B16], [@B17]) and this is thought to be due to insulin-resistance, commonly observed in older adults ([@B18], [@B19]). Despite the opposing pathologies, cancer and AD have common risk factors such as aging, diabetes, obesity, smoking ([@B23]) and lack of exercise, each of which is also associated with insulin-resistance ([@B24]--[@B27]). Yet, as noted, although the AD brain often develops insulin-resistance, tumor cells generally do not. Here, we discuss normal cellular energy homeostasis and how this differs in cancer and AD. Regulation and Function of Insulin and IGF-1 in Health, Cancer and AD {#s2} ===================================================================== The main source of insulin is that secreted from the beta-cells of the pancreas in response to food; this normalizes the levels of blood glucose, by inducing its target tissues, liver, muscle, and fat cells to increase glucose uptake. IGF-I is secreted by the liver in response to growth hormone, and its circulating levels remain constant via its unique interaction with its IGF binding proteins (IGFBPs) ([@B28]). Unlike insulin, IGF-I (and IGF-II) are also made in most cells of the body, where they play key roles in growth, survival and metabolism. During an insulin-resistant state the usual normalizing processes are inhibited, leading to increased levels of circulating insulin and glucose. This also leads to a stimulation of hepatic IGF-I synthesis ([@B29]), and downregulation of IGFBPs-1 and−2, resulting in an increased bioavailability of IGF ([@B30]). The phosphoinositide-3-kinase-(PI3K/Akt) signaling pathway, as depicted in [Figure 1A](#F1){ref-type="fig"}, has been evolutionarily conserved to regulate and maintain appropriate cell growth, survival and metabolism. This schematic presents an overview of glucose utilization management within normal cells. Two major activators of this pathway are insulin and IGFs ([@B31]) which act via specific receptor tyrosine kinases, IGF-IR and the insulin (IR) receptors. The IR can be spliced to produce two isoforms, IR-A and IR-B. Upon ligand binding, the receptors can dimerize forming IR/IGF-IR hybrids which have different biological consequences depending upon the IR isoform present ([@B32], [@B33]). Generally, insulin acts via the IR, and IGF-I and IGF-II act via the IGF-IR and hybrid receptors. IR-A binds IGF-II and insulin, whereas IR-B has a higher affinity for insulin ([@B34], [@B35]). Emerging data have expanded our understanding of the complexity of these receptors and how they signal, in terms of their localization, trafficking and their ability to interact with other molecules ([@B36]). To ensure adequate fuel, insulin/IGF-I bind and activate IR/IGF-IR, causing tyrosine phosphorylation of insulin receptor substrate-1 (IRS-1), leading to Akt activation. This results in translocation of glucose transporter isoforms (GLUTs) ([@B37]) to the cell membrane enabling glucose uptake. Phosphorylation of mTORC1 initiates subsequent negative feedback mechanisms, such as serine/threonine phosphorylation of IRS-1, which are lost in a cancer phenotype ([Figure 1B](#F1){ref-type="fig"}). mTORC1 (as opposed to mTORC2) is also considered a main regulator of autophagy, that maintains tissue homeostasis by degrading "abnormal" cellular contents ([@B38]). Aberrant autophagy occurs in and contributes to both cancer and AD, however, the impact of this is dependent on the stage of disease for both pathologies ([@B39], [@B40]). ![PI3K/Akt pathway in health **(A)**, cancer **(B)** and AD brain **(C)** cells. This is a schematic of the PI3K/Akt cellular pathway which regulates cell proliferation, metabolism and death. These figures attempt to highlight possible differences in cancer and AD compared with health. These indicated differences, as described in human and animal tissues and in cell culture, are meant to represent general concepts not specific cases. **(A)** shows normal regulation **(B)** indicates a cancer phenotype **(C)** illustrates AD as an insulin-resistant state i.e., T3DM. Green lines represent activation and purple lines represent feedback from the activation pathway. Activation of the IGF-1/insulin receptors leads to tyrosine phosphorylation of IRS-1 and activation of mTORC2 and Akt, resulting in glucose uptake. Homeostasis is maintained partly by mTORC1 sensing of metabolic conditions, which, as appropriate, leads to phosphorylation of p53 and S6K1 serine phosphorylation of IRS-1. p53 is a negative regulator of IGF/insulin receptors, IGF-II and glucose transporters. \[A\] Normal cellular homeostasis as described above \[B\] In cancer, negative feedback pathways are switched off leading to upregulation of proliferation, metabolism and cell survival. A modified genetic landscape (e.g., p53, PTEN) enables tumor cells to benefit from a glucose-rich, IGF/insulin-rich environment (insulin-resistance such as in T2DM).In cancer, Akt can phosphorylate and inactivate GSK-3β, which results in increased protein synthesis that supports cell growth. \[C\] In AD brain with insulin-resistance, or if, due to decreased blood flow there is no glucose accessible, the PI3K/Akt pathway is effectively switched off or downregulated. This leads to upregulation of GSK-β that culminates in tau phosphorylation and aggregation and increased amyloid beta production. Lack of intraneuronal glucose would trigger AMPK to activate mTORC1, p53, S6K1 serine phosphorylation of IRS-1. This could be a self-perpetuating cycle.](fendo-11-00403-g0001){#F1} Epidemiologic studies have shown that "higher" normal levels of circulating IGF-I are associated with a 25% increased risk of breast cancer, compared with "lower" normal levels ([@B41]). Overexpression of the IGF ligands and their receptors, IGF-IR, IR (particularly IR-A) and IGF-IR/IR hybrid receptors leads to increased activity of the PI3K/Akt pathway ([@B36], [@B42]--[@B44]). The IGF-IIR is a single, non-signaling, transmembrane receptor, enabling homeostasis by clearing excess IGF-II ([@B45]); thus loss of function mutations in the IGF-II receptor ([@B46], [@B47]) and/or loss of IGF-II gene imprinting ([@B48]) can lead to excess IGF-II available to activate the PI3K/Akt pathway. IGFBPs are often deregulated in cancer; IGFBP-2, for example, is often upregulated which intrinsically downregulates phosphatase and tensin homolog (PTEN) ([@B49], [@B50]) removing the inhibitory brake on the PI3K/Akt pathway. The cells compensate by upregulating glucose transporters, notably GLUT1, which substantially increases glucose importation into the cytoplasm ([@B51], [@B52]) and the cells switch to lactate fermentation (Warburg effect). AD as an insulin-resistant state, by contrast is exemplified in [Figure 1C](#F1){ref-type="fig"}. The brain has a high energy dependence, using about 20% of the body\'s resting energy requirement (\~60% of glucose use) ([@B53]). Insulin crosses the BBB using a saturable transporter. Although GLUT1 and GLUT3 glucose transporters in the brain are insulin independent, the insulin dependent GLUT4 and GLUT8 are present in regions particularly affected in AD ([@B54]--[@B56]). IR (particularly IR-A) and IGF receptors are also strongly expressed in brain areas, such as the hippocampus, olfactory bulb, hypothalamus and cerebral cortex in neurones and glia and are important in memory formation in the hippocampus ([@B55], [@B57], [@B58]). Brain insulin and IGF levels are reduced in the aged brain with decreased insulin signaling and receptor activity ([@B19], [@B59], [@B60]), coinciding with decline in cognitive abilities. An early reduction of glucose uptake/metabolism is seen in pathology-related brain areas in AD and preclinical, pre-symptomatic subjects ([@B61]--[@B63]). Brain insulin-resistance is associated with impaired cognitive function ([@B54]) and is an important feature of AD in patients and in post-mortem tissue ([@B64]--[@B69]). Reduced insulin or IGF signaling leads to deficient uptake of glucose into neurones in those with mild cognitive impairment (MCI) who subsequently convert to AD, as well as being a major contributor to neuronal dysfunction and death in AD ([@B70], [@B71]). Reduced levels of insulin, IGF-I, II and their receptors associate with severity of pathology ([@B19], [@B72]). Furthermore, binding ability of these proteins is decreased, relative to increasing pathology ([@B59], [@B73]). In experimental studies, reduced IGF-I signaling was linked to increased deposition of Aβ ([@B74], [@B75]), phosphorylation of tau ([@B76], [@B77]), increased oxidative stress, neuro-inflammation and neuronal death ([@B78]). Of interest also, is the finding that the (non-toxic) monomeric form of Aβ can activate insulin/IGF-1 receptor signaling, and since these monomers aggregate in early AD, it is suggested that this may form a prelude to the disease process ([@B79]). Notably, systemic administration of IGF-I was able to lower the toxicity of Aβ in normal mice ([@B80]) and restore cognitive function in AD mouse models ([@B81]). There are studies which are not in line with the hypothesis that IGF-I downregulation in AD is causative in the disease process but rather may be protective. The mixed results may partly lie in the fact that total IGF-I poorly reflects its bioactivity as most circulating IGF-I is bound to IGFBPs and will therefore be biologically inactive ([@B82]). There are also several variables between studies, for instance age of onset, stage of disease progression, presence of diabetes, or IGF-I gene polymorphisms. Therefore, overall, in cancer and AD, the control of these pathways is compromised, allowing feed-forward and feed-backward cycles which lead either to cell over proliferation/deregulation or conversely death. Comparing Regulatory Molecules and Their Role in AD and Cancer {#s3} ============================================================== The PI3K/Akt pathway is kept in equilibrium by key regulators, some of these are briefly discussed here in terms of their effects on glucose metabolism in cancer and AD and are depicted in [Figures 1A--C](#F1){ref-type="fig"}. IRS-1 ----- IRS-1 plays a critical regulatory role in transmitting signals from IGF-IR/IR receptors via the PI3K/AKT pathway. It is commonly overexpressed in cancer and this has been associated with poor outcome for breast cancer patients ([@B83]), particularly if the tumor is positive for the estrogen receptor ([@B84]). Tyrosine phosphorylation activates and serine/threonine phosphorylation inhibits IRS-1 activity. Ribosomal protein S6 kinase beta-1 (S6K1) is one kinase responsible for inhibitory phosphorylation of IRS-1([@B85]) and this negative feedback inhibition is lost in many cancers, including breast cancer ([@B86]). In AD, insulin and IGF signaling is adversely affected in important brain areas. Phosphorylation of IRS-1 at serine 616 (pS616) and p-serine 636/639 are early markers of brain insulin-resistance, commonly present in MCI and AD ([@B67]). Aβ oligomers are thought to initiate IGF-I resistance and IRS-1 inactivation and to be associated with increased oligomeric Aβ plaques and memory impairment. Neurones in the temporal cortex in AD have been reported to show reduced levels of active IRS-1 and−2, but increased inactivated IRS-1, particularly at p-serine 312 and 616, and this was associated with NFT ([@B73]). Apart from indicating insulin-resistance and decreased glucose uptake, it suggests a relationship between IRS-1, tau (NFT) and Aβ pathology. p53 Tumor Suppressor Gene ------------------------- Wild-type p53 regulates many cell functions including cell cycle arrest, apoptosis and metabolism ([@B87]). P53 negatively regulates IGF-IR, IGF-II, GLUTs 1 and 4 and positively stimulates IGFBP-3 (pro-apoptotic factor) ([@B88]--[@B91]). In cancer, including breast cancer, p53 is often mutated, resulting in a loss of its tumor suppressor activity ([@B92]--[@B94]). This disrupts regulation of IGF-IR, IGF-II, GLUTs 1, 4, and IGFBP-3, leading to enhanced activation of the PI3K/Akt pathway and glucose uptake. Increased Aβ positively correlates with p53 levels ([@B91], [@B92]). AD brain levels of p53 are thus increased, which promotes tau hyperphosphorylation and ultimately neuronal death ([@B90]). Peptidyl-Prolyl Cis-Trans Isomerase NIMA-Interacting-1 (PIN1) ------------------------------------------------------------- Pin1 is a peptidyl-prolyl cis--trans isomerase (PPIase) able to isomerise p-serine/p-threonine-proline sequences thus effecting conformational change which alters the activity of its target proteins ([@B95]). It is highly expressed in many cancers ([@B96], [@B97]) and facilitates activation of the PI3K/Akt pathway. One way it does this is by increasing Akt stability through serine 473 phosphorylation ([@B98]). In breast cancer, high levels of both Akt-p-S473 and PIN1 predict a poorer prognosis than either alone ([@B99]). PIN1 can also induce a conformational change to the tumor suppressor gene p53 ([@B100]) and its overexpression in the presence of p53 mutations are prognostic for poor clinical outcome in breast cancer ([@B101]). SUMO protease-1 (SENP1) binds to, and deSUMOylates PIN1, and its levels correlate with those of PIN1 in breast cancer ([@B102], [@B103]). PIN1 is inhibited by *BRCA-1*, the tumor suppressor gene ([@B104]) suggesting that PIN1 would play an important role in the development of tumors in which *BRCA1* is mutated. PIN1 also supports increased cell proliferation by promoting glycolysis in tumor cells. This is achieved by stimulation of pyruvate kinase translocation (that catalyses the rate-limiting step during glycolysis) to the nucleus ([@B95], [@B105]). As a consequence of these functions, PIN1 inhibitors have been developed and shown to slow the progression of cancer ([@B96]). In brain, PIN1 is located in neuronal dendrites and postsynaptic densities and its activity and expression are reduced in MCI and AD ([@B106], [@B107]), likely to make neurons more vulnerable to Aβ and increasing synaptic degeneration ([@B108]). Notably, PIN1 enables tau dephosphorylation via protein phosphatase PP2A and co-localizes with hyperphosphorylated tau in AD brain ([@B109]). Low-Density Lipoprotein Receptor--Related Protein 1 (LRP1) ---------------------------------------------------------- The LRP1 receptor is a multifunctional receptor involved in many cellular functions including endocytosis and cell signaling. Notable is its intrinsic link with energy homeostasis; through its binding to the IGF-IR ([@B110]) and the IR ([@B111]), LRP1 plays a central role in insulin/IGF signaling affecting cell proliferation, survival, glucose and lipoprotein metabolism ([@B112], [@B113]). The role that LRP1 plays in cancer is dependent upon the type of tumor and the cellular environment. In breast cancer, early reports indicated that a low expression of LRP1 correlated with more aggressive tumors ([@B114]). More recent work, however, consistently indicates a role for LRP1 in supporting breast cancer cell invasion and metastasis ([@B115], [@B116]) by increasing expression of matrix metalloproteinases (MMPs), MMP-2, and 9 ([@B117]). In the brain, LRP1 is important for cell survival, lipoprotein metabolism and synaptic plasticity, and is highly expressed in neurones. It binds leptin, enabling leptin receptor phosphorylation and Stat3 activation. Deletion of the *Lrp1* gene in the mouse hypothalamus results in increased body weight (obesity) ([@B118]); conditional *Lrp1* brain knock-out produces glucose intolerance ([@B111]). LRP1 interacts with the insulin receptor, regulating insulin signaling and glucose uptake, and influencing GLUT3 and−4 glucose transporter levels ([@B111]). Insulin resistance in peripheral tissues in rodents involves loss of GLUT4 function ([@B119], [@B120]). Centrally, in the rat hippocampus, GLUT4 is vital to memory acquisition, inhibition causing memory impairment ([@B56]). Amyloid requires constant clearance pathways, LRP1 is known for its function as a clearance receptor able to remove amyloid across the BBB ([@B121]), but also to endocytose Aβ for elimination by lysosomes. LRP1 expression is reduced with age in mouse ([@B122]) and human brain ([@B123]), and to a greater degree in AD ([@B122], [@B123]). Notably, hyperglycaemia and increased insulin resistance, as in type-2 diabetes mellitus (T2DM), suppress LRP1 expression and exacerbate AD pathology in mice ([@B111]). Reduced LRP1 levels are associated with increased neuronal death ([@B124]) signifying that LRP1 is required for the neuroprotective effects of insulin signaling ([@B125]). Summary {#s4} ======= The PI3K/Akt pathway is central to the sensing of metabolic and nutritional changes in our environment and is clearly deregulated in both cancer and AD. Considering that most of the risk factors for both, such as obesity, T2DM and smoking are modifiable through lifestyle changes, an effective strategy could be a preventive approach; for instance re-establishing physiological glucose levels by diet. This minireview, however, attempts to briefly explore some of the underlying mechanisms to identify possible therapeutic targets for these conditions, already ongoing. By addressing the apparent inverse relationship between cancer and AD we hope to identify regulatory molecules in the PI3K/Akt pathway important in cell proliferation and glucose utilization. In cancer this leads to upregulation of glucose uptake and cell proliferation, which contrasts with AD where there is lack of glucose availability, increased pathology, and consequent neuronal death. For both breast cancer and AD there has been a drive for the identification of biomarkers for early detection, ultimately to improve long-term survival. Notably, pre-clinical studies have identified IRS-1, p53, PIN1 and LRP1 as individual potential therapeutic targets ([@B126]--[@B133]) for both disease states, and changes in these are in themselves putative biomarkers. These may provide alternative targets for future trials, but the possibility of inverse effects of altering these proteins, as we outline here, suggests that a delicate balance is required within the PI3K/Akt pathway. It is notable therefore that Metformin, an antihyperglycemic agent for diabetes, appears to promise some beneficial therapeutic outcome in both cancer and AD ([@B134], [@B135]). In cancer the mechanism is likely to be via mTOR inhibition and activation of p53 ([@B136]); in T2DM and T3DM-AD, it is probably the reduction of insulin-resistance ([@B137]). Whilst it is challenging to develop specific drugs for the clinical setting, understanding the regulatory aspects of this pathway may enable a co-targeting approach to reduce non-specific toxicity and increase specificity, thus achieving a better outcome. Author Contributions {#s5} ==================== SA-B and CP proposed the concept for the review. CP, SA-B, KB, JH and RB contributed to writing the paper. KB designed the figures. All authors contributed to the article and approved the submitted version. Conflict of Interest {#s6} ==================== The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. **Funding.** We acknowledge support from Cancer Research UK (C18281/A19169) Programme Grant (The Integrative Cancer Epidemiology Programme); BRACE (Bristol Research into Alzheimer\'s and Care of the Elderly). SA-B is a Sigmund Gestetner Senior Research Fellow. The research was also supported by the NIHR Bristol Nutrition Biomedical Research Unit based at University Hospitals Bristol NHS Foundation Trust and the University of Bristol. [^1]: Edited by: Marco Falasca, Curtin University, Australia [^2]: Reviewed by: Fabio Di Domenico, Sapienza University of Rome, Italy; Yves Combarnous, Centre National de la Recherche Scientifique (CNRS), France [^3]: This article was submitted to Cellular Endocrinology, a section of the journal Frontiers in Endocrinology [^4]: †These authors share senior authorship

Background {#Sec1} ========== *Dirofilaria immitis* (Spirurida: Onchocercidae) is a mosquito-borne pathogen spreading worldwide, and the associated infection (i.e. dirofilariosis) is becoming a threat to animals and humans living in endemic areas \[[@CR1]\]. Although definitive hosts are primarily domestic and wild canids, *Dirofilaria immitis* shows low vertebrate host specificity, infecting several mammalian species (e.g. black bears, cats, ferrets, lions, otters, ocelots). In humans, this parasite may cause a severe clinical condition of increasing concern, with adult stages located mostly in the patient's lungs, eyes or other anatomical districts \[[@CR1]\]. However, little is known about the occurrence and risk of infection of *D. immitis* in pinnipeds. Only a few cases of infection in captive pinnipeds have been described so far \[[@CR2]--[@CR7]\]. Accordingly, no epidemiological studies on pinniped populations are available \[[@CR8]\] and adult *D. immitis* have only been found in one hooded seal (*Cystophora cristata*) \[[@CR2]\], one common seal (*Phoca vitulina*) \[[@CR3]\], and in California sea lions (*Zalophus californianus*) kept in zoological parks in Florida \[[@CR4]\], Louisiana \[[@CR5]\] and Japan \[[@CR6]\]. In the above-mentioned reports, nematodes were found upon necropsy in the right ventricle of the heart, pulmonary arteries, vena cavae, portal vein and pericardial sac \[[@CR3]--[@CR6]\]. Clinical signs, only documented in California sea lions, included cardiopulmonary impairment, coughing and laboured breathing \[[@CR4], [@CR5]\]. Indeed, pinnipeds might remain asymptomatic, even when large numbers of parasites inhabit their heart and associated vessels \[[@CR3]--[@CR6]\]. Here we report dirofilariosis in a population of pinnipeds housed at an oceanographic park in Portugal and the South African fur seal, *Arctocephalus pusillus pusillus*, as a new host for *D. immitis*. Methods {#Sec2} ======= In 2013 and 2014, during the necropsy examinations of two adult South African fur seals (*A. p. pusillus*) housed at the Zoomarine park in Albufeira (Algarve, southern Portugal), *D. immitis* nematodes were accidentally found in the pulmonary arteries and right ventricle of two animals (animals A and B). These cases prompted an epidemiological survey to assess the occurrence of *D. immitis* in the park's resident pinniped population (*n* = 16). In 2015, 5 common seals (*Phoca vitulina*), 2 grey seals (*Halichoerus grypus*), 3 California sea lions (*Zalophus californianus*) and 6 South African fur seals (*A. p. pusillus*) were surveyed for *D. immitis* infection. All animals were housed in facilities with pools and dry areas and no ecto- or endoparasitic treatments were administered. The animals originated from either Europe or Canada, and remained in Portugal for at least 10 years. Physical examination was performed to check for the presence of abnormal clinical signs in individual animals. Blood was collected from the epidural intervertebral vein in the phocids (*P. vitulina* and *H. grypus*) and from the interdigital vein of the hind flippers or caudal gluteal vein in the otariids (*Z. californianus* and *A. p. pusillus*), as previously described \[[@CR9]\]. A rapid commercial qualitative antigen WITNESS^®^ HW Heartworm Antigen Test Kit (Zoetis, Europe) was used to assess the presence of *D. immitis* circulating antigens and a modified Knott's technique was performed to detect circulating microfilariae in the pinnipeds' blood. In one of the animals, it was possible to perform an ultrasound to assess the presence of heartworm infection and evaluate cardiac function. During this epidemiological survey two animals died and necropsies were conducted as described in \[[@CR10], [@CR11]\]. Lung and liver samples were collected for histopathological examination, fixed in 10% buffered formalin and embedded in paraffin; sections (3 μm thick) were stained with haematoxylin and eosin for routine microscopic examination. Genomic DNA was extracted from segments (about 10 mm) of the adult worms collected from the necropsies and from the 16 blood samples, using a commercial kit (DNeasy Blood & Tissue Kit, Qiagen, GmbH, Hilden, Germany) and tested by real-time PCR (qPCR), based on SsoFast™ EvaGreen^®^, targeting partial cytochrome *c* oxidase subunit 1 (*cox*1), coupled with melting-curve analysis for the detection and discrimination of *Dirofilaria* spp. \[[@CR12]\]. The real-time PCR products were purified using Ultrafree-DA columns (Millipore, Bedford, USA), sequenced using BigDye® Terminator v3.1 Cycle Sequencing Kit (Applied Biosystems Inc.) in an automated sequencer (ABI-PRISM 377; Applied Biosystems Inc.). All sequences generated were compared to sequences available in GenBank using Basic Local Alignment Search Tool (BLASTn) \[[@CR13]\]. Results {#Sec3} ======= Two *P. vitulina* were antigen positive (12.5%) and one *A. p. pusillus* scored positive for *D. immitis* microfilariae (6.3%) (Table [1](#Tab1){ref-type="table"}). Circulating microfilariae were 300--305 μm long bearing a conical anterior edge and straight rear end (Fig. [1](#Fig1){ref-type="fig"}). Seven (43.8%) out of the 16 animals were positive for *D. immitis* at qPCR (3 *P. vitulina*, 2 *Z. californianus* and 2 *A. p. pusillus*), with melting peaks (mean Tm = 75 °C) corresponding to the species-specific range of *D. immitis* positive control (mean ± SD: 75.7 ± 0.3 °C) (Table [1](#Tab1){ref-type="table"}).Table 1Test results of the 16 pinnipeds surveyed to *Dirofilaria immitis*AnimalSpeciesSexCountry of originBirth yearAntigen testKnott testqPCRSigns1*P. vitulina*MPortugal2012--------2*P. vitulina*FCanada1989Positive--Positive^a^--3*P. vitulina*MCanada1996--------4*P. vitulina*FPortugal2007----Positive--5*P. vitulina*FCanada1989Positive--Positive--6*H. grypus*MPortugal2002--------7*H. grypus*FCanada1990--------8*Z. californianus*MPortugal1996----Positive--9*Z. californianus*MSpain2003--------10*Z. californianus*MBelgium1996----Positive--11*A. p. pusillus*MPortugal1998----Positive--12*A. p. pusillus*MUK1992--PositivePositiveCough, lethargy and exercise intolerance13*A. p. pusillus*MSweden2002--------14*A. p. pusillus*MSweden2002--------15*A. p. pusillus*MPortugal1996--------16*A. p. pusillus*MSpain1996--------Total12M/4F217--*Abbreviations*: *F* female, *M* male^a^Animal in which transthoracic echocardiography was performed revealing the presence of linear mobile hyperechoic structures within the right ventricle and main pulmonary artery, consistent with heartworms Fig. 1Microfilaria of *Dirofilaria immitis* detected using the modified Knott's technique. *Scale-bar*: 50 μm One *P. vitulina* (Table [1](#Tab1){ref-type="table"}, animal no. 5) and one *A. p. pusillus* (Table [1](#Tab1){ref-type="table"}, animal no. 12) died during the study (Table [2](#Tab2){ref-type="table"}). Data from these two animals were gathered with those obtained from animals' A and B dead prior to the epidemiological survey. Overall, during the necropsies of these four animals (A, B, no. 5 and no. 12), adult male and gravid female nematodes were retrieved from the right ventricle and pulmonary arteries (Fig. [2](#Fig2){ref-type="fig"}). Macroscopically, lungs were congested and haemorrhagic, and mild right ventricular hypertrophy was noticed. In these four cases, gross and histopathological abnormalities associated with *D. immitis* infection were present, including pulmonary congestion and haemorrhages (Fig. [3](#Fig3){ref-type="fig"}), pulmonary emphysema, interstitial and exudative pneumonia, catarrhal bronchitis and hepatic congestion (Table [2](#Tab2){ref-type="table"}, Fig. [4](#Fig4){ref-type="fig"}). In the four cases, cardiopulmonary impairment was noticed. The adult nematodes collected from the four individuals were morphologically and molecularly identified as *D. immitis*. All *cox*1 gene sequences obtained from the adult nematodes and from the pinniped's blood were identical (GenBank accession number KX372755), showing 100% nucleotide identity to a *D. immitis* sequence in GenBank (KF553638).Table 2Data of the four necropsies of seals in which *Dirofilaria immitis* nematodes were detectedAnimal IDSpeciesSexBirth yearDeath yearNo. of adult *D. immitis*Location of adult nematodesPathological observations and histological abnormalitiesA*A. p. pusillus*M1988201315Right ventricle and pulmonary arterySevere pulmonary hemorrhages; extensive pulmonary congestion and moderate pulmonary emphysemaB*A. p. pusillus*M1988201410Pulmonary arteryModerate interstitial and exudative pneumonia; catarrhal bronchitis; moderate pulmonary congestion; central lobular hepatic congestion and dilation of the sinusoids5*P. vitulina*F1989201632Right ventricle and pulmonary arteryExtensive pulmonary congestion12*A. p. pusillus*M1992201626Right ventricle and pulmonary arteryExtensive pulmonary congestion Fig. 2Adult nematodes of *Dirofilaria immitis* collected at the necropsies of South African fur seals. **a** Adult nematode in the right ventricle. **b** Adult nematodes (*arrow*) in the pulmonary artery, showing extensive pulmonary congestion. **c** Male and female adult nematodes recovered from the blood clot. *Scale-bar*: 2 cm Fig. 3Macroscopic appearance of the lungs in the necropsy of a South African fur seal, highlighting an extensive pulmonary congestion and pulmonary haemorrhages Fig. 4Histopathology of the lungs of a South African fur seal, stained with haematoxylin and eosin. **a** Severe pulmonary haemorrhages, note the presence of blood in the lumen of the alveoli. **b** Extensive pulmonary emphysema. **c** Catarrhal bronchitis, note the lumen of a bronchus obstructed with mucus (*arrows*) Discussion {#Sec4} ========== This study provides new epidemiological data of dirofilariosis by *D. immitis* in pinnipeds, diagnosed through clinical, molecular and pathological findings. The South African fur seal is herein described as a new host of *D. immitis*. All animals were housed in a tourist attraction (i.e. an oceanographic park) in the Algarve region, southern Portugal, an area that records the highest number of days/year with suitable conditions for *Dirofilaria* transmission \[[@CR14]\], and the highest prevalence of canine dirofilariosis in mainland Portugal \[[@CR15]\]. For the first time, qPCR was successfully used to diagnose *D. immitis* infection in pinnipeds. Indeed, qPCR detected seven infected animals, of which two *P. vitulina* and one *A. p. pusillus* which were also positive for circulating antigen and microfilariae, respectively. qPCR was highly sensitive in diagnosing *D. immitis* in pinnipeds, since it was able to detect one animal that was only antigen-positive (no. 5) and another that was only-microfilaremic (no. 12), although both presented several nematodes (including gravid females) during the necropsy. In addition, it detected also another case (animal no. 2) only positive for the antigen, but in which transthoracic echocardiography revealed linear hyperechoic structures consistent with heartworms within the right ventricle and main pulmonary arteries. Furthermore, qPCR was also able to detect four other animals that were negative in all other diagnostic tests used (microfilariae and/or antigen; Table [1](#Tab1){ref-type="table"}). All pinnipeds were retested by modified Knott's technique and antigen test to rule out potential false negatives. These further analyses displayed identical results. Positivity to qPCR suggests that alive parasites have been in contact with pinnipeds with no information on the current infection status. Indeed, this molecular assay may detect *D. immitis* DNA from a current infection or, theoretically, from a recently cleared infection. In addition, dirofilariosis in pinnipeds may be featured by transient and low intensity microfilaremia, as in the case of infection in cats. This might be the reason for the detection of microfilaremia in only one of the two pinnipeds who had male and female nematodes at necropsies. Additionally, rapid commercial *D. immitis* antigen tests were specifically developed for canine and feline blood samples, thus, their sensitivity and performance might be poor when used in pinnipeds, underestimating the true prevalence. Although pinnipeds are aquatic mammals, they spend large periods of their life in terrestrial environments, and are therefore exposed to mosquito bites. As in the present survey, *D. immitis* could also occur in other parks in countries with endemic areas for canine dirofilariosis. The occurrence of this situation in the Algarve region, a popular summer destination, should be carefully considered due to the zoonotic potential of this parasite. Indeed, although human dirofilariasis has been often underdiagnosed (probably due to the lack of awareness amongst health professionals and to the difficulties in parasite identification), two cases of pulmonary nodules by *D. immitis* \[[@CR16]\] and two cases of subcutaneous dirofilariasis by *Dirofilaria repens* \[[@CR17], [@CR18]\] have already been reported in Portugal. Conclusions {#Sec5} =========== This study emphasizes the need for active surveillance of dirofilariosis in facilities where animals and humans are in close contact, and strengthens the need for routine heartworm preventive measures \[[@CR19]\] and vector control strategies. The high prevalence of *D. immitis* herein reported in a confined area where pinnipeds are kept, may represent a risk interface for zoonotic pathogen transmission. Therefore, a One Health approach applied on a local and global scale \[[@CR20]\] is vital to improve early diagnosis and control of zoonotic pathogens in humans and wildlife. qPCR : real-time PCR The authors are grateful to Mundo Aquático S.A. - Zoomarine and to Fundação para a Ciência e a Tecnologia. Publication of this paper has been sponsored by Bayer Animal Health in the framework of the 12nd CVBD World Forum Symposium Funding {#FPar1} ======= This study was financially supported by Fundação para a Ciência e a Tecnologia (FCT), through a PhD research grant SFRH/BD/85427/2012 and by the Project UID/CVT/00276/2013. The funder had no role in the design of the study, collection, analysis, interpretation of data or in writing the manuscript. Availability of data and materials {#FPar2} ================================== All data generated or analysed during the study are included in the article. Authors' contributions {#FPar3} ====================== AMA and IM conceived the study, performed morphological and serological analysis and drafted the manuscript; VC performed morphological and molecular analysis and drafted the manuscript; CF and NS conducted clinical examination, sample collection and necropsies; JJC performed histopathological analysis; MSL performed molecular and data analysis; DO analysed data and revised the manuscript; LMC conceived the study and reviewed the manuscript. All authors read and approved the final manuscript. Competing interests {#FPar4} =================== The authors declare that they have no competing interests. Consent for publication {#FPar5} ======================= Not applicable. Ethics approval {#FPar6} =============== All technical procedures were reviewed and approved by Zoomarine's Ethical and Animal Welfare Committee. The protocol was performed in accordance to the standards and guidelines of the Alliance of Marine Mammal Parks and Aquariums (AMMPA) and European Association of Aquatic Mammals (EAAM), and to the National legislation regarding animal welfare (DL 276/2001 and DL 314/2003).

Introduction {#sec1_1} ============ Iron deficiency anemia is an important problem commonly associated with patients on maintenance hemodialysis (HD). Hepcidin-25, the major form of mature hepcidin, plays an important role in the regulation of iron metabolism \[[@B1],[@B2]\]. Binding of hepcidin to its receptor ferroportin (FPN) is the principal cellular process for iron efflux, which, in turn, causes the rapid internalization and degradation of FPN \[[@B1],[@B2]\]. In duodenal enterocytes, high hepcidin prevents the movement of dietary iron into the circulation through its binding to FPN. In hepatocytes or macrophages, hepcidin prevents the efflux of stored iron into the circulation. Conversely, low hepcidin increases intestinal iron absorption and iron efflux from macrophages or hepatocytes \[[@B1],[@B2]\]. Hepcidin-25 is synthesized in the liver in response to iron overload or inflammation \[[@B1],[@B2]\]. Excess hepcidin decreases iron availability for erythropoiesis, leading to its inhibition. As described in our previous study \[[@B3]\], serum levels of hepcidin and ferritin have been usually increased in HD patients \[[@B4],[@B5],[@B6]\], and hepcidin levels were correlated positively with ferritin and negatively with erythrocyte count, hemoglobin (Hb) and hematocrit \[[@B3],[@B4],[@B6]\]. Similarly, serum levels of prohepcidin, a prohormone of hepcidin, were correlated negatively with Hb in HD patients \[[@B7]\]. In iron depletion, reflected by low serum ferritin, serum levels of hepcidin-25 were decreased, facilitating iron availability for erythropoiesis in HD patients \[[@B8]\]. Intravenous iron administration has been shown to increase serum levels of hepcidin in HD patients \[[@B9]\]. These data suggest that hepcidin levels are regulated by the iron storage status. The long-term administration of recombinant human erythropoietin (rhEPO) affects serum levels of hepcidin or prohepcidin in HD patients \[[@B5],[@B10],[@B11],[@B12],[@B13]\]. Regarding the early effect of rhEPO on hepcidin, there are a few studies using healthy volunteers showing an early and significant rise \[[@B14],[@B15]\] followed by suppression of hepcidin \[[@B14],[@B15],[@B16]\] after intravenous or subcutaneous administration of rhEPO. So far, there is no study on the early effect of rhEPO administration on serum hepcidin-25 in HD patients. In addition, it is unknown whether the regulation of hepcidin-25 by short- and long-acting rhEPO is different, and whether rhEPO-induced regulation of hepcidin-25 is dependent on ferritin in HD patients. Furthermore, it remains inconclusive whether an alteration of serum hepcidin-25 is predictive of the response to rhEPO in HD patients \[[@B5],[@B10],[@B11],[@B12],[@B13]\]. The present study was undertaken to address these issues in HD patients. Subjects and Methods {#sec1_2} ==================== The Institutional Review Board at our institution approved this study. Informed consent was obtained from all patients. Nine patients on maintenance HD without diabetes mellitus, hepatic disease, or clinical infection were enrolled in the study. To minimize intraindividual variability of serum hepcidin \[[@B17],[@B18]\] and the effect of dialysis on hepcidin \[[@B11]\], two independent studies for the response of hepcidin-25 to the short-acting epoetin-β (EPO) and the long-acting methoxy polyethylene glycol-epoetin-β (PEG-EPO) were performed using the same patients and dialysis condition with a 2-year interval. To examine the early response of hepcidin-25 to the short- or long-acting rhEPO, serum levels of hepcidin-25 were measured at 0, 3, 6, 9 and 18 h after intravenous administration of EPO (3,000 U) or PEG-EPO (100 U). In addition, they were measured on days 3, 5, and 7 after administration of PEG-EPO. During this study, PEG-EPO was withheld for 7 days until the measurement of hepcidin-25 was completed. In two independent studies with EPO or PEG-EPO, all patients received the constant dose of the same agent of rhEPO for 6 months before, during, and 6 months after the measurement of hepcidin. As rhEPO therapy, EPO (750-3,000 U) and PEG-EPO (100-150 U) were intravenously given 3 times a week and once a month, respectively. Based on the baseline levels of serum ferritin before the measurement of hepcidin, the patients were divided into low (serum ferritin of \<15.0 ng/ml) \[[@B19]\] and high ferritin groups (serum ferritin of ≥15.0 ng/ml). The patients with low or high levels of serum ferritin in two independent studies were combined and grouped into low or high ferritin groups, respectively. An alteration of serum hepcidin-25 after intravenous administration of rhEPO was compared between the groups. To determine whether an alteration of serum hepcidin-25 and ferritin is predictive of the effect of rhEPO therapy on anemia, Hb and iron parameters, including serum ferritin and transferrin, before and 6 months after the measurement of hepcidin were compared between the groups. In addition, to examine the influence of inflammation on serum hepcidin-25, serum levels of interleukin (IL)-6 and C-reactive protein (CRP) were compared between the groups. All parameters analysed were measured just before starting HD. In our series, ultrapure dialysate has been used for HD to minimize microinflammation as previously described \[[@B6]\]. Serum levels of hepcidin-25 were measured using a liquid chromatography-tandem mass spectrometric method as previously described \[[@B20]\]. Hb, serum ferritin, transferrin, and CRP were measured by the standard automated laboratory methods, and serum IL-6 by an enzyme-linked immunosorbent assay. Statistical Analysis {#sec2_1} -------------------- Continuous data are expressed as mean ± SD and categorical data as percentage of the total population. Comparisons of two means were performed using the Mann-Whitney U test, and those of two proportions using the Fisher exact test. A software package (SPSS, version 20.0; SPSS Inc., Chicago, Ill., USA) was used for statistical analysis. A p value \<0.05 was considered significant. Results {#sec1_3} ======= In 9 patients (4 males) on maintenance HD, mean age at the time of the study and duration of HD were 59.0 ± 7.0 years and 120.0 ± 95.0 months, respectively. Table [1](#T1){ref-type="table"} shows the clinical parameters of the HD patients before the measurement of hepcidin after administration of rhEPO in two independent studies. There was no significant difference in Hb, serum levels of ferritin, transferrin, CRP, and IL-6 values before the measurement of hepcidin between the studies. Serum levels of hepcidin-25 rose from the baseline of 6.8 ± 9.8 to 15.9 ± 20.9 ng/ml (p \< 0.05) at 6 h, then declined to 13.8 ± 18.0 ng/ml (p \< 0.05) at 9 h, and returned to the baseline at 18 h (4.6 ± 8.7 ng/ml) after EPO administration (fig. [1a](#F1){ref-type="fig"}). Similarly, serum levels of hepcidin-25 rose from the baseline of 10.9 ± 13.6 to 25.0 ± 29.3 ng/ml (p \< 0.01) at 6 h, then declined to 21.5 ± 23.8 ng/ml (p \< 0.01) at 9 h, and returned to the baseline at 18 h (15.2 ± 18.1 ng/ml) after PEG-EPO administration (fig. [1b](#F1){ref-type="fig"}). Serum levels of hepcidin-25 on day 3 (4.5 ± 6.7 ng/ml) tended to be decreased, but did not differ significantly from the baseline. However, they were significantly decreased on day 5 (1.4 ± 1.9 ng/ml, p \< 0.05) and day 7 (0.5 ± 0.4 ng/ml, p \< 0.05) after PEG-EPO administration, compared to the baseline. Based on the baseline levels of serum ferritin before the measurement of hepcidin in two independent studies, the patients were combined and grouped into a low ferritin group (11 patients with serum ferritin levels of \<15.0 ng/ml; 5 and 6 patients in the study treated with EPO and PEG-EPO, respectively) and a high ferritin group (7 patients with serum ferritin levels of ≥15.0 ng/ml; 4 and 3 patients in the study with EPO and PEG-EPO, respectively) (table [2](#T2){ref-type="table"}). There was no difference in gender ratio between the groups. In the low ferritin group, serum levels of hepcidin-25 were low (\<5.0 ng/ml) at all time points of the measurements after rhEPO administration (fig. [2](#F2){ref-type="fig"}). In contrast, they rose from the baseline of 19.8 ± 12.0 to 46.8 ± 19.8 ng/ml (p \< 0.01) at 6 h and 39.3 ± 15.5 ng/ml (p \< 0.01) at 9 h after rhEPO administration in the high ferritin group. To examine the effect of rhEPO on anemia, Hb and iron parameters such as ferritin and transferrin before and 6 months after the measurement of hepcidin were compared between the low and high ferritin groups (table [3](#T3){ref-type="table"}). The levels of Hb before the measurement of hepcidin were similar between the groups. In the low ferritin group, the levels of Hb tended to increase from the baseline of 10.6 ± 2.6 to 11.1 ± 1.5 g/dl at 6 months after the measurement of hepcidin, although this did not reach statistical significance. In contrast, they were significantly decreased from the baseline of 11.2 ± 0.7 to 9.7 ± 0.9 g/dl (p \< 0.05) in the high ferritin group. In the low ferritin group, there was a trend toward an increase in serum levels of ferritin from the baseline of 8.7 ± 3.7 to 16.8 ± 16.5 ng/dl (p = 0.06) at 6 months after the measurement of hepcidin, although this did not reach statistical significance. In contrast, they were significantly decreased from the baseline of 55.3 ± 49.4 to 29.3 ± 21.8 ng/dl (p \< 0.05) at 6 months after the measurement of hepcidin in the high ferritin group. However, serum levels of transferrin before and 6 months after the measurement of hepcidin were similar between the groups. To examine the influence of inflammation on hepcidin, the values for CRP before and 6 months after the measurement of hepcidin were compared between the low and high ferritin groups. Serum levels of CRP before (0.11 ± 0.16 vs. 0.04 ± 0.04 mg/dl in the low and high ferritin groups) and 6 months after the measurement of hepcidin (0.08 ± 0.07 vs. 0.05 ± 0.03 mg/dl in the low and high ferritin groups) did not differ between the groups. Discussion {#sec1_4} ========== Regarding the late effect of rhEPO on hepcidin or prohepcidin in HD patients, serum levels of hepcidin have been shown to be decreased after rhEPO administration \[[@B10],[@B11],[@B13]\]. Similarly, serum prohepcidin levels were decreased after rhEPO administration and accompanied by an increase in the levels of Hb and hematocrit \[[@B10]\]. Serum levels of hepcidin were negatively correlated with the dose of rhEPO in HD patients, suggesting that the long-term administration of erythropoietin suppresses hepcidin levels \[[@B5]\]. Concerning the early effect of rhEPO on hepcidin, several studies using healthy volunteers have shown an early rise in hepcidin \[[@B14],[@B15]\] followed by its suppression within 2-3 days after rhEPO administration \[[@B15],[@B16]\]. Our study showed a similar pattern, since the short- and long-acting rhEPO induced an early and significant rise in serum hepcidin-25 in HD patients. However, the time to reach the maximum levels of hepcidin and return to the baseline after rhEPO was longer than in healthy volunteers \[[@B14]\] probably because of the reduced clearance of hepcidin and rhEPO in HD patients \[[@B21]\]. In our series, an increase in serum hepcidin-25 in response to rhEPO appears to be dependent on ferritin but not on gender or circadian rhythm \[[@B5],[@B20]\]. In a state of low ferritin, indicative of iron deletion, low hepcidin facilitates iron availability for erythropoiesis by increasing iron absorption and efflux \[[@B1],[@B2]\]. In support of our finding, poor response of hepcidin to rhEPO has been associated with HD patients with low serum ferritin \[[@B22]\]. A recent in vitro study has shown that a low dose of rhEPO increases the hepcidin expression, whereas a high dose of rhEPO suppresses the hepcidin expression in human hepatocytes, depending on the activation of the C/EBPα pathway \[[@B23]\]. In addition, an in vivo study has shown that ferritin upregulates the expression of hepcidin in mice \[[@B24]\]. Our data, together with previous data, suggest that, in HD patients, the rhEPO-induced early rise in hepcidin-25 may be dependent on ferritin levels and/or other pathways. On the contrary, our study showed that serum levels of hepcidin-25 were significantly decreased on days 5-7 after PEG-EPO administration compared to the baseline, suggesting that as the late effect, rhEPO suppresses hepcidin in HD patients, which is similar in healthy volunteers \[[@B15],[@B16]\]. Although the precise mechanisms are unknown, the rhEPO-induced suppression of hepcidin at a late phase may be mediated by circulating factors or signals derived from bone marrow cells \[[@B15],[@B25],[@B26]\] and indirect mechanisms affecting the hematopoietic activity in bone marrow cells \[[@B27]\]. It remains inconclusive whether an alteration of serum hepcidin-25 and ferritin is predictive of the response to rhEPO in HD patients. For example, no difference was found in the serum levels of prohepcidin and hepcidin between HD patients hyporesponsive and responsive to rhEPO \[[@B12],[@B28],[@B29]\]. Serum hepcidin was correlated positively with ferritin \[[@B5],[@B29]\] and negatively with the dose of rhEPO \[[@B5]\], but failed to predict the response to rhEPO. However, lower serum levels of hepcidin \[[@B30]\] and ferritin \[[@B21],[@B31]\] have been associated with increased resistance to rhEPO in HD patients. In contrast, our study showed a trend toward an increase of Hb after rhEPO in HD patients with low serum levels of hepcidin-25 and ferritin. Furthermore, it was found that the levels of Hb were significantly decreased in those with high serum levels of hepcidin-25 and ferritin. These data suggest that an alteration of serum levels of hepcidin-25 and ferritin may be predictive of the response to rhEPO in HD patients. In support of our finding, serum levels of hepcidin or prohepcidin after rhEPO administration were negatively correlated with Hb in HD patients \[[@B10]\]. Low serum levels of prohepcidin have been associated with a lower dose requirement of rhEPO in HD patients \[[@B32]\]. In addition, increased serum levels of hepcidin or prohepcidin have been associated with rhEPO-resistant HD patients \[[@B12],[@B26]\]. Our data, together with these previous data, suggest that an alteration of serum levels of hepcidin-25 and ferritin may be predictive of the response to rhEPO in HD patients. Inflammation increases serum levels of ferritin and hepcidin in HD patients \[[@B1],[@B2],[@B8]\], which is related to a hyporesponsiveness to rhEPO \[[@B28]\]. Since our patients had normal serum levels of CRP and IL-6, absence of inflammation may be crucial for a better response to rhEPO in HD patients. In summary, our study showed that in HD patients, the short- and long-acting rhEPOs exert similar biphasic effects on hepcidin-25 in early upregulation followed by late downregulation. An early rise in serum hepcidin-25 in response to rhEPO may be dependent on serum ferritin and predictive of the response to rhEPO in HD patients. Further studies using larger numbers of HD patients are necessary to confirm this finding. Disclosure Statement {#sec1_5} ==================== The authors have no conflicts of interest to disclose. This study has been supported by The Kidney Foundation, Japan (JKFB09-55). {#F1} {#F2} ###### Clinical parameters before the measurement of hepcidin-25 in response to the short- or long-acting rhEPO in HD patients Variable Before EPO (n = 9) Before PEG-EPO (n = 9) -------------------- -------------------- ------------------------ Hb, g/dl 10.3 ± 1.4 11.1 ± 2.7 Ferritin, ng/ml 38.2 ± 50.7 15.3 ± 11.1 Transferrin, mg/dl 254.8 ± 47.7 265.7 ± 36.8 CRP, mg/ml 0.05 ± 0.07 0.1 ± 0.17 IL-6, pg/ml 2.3 ± 1.2 4.3 ± 3.1 ###### Dose of rhEPO and serum levels of ferritin before the measurement of hepcidin-25 Case Study with EPO Study with PEG-EPO ------ ---------------- -------------------------------------- ----- -------------------------------------- 1 2,250 12.1[\*](#T2F1){ref-type="table-fn"} 100 13.2[\*](#T2F1){ref-type="table-fn"} 2 2,250 19.3 100 11.8[\*](#T2F1){ref-type="table-fn"} 3 9,000 6.1[\*](#T2F1){ref-type="table-fn"} 150 30.8 4 4,500 33.1 100 6.5[\*](#T2F1){ref-type="table-fn"} 5 4,500 149.0 100 34.6 6 4,500 3.8[\*](#T2F1){ref-type="table-fn"} 150 4.0[\*](#T2F1){ref-type="table-fn"} 7 9,000 98.1 150 4.6[\*](#T2F1){ref-type="table-fn"} 8 2,250 10.9[\*](#T2F1){ref-type="table-fn"} 100 11.6[\*](#T2F1){ref-type="table-fn"} 9 9,000 11.6[\*](#T2F1){ref-type="table-fn"} 150 20.7 Low ferritin group. ###### Effect of rhEPO on anemia and iron parameters in HD patients with low and high serum levels of ferritin Variable Low ferritin group High ferritin group -------------------- -------------------- --------------------- -------------- --------------------------------------------- Hb, g/dl 10.6 ± 2.6 11.1 ± 1.5 11.2 ± 0.7 9.7 ± 0.9[\*](#T3F1){ref-type="table-fn"} Ferritin, ng/ml 8.7 ± 3.7 16.8 ± 16.5 55.3 ± 49.4 29.3 ± 21.8[\*](#T3F1){ref-type="table-fn"} Transferrin, mg/dl 276.4 ± 25.8 251.4 ± 32.6 232.0 ± 52.5 248.4 ± 58.3 p \< 0.05 vs. data before the measurement of hepcidin-25.

All relevant data are within the Supporting Information file. Introduction {#sec005} ============ Nitric oxide (NO) has been identified as an important biological messenger involved in a number of physiological processes including blood flow regulation, mitochondrial respiration, and cardiac and skeletal muscle contractility. It is produced endogenously from L-arginine and oxygen by NO-synthases (NOS) or the more recently identified reductive (non-NOS) nitrate (NO~3~^-^) → nitrite (NO~2~^-^) → NO pathway \[[@pone.0235047.ref001]\]. The production of NO via the NOS pathway is inhibited under hypoxic conditions, whereas the NO~3~^-^ → NO~2~^-^ → NO pathway is activated under hypoxic conditions. Once ingested, dietary NO~3~^-^ is reduced to bioactive NO~2~^-^ by commensal anaerobic bacteria found in the saliva and then further reduced to NO via multiple enzymatic and non-enzymatic pathways \[[@pone.0235047.ref002],[@pone.0235047.ref003]\]. The NO~3~^-^ → NO~2~^-^ → NO pathway has been shown to affect a number of physiological processes that could lead to an improved exercise response following NO~3~^-^ ingestion \[[@pone.0235047.ref001]\]. However, the evidence supporting the benefits of NO~3~^-^ as an ergogenic aid is equivocal, and there are several factors that can influence its efficacy including, age, training level, dosage, and the mode, duration and intensity of the exercise \[[@pone.0235047.ref004]\]. Research with healthy younger recreationally active adults has repeatedly shown that ingestion of dietary NO~3~^-^ can reduce the oxygen cost of submaximal exercise and improve exercise performance during high intensity exercise \[[@pone.0235047.ref005],[@pone.0235047.ref006]\]. Conversely, research with younger, well trained endurance athletes is equivocal, as some studies have failed to demonstrate an improvement in exercise performance following NO~3~^-^ ingestion \[[@pone.0235047.ref007]--[@pone.0235047.ref011]\] while others have demonstrated an ergogenic benefit \[[@pone.0235047.ref012]--[@pone.0235047.ref015]\]. There are several potential reasons that have been suggested for the conflicting results between trained and untrained individuals. First, it has been hypothesized that well-trained elite younger athletes may consume higher levels of dietary NO~3~^-^ as a result of the increased daily energy intake related to their increased daily energy expenditure \[[@pone.0235047.ref004]\]. Interestingly, Jonvik et al. have shown the median dietary intake of NO~3~^-^ in a large group of Dutch athletes to be 106 mg per day \[[@pone.0235047.ref016]\], a value not that different from what had been reported in the general Dutch population, 99 mg per day \[[@pone.0235047.ref017]\]. A second reason suggested is that these athletes have increases in NOS activity, thus obviating the need for the NO~3~^-^ → NO~2~^-^ → NO pathway \[[@pone.0235047.ref018]\]. Finally, elite endurance athletes exhibit a lower proportion of type II muscle fibers which have been hypothesized to be more responsive to NO~3~^-^ supplementation \[[@pone.0235047.ref019]\]. Nitric oxide metabolism has also been shown to be affected by age, and older adults have been shown to have impairments to the NOS pathway. More specifically, older adults have lower levels of L-arginine, the NOS substrate, and tetrahydrobiopterin, a NOS cofactor \[[@pone.0235047.ref020],[@pone.0235047.ref021]\]. Additionally, aging is positively associated with increases in asymmetric dimethylarginine--an endogenous inhibitor of NOS \[[@pone.0235047.ref022]\]. We have shown that consumption of a supplement high in NO~3~^-^, such as beetroot juice, leads to elevated plasma NO~2~^-^ levels and may help restore NO metabolism in older adults; whereas a diet high in NO~3~^-^ without supplementation was insufficient at increasing plasma NO~2~^-^ levels \[[@pone.0235047.ref023]\]. As such, NO~3~^-^ supplementation has the potential to improve physical function and exercise tolerance in older adults. Our research, along with that of others, has shown NO~3~^-^ supplementation to have positive effects on physical function and exercise tolerance in older adults with chronic diseases \[[@pone.0235047.ref024]--[@pone.0235047.ref027]\]. Research examining the effects of dietary NO~3~^-^ on exercise performance in healthy older adults is limited. Kelly et al. demonstrated that three days of NO~3~^-^ rich beetroot juice consumption did not improve walking performance in a group of 12 healthy, active 60 to 70 year olds \[[@pone.0235047.ref028]\]. Additionally, Siervo et al. failed to find significant improvements in the physical function of healthy older adults following seven days of NO~3~^-^ supplementation \[[@pone.0235047.ref029]\]. Based on the fact that NO metabolism is affected by age, we hypothesized that NO~3~^-^ supplementation would increase exercise tolerance in middle to older-aged adults who engage in regular strenuous physical activity or competitive sports. Our primary objective of this investigation was to evaluate the effect of supplementation with a NO~3~^-^ rich beverage on exercise tolerance in healthy, active, well-trained middle to older-aged adults as compared to placebo supplementation. Methods {#sec006} ======= Subjects {#sec007} -------- Twenty-nine individuals were screened for participation in the study (see below), and 15 of them between the ages of 41 and 64 completed the study. All subjects signed an informed consent approved by the Wake Forest University Institutional Review Board---approval number IRB00022914. Descriptive characteristics of the subjects completing the study are presented in [Table 1](#pone.0235047.t001){ref-type="table"}. Inclusion criteria included the following: competitive runner or cyclist (must have competed in a running or cycling event within the previous year), between the ages of 40 and 65, able to pedal a stationary bike, engage in moderate physical activity for at least 150 minutes per week or engage in strenuous physical activity for at least 75 minutes per week, able to provide own transportation to study testing visits, able to consume study beverages, able to speak and read English and a willingness to provide informed consent and participate in the intervention. Individuals were excluded from the study if they were a current tobacco user, were participating in another research intervention, had contraindications for engaging in vigorous exercise, had a history of hypotension, diabetes, kidney stones, atrophic gastritis, thyroid disorder other than hypothyroidism, cardiovascular disease, chronic obstructive pulmonary disease, impaired liver or kidney function, inflammatory bowel disease or were taking any of the following medications: phosphodiesterase type 5 inhibitors, nitroglycerin or nitrate preparations, or proton pump inhibitors. The 15 subjects that completed the study all reported that they cycled regularly. Fourteen of the 15 had competed in a cycling event the previous year, and three had competed in running events. 10.1371/journal.pone.0235047.t001 ###### Descriptive characteristics of subjects completing the study. {#pone.0235047.t001g} Females Males All ------------------------------------- ------------- ------------- ------------- **n** 4 11 15 **Age (years)** 52 ± 9 48 ± 4 49 ± 6 **Body Mass (kg)** 54.6 ± 3.4 79.8 ± 11.0 73.1 ± 14.9 **Height (m)** 1.62 ± 0.03 1.77 ± 0.07 1.73 ± 0.09 **BMI (kg/m**^**2**^**)** 20.7 ± 1.8 25.3 ± 2.7 24.1 ± 3.2 **V̇O**~**2\ peak**~ **(ml/kg/min)** 51.1 ± 5.0 51.9 ± 5.2 51.7 ± 5.0 **Work Rate Max (watts)** 220 ± 16 366 ± 41 327 ± 76 Design and protocol {#sec008} ------------------- This study was a double blind, placebo-controlled, randomized, cross over study with exercise tolerance as measured by submaximal constant work rate exercise time as the primary outcome. Data were collected at a private university in Winston-Salem, NC. Participants were recruited from the local community through advertisements placed on local running and cycling websites and Facebook pages. Participant recruitment and follow-up was from 12/17 through 12/19. This trial is registered with ClinicalTrials.gov under the title "Nitrate and Exercise Performance in Middle to Older Aged Adults" number NCT03371966. The flow of subjects through the various phases of the study is shown in the CONSORT flowchart in [Fig 1](#pone.0235047.g001){ref-type="fig"}. {#pone.0235047.g001} Subjects completed one screening visit and four subsequent follow-up visits. A schematic of the study visits is show in [Fig 2](#pone.0235047.g002){ref-type="fig"}. During the screening visit (V1), a 3--4 ml venous blood sample was obtained to determine baseline plasma NO~3~^-^ and NO~2~^-^ levels. Subjects then consumed two servings (4 ounces) of a NO~3~^-^ rich dietary supplement (AMPED NOx; Isagenix International, LLC, Gilbert, AZ, USA) containing 9.9 mmoles of NO~3~^-^. Approximately two hours later, a second 3--4 ml venous sample was obtained for the determination of plasma NO~3~^-^ and NO~2~^-^ levels. Subjects that responded to the NO~3~^-^ dosing by exhibiting a 100 percent or greater increase in plasma NO~2~^-^ were asked to return for a second visit (V2). Previous research has shown a high degree of variability in NO~2~^-^ levels following NO~3~^-^ consumption \[[@pone.0235047.ref030]\]. In fact, James et al. raised the question as to whether studies examining exercise performance should aim for a certain percentage increase in nitrite levels \[[@pone.0235047.ref030]\]. Additionally, our previous research has shown that approximately 20 percent of older adults consuming NO~3~^-^ do not exhibit an increase in plasma NO~2~^-^ levels; therefore, we screened subjects to ensure that they responded to the NO~3~^-^ ingestion by exhibiting an increase in NO~2~^-^ levels \[[@pone.0235047.ref024],[@pone.0235047.ref031]\]. Subjects were not aware of the exact purpose of the study, but were informed that we wished to examine the effect of two different beverages on exercise tolerance. {#pone.0235047.g002} During V2, subjects performed an incremental exercise test on an electronically braked cycle ergometer (Lode Corival 2015, Netherlands) for the determination of their maximal work rate and peak oxygen consumption (V̇O~2\ peak~). The test began with a three-minute warm up at either 40 or 50 watts (female and male, respectively) after which time the work rate was increased by 20 or 30 watts (female and male, respectively) each minute until volitional exhaustion. The subject's maximal work rate and V̇O~2\ peak~ were determined based on the highest work rate completed for a full minute. Oxygen consumption (V̇O~2~) was measured continuously throughout the test. At the conclusion of V2, subjects were scheduled for visit 3 (V3) and provided with a list of foods high in NO~3~^-^ that they were to avoid for 48 hours prior to all subsequent visits. Subjects were also instructed to avoid using antibacterial mouthwash for the duration of the study, to maintain similar exercise routines and diets prior to all subsequent visits and to avoid strenuous exercise for 48 hours prior to all subsequent visits. Additionally, subjects were instructed to arrive at the lab at least 2 hours postprandial and to have not consumed alcohol for 48 hours prior to the visit. Visit 3 consisted of a familiarization submaximal constant work rate exercise test on the cycle ergometer at 75 percent of the maximal work rate achieved during the incremental exercise test at V2. The subjects completed a five-minute warm-up at 50 percent of their maximal work rate, after which time the work rate was increased to 75 percent of their maximal work rate. The subjects were asked to maintain that work rate for as long as they could. If the subjects were able to maintain the work rate for 30 minutes, the resistance was increased by five percent of their maximal work rate every 15 minutes thereafter until the subjects were no longer able to maintain a cadence ≥60 rpm at the prescribed work rate. During this test, V̇O~2~ and the tissue saturation index were measured continuously. Every five minutes throughout the test and at the end of the test, measures of blood pressure, heart rate and ratings of perceived exertion were obtained. Following completion of the exercise test, the subjects were randomized to one of the two treatment arms: seven days of a high NO~3~^-^ beverage, a seven-day washout period and then seven days of a low NO~3~^-^ beverage (placebo); or seven days of a placebo beverage, a seven-day washout period and then seven days of a high NO~3~^-^ beverage. Randomization was done using a computerized program to assign individuals, in the order that they arrived, to the treatments arms using simple random assignment. Subjects were given a sealed box containing the seven-day supply of one of the two beverages. Investigators collecting study related data and supplying the beverages to the subjects were unaware of the beverage being supplied to the subjects. For the seven days leading up to visit 4 (V4) and visit 5 (V5), subjects received a text message reminding them to consume their study beverage. Visit 4 consisted of a submaximal constant work rate exercise test at 75 percent of the maximal work rate and followed the same procedures described for V3. Subjects received a text message the morning of their V4 reminding them to consume their beverage two hours prior to the scheduled visit. Upon arrival for V4, a venous blood sample was obtained for the determination of plasma NO~3~^-^ and NO~2~^-^. Once the blood sample had been obtained, the exercise test was performed. At the completion of the exercise test, V5 was scheduled, and subjects were provided with their respective beverages that were given a sealed box containing a seven-day supply of one of the two beverages. Visit 5 followed the same procedures as described for V4. The high NO~3~^-^ beverage provided to the subjects was AMPED NOx, (Isagenix International LLC, Gilbert, Arizona, USA). Subjects were provided with 14 two-ounce bottles, and, during the high NO~3~^-^ beverage supplementation period, the subjects were instructed to consume two bottles each day. Four ounces of the AMPED NOx contained 9.9 mmoles of NO~3~^-^. The low NO~3~^-^ beverage provided to the subjects was V8 Purple Power, Campbells Food Service (Camden, New Jersey, USA). Subjects were provided with three 12-ounce bottles and a four-ounce measuring cup. During the low NO~3~^-^ beverage supplementation period, the subjects were instructed to consume four ounces each day. For both the low and the high nitrate beverage trials, the subjects were instructed to consume the four-ounce dose at the same time of day throughout the seven-day supplementation period. Four ounces of the V8 Purple Power contained 0.5 mmoles of NO~3~^-^. The V8 Purple Power was selected as the placebo because it had minimal levels of NO~3~^-^ and had a similar consistency and color. All bottles supplied to the subjects had the labels removed and subjects were instructed not to share any information about their beverage with the study staff. Four ounces of the AMPED NOx contained 60 calories, 12 grams of total carbohydrates, 2 grams of fiber and 4 grams of sugar. Four ounces of the V8 Purple Power contained 26.7 calories, 6.3 grams of total carbohydrates, 0 grams of fiber and 5.3 grams of sugar. Both beverages were assayed for their NO~3~^-^ levels using the ENO 20 NOx analyzer, Eicom. Study outcomes {#sec009} -------------- The primary outcome for this study was exercise time, in seconds, during the submaximal constant work rate tests at V4 and V5. Exercise time was recorded from the start of the 75 percent of maximal work rate until the subject could no longer maintain a cadence of 60 rpm or above or volitional exhaustion. Standardized encouragement was provided throughout all tests. Secondary outcomes included plasma NO~3~^-^ and NO~2~^-^ levels, V̇O~2~, tissue saturation index, heart rate, rating of perceived exertion, and systolic and diastolic blood pressures. The evaluation of the secondary outcomes V̇O~2~, tissue saturation index, heart rate, rating of perceived exertion, and systolic and diastolic blood pressures was done at three time points during the constant work rate tests at V4 and V5. These were at five minutes after the start of exercise, iso-time exercise and at the end of exercise. Iso-time exercise was defined as the last minute of the shortest exercise time during either V4 or V5. Values from that time were then matched with those obtained at the same time from the longer duration exercise test. End of exercise blood pressures were measured as the last set of data collected during each of the tests. Oxygen consumption and tissue saturation index were the final complete 60-second values collected. Heart rate and rating of perceived exertion were obtained at the end of exercise. All blood samples were obtained from the subject's antecubital vein and collected in a 4 mL lithium heparin vials for the determination of plasma NO~3~^-^ and NO~2~^-^ levels. The blood samples were centrifuged at 4000 rpm (2006 g) for three minutes, the plasma was removed and stored in a -80º C freezer for subsequent analysis. Nitrate and NO~2~^-^ levels were determined using a NOx analyzer ENO-20 (EICOM, Kyoto, Japan), designed specifically for measuring NO~3~^-^ and NO~2~^-^ in fluid samples and based on the colorimetric Greiss assay. Standard curves were obtained for all measurements and used for quantitative measurements. Oxygen consumption data was collected using a COSMED K5 portable cardiorespiratory gas exchange system (COSMED, Rome, Italy). Continuous wave near infrared spectroscopy (NIRS) was used to measure the tissue saturation index. The NIRS probe (PortaMon, Artinis Medical Systems, Elst, Netherlands) was placed over the belly of Vastus Lateralis of the right leg. Probe position was measured and recorded to ensure identical placement on subsequent visits. The probe was attached to the skin using adhesive tape and then wrapped with a black elastic band to prevent movement and negate the influence of ambient light. The tissue saturation index was calculated using the manufacturer's Oxysoft software. Data were sampled at 10 Hz and then averaged every second. The tissue saturation index is reported as differences between the last 30 seconds of the resting value and the last 30 seconds of the fifth minute of the 75% work rate. Perceived exertion was measured with the Borg 10 point scale. Blood pressure was measured manually using the auscultatory method. Only a single measurement was made at each time point throughout the exercise tests. In most instances, blood pressure was measured in the left arm of the subject unless blood had been drawn from that arm for the NO~3~^-^ and NO~2~^-^ determinations. Heart rate was monitored using a Garmin HRM-Dual heart rate monitor (Garmin, Kansas City, USA). Statistical analyses {#sec010} -------------------- Data were analyzed using SPSS version 25.0. Normality of data were first assessed by visual inspection of normal quantile plots. Suspected deviations from normality were subsequently tested using a Shapiro Wilk test. Due to non-normality of data, the Friedman test was used to test for differences in NO~3~^-^ and NO~2~^-^ values at the various sampling time points. Dependent t-tests were used to test for differences in outcomes between the high and low NO~3~^-^ beverage trials. Due to non-normality of end exercise systolic blood pressure and end exercise RPE values, a Wilcoxon test was used to test for differences in outcomes between the high and low NO~3~^-^ beverage trials. Associations between changes in exercise time for the high and low NO~3~^-^ beverage trials and changes in NO~3~^-^ and NO~2~^-^ levels for the high and low NO~3~^-^ beverage trials and changes in V̇O~2~ for the high and low NO~3~^-^ beverage trials were examined using the Pearson correlation coefficient. The relationship between V̇O~2\ peak~ and changes in NO~3~^-^ and NO~2~^-^ from pre to post beverage consumption at VI were also assessed using the Pearson correlation coefficient. All tests were two-sided and significance was set at p \< 0.05. Outcome variables are reported as means with 95% confidence intervals (CI) or medians with 5^th^, 25^th^, 75^th^ and 95^th^ percentiles. Sample size was based on our previous experience testing older adults using constant work rate tests \[[@pone.0235047.ref024]\]. More specifically, because we were able to achieve statistical significance (p = 0.031) with an effect size of 0.5 using 15 older adults in that previous study, that was the sample size we targeted. Results {#sec011} ======= Plasma NO~3~^-^ and NO~2~^-^ levels pre and post consumption at V1 and at the high and low NO~3~^-^ beverage trials are shown in Figs [3](#pone.0235047.g003){ref-type="fig"} and [4](#pone.0235047.g004){ref-type="fig"}, respectively. Following consumption of the high NO~3~^-^ beverage at V1, plasma NO~3~^-^ levels increased by 260 μM (p \< 0.001). Plasma NO~3~^-^ levels at the high NO~3~^-^ beverage trial were not significantly different from post consumption plasma NO~3~^-^ values at V1 (p = 1.0). Plasma NO~3~^-^ levels at the low NO~3~^-^ beverage trial were not significantly different from pre-consumption NO~3~^-^ values at V1 (p = 1.0). Following consumption of the high NO~3~^-^ beverage at V1, plasma NO~2~^-^ levels increased by 0.47 μM (p \< 0.001). Plasma NO~2~^-^ levels at the high NO~3~^-^ beverage trial were not significantly different from post consumption plasma NO~2~^-^ values at V1 (p = 1.0). Plasma NO~2~^-^ levels at the low NO~3~^-^ beverage trial were not significantly different from pre-consumption plasma NO~2~^-^ values at V1 (p = 1.0). The relationship between V̇O~2\ peak~ and changes in NO~3~^-^ and NO~2~^-^ levels from pre to post beverage consumption at V1 were non-significant (r = -0.047 and 0.095, and p = 0.868 and 0.736, respectively). {#pone.0235047.g003} {#pone.0235047.g004} Exercise time between the high NO~3~^-^ versus low NO~3~^-^ beverage trials (1131 \[806--1456\] versus 1060 \[778--1342\] seconds, respectively) was not significantly different (p = 0.31). Individual differences in exercise time between the high NO~3~^-^ and low NO~3~^-^ beverage trials, expressed as a percent of the low NO~3~^-^ beverage trial time, are shown in [Fig 5](#pone.0235047.g005){ref-type="fig"}. Changes in exercise time between the two trials ranged from as great as a 55% improvement with the high NO~3~^-^ beverage to a 40% decrease in time. Interestingly, the two subjects that had the largest decreases in exercise time with the high NO~3~^-^ beverage (24 and 40%) were both diagnosed with hypothyroidism and were each taking 0.75 mg of levothyroxine per day. The correlations between differences in exercise time for the high and low NO~3~^-^ beverage trials and differences in NO~3~^-^ and NO~2~^-^ levels for the high and low NO~3~^-^ beverage trials were not significant (r = 0.157 and -0.173, and p = 0.593 and 0.554, respectively). Additionally, the correlations between changes in exercise time for the high and low NO~3~^-^ beverage trials and changes in V̇O~2~ for the high and low NO~3~^-^ beverage trials at five minutes and iso-time exercise were not significant (r = -0.050 and 0.105, and p = 0.859 and 0.708, respectively). {#pone.0235047.g005} Mean values for systolic and diastolic blood pressure, oxygen consumption, heart rate and rating of perceived exertion at five minutes into exercise, at iso-time exercise and end exercise are shown in [Table 2](#pone.0235047.t002){ref-type="table"}. Diastolic blood pressure at five minutes was 4.2 mm Hg lower during the low NO~3~^-^ beverage trial as compared to the high NO~3~^-^ beverage trial (P = 0.049). Heart rate at iso-time was 2 beats per minute lower during the high NO~3~^-^ beverage trial as compared to the low NO~3~^-^ beverage trial (P = 0.007). None of the other exercise variables were significantly different between the high and low NO~3~^-^ beverage trials. The tissue saturation indexes for the high and low NO~3~^-^ beverage trials were not significantly different from one another (-7.8 \[-10.6 - -5.0\] vs. -7.8 \[-11.0 - -4.6\] %, respectively; p = 0.990). 10.1371/journal.pone.0235047.t002 ###### Exercise responses during the high and low NO3- beverage trials. {#pone.0235047.t002g} High NO~3~^-^ Beverage Trial Low NO~3~^-^ Beverage Trial P ------------------------------------------------------------ ------------------------------ ----------------------------- ------- **Systolic Blood Pressure, mmHg** **5 Minute** 185 \[174--196\] 185 \[176--195\] 0.988 **Iso-Time** 191 \[180--203\] 190 \[179--201\] 0.835 **End Exercise** 196 \[182--210\] 193 \[183--204\] 0.851 **Diastolic Blood Pressure, mmHg** **5 Minute** 79 \[76--82\] 75 \[71--79\] 0.049 **Iso-Time** 77 \[74--81\] 74 \[69--79\] 0.253 **End Exercise** 78 \[74--82\] 75 \[71--80\] 0.443 **Oxygen Consumption ml**^.^**kg**^**-1.**^**min**^**-1**^ **5 Minute** 43.9 \[40.8--47.1\] 43.1 \[40.0--46.3\] 0.499 **Iso-Time** 47.7 \[44.4--50.9\] 46.4 \[42.5--50.3\] 0.549 **End Exercise** 47.5 \[44.5--50.5\] 46.8 \[43.0--50.7\] 0.303 **Heart Rate (bpm)** **5 Minute** 155 \[147--162\] 153 \[143--163\] 0.566 **Iso-Time** 166 \[159--172\] 168 \[161--174\] 0.007 **End Exercise** 168 \[162--174\] 169 \[162--175\] 0.605 **Rating of Perceived Exertion (1--10)** **5 Minute** 5.1 \[4.2--6.1\] 5.6 \[4.6--6.5\] 0.097 **Iso-Time** 8.8 \[8.2--9.4\] 8.7 \[8.0--9.4\] 0.796 **End Exercise** 9.1 \[8.5--9.7\] 9.3 \[8.8--9.8\] 0.168 Values are means and 95% CI at five minutes into exercise, iso-tme exercise and end exercise at 75 percent of maximal capacity are shown for each of the trials. None of the subjects experienced any harms or detrimental effects as a result of participating in the trial. Discussion {#sec012} ========== The results of this study showed that well-trained middle to older aged adults supplementing with a NO~3~^-^ rich beverage for a seven-day period did not experience a statistically significant improvement in exercise tolerance when compared to supplementing with a low NO~3~^-^ beverage. However, examination of individual subject responses showed a highly variable response to constant work rate exercise with some subjects showing large improvements in performance with high NO~3~^-^ supplementation and others showing a reduced exercise tolerance, in particular those with hypothyroidism. Differences in exercise blood pressure and oxygen consumption between the high and low NO~3~^-^ trials measured at different time points were minimal. The only significant differences were diastolic blood pressure at five minutes was higher and heart rate at iso-time was lower during the high NO~3~^-^ trial. Constant workrate exercise time {#sec013} ------------------------------- In this cohort of highly fit middle to older aged adults, we did not find a statistically significant increase in constant workrate exercise time. However, the 79 second increase in exercise time with the high NO~3~^-^ beverage may be a meaningful improvement for this population. Future research using a time trial protocol or multiple constant work rate tests is needed to establish the practical significance of this improvement. Previous studies with younger, recreationally active subjects using constant work rate tests have reported the improvements in exercise time during constant work rate tests to range between 12 and 25 percent \[[@pone.0235047.ref005],[@pone.0235047.ref006],[@pone.0235047.ref032]\]. Studies of older adults with various comorbidities have also reported similar findings. Studies with COPD patients, peripheral artery disease patients and those with heart failure have reported 8, 18 and 24 percent improvements in exercise time, respectively, with NO~3~^-^ supplementation \[[@pone.0235047.ref024]--[@pone.0235047.ref026]\]. It should be noted, however, that these are not consistent findings. Other studies with older, clinical populations have failed to find improvements in exercise performance with NO~3~^-^ supplementation \[[@pone.0235047.ref033],[@pone.0235047.ref034]\]. As previously mentioned, the variation in the degree of improvement in response to the NO~3~^-^ supplementation seen among the subjects in this study was high. Previous studies with well-trained subjects have reported similar findings, and it has been suggested that with respect to exercise performance there are responders and non-responders to NO~3~^-^ supplementation \[[@pone.0235047.ref008],[@pone.0235047.ref010],[@pone.0235047.ref011]\]. While it is generally believed that NO~3~^-^ supplementation will improve exercise performance in younger recreationally active individuals, the results are equivocal when examining NO~3~^-^ supplementation and exercise performance in athletes \[[@pone.0235047.ref004]\]. There are studies showing improvements in performance with NO~3~^-^ supplementation in athletes \[[@pone.0235047.ref012]--[@pone.0235047.ref015]\], whereas others have failed to find improvements \[[@pone.0235047.ref007]--[@pone.0235047.ref011]\]. Interestingly, while Boorsma \[[@pone.0235047.ref008]\] and Christensen \[[@pone.0235047.ref010]\] both failed to find improvements in overall group mean performance, both studies reported that 20 to 25 percent of their subjects were positive responders to NO~3~^-^ supplementation when examining exercise performance. Presently, it is unclear as to the reason why some well-trained individuals will respond to NO~3~^-^ supplementation, whereas others will not; however, a number of reasons have been proposed. Wilkerson et al. suggested that highly trained individuals have increased NOS activity which might attenuate the influence of the NO~3~^-^ → NO~2~^-^ → NO pathway \[[@pone.0235047.ref011]\]. They reported baseline NO~2~^-^ levels close to 0.4 μM---more than double what we measured in our subjects and higher than what is typically reported. This difference may be due to the older age of our subjects and the associated reductions in nitrate metabolism that occur with aging. Porcelli and colleagues found that after NO~3~^-^ supplementation there were inverse relationships between the subject's V̇O~2\ max~ and reduced oxygen cost of submaximal exercise and improvement in three kilometer time trial performance \[[@pone.0235047.ref035]\]. They also found that subjects with a higher V̇O~2\ max~ had attenuated increases in plasma NO~2~^-^ and NO~3~^-^ following NO~3~^-^ supplementation. We failed to find a significant relationship between V̇O~2\ max~ and changes in either plasma NO~2~^-^ or NO~3~^-^ following NO~3~^-^ supplementation, which again may have been due to the age or the homogeneous fitness level of our subjects. It has also been suggested that physiological adaptations to long-term endurance training may diminish the efficacy of NO~3~^-^ supplementation \[[@pone.0235047.ref011]\]. As previously mentioned, all our subjects were well-trained individuals with high aerobic fitness levels. Because endurance training increases capillary density in skeletal muscle, this may preserve muscle oxygenation and consequently diminish the likelihood of developing a hypoxic environment in the muscle. The reduction of NO~2~^-^ to NO is enhanced in hypoxic environments, whereas the production of NO from the NOS pathway requires oxygen \[[@pone.0235047.ref001]\]. Therefore, the increased availability of oxygen may limit the reduction of NO~2~^-^ to NO and favor the NOS pathway. Nitrate supplementation has also been shown to increase muscle force production \[[@pone.0235047.ref036]\]. As such, differences in the proportion of type I and II fibers in higher fit individuals could also account for some of the variation. Finally, polymorphisms in the NOS gene could also account for some of the variation in exercise performance and/or the response to NO~3~^-^ supplementation. Montesanto et al. noted an association between inducible NOS genotype and physical performance suggesting that genetic variability of NOS may affect exercise performance \[[@pone.0235047.ref037]\]. Effects of dietary nitrate supplementation on plasma nitrite levels {#sec014} ------------------------------------------------------------------- A unique aspect of this study was that we screened subjects for their ability to convert NO~3~^-^ to NO~2~^-^ prior to randomizing them into the study. There is significant variation in the plasma levels of NO~3~^-^ and NO~2~^-^ following consumption of dietary NO~3~^-^ \[[@pone.0235047.ref030]\]. We, along with others, have shown that nearly 20 percent or more of adults do not exhibit a significant increase in plasma NO~2~^-^ levels following consumption of a high NO~3~^-^ beverage \[[@pone.0235047.ref011],[@pone.0235047.ref024],[@pone.0235047.ref031]\]. Additionally, Wilkerson et al. reported that the exercise performance was only improved in those subjects who had increases in NO~2~^-^ levels following supplementation \[[@pone.0235047.ref011]\]. There are several possible reasons for the high variability in the conversion of dietary NO~3~^-^ to NO~2~^-^. First, differences in the abundance of species of the oral bacteria responsible for the conversion of NO~3~^-^ to NO~2~^-^ has been suggested to account for some of this variability \[[@pone.0235047.ref030]\]. Additionally, Porcelli and colleagues have shown an inverse relationship between aerobic fitness levels and increases in NO~3~^-^ and NO~2~^-^ following NO~3~^−^ supplementation \[[@pone.0235047.ref035]\]. We did not find such a relationship in this study, which may have been due to the lack of variability in the aerobic fitness levels of our subjects. Our subjects were all well-trained individuals, and when comparing their fitness levels to age and sex-matched norms, eight were at or above the 95^th^ percentile, three were at or above the 90^th^ percentile and four were at or above the 75^th^ percentile \[[@pone.0235047.ref038]\]. In this study, subjects ingested 9.9 mmol of NO~3~^-^ daily for seven days. Despite seven days of NO~3~^-^ consumption, NO~2~^-^ levels were not significantly different between the V1 post NO~3~^-^ consumption values and those obtained on day seven of the high NO~3~^-^ supplementation trial. While it might seem that a single dose of NO~3~^-^ may be as beneficial as seven days of supplementation, previous trials showing improved economy of oxygen utilization with NO~3~^-^ have used supplementation periods of at least three days \[[@pone.0235047.ref039],[@pone.0235047.ref040]\]. It is unclear as to whether a single dosing would improve oxygen utilization efficiency. Previous studies have shown this level of consumption sufficient to increase NO~2~^-^ levels and improve exercise performance in younger highly trained athletes \[[@pone.0235047.ref013]\] and younger recreationally active men \[[@pone.0235047.ref005]\]. Following consumption of the high NO~3~^-^ beverage, plasma NO~3~^-^ and NO~2~^-^ levels were increased by 934 and 428 percent, respectively, similar to previously reported finding in both younger \[[@pone.0235047.ref015]\] and older adults \[[@pone.0235047.ref024]\]. Exercise economy {#sec015} ---------------- Despite increases in NO~2~^-^, we failed to note differences in submaximal V̇O~2~, and we did not find a relationship between changes in submaximal V̇O~2~ and changes in exercise duration when comparing the high and low NO~3~^-^ beverage trials. These results would suggest that an improvement in exercise economy was not the mechanism responsible for the increase in exercise duration noted in some of our subjects. Rokkendal-Lausch et al. reported similar findings in their young well-trained cyclists \[[@pone.0235047.ref015]\]. They found that while NO~3~^-^ supplementation did improve time trial performance, it had no effect on exercise economy as measured by the power output to V̇O~2~ ratio. A number of other studies with well-trained athletes and healthy older adults have reported similar findings \[[@pone.0235047.ref008],[@pone.0235047.ref010],[@pone.0235047.ref029]\]. In contrast to our findings, Lansley et al. \[[@pone.0235047.ref014]\] and Wilkerson et al. \[[@pone.0235047.ref011]\] both reported improved exercise economy during time trial test with younger well-trained subjects following NO~3~^-^ supplementation. Additionally, a reduced O~2~ cost of exercise following NO~3~^-^ supplementation has been reported with young healthy recreationally active subjects \[[@pone.0235047.ref006]\]. Blood pressure {#sec016} -------------- In addition to a lack of differences in submaximal exercise V̇O~2~, we did not find differences in blood pressure or blood flow between the two trials. The exception to this was the diastolic blood pressure at minute 5 which was higher during the high NO~3~^-^ beverage trial. It is unclear as to why this may have occurred. The overall lack of significant blood pressure differences is supported by Oggioi et al. who failed to find differences in either systolic or diastolic blood pressure of healthy older adults at rest or during an incremental exercise test when comparing a NO~3~^-^ rich beverage to a placebo \[[@pone.0235047.ref041]\]. Siervo et al. also reported no differences in either the resting systolic or diastolic blood pressure of healthy older adults when comparing a NO~3~^-^ rich or a NO~3~^-^ depleted beverage \[[@pone.0235047.ref029]\]. In contrast, Kelly et al. found decreases in both resting systolic and diastolic blood pressure of healthy older adults when comparing a NO~3~^-^ rich to a NO~3~^-^ depleted beverage \[[@pone.0235047.ref028]\]. Significant reductions in systolic and diastolic blood pressure directly linked to NO~3~^-^ ingestion have been reported in young healthy adults \[[@pone.0235047.ref042]\] and older adults with comorbidities \[[@pone.0235047.ref024],[@pone.0235047.ref026]\]. Reasons for the differences among the various studies are not clear. However, it may be due to one or more of the reasons cited above for the differences in exercise time reported among the various studies. Nitrate and hypothyroidism {#sec017} -------------------------- It is unknown whether NO~3~^-^ supplementation negatively impacts hypothyroid patients' exercise tolerance, or if the large decreases we noted in the two subjects with hypothyroidism was a fortuitous finding of this study. Previous research has shown that acute NO~3~^-^ supplementation does not affect thyroid levels in healthy individuals \[[@pone.0235047.ref043],[@pone.0235047.ref044]\], and we are unaware of any data showing acute NO~3~^-^ supplementation to negatively affect exercise performance in hypothyroid patients. However, there is indirect evidence to suggest that NO~3~^-^ supplementation may negatively affect thyroid hormone levels and, therefore, exercise performance \[[@pone.0235047.ref045],[@pone.0235047.ref046]\]. Epidemiological reports suggest a weak link between chronic increases in dietary NO~3~^-^ and increases in the prevalence of hypothyroidism and decreased thyroid hormone levels \[[@pone.0235047.ref045]\]. Furthermore, exercise intolerance has been noted in both treated and untreated hypothyroid patients \[[@pone.0235047.ref046]\]. Further studies are needed to more clearly define the role NO~3~^-^ plays in the exercise response of individuals with hypothyroidism. Strengths and limitations {#sec018} ------------------------- The strengths of our study include the following: It was a double-blind, randomized crossover trial. We had subjects participate in a familiarization trial prior to the high and low NO~3~^-^ beverage trials. We recruited both men and women into the study, and we verified the NO~3~^-^ content of both study beverages using an analyzer designed specifically for measuring NO~3~^-^ and NO~2~^-^. Despite the many strengths of the study, it was not without its limitations. First, two different beverages were used, and they were packaged differently. While we measured the NO~3~^-^ content of both beverages, we did not measure levels of other ergogenic bioactive compounds such as betaine, polyphenols, quercetin or resveratrol. A second limitation of the study is we used an open-ended test rather than a time trial with our subjects. Open-ended test where subjects are required to exercise at a constant work rate for as long as possible have been criticized due to the greater variation in performance time versus closed-end tests where subjects are required to cover a given distance in the quickest time possible or ride for a given time period and told to complete the greatest distance possible \[[@pone.0235047.ref047]\]. However, it should be noted that not all studies using open-ended tests have reported large variations in exercise time \[[@pone.0235047.ref048]\], and studies have reported differences in exercise time when evaluating the effects of other nutritional supplements \[[@pone.0235047.ref049]\]. We chose the open-ended test as it has been shown to be responsive to NO~3~^-^ supplementation, and we wished to hold the work rate constant to determine if there was an increase in exercise economy. Conclusions and recommendations {#sec019} =============================== In conclusion, we found that chronic supplementation with a NO~3~^-^ rich beverage did increase NO~3~^-^ and NO~2~^-^ levels in middle to older-aged adults; however, it did not appear to have a significant effect of exercise tolerance. Further research is needed to determine if dietary NO~3~^-^ will affect exercise performance in individuals with hypothyroidism or if this was a singular finding in this investigation. Additional research is also needed to determine what factors differentiate exercise responses between high and low responders relative to NO~3~^-^ supplementation. Finally, the effect of nitrate supplementation on time trial performance in a group of high-fit, well-trained older adults should be investigated. Supporting information {#sec020} ====================== ###### CONSORT 2010 checklist of information to include when reporting a randomised trial\*. (DOC) ###### Click here for additional data file. ###### (PDF) ###### Click here for additional data file. ###### (XLSX) ###### Click here for additional data file. The authors thank the subjects for their time, their willingness to adhere to the study protocol and repeatedly exercise to exhaustion, and their knowledge that every pleasure\'s got an edge of pain. [^1]: **Competing Interests:**This work was supported by a grant from Isagenix International LLC, Gilbert, AZ awarded to MJB. MJB, GDM, DBK-S and SB received funding from Isagenix LLC to conduct the study. Contents of the manuscript are the sole responsibility of the authors. This does not alter our adherence to PLOS ONE policies on sharing data and materials. The funding source had no involvement with the study design, data collection, data analysis and interpretation or the decision to submit this work for publication. The results of the study are presented clearly, honestly, and without fabrication, falsification or inappropriate data manipulation. Dr. Kim-Shapiro is co-inventor on a patent related to the use of nitrite for cardiovascular conditions, and owns stock in and serves on the scientific advisory board for Beverage Operations LLC, which has licensed Wake Forest University intellectual properties and thus has a financial interest in Beverage Operations LLC. None of the other authors have any financial, non-financial, professional or personal competing interests to declare.

Introduction {#sec1} ============ Lung cancer remains the most common cause of cancer death worldwide, and the 5-year survival rates of non--small cell lung cancer (NSCLC) remain quite poor despite advances in diagnosis and treatment ([@B1], [@B2]). Further, many patients will develop recurrence or progression following primary treatment. The absolute risk of any recurrence at 5 years post-treatment ranges from 33% to 52%, with the majority occurring at a distant site ([@B3], [@B4]). Among prognostic factors for predicting outcomes in NSCLC, tumor stage based on the American Joint Committee on Cancer (AJCC) staging system is currently considered the best for predicting outcomes ([@B5]). More accurate clinical, imaging, and molecular biomarkers will be extremely useful for stratifying patients who are at a higher risk of recurrence and who might benefit from adjuvant or more aggressive treatment options ([@B6]). Maximum standardized uptake value (SUV~max~) on fluorine-18F fluoro-2-deoxy-D-glucose (FDG) positron emission tomography (PET) imaging has also been shown to predict recurrence or death in NSCLC ([@B7]). However, this is a single-voxel metric; we hypothesized that applying a radiomics approach to extract more complex information (eg, texture) from standard medical images could provide additional prognostic information ([@B8], [@B9]). While recent work has evaluated the potential for radiomics features to augment traditional metrics of response ([@B10]--[@B12]), the majority of studies to date have focused on only the metabolic tumor volume (MTV) on PET and, to the best of our knowledge, no study has investigated the peritumoral region. Tumor invasion from the main mass can be defined by infiltration of stroma, blood vessels, or visceral pleura ([@B13]). Recent studies have also shown the potential for tumor cells to spread into air spaces in the lung tissue adjacent to the tumor volume ([@B14]). It is well known that these features may present as border spiculation, vascular convergence, or pleural attachment surrounding the tumor on anatomical imaging, and that they may result in subtle heterogeneous uptake on PET imaging ([@B15]). We investigated the potential of FDG-PET radiomics to predict recurrence in NSCLC by (1) assessing the variability in radiomic feature extraction from PET images and (2) building and validating a radiomics model to predict time to recurrence. We hypothesize that computational imaging features in the tumor and surrounding area on FDG-PET can augment clinical features to improve recurrence prediction. Methodology {#sec2} =========== Patient Selection {#sec2-1} ----------------- We retrospectively analyzed a total of 291 patients with NSCLC from 2 distinct cohorts of prospectively acquired patients (n = 145 and n = 146). The study was approved by our Institutional Review Board, and all subjects signed informed consent before participation. Our study was also compliant with the Health Insurance Portability and Accountability Act. The training cohort consisted of subjects from a pool of patients with early-stage NSCLC referred for surgical treatment at 2 local medical centers between 2008 and 2012 with preoperative PET/computed tomography (CT) performed before surgery (n = 145). This data set is publicly available on The Cancer Imaging Archive ([@B16], [@B17]). We used a second cohort (n = 146) for model validation. This was a cohort from 3 local medical centers between 2010 and 2016. Subjects were selected from patients undergoing evaluation for lung cancer by PET/CT imaging before definitive treatment as part of an observational biomarker study. In both the training and validation cohorts, there were no patients that received neoadjuvant therapy. The AJCC seventh edition system was used for staging. Pathological staging was used in the training cohort and a combination of clinical and pathological staging in the validation cohort. Demographic differences between the training and validation cohorts were assessed using the Wilcoxon rank-sum test for continuous variables and the χ^2^ test for categorical variables. All patients were followed per standard clinical protocol with clinical examination and imaging. We analyzed the combined endpoint of disease recurrence or progression. For stage I--IIIA subjects, we defined recurrence as either local, regional, or distant. For patients with stage IIIB--IV disease, we defined an event as any progression of disease. Time to event or last known follow-up was recorded from the date of pretreatment PET imaging. Image Acquisition {#sec2-2} ----------------- Pretreatment FDG-PET/CT scans were acquired using a standard clinical protocol at 1 of 3 local medical centers. Images were acquired using either a GE Discovery VCT (GE Health care, Waukesha, WI), a GE Discovery LS PET/CT (GE Healthcare, Waukesha, WI), a Siemens Biograph mCT (Siemens Healthcare, Erlangen, Germany), or a Phillips Allegro/Gemini TF PET/CT (Phillips Healthcare, Cleveland, OH). Patients underwent scanning following fasting for a minimum of 6--8 h. A dose of 12--17 mCi of FDG was administered and patients underwent scanning from the skull base to mid-thigh using bed positions acquired every 2--5 minutes ∼45--60 minutes after injection. Manufacturer-specific CT-based attenuated correction was performed using ordered subset expectation maximization reconstruction. Region of Interest Delineations {#sec2-3} ------------------------------- Pretreatment PET images were converted to SUV units normalized by body weight. Two research assistants (S.M. and S.B.) were trained by a board-certified physician in Nuclear Medicine (G.D.) in using MIM Version 6.6 (MIM Software Inc., Cleveland, OH) to contour tumor MTVs using the semiautomatic PET-edge gradient-based segmentation tool. Both observers contoured all images independently in the training cohort. A subset of 21 images considered difficult to contour were reviewed by the same physician and re-delineated if necessary. To assess intraobserver variability, observer 1 (S.M.) contoured all images a second time after a delay of 3 months. We calculated the Dice similarity coefficient (DSC), mean absolute distance (MAD) of the boundary, and absolute volume difference between each set of contours to assess inter- and intraobserver variability of the MTV regions in the training cohort. Observer 1 alone contoured all images in the validation cohort. We then generated a 3-dimensional penumbra region extending outward 1 cm from the surface of the MTV to sample surrounding uptake by using a 3D distance transform with a threshold of 1 cm. This distance was intuitively chosen to sample enough surrounding tissue given the voxel sizes of the PET images, while avoiding oversampling normal tissue. In addition to the MTV alone, we also evaluated the following 2 additional regions: the MTV plus penumbra and the penumbra only (excluding the MTV). Feature Extraction {#sec2-4} ------------------ We extracted radiomics features in the MTV, penumbra, and MTV plus penumbra regions in both cohorts using The Quantitative Image Feature Engine ([@B18]) implemented in MATLAB R2016B (The MathWorks, Natick, MA). In the MTV, features included size (n = 4), sphericity (n = 1), local volume-invariant integral (LVII) shape (n = 39), histogram intensity (n = 12), and gray-level co-occurrence matrix (GLCM) texture (n = 144) ([@B19], [@B20]), for a total of 200 features. Because the penumbra region was generated from the MTV, 44 size and shape measures were not calculated in the penumbra and MTV plus penumbra regions (because they would not be independent measurements), for a total of 156 features in each. This resulted in a total of 512 features for analysis as summarized in [Table 1](#T1){ref-type="table"}. We set a fixed intensity bin size of 0.2 SUV for texture feature calculation to allow a meaningful comparison between images on the same SUV scale. This discretization may also reduce the differences between multiple scanners used in this study ([@B21]). ###### Number of Extracted Features Region of Interest Feature Type Number of Features Total Number of Features in ROI ------------------------------ -------------- -------------------- --------------------------------- Metabolic Tumor Volume (MTV) Size 4 200 Sphericity 1 LVII shape 39 Intensity 12 GLCM texture 144 Penumbra Intensity 12 156 GLCM texture 144 MTV + Penumbra Intensity 12 156 GLCM texture 144 Total Number of Features 512 We then calculated intraclass correlation coefficients (ICCs) across the 3 sets of outlines for each radiomic feature to assess inter- and intraobserver variability. Robust features, defined as those with ICCs \>0.8 in the training cohort, were selected for further analysis ([@B22], [@B23]). Model Building and Validation {#sec2-5} ----------------------------- All radiomic features were normalized (Z-score transformation) before feature selection and model building. We further optimized the features through a generalized linear model via the least absolute shrinkage and selection operator (LASSO) ([@B24]) Cox regression using the *glmnet* package in R software version 3.4.3 ([@B25]). LASSO is a shrinkage and variable selection method for high-dimensional data, which was used to select top features to predict time to recurrence in the training cohort. The robust radiomic features and the 2 known clinical predictors (stage and SUV~max~) were provided to LASSO. Alpha, the regularization parameter, was set to 1 (LASSO penalty) to minimize the number of selected features by shrinking most of the coefficients to zero and to minimize potential overfitting in the training cohort. In total, 100 randomizations of 4-fold cross-validation was used to reduce the effect of randomness in fold selection. The mean cross-validated error curves were averaged for each tuning parameter lambda value across all randomizations. The lambda and corresponding radiomic features associated with the minimum error were selected. We built univariate and multivariate Cox proportional hazards models in the training cohort using the most frequently selected radiomic and/or clinical features. We evaluated the Akaike information criterion (AIC) to compare the quality of the different models, with lower AICs representing a higher quality model. We assessed the likelihood ratio *P*-value for the derived models to show recurrence prediction significance. HRs and 95% CIs were reported for individual variables. To evaluate nested models combining the clinical and/or radiomic features, the likelihood ratio test was used to compare the goodness of fit. To verify prediction validity, we locked the coefficients of the variables in the top model generated from the training cohort and evaluated it in the validation cohort. The prognostic value was assessed using the concordance index with Noether\'s test to determine significance from random (0.5). We performed Kaplan--Meier analysis to separate high- and low-risk groups based on the median risk score in the training cohort. We performed a Student\'s *t* test for dependent samples to compare concordance indices between the models. All statistical analyses and model building were performed using R. Statistical significance was assessed at the *P* \< .05 level. Results {#sec3} ======= Patient Demographics {#sec3-1} -------------------- The training and validation cohorts were similarly matched with regard to median age (*P* = .057) and tumor location (*P* = .571) ([Table 2](#T2){ref-type="table"}). The training cohort had a higher proportion of males (*P* = .005) and adenocarcinoma histology (*P* = .035). There was a slightly higher proportion of stage IV patients in the validation cohort (*P* \< .001), resulting in a larger percentage of patients who recurred/progressed (*P* = .038). The median time to recurrence was 14 months (range, 2--97) in the training cohort and 15 months (range, 1--59) in the validation cohort. The median follow-up time for censored patients without an event was 50 months (range, 1--115) in the training cohort and 32 months (range, 1--76) in the validation cohort. ###### Baseline Patient and Lesion Characteristics Training (n=145) Validation (n=146) *P*-value ------------------------------------------------ ------------------------------------- -------------------- ----------- -------- Age, years 69 (42--87) 71 (41--96) .057 Gender Male 109 (75%) 87 (60%) .005 Tumor Location Right upper lobe 52 (36%) 50 (34%) .571 Right middle lobe 14 (10%) 9 (6%) Right lower lobe 21 (14%) 26 (18%) Left upper lobe 38 (26%) 34 (23%) Left lower lobe 20 (14%) 27 (19%) Tumor Histology Adenocarcinoma 113 (78%) 103 (71%) .035 Squamous cell 29 (20%) 30 (21%) Non--small cell cancer not otherwise specified 3 (2%) 13 (9%) Tumor Stage 0^[a](#TF2-1){ref-type="table-fn"}^ 4 (3%) 0 (0%) \<.001 I 89 (61%) 100 (68%) II 28 (19%) 13 (9%) III 21 (14%) 17 (12%) IV 3 (2%) 16 (11%) Recurrence/Progression Yes 40 (28%) 57 (39%) .038 No 105 (72%) 89 (61%) Variables shown as median (range) or number (%). ^a^ Pathological stage 0 disease is defined as a carcinoma in situ (TisN0M0) as per the American Joint Committee on Cancer (AJCC) 7th edition staging system. Segmentation Variability {#sec3-2} ------------------------ [Table 3](#T3){ref-type="table"} shows the Dice Similarity Coefficient (DSC), Mean Absolute Boundary Distance (MAD), and absolute volume difference between observers in the training cohort. Overall, semiautomatic segmentations were highly reproducible with an average DSC \>0.9, MAD \<1 mm, and volume differences \<1 mL. When we inspected images with low DSC, high MAD, and/or high volume differences, we found that lesions that had the largest degree of variability tended to have a low uptake (eg, SUV~max~ \<2), heterogeneous uptake, and/or were adjacent to structures with a similar metabolic uptake as the tumor (eg, the heart or mediastinum), making the precise boundary of the tumor difficult to determine. These features were evident in ∼20% of the cases. ###### Inter- and Intraobserver Variability in Metabolic Tumor Volume (MTV) PET-edge Segmentations Observer^[a](#TF3-1){ref-type="table-fn"}^ Dice Similarity Coefficient (DSC) Mean Absolute Boundary Distance (MAD, mm) Absolute Volume Difference (mL)^[b](#TF3-2){ref-type="table-fn"}^ -------------------------------------------- ----------------------------------- ------------------------------------------- ------------------------------------------------------------------- A vs a (Intra) 0.916 (0.090) 0.548 (0.544) 0.71 (1.66) A vs B (Inter) 0.917 (0.087) 0.559 (0.507) 0.58 (0.92) a vs B (Inter) 0.904 (0.105) 0.628 (0.631) 0.79 (1.46) All values are the mean (standard deviation). ^a^ Observer 1 contoured each tumor twice (A and a) and observer 2 contoured each lesion once (B). ^b^ For reference, the average \[range\] volumes of all MTV contours by the three observers were 15.4 \[0.4--297.8\], 15.3 \[0.4--296.9\], and 15.3 \[0.3--296.0\] mL. Feature Variability {#sec3-3} ------------------- [Table 4](#T4){ref-type="table"} shows the ICCs of the 4 different classes of radiomic features in each of the 3 regions of interest. We found that a total of 435 of the 512 features (85%) had an ICC \>0.8 ([Table 5](#T5){ref-type="table"}) and were considered robust to differences in the segmentations ([@B22], [@B23]). ###### Intraclass Correlation Coefficients for All FDG-PET Radiomic Features Feature Type MTV Penumbra MTV + Penumbra -------------- ------------------- ------------------- ------------------- ------------------- ------------------- ------------------- Size 0.996(0.99--1.00) 0.994(0.99--1.00) -- -- -- -- Intensity 0.977(0.89--1.00) 0.972(0.84--1.00) 0.931(0.48--0.99) 0.916(0.36--0.99) 0.995(0.98--1.00) 0.995(0.98--1.00) Shape 0.867(0.37--0.98) 0.847(0.39--0.98) -- -- -- -- Texture 0.898(0.50--0.99) 0.893(0.48--0.99) 0.892(0.14--0.99) 0.925(0.50--0.99) 0.981(0.28--1.00) 0.977(0.66--1.00) All values are shown as the mean (range). ###### Number (percent) of Robust FDG-PET Radiomic Features Selected in Each Category by Virtue of an ICC \> 0.8 Feature Type MTV Penumbra MTV + Penumbra -------------- ----------- ----------- ---------------- ----------- ------------ ----------- Size 4 (100%) 4 (100%) -- -- -- -- Intensity 12 (100%) 12 (100%) 11 (92%) 11 (92%) 12 (100%) 12 (100%) Shape 27 (68%) 30 (75%) -- -- -- -- Texture 115 (80%) 115 (80%) 118 (82%) 131 (91%) 144 (100%) 142 (99%) Feature Selection and Model Training {#sec3-4} ------------------------------------ Across the 100 randomizations, the average minimum cross-validation error was 10.5% at a lambda value of 0.1296 in the training cohort. This lambda generated 2 features with nonzero coefficients, stage, and 1 MTV plus penumbra GLCM texture feature (maximum probability). Although SUV~max~ has previously been shown to be associated with recurrence in NSCLC, it was not selected by LASSO as a top feature. However, it was found to be a significant univariate predictor in our cohort ([Table 6](#T6){ref-type="table"}), consistent with previous studies ([@B7]). ###### Cox Proportional Hazards Model Statistics for Univariate Features in the Training Cohort Feature Akaike Information Criterion Likelihood Ratio *P*-value HR \[95% CI\] Concordance \[95% CI\] --------------------------------------------------------------------- ------------------------------ ------------------ ----------- -------------------- ------------------------ Stage 341.7 19.98 \<.001 2.15\[1.56--2.95\] 0.68\[0.60--0.76\] Gray-level Cooccurrence Matrix Maximum Probability (MTV + Penumbra) 347.5 14.18 \<.001 0.41\[0.23--0.74\] 0.66\[0.57--0.74\] SUV~max~ 353.7 7.99 .005 1.06\[1.02--1.10\] 0.67\[0.58--0.75\] [Figure 1](#F1){ref-type="fig"} visualizes the Pearson correlation coefficients of the top features. For reference, correlation of the top features with MTV volume and SUV~max~ is also shown. All correlations were low and the radiomic feature showed no correlation with stage, volume, or SUV~max~. {#F1} Univariate Cox regression model statistics, including the AIC, likelihood ratios, *P*-values, and HRs, are shown for the top features in [Table 6](#T6){ref-type="table"}. Both features were significant univariate predictors of time to recurrence. Overall, stage was the best univariate predictor. Because stage was the best univariate predictor, the likelihood ratio test was performed to assess significant improvements to this well-established clinical model for recurrence prediction. Additional features were added to determine significant improvements to the model. Adding the MTV plus penumbra texture feature to stage significantly improved the model (*P* = .006). This multivariate model was a significant predictor of time to recurrence in the training cohort (likelihood ratio = 27.59, *P* \< .001, concordance = 0.74 \[95% CI: 0.66-0.81\]). Both stage (HR = 1.92 \[95% CI: 1.37--2.67\], *P* \< .001) and the radiomic texture feature (HR = 0.52 \[95% CI: 0.30--0.91\], *P* = .02) were significant covariates in the multivariate model. Adding SUV~max~ to stage did not significantly improve the clinical model performance (*P* = .22). It also did not significantly improve performance in the combined stage and radiomic model (*P* = .73). Model Validation {#sec3-5} ---------------- Univariate results were confirmed in the validation cohort ([Table 7](#T7){ref-type="table"}), with all features being significant predictors of time to recurrence. The locked multivariate model from the training cohort, which included stage and the radiomic texture feature, was a significant predictor in the validation cohort (concordance = 0.74 \[95% CI: 0.67--0.81\], Noether\'s *P* \< .001). We separated the patients into high- and low-risk groups on the basis of the median risk score in the training cohort. Kaplan--Meier time-to-recurrence curves for the multivariate model in both cohorts are shown in [Figure 2](#F2){ref-type="fig"}. Recurrence was lower in the group below the median model risk score. ###### Cox Proportional Hazards Model Statistics for Univariate Features in the Validation Cohort Feature Akaike Information Criterion Likelihood Ratio *P*-value HR \[95% CI\] Concordance\[95% CI\] --------------------------------------------------------------------- ------------------------------ ------------------ ----------- -------------------- ----------------------- Stage 475.6 35.7 \<.001 2.13\[1.69--2.68\] 0.69\[0.63--0.76\] Gray-level Cooccurrence Matrix Maximum Probability (MTV + Penumbra) 497.2 14.14 \<.001 0.50\[0.33--0.76\] 0.66\[0.60--0.72\] SUV~max~ 506.1 5.24 .02 1.03\[1.01--1.05\] 0.67\[0.61--0.73\] {#F2} The multivariate model including stage and the radiomic feature significantly outperformed the best performing clinical model of stage in the training (*P* = .036) and validation (*P* = .033) cohorts. The combined model also outperformed the radiomic feature alone in both the training cohort (*P* = .019) and the validation cohort (*P* \< .001). [Figure 3](#F3){ref-type="fig"} exemplifies 2 patients with similar SUV~max~ that would typically be considered to be at a high risk of recurrence. Yet, the combined model including radiomics correctly predicted the recurrence status of each patient on the basis of the median risk value. Based on qualitative inspection, the high-risk patient had more heterogeneous uptake in the penumbra region compared with the low-risk patient. {#F3} Discussion {#sec4} ========== We show here evidence that texture in the MTV and nearby surrounding region can predict recurrence in NSCLC. Furthermore, augmenting this radiomic feature with stage significantly improved performance over stage alone, which was validated in an independent data set. This model also showed potential value in risk-stratifying patients with NSCLC who are at high versus low risk of recurrence or progression. A general rule in modeling studies is that 10 patients are needed for every feature selected in the model ([@B8]). To minimize overfitting, our final model consisted of only 2 features. However further studies on larger sample sizes with additional features may improve prognostic performance and applicability to other cohorts. The radiomic feature selected was a GLCM texture feature in the combined MTV plus penumbra volume. This feature, which describes local texture variations, suggests that patients whose PET images show a more heterogeneous texture, specifically in the penumbra region surrounding the MTV, are more likely to recur. This suggests the importance of image data in the surrounding region for recurrence prediction. This region may contain uptake not measured in the MTV (and not by the SUV~max~) and could indicate areas of disease adjacent to the primary mass. The texture being detected in this region may be indicative of an invasive component of the tumor, for example, spiculations or tumor spread through blood vessels, but this requires further investigation ([@B15]). Notably, size or shape features, including the commonly used metrics of maximum axial diameter and 3D volume, were not selected as predictive features. SUV~max~ was also not selected, and adding it to clinical or combined models did not significantly improve performance. This suggests that texture features may provide more useful information than traditional metrics for predicting recurrence/progression. Previous work in the field of radiomics has evaluated FDG-PET features for outcome prediction in lung cancer. Jansen et al. found the GLCM energy texture feature was a significant predictor of overall survival in oligometastatic NSCLC ([@B26]). Others have shown that texture features may be beneficial for predicting local control, distant metastasis, and disease-free survival in lung cancer ([@B10]--[@B12]). However, the majority of studies to date have focused on only the MTV. To the best of our knowledge, ours is the first study that evaluates the lung tumor penumbral region of PET images for recurrence prediction. Future work integrating CT imaging features or molecular data may improve prognostic performance. Our study investigated PET/CT images from multiple scanners and institutions, potentially introducing variability in image data and quality and therefore the construction of a predictive model. We used a standard acquisition protocol across all institutions to minimize this variability ([@B27], [@B28]). This may still result in signal variations in the tumor and penumbra regions; therefore, further studies investigating single scanners are warranted and may improve model performance. Previous work has also shown that PET radiomic features are dependent more on delineation variability than on reconstruction algorithm ([@B29]) and that texture features are less affected by difference in scanners ([@B30]). Many radiomic features also show high test--retest stability with repeat PET imaging ([@B31]). The PET-edge segmentation tool we used for tumor segmentation showed high reproducibility with associated radiomic feature robustness. Segmentations were performed with commercially available software (MIM Software, Inc.), making it an easily deployed and integrated system. Our work is also applicable in a "real world," nonresearch setting, where different scanners and images of variable quality are routinely used for clinical assessment. However, additional external validation of this radiomics model is warranted to determine the impact of different scanners and acquisition protocols on model predictions. Our study has several limitations. The primary limitation is that the penumbra region was not restricted to the lung volume, that is, it may at times have included the adjacent chest wall, major blood vessels, and/or mediastinum. However, as features were selected from within this region, it is providing relevant information for the prediction of recurrence. The effect of this and the efforts to minimize it remain the subject of further investigation. Owing to differences in breathing between the PET and CT images, accurate registration of the lung boundary is challenging. We also investigated only a single distance of 1 cm for the penumbra region; it is possible that larger or smaller distances could improve or degrade performance. Another limitation is the inherent low resolution of the PET images, limiting the amount of information we can analyze for each tumor owing to lower voxel quantities for smaller tumors. Finally, the sample sizes analyzed were relatively small, and validation of this model in larger data sets is warranted. In conclusion, a PET texture feature in the metabolic tumor volume and surrounding region augmented staging for NSCLC recurrence prediction. This model may be useful in identifying patients who are at a higher risk of recurrence or progression and may assist physicians in determining what patients may benefit from adjuvant or personalized treatment options at the time of diagnosis. Abbreviations: NSCLCNon--small cell lung cancerMTVmetabolic tumor volumeFDGfluoro-2-deoxy-D-glucosePETpositron emission tomographyCTcomputed tomographyDSCDice similarity coefficientMADmean absolute distanceGLCMgray-level co-occurrence matrixLASSOleast absolute shrinkage and selection operatorAICAkaike information criterionHRhazard ratioCIconfidence intervalICCsintraclass correlation coefficients Equal contribution: "S.N and V.S.N contributed equally to this work." The authors would like to acknowledge MIM Software Inc. for their assistance with segmentation software, Jalen Benson and Weiruo Zhang for their assistance with clinical data curation, and the Stanford Data Studio for statistical consulting. The authors would like to acknowledge funding from the Natural Sciences and Engineering Research Council of Canada (NSERC) Postdoctoral Fellowship and the National Cancer Institute (NCI) R01 CA160251, U01 CA187947, and U01 CA196405. Disclosures: Dr. Sandy Napel is a Consultant for Carestream Health Inc., on the Medical Advisory Board for Fovia Inc., a Scientific Advisor for EchoPixel Inc., and a Scientific Advisor for RADLogics Inc. However, he is not an employee of any of these companies and none of these conflicts are related to the data used and research completed in this manuscript. Conflict of Interest: The authors have no conflict of interest to declare.

{#sp1 .160} {#sp2 .161} {#sp3 .162} {#sp4 .163}

Introduction {#s1} ============ The ecological success of coral reefs in nutrient poor waters relies on the nutrient-exchange between cnidarians and dinoflagellate algae of the genus *Symbiodinium* living in the host\'s tissues (Muscatine and Porter, [@B45]; Falkowski et al., [@B17]; Hatcher, [@B26], [@B27]). In this association, the endosymbiotic algae translocate the majority of their photosynthetically-fixed carbon to the host, which in turn provides inorganic nutrients from its metabolism to sustain algal productivity (Muscatine, [@B42]; Muscatine et al., [@B44]; Falkowski et al., [@B17]; Rädecker et al., [@B51]). The efficient recycling of organic as well as inorganic nutrients within this symbiosis underpins the high productivity of coral reefs in the absence of major sources of allochthonous nutrients (Muscatine and Porter, [@B45]; Wang and Douglas, [@B64]). Yet, this ecosystem is in global decline as anthropogenic environmental change impedes the role of cnidarians as key ecosystem engineers (Knowlton, [@B30]; Wild et al., [@B66]). Mass bleaching events, i.e. the disruption of the cnidarian---*Symbiodinium* symbiosis signified by the expulsion of symbionts and physical whitening of corals on broad scales, are among the dominant drivers of this decline (Bellwood et al., [@B8]; Hughes et al., [@B29]). Understanding the causes of this symbiotic breakdown requires considering these symbiotic organisms as holobionts: complex metaorganisms that arise from the interactions of the hosts and their associated microorganisms such as protists, bacteria, and archaea (Rosenberg et al., [@B54]). A crucial attribute of cnidarian holobionts is the ability to assimilate and recycle nutrients (Suggett et al., [@B57]). In particular, nitrogen cycling appears to be key to the functioning of these holobionts (Rädecker et al., [@B50]; Pogoreutz et al., [@B49]), since the growth of *Symbiodinium* is nitrogen-limited in a stable symbiosis (Muscatine et al., [@B43]; Belda et al., [@B6]; Falkowski et al., [@B16]; Rädecker et al., [@B50]; Aranda et al., [@B3]). Nitrogen limitation might stabilize symbiont populations and facilitate the translocation of photosynthates to the host (Ezzat et al., [@B15]), a process providing most of the energy required for the host\'s metabolism (Falkowski et al., [@B17]; Tremblay et al., [@B61]). Despite the importance of disentangling the individual contribution of host and symbionts to holobiont nutrient cycling (Yellowlees et al., [@B70]; Starzak et al., [@B56]; Leal et al., [@B34]; Rädecker et al., [@B50]), studying these processes in scleractinian corals has proven difficult due to the complex and interwoven nature of the coral holobiont. As most corals are associated with a diverse *Symbiodinium* community and are difficult to maintain in a symbiont-free stage (Baker, [@B4]; Wang et al., [@B63]), identifying underlying processes within these symbiotic interactions is challenging. In contrast, the emerging model organism Aiptasia (*sensu Exaiptasia pallida;* Grajales and Rodríguez, [@B22]) promises to be an easy and cost-effective tool to study cnidarian---*Symbiodinium* interactions. While this sea anemone differs from scleractinian corals in some key functional traits, most notably the lack of a calcareous skeleton, it features distinct advantages for the study of cnidarian---*Symbiodinium* symbioses (Voolstra, [@B62]; Baumgarten et al., [@B5]; Röthig et al., [@B55]): (I) it can be reared in clonal lines, enabling the study of processes in the absence of biological variation (Weis et al., [@B65]); (II) animals can be easily maintained in a symbiont-free stage, allowing the study of host processes in the absence of symbionts (Voolstra, [@B62]); (III) symbiont-free Aiptasia can be re-infected with specific symbiont strains, enabling the comparison of different symbionts (including those commonly associated with corals) in the same host background *in hospite* (Wolfowicz et al., [@B67]); (IV) natural populations of Aiptasia can be found in a range of environmental conditions and in association with different symbionts, offering a natural laboratory to study adaptation and coevolution in this symbiosis (Thornhill et al., [@B59]; Voolstra, [@B62]); (V) an extensive array of genetic resources is available for Aiptasia, allowing to link genetic and physiological traits (Baumgarten et al., [@B5]). These distinct advantages will prove especially powerful to study metabolic interactions between host and symbionts when combined with state of the art imaging techniques such as nano-scale secondary ion mass spectrometry (NanoSIMS). Coupled with stable isotope labeling, this technology enables imaging of metabolic processes at subcellular resolution and consequently quantification of nutrient assimilation at the single-cell level for each symbiotic partner (Kopp et al., [@B33]; Pernice et al., [@B47]). NanoSIMS has opened doors to an unprecedented level of information across all fields of biology and has previously been successfully applied to corals (Lechene et al., [@B35]; Pernice et al., [@B48], [@B47]; Kopp et al., [@B33]; Musat et al., [@B41]). In this study, using the combined advantages of the Aiptasia model system and high resolution NanoSIMS, we sought to investigate the relative contribution of cnidarian host identity and associated *Symbiodinium* type to assimilate dissolved inorganic nitrogen and carbon both at the organismal and at the cellular level. By doing this, we aim to promote the use of Aiptasia as a model for the study of metabolic interactions in the cnidarian-*Symbiodinium* symbiosis. Materials and methods {#s2} ===================== Maintenance of Aiptasia ----------------------- Four different host--symbiont pairings were maintained in separate batches. These combinations involved two different host clonal lines \[CC7 (Sunagawa et al., [@B58]) and H2 (Xiang et al., [@B69])\] as well as two different symbiont populations (A4 and B1 dominated; Grawunder et al., [@B23]). While CC7 Aiptasia can form stable associations with a diversity of *Symbiodinium* types, H2 Aiptasia show high fidelity to their native *Symbiodinium* community suggesting a higher selectivity and/or specificity with their symbionts (Thornhill et al., [@B59]). This specificity of H2 Aiptasia hinders reinfection with other symbionts thereby preventing a full factorial design in this study. Nevertheless, these host clonal lines provide an ideal basis for the comparison of symbiont diversity and specificity. To allow comparison of symbiont types within the same host line and to compare performance of the same symbiont type within different host lines, CC7 Aiptasia were bleached and reinfected with type B1 (strain SSBO1) symbionts, previously isolated from H2 Aiptasia. For this, aposymbiotic CC7 Aiptasia were generated and reinfected as described by Baumgarten et al. ([@B5]). In brief, animals were repeatedly bleached by incubation in 4°C sterile seawater for 4 h, followed by 1--2 days at 25°C in sterile seawater containing the photosynthesis inhibitor diuron. Aposymbiotic animals were maintained for at least 1 month prior to reinfection to confirm absence of residual symbionts. For reinfection, aposymbiotic animals were subjected to three cycles of incubation for 1 day in sterile seawater containing 10^5^ *Symbiodinium* cells mL^−1^ followed by *Artemia salina* nauplii feeding the next day. Thus, the four combinations were: aposymbiotic CC7 Aiptasia, CC7 Aiptasia with its native A4 symbionts; CC7 Aiptasia reinfected with B1 symbionts and H2 Aiptasia with its native B1 symbionts (Figures [1A--D](#F1){ref-type="fig"}). Animals were reared in autoclaved seawater (35 PSU, 25°C, \~80 μmol photons m^−2^ s^−1^ on a 12:12 h light:dark schedule) and fed with freshly hatched *A. salina* nauplii three times per week. Notably, while these light levels are low compared to shallow coral reef environments, they were chosen to support optimal growth of animals and are in the range of previous studies working with Aiptasia (Muller-Parker, [@B40]; Lehnert et al., [@B36]; Hillyer et al., [@B28]). Animal cultures were propagated under these conditions for more than 1 year to ensure anemones recovered from bleaching and reinfection procedures and to confirm the stability of native and introduced symbiotic associations. Stability of *Symbiodinium* communities was monitored using qPCR as outlined by Correa et al. (Correa et al., [@B12]). Any feeding was abandoned 3 days prior to measurements to exclude potential confounding effects. Thereby this experimental design allowed us to disentangle the contribution of host and symbionts to holobiont nutrient cycling in three comparisons: (I) between different symbionts within the same host line, (II) between different hosts lines with the same symbiont, and (III) between symbiotic and aposymbiotic states within the same host line. {#F1} Oxygen flux measurements ------------------------ Net photosynthesis and respiration rates were measured via oxygen (O~2~) evolution and consumption measurements during light and dark incubations, respectively. For this purpose, four specimens of each host--symbiont combination were transferred into 25 ml glass chambers filled with sterile seawater. Specimens were left to settle for 30 min in the dark, before magnetic stirrers were turned on to prevent stratification of the water column. Subsequently, O~2~ concentrations were recorded once per second over the course of 30 min incubations in the light (\~80 μmol photons m^−2^ s^−1^, 25°C) and dark (\<1 μmol photons m^−2^ s^−1^, 25°C) using FireSting O2 optical oxygen meters (PyroScience, Germany). Following incubation all specimens were immediately flash frozen and stored at −20°C until further analysis. Net photosynthesis (inferred from light incubations) as well as respiration (inferred from dark incubations) rates were corrected for seawater controls and normalized to total protein content and *Symbiodinium* densities of specimens. O~2~ fluxes of net photosynthesis and respiration rates were transformed into their carbon equivalents using the photosynthetic and respiration quotients of 1.1. and 0.9 as proposed by Muscatine et al. ([@B44]). Gross photosynthesis rates (expressed as pmol C symbiont cell^−1^ h^−1^ and μmol C mg host protein^−1^ h^−1^, respectively) were calculated according to: g r o s s p h o t o s y n t h e s i s r a t e = n e t p h o t o s y n t h e s i s r a t e \+ \| r e s p i r a t i o n r a t e \| . Quantification of $\text{NH}_{4}^{+}$ uptake and release -------------------------------------------------------- Net ammonium ($\text{NH}_{4}^{+}$) uptake rates were assessed at the holobiont level during light (\~80 μmol photons m^−2^ s^−1^, 25°C) and dark (\<1 μmol photons m^−2^ s^−1^, 25°C) conditions using the depletion technique (Godinot et al., [@B20]). Four specimens of each host--symbiont combination were incubated for 60 min in 25 ml chambers filled with $\text{NH}_{4}^{+}$-enriched artificial seawater (ASW) with a final concentration of 5 μM (Harrison et al., [@B25]). Ten milliliters water samples were collected before and after the incubation, filtered (45 μm) and immediately analyzed for $\text{NH}_{4}^{+}$ concentrations using an autoanalyzer (SA3000/5000 Chemistry Unit, SKALAR, Netherlands). Differences in $\text{NH}_{4}^{+}$ concentrations were corrected for seawater controls and normalized to incubation time, total host protein content and *Symbiodinium* densities of specimens to obtain net uptake rates during both light and dark incubations. Protein content, *Symbiodinium* density, and chlorophyll concentrations ----------------------------------------------------------------------- Frozen specimens were defrosted in 500 μl sterile saline water and homogenized using a Micro DisTec Homogenizer 125 (Kinematica, Switzerland). An aliquot of the homogenate was immediately analyzed for total protein content as well as symbiont concentrations, respectively. For total host protein content, *Symbiodinium* cells were removed by brief centrifugation and the supernatant was analyzed with the Micro BCA Protein Assay Kit (Thermo Scientific, USA) using 150 μl of 15x diluted tissue slurry as per manufacturer instructions. Likewise, *Symbiodinium* density was quantified by flow cytometry (BD LSRFortessa, BD Biosciences, USA) using 100 μl of strained tissue slurry. Cells were excited at a wavelength of 488 nm and fluorescence emission was recorded at 695/40 nm. *Symbiodinium* cell densities were quantified in triplicate measurements (20 μl each) based on forward-scattered light and chlorophyll autofluorescence signals of recorded events. Isotope labeling and sample preparation --------------------------------------- To corroborate nitrogen and carbon assimilation rates on the holobiont level, an isotopic labeling experiment was conducted for subsequent nanoscale secondary ion mass spectrometry (NanoSIMS) analysis. Individual specimens of each host--symbiont combination were incubated for 24 h (12:12 h light dark cycle) in 25 ml incubation chambers containing ASW. For isotopic enrichment, freshly prepared ASW, essentially free from bicarbonate and ammonium, was supplemented with NaH^13^CO~3~ (isotopic abundance of 99%) as well as ^15^NH~4~Cl (isotopic abundance of 99%) at a final concentration of 2 mM and 5 μM, respectively (adapted from Harrison et al., [@B25]). Following incubation, all specimens were immediately transferred to a fixative solution (2.5% glutaraldehyde, 1 M cacodylate) and stored at 4°C until further processing (within 14 days). Individual tentacles were collected from each anemone under a stereomicroscope for further sample preparation adapted after Pernice et al. ([@B48]) and Kopp et al. ([@B32]). First, samples were post-fixed for 1 h at RT in 1% OsO~4~ on Sörensen phosphate buffer (0.1 M). Samples were dehydrated in a series of increasing ethanol concentrations (50, 70, 90, 100%) followed by 100% acetone. Tissues were then gradually infiltrated with SPURR resin of increasing concentrations (25, 50, 75, 100%). Subsequently, tissues were embedded in SPURR resin and cut into 100 nm sections using an Ultracut E microtome (Leica Microsystems, Germany) and mounted on finder grids for Transmission Electron Microscopy (ProsciTech, Australia). NanoSIMS imaging ---------------- Gold-coated sections were imaged with the NanoSIMS 50 ion probe at the Center for Microscopy, Characterisation and Analysis at the University of Western Australia. Surfaces of samples were bombarded with a 16 keV primary Cs^+^ beam focused to a spot size of about 100 nm, with a current of \~2 pA. Secondary molecular ions ^12^C^12^C^−^, ^12^C^13^C^−^, ^12^C^14^N-, and ^12^C^15^N^−^ were simultaneously collected in electron multipliers at a mass resolution (M/ΔM) of about 8,000, enough to resolve the ^12^C^13^C^−^ from the ^12^C~2~^1^H^−^ peak and the ^13^C^14^N^−^ and ^12^C^15^N^−^ peaks from one another. Charge compensation was not necessary. Five images of different areas within the gastrodermis of the tentacle (25--45 μm raster with 256 × 256 pixels) were recorded for all targeted secondary molecular ions by rastering the primary beam across the sample with a dwell-time of 10--20 ms per pixel. After drift correction, the ^13^C/^12^C or ^15^N/^14^N maps were expressed as a hue-saturation-intensity image (HSI), where the color scale represents the isotope ratio. Image processing was performed using the ImageJ plugin OpenMIMS (National Resource for Imaging Mass Spectrometry, <https://github.com/BWHCNI/OpenMIMS/wiki>). Enrichment of the isotope labels was quantified for 20 Regions of Interest (ROIs) (circles of 2--10 μm) per category (symbiont cells, gastrodermal host tissue and gastrodermal vesicles) for each host--symbiont combination, and expressed using δ^13^C and δ^15^N notation. Gastrodermal host tissue was quantified in the form of ROIs placed adjacent to symbiont cells as clear cell boundaries were not always identifiable. Unlabeled Aiptasia served as unlabeled controls. δ^13^C and δ^15^N enrichment (expressed in %0) was quantified as follows: δ 13 C = ( ( C s a m p l e C u n l a b e l l e d ) \- 1 ) × 1 0 3  and , δ 15 N = ( ( N s a m p l e N u n l a b e l l e d ) \- 1 ) × 1 0 3 , where N is the ^15^N/^14^N ratio of sample or unlabeled control and C is the ^13^C/^12^C ratio (measured as ^12^C^13^C^−^/^12^C^12^C^−^ ions) of sample or unlabeled control, respectively. In this context, it is important to note that carbon and nitrogen incorporation at the cellular level was likely underestimated in our study as sample preparation for NanoSIMS may result in partial extraction of biomolecules. Statistical analysis -------------------- All statistical analyses were conducted with R version 3.2.5 (R Development Core Team, [@B53]). Data were tested for normal distribution using the Shapiro-Wilk test. All measurements at the holobiont level (symbiont densities, gross photosynthesis, respiration, net $\text{NH}_{4}^{+}$ uptake) followed normal distribution and were analyzed with a one-way analysis of variance (ANOVA) using host-symbiont combination as explanatory variable; only gross photosynthesis rates normalized by symbiont density were right-skewed and did not follow a normal distribution and hence were analyzed with a generalized linear model (GLM) using host-symbiont combination as the explanatory variable. Similarly, δ^13^C and δ^15^N enrichment data did not follow a normal distribution and were analyzed in two-factorial GLMs using additive as well as interactive effects of host-symbiont combination as well as holobiont compartment (host, lipid body, symbiont). All GLMs were fitted with Gamma distribution and "log" function to optimize the fit of the model. Fit of model residuals were confirmed using the qqPlot() function as implemented in the "car" package for R (Fox and Weisberg, [@B18]). An overview of replication and model results is provided in the Supplementary Information Table [S1](#SM1){ref-type="supplementary-material"}. Adjustment for multiple comparisons between host---symbiont combinations and holobiont compartments was done following the Bonferroni procedure. Significant differences identified via the post hoc comparison are indicated in the figures as different letters above bars. Results {#s3} ======= Symbiont densities ------------------ Aiptasia of the clonal line CC7 with their native symbiont community (*Symbiodinium* type A4) contained a significantly lower density of symbionts when normalized to host protein content compared to the other two symbiotic host--symbiont combinations (Figure [1E](#F1){ref-type="fig"}, see Supplementary Information Table [S1](#SM1){ref-type="supplementary-material"} for an overview of statistical model results). Notably, symbiont densities in Aiptasia were of the same order of magnitude as previously reported for scleractinian corals (Cunning and Baker, [@B13]; Ziegler et al., [@B71]). As expected, no symbionts were detected in aposymbiotic Aiptasia. Carbon assimilation and translocation ------------------------------------- Host--symbiont combinations of Aiptasia showed distinct differences in carbon fixation both at the holobiont (Figure [2](#F2){ref-type="fig"}) as well as at the cellular level (Figure [3](#F3){ref-type="fig"}). While fixation rates were highly variable between the three combinations of symbiotic Aiptasia, no carbon fixation was detectable in aposymbiotic Aiptasia, confirming that carbon assimilation was photosynthetically driven. At the holobiont level, gross photosynthesis was highest in Aiptasia of the clonal line CC7 with their native Clade A symbionts after normalization to symbiont cells (Figure [1A](#F1){ref-type="fig"}) or host protein content (Figure [1E](#F1){ref-type="fig"}). In contrast, CC7 Aiptasia symbiotic with Clade B (type B1, SSBO1) *Symbiodinium* showed the lowest gross photosynthesis rates of all symbiotic Aiptasia combinations. In particular, rates were lower than H2 Aiptasia hosting the same type B1 dominated symbiont community. Photosynthetic carbon fixation was more than three-fold higher than dark respiratory carbon consumption in all symbiotic Aiptasia groupings. Overall, dark respiration rates largely followed patterns of gross photosynthesis rates when normalized to *Symbiodinium* content, with CC7 Aiptasia symbiotic with type A4 having higher respiration rates than the other two symbiotic Aiptasia combinations hosting type B1 *Symbiodinium*. In contrast, no significant differences in respiration rates were detectable between the fours host---symbiont combinations when normalized to host protein content (Figures [2B,F](#F2){ref-type="fig"}). {#F2} {#F3} Isotope labeling and NanoSIMS imaging revealed that the observed differences in carbon fixation at the holobiont level translated into an intricate picture at the cellular level (Figures [3A--H](#F3){ref-type="fig"}). First, δ^13^C enrichment was evident in both host and symbiont cells in all symbiotic Aiptasia groupings (Figures [3B--D](#F3){ref-type="fig"}). Second, although enrichment was highest in *Symbiodinium* cells, localized regions of \<5 μm diameter in the host tissue (referred to as "lipid bodies" from this point on) also showed significantly higher rates of enrichment compared to the surrounding host tissue. Third, Clade B *Symbiodinium* showed no differences in ^13^C-incorporation depending on the host, and incorporation rates were 30--40% lower than in Clade A symbionts. Host lipid bodies, on the contrary, showed a reversed picture with Clade B associated H2 Aiptasia having the highest and Clade B associated CC7 Aiptasia having the lowest ^13^C assimilation rates, despite harboring the same symbiont types. $\text{NH}_{4}^{+}$ assimilation and release -------------------------------------------- Similar to carbon fixation, strong differences in $\text{NH}_{4}^{+}$ assimilation were evident between the experimental groups of Aiptasia at both holobiont and cellular levels. At the holobiont level, all four host---symbiont combinations showed higher $\text{NH}_{4}^{+}$ uptake/release rates during the light (Figures [2C,G](#F2){ref-type="fig"}), compared to dark conditions (Figures [2D,H](#F2){ref-type="fig"}). When normalized to host protein content, aposymbiotic Aiptasia showed the highest net release of $\text{NH}_{4}^{+}$ at the holobiont level both during light (Figure [2G](#F2){ref-type="fig"}) and dark incubations (Figure [2H](#F2){ref-type="fig"}). Albeit significantly lower, symbiotic H2 Aiptasia also had a net release of $\text{NH}_{4}^{+}$ into the surrounding seawater during the light incubations, yet took up $\text{NH}_{4}^{+}$ during dark incubations. In contrast, both groups of symbiotic CC7 Aiptasia showed a net uptake of $\text{NH}_{4}^{+}$ by the holobiont during both light and dark conditions. Further, the uptake rate was affected by the associated symbiont community, with Clade A dominated CC7 holobionts taking up more $\text{NH}_{4}^{+}$ than their Clade B infected counterparts (Figures [2C,G](#F2){ref-type="fig"}). Although $\text{NH}_{4}^{+}$ assimilation ranged from net uptake to net release in the different experimental groups, NanoSIMS imaging confirmed that all four host--symbiont combinations incorporated ^15^N into their cells (Figures [3I--P](#F3){ref-type="fig"}). While δ^15^N signatures were highest in *Symbiodinium* cells, ^15^N assimilation was also observed within the cnidarian host tissue including that of aposymbiotic Aiptasia. Similar to δ^13^C patterns, δ^15^N enrichment in *Symbiodinium* cells aligned with algal symbiont type rather than host identity, and Clade B symbionts showed lower rates of incorporation than Clade A. Conversely, ^15^N incorporation into host cells was not significantly different between symbiotic Aiptasia groupings, irrespective of their symbiont type. Aposymbiotic CC7 Aiptasia had the lowest overall ^15^N incorporation into their tissue, yet showed small (\<5 μm in diameter) and localized regions of high enrichment. In contrast, the afore mentioned lipid bodies of high δ^13^C-enrichment showed consistently lower δ^15^N-signatures than surrounding host tissue in all three symbiotic Aiptasia strains. Discussion {#s4} ========== Aiptasia has proven to be a powerful emerging tool for the genetic and molecular study of the cnidarian---alga symbiosis (Sunagawa et al., [@B58]; Thornhill et al., [@B59]; Voolstra, [@B62]; Baumgarten et al., [@B5]; Bellis et al., [@B7]; Dani et al., [@B14]; Matthews et al., [@B38]). Beyond these realms, only a few studies have begun to exploit the advantages Aiptasia has to offer (Tolleter et al., [@B60]; Starzak et al., [@B56]; Leal et al., [@B34]; Biquand et al., [@B9]; Gegner et al., [@B19]). Here, we set out to showcase the use of Aiptasia as a model to study nutrient cycling in the cnidarian---alga symbiosis. While NanoSIMS has been successfully used previously to study nutrient uptake in corals (Ceh et al., [@B10]; Kopp et al., [@B32]; Lema et al., [@B37]), the flexibility of the Aiptasia model enabled us to decouple the relative contribution of host and symbionts to nutrient cycling. We could identify differences in nutrient assimilation across different host--symbiont associations, both at the holobiont as well as the cellular level. Yet, only the integration of both levels of biological organization allowed us to comprehensively disentangle some of the intricacies of nutrient cycling in the Aiptasia holobiont. Carbon cycling in Aiptasia -------------------------- All three groups of symbiotic Aiptasia showed high rates of gross photosynthesis that exceeded their respiratory carbon requirements thereby supporting net productivity of the holobiont required for stable symbiotic associations (Muscatine et al., [@B44]; Muller-Parker, [@B40]). Yet, differences in gross photosynthesis between host--symbiont combinations were evident at the holobiont level. Gross photosynthesis rates differed between the same host infected with different algal symbionts and between different hosts infected with the same algal symbionts. Thereby our results support the findings by Starzak et al. ([@B56]) who reported differences in carbon flux depending on symbiont type and between heterologous and homologous symbionts in Aiptasia, confirming previous observations that carbon fixation depends on the interaction of both host and symbionts (Goulet et al., [@B21]; Pernice et al., [@B47]; Starzak et al., [@B56]; Leal et al., [@B34]). At the cellular level, we observed particular areas of δ^13^C enrichment (hotspots) in the host tissue similar to previous observations (Pernice et al., [@B47]; Kopp et al., [@B31]). This high δ^13^C enrichment is further coupled with lower δ^15^N enrichment, suggesting that these hotspots likely constitute a form of carbon storage compartments in the host tissue. Based on shape, size, and location in the tissue, these compartments are most likely lipid bodies (Peng et al., [@B46]). These cellular organelles are abundant in symbiotic cnidarians as they allow for short-term carbon storage and remobilization depending on cellular carbon availability (Chen et al., [@B11]). Hence, amount, size and enrichment of these lipid bodies may be an excellent proxy to assess the amount of carbon translocated by *Symbiodinium* to the host, but further studies are needed to unequivocally determine their nature. Lipid body enrichment in the host was highest in H2 Aiptasia and lowest in CC7 Aiptasia, both associated with *Symbiodinium* type B1. Yet, δ^13^C enrichment in algal cells was unaffected by host identity. At the same time, our results revealed that Clade A and B symbionts had distinctly different δ^13^C enrichment, even in the same clonal Aiptasia host line. These differences are likely the consequence of differential metabolic requirements by the specific symbionts. Thus, δ^13^C enrichment may be a powerful tool to differentiate between symbiont types *in hospite*. Taken together, observed differences in gross carbon fixation at the holobiont level were reflected in the combined δ^13^C enrichment (host tissue + lipid bodies + symbionts) at the cellular level. However, NanoSIMS data revealed that the differences in carbon fixation at the holobiont level were not evenly reflected across all compartments at the cellular level. δ^13^C enrichment in algal cells differed depending on symbiont type (i.e., showed stable δ^13^C enrichment within the same symbiont type), but was unaffected by host identity. In contrast, patterns of carbon fixation rates were directly reflected in the enrichment of the host lipid bodies and host-dependent. Given that both host lineages showed similar respiration rates and symbiont densities when infected with the same symbiont, their differences in host δ^13^C enrichment imply differences in carbon translocation rates. Thereby, this differential enrichment pattern between host and symbionts may have important implications for our understanding of symbiosis functioning. The fact that δ^13^C enrichment in algal cells differed only depending on symbiont type but was unaffected by host identity implies that symbionts retained the same amount of fixed carbon regardless of overall fixed carbon availability. Hence, only excess carbon, not consumed by algal metabolism, appears to be available for translocation to the host. Therefore, factors reducing the availability of excess carbon in the symbiont, may potentially deprive the host of its main energy source, despite harboring viable symbionts in its tissue. This "selfish" aspect of the symbiosis may pose a potential threat to the stability of the holobiont under conditions of reduced fixed-carbon availability, such as those imposed by environmental stress (Anthony et al., [@B2], [@B1]). Nitrogen cycling in Aiptasia ---------------------------- The observation of drastically different carbon fixation and translocation rates between different host--symbiont combinations raises questions regarding the underlying regulatory mechanisms of carbon cycling within these symbioses (Suggett et al., [@B57]). Importantly, nitrogen availability *in hospite* has been proposed to be among the environmental controls of these processes (Wooldridge, [@B68]; Ezzat et al., [@B15]; Rädecker et al., [@B50]; Pogoreutz et al., [@B49]). Indeed, drastic differences in nitrogen assimilation became evident when comparing different host--symbiont combinations. Strikingly, the two different host lines Aiptasia H2 and CC7 showed net $\text{NH}_{4}^{+}$ release and $\text{NH}_{4}^{+}$ uptake during the light, respectively, even when hosting the same algal symbionts. These findings suggest that the *in hospite* nutrient availability for the symbiont may be drastically different depending on the associated Aiptasia host. Hence, differences in gross photosynthetic activity and translocation may be partly attributed to variations in availability of nitrogen derived from the host metabolism. Interestingly, while CC7 Aiptasia showed light-enhanced $\text{NH}_{4}^{+}$ uptake as previously reported for corals (Grover et al., [@B24]), H2 Aiptasia showed a net release of $\text{NH}_{4}^{+}$ during the light, contrasted by slight uptake during the dark. While we cannot explain the discrepancy at this point, it may reflect differences in internal nitrogen requirement, response times to increased nitrogen availability, and/or uptake efficiency. These differences illustrate the drastic effects of host identity on nitrogen assimilation of the holobiont. At any rate, our results highlight the functional diversity and specificity of cnidarian---*Symbiodinium* symbioses, prompting research across a range of host--symbiont combinations. In agreement with previous studies, δ^15^N enrichment was highest in *Symbiodinium* cells reflecting their efficient nutrient uptake capacity (Kopp et al., [@B33]; Pernice et al., [@B47]; Aranda et al., [@B3]). However, biomass of the host largely exceeds that of their symbionts. Hence, anemone hosts likely accounted for a large fraction of the nitrogen assimilation in the holobiont, despite having a lower δ^15^N enrichment. At this point it is not possible to distinguish whether the increased δ^15^N enrichment of host tissues in symbiotic animals are due to direct $\text{NH}_{4}^{+}$ fixation by the host or the translocation of fixed nitrogen by the symbiont. However, nitrogen assimilation was observed even in the absence of algal symbionts, as evidenced by aposymbiotic Aiptasia. Although these animals showed a high net release of $\text{NH}_{4}^{+}$ at the holobiont level, NanoSIMS imaging confirmed the incorporation of ^15^N within localized hotspots of their tissue at low rates. While the exact nature of these hotspots remains unknown at this point, our results confirm that Aiptasia also has the ability to assimilate inorganic nitrogen from seawater as previously reported for corals (Pernice et al., [@B48]). However, it remains to be determined whether this capability is intricate to the host cellular machinery or a function of associated bacterial symbionts or both (Ceh et al., [@B10]). In contrast to δ^15^N enrichment of their hosts, *Symbiodinium* types showed characteristic δ^15^N enrichment patterns regardless of the identity of the host. Hence, δ^15^N enrichment may prove a useful tool to identify symbiont identity *in hospite*, especially when combined with δ^13^C measurements. Different to carbon fixation measurement, patterns of $\text{NH}_{4}^{+}$ uptake on the holobiont level were not directly reflected in the overall δ^15^N enrichment at the cellular level. Specifically, symbiont-free CC7 Aiptasia as well as symbiotic H2 Aiptasia showed net release of $\text{NH}_{4}^{+}$ from the holobiont during light conditions, yet NanoSIMS analysis confirmed the incorporation of ^15^N from surrounding seawater. While these differences may be partly attributed to differences in incubation time and light availability for the two measurements, they further suggest that uptake and release of $\text{NH}_{4}^{+}$ appear to be in a dynamic equilibrium in Aiptasia. Hence, the stable δ^15^N enrichment of the same symbiont type in CC7 and H2 suggests that the contribution of nitrogen derived from host metabolism was negligible compared to the incorporation of nitrogen from seawater under these conditions. Under natural oligotrophic conditions, however, host metabolism may make a significant contribution to the nitrogen supply of the symbiont. Deciphering the role of nutrient cycling in cnidarian holobionts ---------------------------------------------------------------- Our results show (I) that nutrient cycling is drastically altered between symbiotic and aposymbiotic Aiptasia; (II) that different *Symbiodinium* types possess different metabolic capabilities within the same Aiptasia strain and (III) that different Aiptasia strains affect the metabolic performance of the same algal symbiont. Taken together, our findings support the idea of a symbiosis in which partners directly compete for available nutrients and only excess nutrients are exchanged. Consequently, this inherent instability may render this symbiosis highly susceptible to environmental change. Noteworthy, the observed levels and patterns of nutrient assimilation show strong similarities with those previously reported in corals (Grover et al., [@B24]; Pernice et al., [@B48]; Kopp et al., [@B32]), thereby supporting the suitability of Aiptasia as a model for the study of the coral---*Symbiodinium* symbiosis. Although our results require further validation with regard to their wider applicability beyond the Aiptasia model system, our findings showcase the distinct advantages of a model system approach for the study of nutrient cycling in the cnidarian---*Symbiodinium* symbiosis. Nevertheless, questions remain regarding the precise nature of nutrients exchanged in this symbiosis and the underlying processes involved. In particular, future studies should focus on the role of carbon translocation in establishing and maintaining this symbiosis to decipher the intricacies of this symbiosis (Mies et al., [@B39]). The methodological approach outlined in this study offers a powerful toolset to address such questions. Although optimized to trace carbon and nitrogen assimilation within coral or Aiptasia holobionts (Pernice et al., [@B48]; Kopp et al., [@B32]), NanoSIMS can be easily modified depending on the experimental requirements. Specific labeled compounds can also be used as tracers to follow the translocation and uptake of specific molecules in complex systems by coupling the spatial resolution of NanoSIMS with the molecular characterization afforded by time-of-flight secondary ion mass spectrometry (ToF-SIMS) (Raina et al., [@B52]). As shown here, detailed cellular insights gained from NanoSIMS will prove most powerful when integrated with traditional holobiont based measurements to identify the complexity of processes. Future research efforts combining a model system approach with field-based coral studies, will help to transform our understanding of the mechanisms underlying this symbiosis and may prompt new solutions to prevent further loss and degradation of reef ecosystems. Preprint availability {#s5} ===================== The original version of the manuscript has been deposited on the bioarxiv.org preprint server (manuscript id: 223933). The manuscript is available under <https://www.biorxiv.org/content/early/2017/11/22/223933>. Author contributions {#s6} ==================== NR, MA, and CV: conceived and designed the experiment; NR, J-BR, MP, and GP: conducted the experiment; PG and MK: carried out NanoSIMS data acquisition. All authors wrote, revised and approved the manuscript. Conflict of interest statement ------------------------------ The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest. The authors would like to thank Dr. Rachid Sougrat and Ptissam Bergam from the KAUST imaging core lab for their help with sample preparation. CV and NR acknowledge funding from the KAUST CPF funding. Further, research in this publication was supported by KAUST baseline research funds to CV. The authors would like to acknowledge the Australian Microscopy & Microanalysis Research Facility, AuScope, the Science and Industry Endowment Fund, and the State Government of Western Australian for contributing to the Ion Probe Facility at the Centre for Microscopy, Characterisation and Analysis at the University of Western Australia. J-BR was supported by Australian Research Council fellowship DE160100636. Supplementary material {#s7} ====================== The Supplementary Material for this article can be found online at: <https://www.frontiersin.org/articles/10.3389/fphys.2018.00214/full#supplementary-material> ###### Click here for additional data file. [^1]: Edited by: Zhijun Yu, Hebei Normal University, China [^2]: Reviewed by: Noga Stambler, Bar-Ilan University, Israel; Jiannan Liu, Hebei University of Engineering, China [^3]: This article was submitted to Invertebrate Physiology, a section of the journal Frontiers in Physiology

Background {#Sec1} ========== The prevalence of chronic diseases is globally on the rise, with cardiovascular diseases, respiratory disease, diabetes, cancer, and other chronic illnesses being major contributors to disability \[[@CR1],[@CR2]\]. In Canada, two out of five people have at least one chronic disease. Chronic disease is a major driver of health care expenditure, reaching approximately \$68 billion in Canada in 2010 \[[@CR3]\]. The current health care system is oriented towards episodic and acute care, making it unprepared to address the multi-faceted and complex needs of those with chronic diseases \[[@CR4],[@CR5]\]. Given the need for continuity, comprehensiveness and coordination, primary care has been suggested as potentially playing a central role in effective management and integration of care \[[@CR6]\]. However, literature on current practice suggests that patients often receive inadequate care, with limited physician involvement in disease management, and little coordination and communication among care providers \[[@CR7]\]. In response to these challenges and the call for redesigning care delivery for chronic diseases, Wagner and colleagues developed the Chronic Care Model (CCM) \[[@CR8],[@CR9]\]. The CCM was developed to bridge the gap and translate knowledge between evidence-based chronic disease care and actual care practices. The framework which is centered in primary care, 'conceptualizes care as prepared practice teams in productive interactions with informed, activated patients' \[[@CR10]\]. It posits six interrelated elements that are key to high quality chronic disease care: self-management support, redesigning delivery systems, decision support that is system wide, clinical information technology, linkages to community resources, and health care system organization \[[@CR10],[@CR11]\]. The components seek organizational change at the systems' level, promoting and delivering care that is evidence-based through using clinical tools such as guidelines, utilizing information systems that improve patient data sharing across the organization and between providers, engaging and empowering patients in their care, and mobilizing community resources to meet patient needs \[[@CR11]\]. The CCM and its various components have been widely adopted and evaluated, with results showing that it improves patient care and clinical outcomes, and reduces care utilization and costs \[[@CR12]-[@CR16]\]. Despite the extensive evaluation of quality improvement (QI) initiatives, and research on CCM-based interventions, particularly across the United States, little is known about different primary care experiences with its implementation, and the factors that influence its successful uptake \[[@CR10],[@CR14],[@CR17],[@CR18]\]. The model provides no clear blueprint on how each component can be implemented in practice, and there is considerable heterogeneity in the types of interventions implemented in primary care in support of the CCM \[[@CR10]\]. Previous reviews synthesizing evidence on the CCM have focused on associated care changes, clinical outcomes and cost-effectiveness \[[@CR10],[@CR14],[@CR19]\]. Although a recent systematic review by Holm and Severinsson identified barriers and facilitators of successful CCM implementation in primary care, it was specific to depression management in the US \[[@CR20]\]. An understanding of the barriers and facilitators of implementing the CCM, in different care settings is important for several reasons. A barrier in this context is defined as any factor that hinders or impedes care change processes of CCM implementation. First, there are numerous contextual factors that enable organizational change and successful translation of evidence into practice \[[@CR21],[@CR22]\]. Some of the factors previously identified include: evidence fit and relevance to the organizational context, staff relationships and collaboration, availability of resources, strong and committed leadership, and a culture supportive of change \[[@CR22]-[@CR24]\]. Second, given the complex and multifaceted nature of the model, primary care organizations can face difficulties with its implementation \[[@CR12]\]. This is particularly the case given that there are no guidelines available on how to effectively operationalize CCM elements across different settings \[[@CR25]\]. We therefore aimed to identify and review evidence on the challenges and barriers encountered while implementing the CCM in primary care. Methods {#Sec2} ======= We conducted a systematic literature review to synthesize findings of studies that implemented the CCM in primary care, in order to identify facilitators and barriers encountered during implementation. Barriers and facilitators were interpreted using the Consolidated Framework for Implementation Research (CFIR) \[[@CR26]\]. As this research did not involve human subjects, we did not seek ethics clearance for the project. Data sources {#Sec3} ------------ This study identified English-language, peer-reviewed research articles, describing the CCM in primary care settings. Searches were performed in three data bases: Web of Knowledge, PubMed and Scopus. These databases include Medline, EMBASE and the National Library of Medicine. The PubMed and Scopus search strategy used the following MeSH terms to describe 'primary care': primary health care, general practice and family practice. Since there were no MeSH terms for Chronic Care Model, the term was put under quotation marks during the search. In order to ensure a comprehensive search that included all studies that implemented the CCM, MeSH terms for 'implementation' were not used in the search. This strategy was also used to avoid excluding studies that might not have identified the term in their titles and abstracts. Search terms and concepts were combined using the Boolean and Proximity operator 'OR', while concepts were combined using 'AND' and 'Near' (Table [1](#Tab1){ref-type="table"}).Table 1**Key words used in search strategiesConcept\*Relevant key words\*\***Primary Health CareCare, Primary Health; Health Care, Primary; Primary Care; Care, Primary; Primary Healthcare, Healthcare, PrimaryGeneral PracticeGeneral PracticeFamily PracticeFamily Practices; Practice, Family; Practices, FamilyChronic Care Model'Chronic Care Model'\*Concepts were combined using the Boolean & Proximity operators AND or NEAR (as databases allow) and the search terms within each concept were combined with OR.**\*\***Keywords were searched using truncation and phrase symbols when appropriate to ensure precise and comprehensive results. A second strategy adapted from Coleman and colleagues involved searching articles from Web of Knowledge Science Citation Index, which cited the five foundational CCM articles by Wagner and colleagues and Bodenheimer and colleagues \[[@CR8]-[@CR10],[@CR14],[@CR27],[@CR28]\]. In addition, hand searching of the reference lists in all articles that met the inclusion criteria outlined below was performed to identify any missed relevant articles. Search terms used in both search strategies are described in Table [1](#Tab1){ref-type="table"}. Study selection {#Sec4} --------------- Citations were downloaded and screened in Refworks, an online citation manager tool. Article abstracts and titles were read based on the exclusion and inclusion criteria detailed below. If the reviewer could not determine whether to exclude an article based on its abstracts and title, then it was retrieved for full text reading. Figure [1](#Fig1){ref-type="fig"} displays the process involved in study selection.Figure 1**Exclusion and inclusion criteria for article selection.** Exclusion criteria:Articles published before 2003 and in languages other than English; this year was chosen as the search cut-off to follow the publication date of the last CCM foundational paper by Bodenheimer and colleagues \[[@CR10]\], thus reflecting studies that implemented a more mature conceptualization of the modelArticles that solely described the CCM conceptually, i.e., did not report on an actual implementation of the model, commentaries and opinion pieces, study protocols, reviews including: systematic and narrative reviews, and meta-analysesThe target population of the study was not adults aged 18+ with chronic conditionsArticles arguing or providing the rationale for implementation of CCM in primary care, but which were not based on empirical studies. Inclusion criteria:Articles describing or evaluating the implementation of the CCM. Implementation had to refer to efforts which used change strategies to promote use of evidence-based practices or programs \[[@CR29]\]Implementation of the CCM had to occur in primary care, which is defined as integrated and accessible healthcare, delivered in the context of family and community \[[@CR30]\].Articles identifying barriers and/or facilitators of CCM implementation. Data abstraction {#Sec5} ---------------- The methods used for the study selection and data abstraction in this systematic review are aligned with those in the PRISMA statement. The PRISMA statement provides an evidence-based checklist intended to improve the standards of reporting in systematic reviews \[[@CR31]\]. Given that the focus was on implementation, rather than study outcomes, not all aspects of the PRISMA statement were adopted. Data abstraction involved two stages. First, articles were categorized by reference, study design and methods, participants and setting, study objective, CCM components used, and description of the intervention. The next stage of data abstraction involved qualitative analysis using the Consolidating Framework for Research Implementation (CFIR), which has five domains: intervention characteristics, outer setting, inner setting, characteristics of the individuals involved, and the process of implementation \[[@CR26]\]. It provides a conceptual framework which can be used to understand factors that influence successful implementation in health care, and is based on theories identified by Greenhalgh and colleagues' widely cited systematic review \[[@CR26],[@CR32]\]. The CFIR was selected because it includes multiple constructs and theories from peer reviewed studies on evidence-based knowledge dissemination and translation, organizational change and implementation, and uptake of research. It has also been suggested as a framework that can be used to guide the implementation of CCM components in interventions: therefore, it was deemed most appropriate for our study \[[@CR26]\]. Table [2](#Tab2){ref-type="table"} provides summarized descriptions of the CFIR domains.Table 2**Description of CFIR domains and constructs** \[[@CR26]\]**DomainDefinitionIntervention characteristics**The characteristics of the intervention being implemented include whether: the intervention is perceived to be developed external or internal to the organization, there is evidence supporting its effectiveness, and its implementation will be advantageous to its alternatives. Other characteristics include how the intervention is presented, its adaptability, complexity and whether it can be tested on a smaller scale.**Outer setting**The external context of the organization includes: patient needs and the ability to meet them, networks with other organizations, pressure to implement the intervention and external policies and incentives to adopt the intervention.**Inner setting**Features of the organization including its structural characteristics (such as size, age of the organization and division of labour), networks and communication (such as connections and information sharing between individuals, units and services), cultural norms and values, implementation climate, organizational capacity and readiness for change.**Characteristics of individuals**Staff knowledge and belief about the intervention, their ability to execute their respective aspects of the implementation, and their individual stage of change. Other characteristics include individual identification with the organization and other personal attributes.**Process**Active change process, the purpose of which is to promote uptake of the intervention by the organization. This is influenced by the level of planning prior to implementation, and engaging organization stakeholders through appointing implementation leaders and champions of the intervention. This includes the ability to execute the implementation of the intervention as planned and to continuously reflect on and evaluate the quality of implementation and intervention as it progresses. Using qualitative content analysis, implementation barriers and facilitators in 22 articles were mapped on to the CFIR framework. When articles described barriers or facilitators of CCM implementation, they were regarded as "attributive statements", which were coded under the appropriate constructs and domains. These statements were often found in the discussion and results section of the articles. If the statement was beyond the domains and constructs of the CFIR, then it was still documented. Our approach was modeled after the data abstraction method used in a systematic review by Mair and colleagues \[[@CR33]\]. The data abstraction and coding was performed by one reviewer. Interpretative and inductive reasoning were used to map out the attributive statements to the framework. Results {#Sec6} ======= Twenty two studies were included in this review. Study descriptions and methodological procedures were summarized in terms of design, measurements, sample size and context, as shown in Table [3](#Tab3){ref-type="table"}. In Table [4](#Tab4){ref-type="table"} statements reflecting implementation barriers and facilitators from each article were analyzed and coded to their respective domains and constructs under the CFIR framework.Table 3**Overview of studies on the CCM in primary careReference/LocationStudy design, methodsParticipants (n)/ study settingObjectiveCCMIntervention**\[[@CR34]\] **Mexico**Quantitative, pilot study, survey assessing chronic care delivery, and measurement of clinical outcomePrimary care teams (n = 10): physicians, nurses and other professionals were randomly selected and assigned to intervention or control groupEvaluate whether implementation of diabetes quality improvement (QI) project improved patient outcomesA, B, C, D, E, FImplementation of QI strategy for diabetes care based on three learning sessions, followed by Plan, Do, Study, Act (PDSA) practice\[[@CR35]\] **USA**Quantitative, pilot studyRegistered nurse, general internists and multi-morbid patients in an urban primary care practiceAssess feasibility of implementing the Guided Care ModelA, C, D, E, FGuided Care Nurse worked with two physicians to conduct geriatric evaluation, disease management and to coordinate care.\[[@CR36]\] **\[USA\]**Quantitative, nonrandomized-prospective clinical trial, survey measuring primary care experiencesOlder community patients (n = 150), Registered nurse, general internists (n = 4) in an urban primary care practiceEvaluate intervention to enhance the quality of primary care experiences in chronically ill older persons based on Guided Care modelA, C, D, E, FGuided Care Nurse provided geriatric assessment, a comprehensive care plan, proactive follow-up, coordination of care, and access to community resources\[[@CR19]\] **\[USA\]**Mixed methods study, triangulation of measured clinical processes and outcomes, provider surveys and semi-structured interviewsTeam leaders and members (n = 106) in 19 community health centres (CHC)s participating in diabetes QI collaborativeEvaluate whether the Diabetes Health Disparities Collaborative can improve the quality of care in CHCsA, B, C, D, E, FCHCs formed QIs teams which attended collaborative learning sessions and adapted QI plans using the PDSA design\[[@CR37]\] **USA**Quantitative study, self-administered questionnaires on CHC staffStaff (n = 622) of CHCs (n = 145) participating in QI initiativeAssess predictors of changes in staff morale and burnout at CHCs participating in Health Disparities CollaborativeA, B, C, D, E, FCHCs participated in quarterly regional or national learning sessions and developed QI teams which utilized the PDSA model\[[@CR38]\] **\[USA\]**Quantitative, matched control study, organizational survey, and measurement of care processCHCs (n = 19) in Health Disparities Cancer Collaboratives, and controls (n = 22) in underserved populationAssess whether CHCs in collaboratives were more likely to implement cancer care process changesA, B, C, D, E, FCHCs formed teams to learn how to implement change, facilitated by an expert faculty. Health centers reported and shared QI experiences during monthly teleconferences and three in-person learning sessions\[[@CR39]\] **USA**Qualitative study, semi-structured interviews, using grounded theory approachPrimary care physicians (n = 24) in multi/single specialty groups or single practicesExamine primary care physicians' views on obstacles to providing depression care and CCM-based interventionsA, B, C, D, EDepression screening, structured assessment, patient education, mental healthcare integration, consults and care management\[[@CR40]\] **USA**Qualitative study, semi-structured interviews, observational notesLeaders and front-line physicians and nurses (n = 53) in a large multispeciality health group (clinics, n = 5)Evaluate care changes and processes used to implement CCMA, B, C, D, E, FProject leaders and multidisciplinary teams were created to guide implementation, and individual care teams piloted the intervention\[[@CR41]\] **USA**Quantitative studyPhysicians (n = 17) and nurse practitioners (n = 5) in a metropolitan family practice clinicDescribe steps to successfully implement clinic-in-a-clinic diabetes self-management that uses PDSAA, B, C, D, E, FEducation, behaviour change support, goal setting and follow up provided by nurse practitioner to Type 2 diabetes patients who require more intensive counselling on diabetic self management issues\[[@CR42]\] **USA**Quantitative, quasi-experimental with concurrent non-randomized controls, measuring intermediate diabetes outcomesGeneral internists, nurse practitioners, pharmD, clinical health psychologist and nurses in a primary care clinic in a tertiary care academic medical centreEvaluate intermediate outcome measures of diabetic patients in shared medical appointments (SMA) in comparison to control patients.A, B, C, D, EUtilised diabetes registry to identify target patients. Provided decision support by practice guidelines and by including a diabetes specialist in the team. Multidisciplinary team provided didactic group education and individual learning in shared medical appointments\[[@CR43]\] **USA**Quantitative study, measuring patient participation and changes in diabetes related outcomesDiabetic patients (n = 275) in a CHC serving low-income LatinosAssess patient engagement in self management activities and changes in glycosylated hemoglobin (HbA1c).BImplementation of diabetes education classes, chronic self-management classes, weekly drop-in sessions, individual counseling, daily exercise classes and bilingual services\[[@CR44]\] **USA**Qualitative study, structured interview based on ecological systems theoryTeam leaders and members of CHCs collaborative (n = 14)Identify strategies that contributed to CHCs' successes and challenges in diabetes QIA, B, C, D, E, FCHCs assembled teams to participate in the collaborative. They were responsible for coordinating and reporting activities, and electronic registries. The CCM was implemented by a champion panel made of diabetic patients.\[[@CR45]\] **USA**Qualitative study, telephone interviewsManagers, mental health specialists and care managers in health care organizations (n = 5)To understand the experiences of project participants in implementing depression improvement model.A, B, C, D, ECare management, an improved interface between mental health consultants and primary care clinicians, and preparation of primary care clinicians and practices to provide systematic depression management\[[@CR46]\] **USA**Quantitative study, measured fidelity to and intensity of CCM implementationHealth care organizations (n = 42) part of QI collaboratives (n = 3)Measure organizations' implementation of CCM interventions for chronic care QIA, B, C, D, E, FHealth care organizations attended three learning sessions together to collaboratively improve performance and focus on implementing small rapid change cycles in their practices\[[@CR47]\] **USA**Quantitative studyCommunity based primary care physicians' offices.Evaluate the Assessing Care of Vulnerable Elderly Persons (ACOVE) intervention for adults with geriatric conditionsA, B, C, D, ECase finding, collection of condition-specific clinical data, medical record prompts to encourage performance of essential care processes, patient education and activation, and physician decision support and education\[[@CR18]\] **Canada**Quantitative study, survey questionnaire evaluating physician normative practices consistent CCMPhysicians (n = 195) in walk-in clinics (n = 29), solo family practices (n = 29), group family practices (n = 104), CHCs (n = 14) and primary care networks (n = 27)Examine implementation of CCM in different primary care practicesA, B, C, D, E, FN/A\[[@CR48]\] **USA**Quantitative studyDiabetic patients (n = 70) over 65 years old in a private medical clinicDetermine whether patients in shared medical appointment meet the American Diabetes Association standards in diabetes self-management educationA, C, DImplementation of a diabetes self management program using shared medical appointments\[[@CR49]\] **USA**Quantitative study, questionnaire measuring organization characteristics and care management processesAdministrative leaders of physician organizations (n = 957), including medical groups (n = 621), independent practice associations (n = 336) across the USExamine the relationship between measures of primary care orientation and the implementation of the CCMA, B, C, D, FN/A\[[@CR50]\] **Belgium**Mixed methods study, CCM implementation survey, analysis of meeting reportsGeneral practitioner (n = 83), dietician (n = 1), pharmacist (n = 46), podiatrist (n = 5) and nurses (n = 90) providing care to type 2 diabetes patients (n = 2300)Assess degree of implementation of CCM, and facilitators and barriers encounteredA, B, C, D, E, FDevelopment and implementation of education program for patients on diet or oral therapy, establishment of a local steering group, appointment of program manager, provider education and regional audit\[[@CR51]\] **Canada**Qualitative study, structured interview with staffHealth administrators, physician leaders, nurses and physicians (n = 12) in a large integrated academic institution.Examine strategies that promote physician involvement in planning and developing of heart failure care deliveryA, B, C, D, E, FDetailed analysis of existing heart failure management strategies, a review of best practice strategies and potential future best direction for increased effectiveness\[[@CR52]\] **Netherlands**Qualitative study, semi-structured interview of project managersProject directors and managers (n = 16), in health care provider groups (n = 5)Understand the development, implementation and execution of disease management programs by project leaders and cliniciansA, B, D, EImplementation of nation-wide disease management program in health organization in the Netherlands\[[@CR53]\] **\[USA\]**Qualitative, case study analysis using interviewsStaff and patients from disease-specific shared medical appointments groups (N = 3)To describe the roles of nurse practitioners in shared medical appointment group visitsA, B, C, D, E, FImplementation of nurse practitioners in shared medical appointmentsQuality improvement; QI, Chronic Care Model; CCM, Plan Do Study Act model; PDSA, Guided Care Nurse; GCN, Community Health Center; CHC; N/A; not available.CCM components.A = Delivery system redesign.B = Self management support.C = Decision support.D = Clinical information system.E = Health system organization.F = Community linkages.Table 4**Thematic analysis shows the barriers and facilitators identified by the studies mapped on to their corresponding CFIR domains and constructsConstructDomainFacilitator \[reference number\]Barrier \[reference number\]1. Intervention characteristicA. Intervention sourceB. Evidence strength & quality**"Limited guidance on prepared practice team development" \[[@CR40]\]**C. Relative advantage**"Patient screened by staff before seeing physician " \[[@CR39]\], "Structured assessment in patient education" \[[@CR39]\]**D. Adapability**"Integrating Guided Care nurse in work flow" \[[@CR36]\], "Processes integrated in to existing clinical operations" \[[@CR43]\], "CCM adaption within context of daily practice" \[[@CR48]\], "Program tailored to region needs" \[[@CR50]\], "Adapting communication system to local context" \[[@CR52]\], "Integrated project to routine care" \[[@CR52]\]**E. TrialabilityF. Complexity**"Intervention was too complex, targeted different components resulting in many priorities" \[[@CR50]\]**G. Design quality & packaging**"Nurse training for components of intervention" \[[@CR35]\], "Curriculum should be specific to CCM intervention" \[[@CR36]\], "Different intervention model options were offered" \[[@CR19]\], "Structured learning sessions and support by health collaborative" \[[@CR44]\], "Guideline development" \[[@CR50]\]"Intervention was too disease specific and did not address chronic care principles" \[[@CR45]\]**H. Cost**"Low-cost program relied on community health workers, mentors and non-clinical staff" \[[@CR43]\], "Financially viable" \[[@CR48]\], "Sufficient funding" \[[@CR37]\]**2. Outer settingA. Patient needs & resources**"Community health workers important in addressing patient needs" \[[@CR43]\], "Program accessible and offered peer support" \[[@CR43]\]"Need for patient resources" \[[@CR19]\], "Patients uninsured or Medicare insured" \[[@CR38]\], "Language barriers" \[[@CR38]\], "Language and literacy issues" \[[@CR44]\]**B. CosmopolitanismC. Peer pressureD. External policies & incentives**"Poor organization of primary care in region" \[[@CR50]\]**3. Inner settingA. Structural characteristics**"Development of prepared practice teams" \[[@CR40]\], "Electronic medical record (EMR) implementation and clinic remodelling" \[[@CR39]\], "Recruitment of multilingual staff and interpreters to address language barriers" \[[@CR44]\], "Worked with human resources to change organizational policies" \[[@CR44]\], "Role of specialist in supporting and supervising other staff" \[[@CR45]\], "Addition of technology system" \[[@CR52]\], "Nurse practitioner role in implementation" \[[@CR53]\]"Staff turnover" \[[@CR19]\], "Large size of medical group" \[[@CR40]\], "Unions unsupportive of staff role change" \[[@CR40]\], "Medical director turnover" \[[@CR38]\], "Need to expand role of provider" \[[@CR44]\], "Staff turnover and loss meant very few staff could assume additional responsibilities" \[[@CR44]\], "Lack of staff expertise in team approach to implementation" \[[@CR48]\], "Lack of flexibility in reorganizing model of care" \[[@CR52]\], "Smaller organizations had difficulty addressing barriers" \[[@CR45]\]**C. Culture**"Support from primary care physicians" \[[@CR35]\], "Support from physicians" \[[@CR36]\], "Recognition of benefit of care managers" \[[@CR39]\], "Stable work relationships" \[[@CR40]\], "Recognition of patient role in self management" \[[@CR44]\], "Persistence despite extra work" \[[@CR44]\], "Organizational culture and enthusiasm for care improvement" \[[@CR45]\], "Promoting multidisciplinary approach" \[[@CR51]\], "Change to patient-centred care" \[[@CR52]\], "Receiving personal recognition" \[[@CR37]\]"Providers need for clear structure and autonomy" \[[@CR19]\], "Organizational culture unsupportive of change" \[[@CR40]\], "Lack of commitment or tradition of working in interdisciplinary teams" \[[@CR50]\], "Difficulty changing provider care to patient-centered care" \[[@CR52]\], "Rigid role expectations and thought processes" \[[@CR52]\]**D. Implementation climate**"Clear, shared long term commitment for change" \[[@CR40]\], "Recognized need for change" \[[@CR40]\], "Work credit to ensure staff buy-in" \[[@CR42]\], "Institutional commitment for change" \[[@CR45]\], "Commitment to follow guidelines" \[[@CR48]\], "Provider dissatisfaction with current system" \[[@CR50]\], "Financial reimbursement for attending meetings" \[[@CR51]\], "Organizational will to promote change and manage change" \[[@CR51]\] "Career promotion opportunities" \[[@CR37]\], "Incentives such as skill development" \[[@CR37]\]"Lack of physician interest in addressing communication barriers with specialists" \[[@CR39]\], "Disagreement on need for standardized care" \[[@CR40]\], "Lack of commitment and interest by chief physician" \[[@CR40]\], "Lack of committed vision" \[[@CR45]\], "Difficult to motivate providers due to program uncertainty" \[[@CR50]\], "Lack of provider commitment" \[[@CR50]\]**E. Readiness for implementation**1. "Used pre-existing available resources: information system and education program" \[[@CR34]\], "Buy-in from senior management" \[[@CR19]\], "Previous implementation of structured assessment in EMR" \[[@CR39]\], "Importance of project leaders" \[[@CR52]\], "Sufficient staff personnel" \[[@CR37]\]"Low staff and space resources" \[[@CR43]\], "Lack of reimbursement strategy" \[[@CR45]\], "Lack of financial resources" \[[@CR50]\], "Software builder did not meet goals" \[[@CR52]\], "Limited financial resource" \[[@CR34]\], "Hidden and unexpected implementation expenditures" \[[@CR52]\]**4. Individual characteristicsA. Knowledge & beliefs about intervention**"Increase awareness and education about program to providers" \[[@CR41]\], "Observation of program processes by providers" \[[@CR42]\], "Patient registry received interest in providers" \[[@CR44]\], "Clinical assessment tool accepted and endorsed" \[[@CR45]\], "Information campaign to increase awareness and knowledge" \[[@CR50]\], "Education about project goals & process" \[[@CR51]\], "Demonstration of project benefit to physicians" \[[@CR51]\], "Staff morale and burnout reduction associated with reports of improved care outcomes" \[[@CR37]\]"Needed more information on structured assessment" \[[@CR39]\], "Unconvinced of usefulness of structured assessment for diagnoses" \[[@CR39]\], "Lack of program information from providers that were not full time" \[[@CR41]\], "Physician buy-in and adoption of intervention was not uniform" \[[@CR47]\], "Fear of losing patient control to education program" \[[@CR50]\], "Time needed for provider trust in program" \[[@CR50]\], "Clinicians sensitive to workload and time commitment" \[[@CR45]\]**B. Self-efficacy**"Fatigue and apathy from pace of change" \[[@CR40]\], "Decreased staff participation in intervention results in low morale" \[[@CR37]\]**C. Individual identification with organizationD. Personal attributes5. ProcessA. Planning**"Realistic expectations for measureable results" \[[@CR40]\], "Consultation with focus groups for change process priorities" \[[@CR50]\], "Physician involvement in planning" \[[@CR51]\], "Utilized patient and physician experience in project development" \[[@CR51]\], "Goals of QI as drivers of planning " \[[@CR52]\]"Lack of details on care change goals & outcomes" \[[@CR40]\], "Too many priorities and uncoordinated change processes" \[[@CR40]\],"Need for stronger program goals delineation" \[[@CR41]\], "Lack of clear program aim at the start of campaign" \[[@CR50]\]**B. Engaging**"Supportive administration and intervention champion" \[[@CR19]\], "Strong physician leadership" \[[@CR40]\], "Supervisor support" \[[@CR40]\],"Strong registered nurse leadership" \[[@CR40]\], "Clear goals by leaders" \[[@CR40]\], "Strong supportive leader" \[[@CR45]\], "Commitment & support of senior leaders" \[[@CR50]\], "Recruitment of physician champion" \[[@CR51]\], "Engaging champions with physicians" \[[@CR51]\], "Presence of strong champion" \[[@CR37]\]"Need for more senior management support" \[[@CR19]\], "Need for intervention champion" \[[@CR19]\], "Lack of accountability by leadership" \[[@CR40]\], "Leaders face multiple uncertainties and distractions" \[[@CR40]\], "Champion provider had limited time with patients" \[[@CR44]\], "Change difficult without leadership endorsement" \[[@CR44]\], "Lack of active provider champion" \[[@CR44]\]**C. Executing**"Coordination of program components" \[[@CR41]\], "Target screening of at risk patients" \[[@CR39]\], "Pre-visit screening by staff before seeing physicians" \[[@CR39]\], "Pre-visit by nurse and clerical staff" \[[@CR40]\], "Approached patient as a team" \[[@CR44]\], "Health care organizations part of collaborative had high CCM fidelity and moderate intensity" \[[@CR46]\], "Flexible meeting times and locations" \[[@CR51]\], "Fair distribution of tasks" \[[@CR37]\]"Inadequate time to work on intervention" \[[@CR19]\], "Difficulty with patient registry" \[[@CR19]\], "Need for technical support" \[[@CR19]\], "Competing demand of simultaneous EMR implementation" \[[@CR40]\], "Physicians not engaged in change processes" \[[@CR40]\], "Patient registry lacked IT support" \[[@CR44]\], "Difficult to implement all CCM elements at high intensity in 12 months" \[[@CR46]\], "Screening all patients time was consuming" \[[@CR39]\], "Time constraints in appropriate assessment" \[[@CR48]\], "Buy-in from staff not sufficient to sustain program" \[[@CR48]\], "Increase in administrative burden" \[[@CR50]\], "Patient involvement in own care was difficult" \[[@CR52]\]**D. Reflecting & evaluating**"Periodic reviews and feedback of performance" \[[@CR36]\], "Staff provided feedback on process design" \[[@CR41]\], "Continuous assessment and revisions of program" \[[@CR41]\], "Support from monthly feedback and learning sessions" \[[@CR44]\]"Insufficient time to measure change" \[[@CR40]\], "Lack of useful measure of change" \[[@CR40]\], "Lack of EMR and billing codes were barriers for measurement of processes and outcomes" \[[@CR48]\], "Implementing and measurement was labour intensive" \[[@CR48]\] Facilitators {#Sec7} ------------ ### Networks and communication {#Sec8} Strong networks and increased communication between health care providers and organizations were fostered by collaboration across disciplines and specializations during care change processes \[[@CR39],[@CR40],[@CR44],[@CR50],[@CR51]\]. Communication was reportedly supported by regular group meetings with supervisors and managers to discuss implementation issues, computerized information sharing and clinical assessment tools \[[@CR41],[@CR45],[@CR52]\]. ### Culture {#Sec9} An organizational culture that promotes multidisciplinary, or patient centered care, was identified as important during implementation \[[@CR45],[@CR51],[@CR52]\]. Support from clinical providers and the recognition of their importance in care change efforts was found to increase uptake of the CCM in primary care \[[@CR35],[@CR37],[@CR39]\]. ### Implementation climate {#Sec10} Studies found that implementation climate was influenced by commitment and recognition for the need for change from the organization \[[@CR40],[@CR45]\]. Willingness to advance and manage change was evident through incentivizing provider buy-in using financial reimbursement and work credit for project involvement \[[@CR37],[@CR42],[@CR51]\]. ### Structural characteristics {#Sec11} Operationalization of CCM components was facilitated by health care providers, particularly specialists and non-physician staff such as nurse practitioners, who had to expand their responsibilities and scope of practice \[[@CR45],[@CR53]\]. This sometimes required changing organizational policies and development of care teams to meet implementation needs \[[@CR40],[@CR44]\]. ### Engaging {#Sec12} Strong, committed and engaging leadership in the form of supportive administration and supervisors, with clear goals, was cited as a facilitator \[[@CR40],[@CR45],[@CR50]\]. This included the appointment of an intervention champion to promote uptake of the model within the organizations \[[@CR19],[@CR37],[@CR51]\]. Leadership roles were not limited to physicians, other health care providers such as nurse practitioners were found to play a major role in implementation \[[@CR40]\]. ### Knowledge and beliefs about the intervention {#Sec13} Provider knowledge about CCM interventions was promoted through observing the execution process by other staff and education about project goals \[[@CR42],[@CR50],[@CR51]\]. Strategies used to foster beliefs of the CCM effectiveness in care providers, particularly physicians, included demonstration of its benefits to their practice and sharing reports of patient improvements \[[@CR37],[@CR51]\]. Barriers {#Sec14} -------- ### Executing {#Sec15} Many studies identified barriers related to executing intervention processes. Implementing the multiple components of CCM into practice created additional responsibilities for staff who were limited by time constraints \[[@CR19],[@CR40],[@CR48],[@CR50]\]. Pearson & colleagues found that operationalizing the model elements at a high level of intensity, within a short time frame to be challenging \[[@CR46]\]. Sustainability of the intervention was found to be difficult in some studies; in some instances, staff buy-in, an important aspect of implementation, was not enough to ensure program longevity \[[@CR48]\]. ### Structural characteristics {#Sec16} Characteristics of the healthcare organization such as its size, whether it adopted a team-based approach and its flexibility in reorganizing care, were seen to influence the success of CCM adoption \[[@CR40],[@CR45],[@CR48],[@CR52]\]. Institutional factors such as staff turnover and loss meant increased burden of responsibilities on existing providers \[[@CR19],[@CR44]\] 10). Leadership turnover, particularly that of a medical director, was cited as a barrier towards implementing care change processes \[[@CR38]\]. ### Readiness for implementation {#Sec17} Organizational readiness for the CCM was found to be impacted by the lack of interest and commitment from leadership and unavailability of resources for implementation \[[@CR40],[@CR45]\]. Lack of resources that influenced readiness included low funding, lack of provider reimbursement strategies and low staff numbers \[[@CR34],[@CR43],[@CR45],[@CR50]\]. ### Engaging {#Sec18} Many studies found that execution of the intervention processes was challenging without support and accountability from senior leadership \[[@CR19],[@CR20],[@CR44]\]. Without the presence of an intervention champion, endorsement of the CCM initiative was found to be limited in healthcare providers \[[@CR19]\]. ### Knowledge and beliefs {#Sec19} Provider buy-in was greatly influenced by knowledge and beliefs about the intervention, particularly if they had misconceptions, were unconvinced of its effectiveness or lacked information \[[@CR39],[@CR47],[@CR50]\]. Acceptance of the interventions by clinicians required time, and was also affected by the workload associated with implementing and executing the intervention components \[[@CR45],[@CR50]\]. Discussion {#Sec20} ========== This review identified multiple barriers and facilitators of implementing the CCM across various primary care settings. The major emerging themes were those related to the *inner setting* of the organization, the *process* of implementation and *characteristics of the individual* healthcare providers. These included: culture of the organization, its structural characteristics, networks and communication, implementation climate and readiness, supportive leadership, and provider attitudes and beliefs. Every primary care organization possesses its own cultural norms, practices and leadership. It is impossible to achieve change without adopting an approach that considers the individual and the team of providers, the organization setting and the greater system within which it is embedded \[[@CR54]\]. Wolfson and colleagues attributed the success of QI in different primary care practices to facilitators in various levels of the organization including: presence of an initiative champion; physician, staff and patient cooperation; leadership investment; team practice and progress tracking \[[@CR55]\]. The uptake of CCM elements in the studies required a primary care culture supporting willingness to change and quality improvement at the individual clinician, team and organizational levels. Implementation was most successful when there was a shared vision and a recognized need across the organization for new care change approaches to promote effective execution of the CCM \[[@CR35],[@CR36],[@CR39],[@CR44],[@CR52]\]. Transforming care practices in an organization requires a supportive culture of change and learning \[[@CR23]\]. Clinical provider beliefs and attitudes about evidence-based practice can influence the culture and learning environment, particularly when the provider perceives the evidence as unreflective of their day-to-day clinical decision making. This suggests the need to involve clinicians in early stages of intervention development and implementation \[[@CR22]\]. Interventions that incorporated providers, patients and their experiences in the planning phase of the intervention were more successful in operationalizing the CCM \[[@CR50],[@CR51]\]. This approach may bridge the cultural divide between leadership and clinical providers, which can hinder quality improvement efforts if left unaddressed. On the other hand, literature shows that lack of a group-oriented culture, as well as hierarchical relationships where the leadership is unsupportive of change, are negatively associated with implementation of care change processes \[[@CR55]\]. Marshall and colleagues highlight the importance of culture and cultural change when implementing clinical governance in primary care. Cultural traits that support implementation efforts include commitment to accountability by the organization, willingness for collaborative work and learning, and ability to evaluate and reflect on mistakes \[[@CR56]\]. Implementing and managing change processes in primary care can require time and flexibility. Organizational transformation can be slow and resistant to change, while spread of best-practice can be a challenge \[[@CR57]\]. In some cases, even when an organization's culture is supportive of the CCM, the inner structures of the primary care organization, such as a lack of staff and financial resources or a lack of clinician expertise, can impede organizational readiness for implementation and cause unexpected setbacks \[[@CR34],[@CR48],[@CR52]\]. A study evaluating the implementation of evidence-based practice revealed that the current primary care system is not adaptive to rapid change, or accommodating to the additional duties associated with adopting new interventions. What this suggests is the need to set realistic implementation goals that are reflective of the organization and staff capacity for changes associated with the CCM. This requires comprehensive planning at all stages of component adaptation, to mitigate impeding factors such as rigid bureaucracies and organizational policies. On the other hand, clinical leaders and champions can be drivers of change by ensuring the availability of resources and providing adequate staff supports \[[@CR58]\]. Indeed, leadership support for change has been shown to be positively associated with QI outcomes and sustainability in primary care \[[@CR24]\]. Implementation of CCM in primary care requires tailoring interventions to the local context, as well as altering the context, for the process to be successful. The two can co-adapt and evolve during the implementation process, thereby ensuring sustainability \[[@CR59]\]. The majority of the studies included in the review identified the impact of the CCM on existing routines, practices, and culture of the organization. There was variability in how each organization adapted the CCM, i.e., translating the framework's components into practice resulted in implementation heterogeneity. What became clear is that a standardized one-size-fits-all approach was difficult to put into practice when the components were conceptualized differently by each primary care organization. Tailoring the intervention necessitates accounting for innovation-promoting and hindering factors at different levels of the organization, as well as reconfiguring aspects of the primary care setting itself \[[@CR50]\]. This systematic review has several limitations. First, our search strategy meant that we did not assess: grey literature, studies that have not been published in peer-reviewed journals or those published in languages other than English; therefore, articles that were relevant to our review may not have been included. The search may have excluded studies that implemented CCM-based interventions but which were not explicitly referred to as such in the articles. In addition to the challenge of consistently identifying such studies, it would be difficult to be certain that implementation issues were reflective of issues relevant to the CCM. Another limitation is that the articles that were included were selected and assessed by one reviewer, thus limiting the reliability of the selected studies. Given that the articles were abstracted qualitatively by a single data abstractor, there is a possibility of bias in how the attributive statements were mapped under CFIR constructs and domains. While abstraction and coding was carried out by one reviewer, extensive and continuous discussion took place between both authors occurred during the study selection and data abstraction process. While using the CFIR as a guiding framework is a strength of our review, the numerous construct and sub-constructs meant that areas with few facilitators and barriers identified received less consideration (although these were captured in Table [4](#Tab4){ref-type="table"}). Conclusion {#Sec21} ========== These findings highlight the need to evaluate factors that influence successful implementation of CCM in primary care. The CFIR can be used to guide the formative evaluation processes of CCM interventions. Assessment of organizational capacity and needs is important prior to and during the implementation of the intervention, in order to gain a better understanding of health care providers and organizational perspective. The barriers and facilitators identified under the CIFR domains can be used to build knowledge on how to adapt the CCM to different primary care settings. **Competing interests** The authors declare that they have no competing interests. **Authors' contributions** MK and PS conceived of the paper. MK drafted the initial version and PS drafted revisions of this paper. Both authors read and approved the final manuscript. We would like to acknowledge University of Waterloo librarian Rebecca Hutchinson for assisting us with developing a search strategy for the systematic review.

Background {#sec1-1} ========== Gestational diabetes is a condition characterized by glucose intolerance that particularly develops during pregnancy. There has always been a dilemma among the clinicians to diagnose gestational diabetes. Several criteria have been proposed by various societies across the world. To bring in uniformity the WHO brought out the guidelines in 1999 that was updated in 2006 and this was being followed largely. In 2013, the WHO has revised and formulated the new set of diagnostic criteria for gestational diabetes.\[[@ref1]\] These new range of values were devised based on the risk of adverse feto-maternal outcomes. Irrespective of the ambiguity in the diagnostic values, there is ample evidence to suggest the increasing prevalence of gestational diabetes.\[[@ref2][@ref3]\] There is wide region to region variation in prevalence ranging from 2.5% to 21%.\[[@ref4][@ref5][@ref6][@ref7]\] The prevalence has also been noted to be different across the sociodemographic and economic strata. Various studies point out the ethnic differences in the prevalence of gestational diabetes. Asian women, especially those having Indian ethnicity, have been undoubtedly proved to be more at risk for this condition.\[[@ref2][@ref8]\] In India the prevalence of gestational diabetes varies widely geographically. Studies conducted in southern part of the country reveal a bigger burden of gestational diabetes than the northern states.\[[@ref7][@ref9][@ref10]\] The regional differences in the prevalence of gestational diabetes is in correlation with that of type 2 diabetes. The risk of developing type 2 diabetes mellitus is high for the patients with gestational diabetes.\[[@ref11]\] Kerala is a state which is currently facing a huge burden of diabetes mellitus. Increasing prevalence of gestational diabetes will make a major contribution to the diabetic pool of this state. Moreover, a significant proportion of the women belonging to the reproductive age group in Kerala are overweight and obese.\[[@ref12]\] High BMI is a proven risk factor to develop gestational diabetes.\[[@ref13][@ref14][@ref15]\] In the setting of high prevalence of type 2 diabetes, these women also are prone to have a positive family history which also adds to the risk of gestational diabetes.\[[@ref13]\] It is inevitable for the state\'s health system to be equipped to face the consequences of the burgeoning burden of gestational diabetes in the near future. Understanding the various outcomes of gestational diabetes would be the key to initiate the cascade of preparatory steps to tackle them. Studies are available from various parts of the country signifying the adverse maternal and neonatal outcomes of gestational diabetes. However, similar studies from Kerala setting are scarce. This study therefore aims to study the frequency of the occurrence of various maternal and fetal outcomes among gestational diabetes patients. Materials and Methods {#sec1-2} ===================== This retrospective cohort study included a total of 180 individuals. These participants were taken from the database of INDADE (Indo Danish Collaboration on Diabetes Epidemiology) study, a community prospective cohort study conducted in rural Kerala. This study is a collaborative research program between Health Action by People (HAP), a non-profit research organization in Kerala and University of southern Denmark. The INDADE study included 25,000 subjects and it was aimed to understand the epidemiology of diabetes mellitus. The residents of a Local Self Government Unit (Ottoor Panchayath) in southern Kerala were the study participants. They were followed up for a period of 4 years, from 2007 to 2011. Ottoor Local Self Government unit belongs to the rural area of Thiruvananthapuram district in Kerala. After recording their baseline anthropometric measurements and fasting blood sugar levels, individuals aged above 20 years were being followed up. Annual medical examination and biochemical investigations were performed among the diabetic individuals. The current analysis was conducted in 2010, for which the baseline data collected in 2007 were retrieved from the records of INDADE study. These included their personal identification details, educational qualification, occupation, anthropometric measurements like weight, height, waist and hip circumference, and FBS values. The participants included 60 women with gestational diabetes and 120 women who were pregnant but without gestational diabetes. The details of their antenatal period and the pregnancy outcomes were collected through telephonic interview using a semi-structured questionnaire. Interviewers were trained to conduct interviews in a uniform manner. The questionnaire included their glycemic status, details of the treatment, mode of delivery, delivery complications, pre-mature births, still births, abortions, birth weight of the child, neonatal ICU admissions and persistence of hyperglycemia after delivery. Gestational diabetes was the major exposure variable. All quantitative variables have been summarized using mean (SD) and qualitative variables in proportions. We analyzed the feto-maternal outcomes of the patients with gestational diabetes. Bivariate analysis was done using the Chi-square test. A *P* value less than 0.05 was considered to be statistically significant. Multivariable logistic regression was done to compute the risk for various outcomes in gestational diabetes as compared to those without gestational diabetes. Analysis was done using SPSS software, version 16. Results {#sec1-3} ======= A total of 180 individuals were included in the study of which 60 had gestational diabetes and 120 did not. The mean (SD) age of the study group was 32 (7.8) years. Majority of them \[160 (89.8%)\] had an educational qualification above the high school level. Most of the participants \[157 (87.2%)\] were home makers. The baseline characteristics which were measured in 2007, of those with gestational diabetes have been compared with those without gestational diabetes in [Table 1](#T1){ref-type="table"}. ###### Baseline characteristics of gestational diabetic individuals as compared to those without gestational diabetes  The baseline fasting blood sugar value done at the start of the study was on the higher side for gestational diabetes patients with a mean (SD) of 91 (37.6) mg/dl as compared to 84 (16.4) mg/dl for those without gestational diabetes. However, this difference was not statistically significant. The mean BMI of this study group was 24.7 (3.5) kg/m^2^. More than one fourth of these individuals \[50 (27.8%)\] had BMI above 25 kg/m^2^. The mean (SD) waist-hip ratio of the diabetic individuals was higher \[0.87 (0.16)\] than that of the non-diabetic individuals \[0.86 (0.08)\], but the difference was not statistically significant. Nearly half \[28 (48.3%)\] of 60 individuals who had gestational diabetes had a positive family history for type 2 diabetes. Most of the gestational diabetic patients \[42 (70%)\] were multi-gravida. Majority of them \[41 (68.3%)\] had developed gestational diabetes during the first pregnancy itself. Insulin therapy had to be initiated for 26 (43.3%) patients. A few of the patients with gestational diabetes \[12 (21.1%)\] also had associated hypertension during pregnancy. Mode of delivery was by caesarean section for 19 (32.8%) individuals. It was noted that more than half of those who underwent caesarean section \[10 (52.6%)\] were on insulin therapy. The outcome of the pregnancy ended in abortion for 11 (19%) of individuals. Birth weight of the newborns was above 3 kg in more than half \[31 (51.7%)\] of these individuals. In-born nursery (IBN) admission was warranted for neonates of seven (12.1%) patients. The various feto-maternal outcomes of the 60 individuals with gestational diabetes are given in [Table 2](#T2){ref-type="table"}. The adjusted risk for progression to overt diabetes after delivery, babies with birth weight more than 3 kg and IBN admissions of the neonates, was significantly more for those with gestational diabetes \[[Table 3](#T3){ref-type="table"}\]. ###### Feto-maternal complications among patients with gestational diabetes (*n*=60)  ###### Adjusted relative risk for developing complications among GDM patients as compared to non GDM individuals  The progression to type 2 diabetes was observed in six (10%) of the gestational diabetes patients. They were significantly older than the rest of the gestational diabetic with the mean (SD) age being 37 (7.2) years (*P* = 0.006). The mean (SD) baseline fasting blood glucose value for them was 130 (92.3) mg/dl which was much higher than the rest of the group (*P* = 0.001). Out of these six individuals who progressed to type 2 DM, four (66.7%) had given birth to babies with birth weight more than 3 kilograms. The mode of delivery was through caesarean section for 5 (83.3%) out of 6 of these patients. This association was found to be statistically significant (*P* = 0.005). Discussion {#sec1-4} ========== This study brings into light the various maternal and neonatal outcomes of gestational diabetes. Delivering the child through surgical intervention and progression to type 2 diabetes mellitus were the major maternal outcomes. Major neonatal outcomes included increased birth weight, IBN admissions, premature deliveries and abortions. Evidence generated from other parts of the country supports the results in this study.\[[@ref10][@ref11]\] Long-term progression to type 2 diabetes has far reaching implication in this study setting. The prevalence of type 2 diabetes is as high as 20% in Kerala.\[[@ref16]\] It is a known fact that the intrauterine hyperglycemic status in gestational diabetes increases the risk for type 2 diabetes in future.\[[@ref11][@ref17][@ref18]\] Even the molecular aspects of progression to type 2 diabetes has been studied in detail among gestational diabetes patients. The endothelial irregularities found in diabetic vasculopthic patients was detected in the fetal placenta of gestational diabetic individuals.\[[@ref19]\] This suggests that the origin of diabetic complication is seen right from the fetus of gestational diabetic women. The population of Kerala is highly vulnerable for the development of type 2 diabetes with the alarming prevalence of other risk factors such as obesity, unhealthy dietary habits, sedentary lifestyle and a positive family history.\[[@ref20]\] Hence, the early detection and control of gestational diabetes become a mandate in controlling the unprecedented increase in the type 2 diabetes burden of the state. The rate of caesarean section in the state is considerably high than the rest of the country.\[[@ref21][@ref22]\] This study highlights that more than 32% of the gestational diabetes patients had to undergo caesarean section to terminate their pregnancy. Big baby which is an outcome of gestational diabetes is an indication for caesarean section. Similar studies done in other parts of the world also concludes that women with gestational diabetes are more at risk for caesarean sections.\[[@ref23]\] More than half of those who underwent caesarean section were on insulin therapy for uncontrolled glycemic status. Regular monitoring of blood glucose and its strict control reduces the risk of surgical interventions and can facilitate normal deliveries. Simple interventions like this could reduce the rate of caesarean section in the state. Health indicators in Kerala are the best in the country and are at par with the developed nations. The most important health indicator suggestive of the system of health care of a region is the infant mortality rate (IMR). It reflects the development of the region as a whole. According to Sample Registration System (SRS) 2011 data, the IMR in Kerala is 12 which is much lower than the national average of 44.\[[@ref24]\] Neonatal deaths takes up the major share of infant mortality. A study done in central Kerala reveals that neonatal mortality accounts for 58% of childhood mortality in that area. Preterm births are an important cause of neonatal mortality. This study reveals the prevalence of preterm delivery to be 10% among gestational diabetes patients. IBN admissions were required for 12% of the neonates of these patients. The need for improving the neonatal care facilities in the states health care infrastructure is being brought into light here. The development of neonatal health care facilities is inevitable to further bring down the state\'s IMR from the static figure of 12. Conclusion {#sec1-5} ========== This study concludes that the major feto-maternal outcomes of gestational diabetes are long-term progression to type 2 diabetes, increased birth weight and increased IBN admissions for neonates. In the present health milieu of the state these findings are of utmost importance. The burden of gestational diabetes is only going to increase in the near future. Cautious attention therefore must be given to devise measures to tackle the challenges of gestational diabetes. Awareness to achieve good glycemic control and to prevent progression to type 2 diabetes must be given to these women. The Government should largely take up the responsibility of improving the neonatal care facilities of the state to handle the neonatal outcomes of gestational diabetes. **Source of Support:** Nil. **Conflict of Interest:** None declared.

Background {#Sec1} ========== Malaria is among the major vector-borne diseases that exact the heaviest toll from populations of sub-Saharan Africa. Worldwide, it is estimated that about 3.4 billion people are at risk of malaria. Within the past few years, approximately 207 million yearly cases of malaria occurred globally with the highest burden (80% of cases and 90% of deaths) recorded in Sub-Saharan Africa. The huge proportion of deaths (77%) occurs in children under 5 years \[[@CR1]\]. The whole Cameroonian population is at risk of malaria transmission, of which 71% are at high risk and 29% at low risk. *Anopheles gambiae* is the main vector responsible for malaria transmission in the sub-Saharan Africa region \[[@CR2]\]. The combined efforts of distributing long lasting insecticide treated mosquito nets, indoor residual sprays, and effective case management with potent antimalarials have dramatically declined the malaria related illness and mortality \[[@CR3]\]. However, the phenomenon of insecticide resistance poses a serious threat to sustainable insecticide-based vector control in many African countries. Besides, the lack of effective anti-vector tools designed specifically to prevent outdoor transmission and the serious threat posed by insecticides to the environment as well as the human health emphasize the need to develop new vector control approaches that can complement the existing interventions. In that framework, there are increased research efforts to develop natural and environment friendly interventions including plant-based mosquito repellents. Of note, plant-based repellents are generally target-specific and relatively non-toxic, and are still extensively used traditionally as first intention and affordable tools in malaria endemic rural communities for protection against mosquito bites \[[@CR4]\]. Therefore, the indigenous knowledge and use of plants as repellents should be documented and preserved as a cheap and sustainable way of preventing mosquito\'s bites or entering homes in Cameroon and beyond. This documentation represents a prerequisite for future research on efficacy and safety as well as identification of single chemical entities with mosquito repellent activity which could lead to the development of standardized bio-pesticides. So far, only a limited number of studies have been conducted in Cameroon on traditional use of medicinal plants to repel insects/mosquitoes \[[@CR5]\]. The question is therefore to know if there is a consensus extent in indigenous use of plants from the Cameroonian biodiversity to repel mosquitoes/insects. To fill this gap in, we hypothesized that plants from six selected localities of the East, South, West and Centre regions of Cameroon are culturally used to repel mosquitoes/insects. Thus, the present paper documents the indigenous knowledge on plants used as mosquito/insect repellents in the six selected localities in order to further evaluate their potential for new plant-based repellent products. Methods {#Sec2} ======= Selection of the study sites {#Sec3} ---------------------------- The study sites were selected among the malaria endemic regions of Cameroon. In fact, the 71% Cameroonian population that are at high risk of malaria transmission \[[@CR2]\] live in the endemic and perennial zones of continuous transmission (7 to 12 months) that cover the Southern Cameroonian Equatorial forest, the High Western Plateau altitude and the Coastal region, including all the selected study sites where about a hundred infective bites per man per month can be registered \[[@CR6]\]. Besides, the presence of indigenous pygmies \[Lolodorf, Bipindi (South region), Dimako (East region)\] who have great knowledge of the forest, plants and their medicinal properties \[[@CR7], [@CR8]\], and the presence of herbal practitioners/medicinal plant users in Londji (South region), Mbouda-Babete (West region), and Kon-Yambetta (Centre region) also motivated their selection as study sites (Fig. [1](#Fig1){ref-type="fig"}).Fig. 1Study sites of the ethnobotanical survey of insect repellent medicinal plants. Study sites were selected based on their ethno-cultural and floristic characteristics and included Dimako (East region), Kon-Yambetta (Centre region), Lolodorf, Bipindi, and Kribi-Londji (South region), and Mbouda-Babete (West region) Lolodorf, Bipindi, and Kribi-Londji {#Sec4} ----------------------------------- Lolodorf, Bipindi, and Kribi-Londji are districts of the Ocean Division in the South Region of Cameroon nearby the western coast of Africa. The mean temperature varies from 25 to 26 °C with two rainy and two dry seasons. Lolodorf is located between Ngoumou and Bipindi, in a zone of the Atlantic Littoral Evergreen Forest, extending over the equatorial climatic zone. Londji is a small locality located 20 km to North of Kribi District also characterized by a mangrove forest ecosystem, and has vast natural beaches and the largest fishermen village and fish market in Cameroon \[[@CR9]\]. The Londji bay is an extension of the Atlantic Ocean on the continent and the mouth of the Lokoundje and Nyong rivers. The village showcases a vast diversity of heterogeneous but integrated populations and ethnic groups, as the dynamism of the fishing activity has attracted people from West Africa throughout the years. Londji prides itself in the sustainable preservation of the mangrove areas and wetlands. Moreover, it is crossed by dotted rivers with mangrove plots that promote the development of insects. The humidity remains high throughout the year and causes increased insects breeding sites \[[@CR10]\]. Bipindi and Lolodorf are notable for being the home of Pygmy clans and their camp settlements and hunting areas, such as those of Lala, Bakola and Bagyeli pygmies. There are also settled Bantu people such as Ngumba and Kwassio tribes. Farming, tapping palm wine, gathering fruits and nuts for food, fishing, and hunting are the main occupations within these communities. Malaria and sleeping sickness both vector-borne diseases are common problems in this humid climate. Bagyeli and Bakola pygmies have great knowledge of using plants for their medicinal properties \[[@CR7], [@CR8]\]. The present study was partly conducted in Ngoyang, Mbikiliki, Mougoué, Ngovayang, and Bigambo, all surrounding villages of Lolodorf as well as in Bipindi and Bidjouka, two villages of Bipindi District. Dimako and surroundings {#Sec5} ----------------------- Dimako is a small town situated in the Upper Nyong Division of the East Region of Cameroon. It lays a little way south of East Region capital of Bertoua at only about 28 km away. Vegetation is characteristic of a tropical rain forest, dotted with fallows, which covers 90% of the municipal territory. The remaining 10% is covered by a shrubby savannah in the north of Dimako. A greater part of the forest is described as semi-caducus forest made of Sterculiaceae and Ulmaceae plants. Inhabitants of the Dimako District are very closely linked to the forest and its resources. Dimako District accounts 30 villages including Loussou and Mayos inhabited by Baka Pygmies. Due to the humid mosquito- and black fly-infested forests \[[@CR11]\], the area was also selected as study site. Kon-Yambetta {#Sec6} ------------ Kon-Yambetta is a District of the Mbam-et-Inoubou Division (Centre region) located about 150 km (Northwest) from Yaoundé, in the grasslands between Bafia and Ndikinimeki. It is the transition zone between the forest and the savannah. The area has an equatorial type climate with two raining and two dry seasons. The inhabitants practice peasant agriculture dominated by the cocoa and diverse subsistence crops. Environmental degradation and mismanagement has increased the development of vectors (mosquitoes, mango flies) of diseases such as malaria. This area has Guinea savannah-type flora where the yambetta and bamoun people rely on herbs for first intention treatment when ill \[[@CR12]\]. Mbouda (Babete) {#Sec7} --------------- Babete is one of the 6 villages of Mbouda District (Bamboutos Division, West Region) located at approximately 3.2 km from Mbouda. It is characterized by high lands, cool temperatures, heavy rainfall, and savanna vegetation. The climate is classified as a tropical savanna, with a subtropical moist forest biozone \[[@CR13]\]. This area is known for use of medicinal plants as source of essential oils as well as dry fruits and natural drinks. Such products are commonly manufactured and sold by the nuns of the Saint Benoît Monastery \[[@CR13]\]. Interview and collection of plants specimens {#Sec8} -------------------------------------------- In prelude to the survey, legal authorities (village heads) from each study site were approached and the aims and objectives of the survey were discussed for authorization to investigate within their communities. Volunteers and recommended herbal practitioners were identified as potential informants and subsequently participated in personal interviews. During a face-to-face interaction, the purpose and the procedure of the study, as well as the expected benefits and rights to their community were explained to them. The interviews were conducted in French or in the local languages with the assistance of volunteer interpreters when needed. The information surveys took place between May and June, 2015. A total of 182 informants from 38 households including local herbal practitioners were surveyed with a set of pre-tested semi-structured questionnaires in all the six study areas and selected for further interaction. Prior informed consent was obtained verbally from informants before they were interviewed. During the interviews, semi-structured questionnaires were used to obtain information on plants used to repel mosquitoes/insects, including parts used, local name, modes of preparation and administration (Additional file [1](#MOESM1){ref-type="media"}). Some repellent plants were also identified by visiting the field with the informant (field interview); listening to him and asking questions; presenting plant specimens to informants and asking questions (plant interview); or using a checklist (checklist interview; plant name-common or vernacular-was addressed to the informants in order to gather supplementary field information). Group interview have sometimes developed spontaneously with community members and this helped to supplement the interview. Specimens were collected from the plants claimed by informants to have repellent activity. Such plants were identified by a taxonomist and voucher specimens deposited under specific identification numbers at the Cameroon National Herbarium in Yaoundé. Data analysis {#Sec9} ------------- For each cited plant species, the frequency of citation (FC) per locality was determined as:$$\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$ \mathrm{F}\mathrm{C} = \left(\mathrm{Number}\ \mathrm{of}\ \mathrm{citations}\ /\mathrm{Total}\ \mathrm{number}\ \mathrm{of}\ \mathrm{citations}\ \mathrm{for}\ \mathrm{all}\ \mathrm{recorded}\ \mathrm{species}\right)\ \mathrm{x}\ 100 $$\end{document}$$ The value of FC obtained directly correlates with the broad use of the plant species \[[@CR14]\]. Results {#Sec10} ======= Profile of the informants {#Sec11} ------------------------- A total of 182 informants from the six localities agreed to participate in this study. Their profile is given in Table [1](#Tab1){ref-type="table"}.Table 1Profile of informants in the study localitiesLocalityNumber of informantsSexAge range (years)Informants with knowledge about repellent plantsMenWomenLolodorf and surroundings5348525-7051Bipindi and surroundings3530525-5525Londji1313060-650Kon-yambetta3229325-7518Dimako and surroundings3636025-5536Babete1321140-6013Total182158 (86.8%)24 (13.2%)143 (78.6%)Informants were recruited following a pre-tested semi-structured questionnaire in all the six study areas. Prior informed consent was obtained verbally from informants before they were interviewed The age of respondents averaged 50 years, 86.8% being men and 13.2% women. Globally, 78.6% of respondents had knowledge about mosquito and/or insect repellent plants. The level of this knowledge varied from one locality to another with 100% in Dimako and Babete, 96.2% in Lolodorf and surroundings, 71.4% in Bipindi and surroundings, and 56.3% in Kon-yambetta. No specific record about mosquito and/or insect repellent plants was obtained from the Kribi-Londji site. Diversity of medicinal plants and knowledge on mosquito repellent plants {#Sec12} ------------------------------------------------------------------------ A total of 16 plant species belonging to 15 genera and 14 families were reported to be commonly used as insect repellents in the targeted communities as shown in Table [2](#Tab2){ref-type="table"}. They consisted of 50% of trees, 31.2% of shrubs, and 18.8% of herbs. Of these, 56.2% were wild while 43.8% were usually cultivated. Lamiaceae, Leguminosae, Musaceae and Poaceae (2 species each) were the most represented families, while the others consisted of a single species each.Table 2Plants commonly used to repel insects in six Cameroonian local communitiesPlant familySpeciesSpecimen numberVernacular nameOriginBotanyPart usedMode of PreparationMode of administrationAnnonaceae*Greenwayodendron suaveolens* Engl. & Diels1879 HNCMpolè (Bakola, Bagyeli)WildTreeBarkSmash fresh barkSkin applicationArecaceae*Elaeis guineensis* Jacq.34163 HNCLélendé (Bakola); Mimbang malendé (Bagyeli); Alen (Ewondo); Zam-na-malendé (Ngumba)CultivatedTreeFlower1. Burn dry flowers\ 2. Burn dry flowers + fresh leaves of *C. odorata.*\ 3. Burn dry flowers + leaves of *S. officinarum* + suspended termitariumSmokesBurseraceae*Canarium schweinfurtii* Engl.16929 SFR/CAMBélè (Ngumba); Ottou (Ewondo); Bologa (Baka); Beul (Bakoum); Waate (Yambeta)WildTreeSapBurn dry or fresh sap/mix ash with body milk/lotion.Smokes/skin application of resulting potionCompositae*Chromolaena odorata* (L.) R.M.King & H.Rob.57408 HNCGomgom (Ngumba)WildHerbLeaf and whole plant1. Burn whole plant\ 2. Burn fresh leaves + dry flowers of *E. guineensis.*SmokesLamiaceae*Ocimum gratissimum* L.5817SFR/CamMassissip (Ngumba)CultivatedHerbLeafSmash leavesSkin application*Premna angolensis* GürkeKinoubeunoubeu (Bafia)WildShrubLeafHang fresh leavesHang inside the houseLeguminosae*Cylicodiscus gabunensis* Harms21574 SRF/CamOkan; Adoum (Ewondo)WildTreeBarkBurn dry bark + dry flowers of *E. guineensis*Smokes*Erythrophleum ivorense* A.Chev.45333 HNCMbamda (Bafia)WildTreeBarkBurn fresh barkSmokesMusaceae*Musa* ^*x*^ *paradisiaca* L.42391 HNCEkouan (Ewondo)CultivatedTreeLeafBurn dry leaves + flowers of *E. guineensis*Smokes*Musa sp.*Odué (Ewondo)CultivatedTreeLeafBurn dry leaves + flowers of *E. guineensis*SmokesPiperaceae*Piper umbellatum* L.42274 HNCNdembèlembè (Baka)WildShrubLeafSmashed fresh leavesSkin applicationPoaceae*Saccharum officinarum* L.25820 SRF/CamNkogo'o (Bakola)CultivatedShrubLeafBurn fresh leaves + dry flowers of *E. guineensis +* suspended termitariumSmokes*Cymbopogon nardus* (L.) RendleCitronella grass (English)CultivatedHerbWhole1. Hang whole plant\ 2. Distillate whole plant for essential oilHang/spray oil inside the houseRutaceae*Citrus limon* (L.) OsbeckNyopiang (Ngumba)CultivatedTreeFruitPressed fruit for juice.Skin applicationSiparunaceae*Glossocalyx longicuspis* Benth.46245 HNCAnyanzoa (Ngumba)WildShrubLeafHang/Crumple fresh leavesHang/spread inside the house.Zingiberaceae*Aframomum alboviolaceum* (Ridl.) K. Schum.34888 HNCNdjii (Baka)WildShrubLeafSmash fresh leavesApply on skinInformation on medicinal plants uses was collected through visiting the field with informants, presenting plants specimens, and/or checklist interviews on medicinal plants used to repel insects, including mosquitoes. The plant parts used as well as the modes of preparation and administration were recorded Overall, the majority of plant species were recorded in Lolodorf and Bipindi. The most cited were *Canarium schweinfurthii* recorded in four localities (Lolodorf, Bipindi, Kon-Yambetta, Dimako), and mentioned by 58 informants, followed by *Elaeis guineensis* cited in three localities (Lolodorf, Bipindi, Kon-Yambetta) and mentioned by 38 informants, *Citrus limon* and *Chromolaena odorata* cited in two localities (Lolodorf, Bipindi) and mentioned by 11 and 16 informants respectively. *Afromomum alboviolaceum* (Dimako), *Cymbopogon nardus* (Babete), *Cylicodiscus gabunensis* (Lolodorf), *Erythrophleum ivorense* (Dimako), *Glossocalyx longicuspis* (Lolodorf), *Musa paradisiaca* (Lolodorf), *Musa sp.* (Lolodorf), *Ocimum gratissimum* (Bipindi), *Piper umbellatum* (Dimako), *Premna angolensis* (Kon-Yambetta), and *Saccharum officinarum* (Lolodorf) were recorded only in one locality, but no clear patterns could be identified to explain this variation (Table [3](#Tab3){ref-type="table"}). This inter-community variation in knowledge and use of insect repellent medicinal plants could result from the fact that ethnobotanical knowledge is not shared among communities. This important gap could also be due to a lack of individual motivation, experience, or curiosity of inhabitants to search beyond the customary habits of the surveyed communities \[[@CR15]\]. The ethnobotanical knowledge appeared to be commonly shared among people from the same setting rather than between communities.Table 3Frequency of Citation (FC in %) of plant species in the survey areasPercent Frequency of Citation (in %)Plant familySpeciesLolodorfBipindiKon-YambettaDimakoLondjiBabeteAnnonaceae*Greenwayodendron suaveolens*012.60000Arecaceae*Elaeis guineensis*25.214.632.5000Burseraceae*Canarium schweinfurthii*32.236.949.834.200Compositae*Chromolaena odorata*8.817.90000Leguminosae*Cylicodiscus gabunensis*2.900000*Erythrophleum ivorense*00011.800Lamiaceae*Ocimum gratissimum*011.20000*Premna angolensis*0017.7000Musaceae*Musa paradisiaca*2.900000*Musa sp.*2.900000Piperaceae*Piper umbellatum*00021.100Poaceae*Saccharum officinarum*8.800000*Cymbopogon nardus*00000100Rutaceae*Citrus limon*8.86.80000Siparunaceae*Glossocalyx longicuspis*7.500000Zingiberaceae*Aframomum alboviolaceum*00032.900The percent frequency of citation was calculated as the number of citations/total number of citations for all recorded species Apart from Londji where no specific repellent plant was reported, the number of plant species used as repellent varied depending on locality. In fact, informants from Londji claimed to burn any plant to repel insect. This is a gap in ethnobotanical knowledge that needed to be spelled out. In Lolodorf and surroundings, 9 plants species were documented (*Elaeis guineensis*, *Chromolaena odorata*, *Canarium schweinfurthii, Cylicodiscus gabunensis*, *Glossocalyx longicuspis*, *Musa paradisiaca*, *Musa sp*., *Saccharum officinarum*, *Citrus limon*). The most used species were *Canarium schweinfurthii* (FC = 32.3%), followed by *Elaeis guineensis* (FC = 25.2%). *Musa paradisiaca*, *Musa sp.* and *Cylicodiscus gabunensis* (FC = 2.9%) were less mentioned as repellent. In Bipindi and surroundings, we recorded 6 species namely *Greenwayodendron suaveolens, Elaeis guineensis, Chromolaena odorata, Canarium schweinfurthii, Ocimum gratissimum*, and *Citrus limon*. Of note, *Canarium schweinfurthii* (FC = 36.9%) and *Chromolaena odorata* (FC = 17.9%) were the most cited species, compared to *Citrus limon* (FC = 6.8%). Three repellent species were reported in Kon-Yambetta, including *Canarium schweinfurthii* (FC = 49.8%), *Elaeis guineensis* (FC = 32.5%) and *Premna angolensis* (FC = 17.7%). In Dimako and surroundings four species were recorded including *Canarium schweinfurthii* (FC = 34.2%), *Afromomum alboviolaceum* (FC = 32.9%), *Piper umbellatum* (FC = 21.1%) and *Erythrophleum ivorense* (FC = 11.8%). Finally, in Babete, only *Cymbopogon nardus* was reported with FC value of 100% (Table [3](#Tab3){ref-type="table"}). Plant parts used in the study sites and their modes of administration {#Sec13} --------------------------------------------------------------------- With respect to the parts of the plants used for repelling insects, results from the present study indicated that the local communities preferentially use the leaf (52.9%), followed by the bark, and whole plant (Fig. [2](#Fig2){ref-type="fig"}). This preference might be based on the fact that leaves are readily available or more likely that their volatile components have the desired activity. Viewing from another angle, this preference might also result from the desire of the communities to preserve the biodiversity on which they depend for food and remedies and from which harvesting plant barks or whole plants should rather contribute to the extinction.Fig. 2Plant parts used to repel insects. The information surveyed indicated the leaf as the preferred plant part used in repelling insects in the study areas From this study, it was recorded that communities adopt many modes of administration of plants products. Amongst those, 50% were reported to be burnt to produce smokes inside the houses, 31.2% mashed and thereafter applied on the body, and 18.8% hung in the houses. Of note, the only record from the Kribi-Londji site indicated the use of burning smokes from any plant to repel insects. Amongst the species reported, some were used in combination (Table [2](#Tab2){ref-type="table"}). The flowers of *Elaeis guineensis* came out to be the primary ingredient to combine with the leaves of *Chromolaena odorata, Saccharum officinarum*, *Musa paradisiaca*, *Musa sp.*, and bark of *Cylicodiscus gabunensis* to afford blended repellent prescriptions. Based on informants' reports, the appropriate plant parts were collected only when needed and at any time. Moreover, no ritual was reported to be observed during plant collection. About the application time, the herbal preparations were preferably applied during sunset. The repellent effect was reported by the informants to be efficient mainly throughout the night. However, the specific dosage of plant materials used to repel insects was not clearly mentioned as they were administered as long as insects' threat was pending. Discussion {#Sec14} ========== This study attempted to record knowledge about plant species used as repellents against insects/mosquitoes in 6 localities of Cameroon to identify promising candidate plants that might be formulated as insect repellents. The results achieved indicated 16 plant species used by the people of Lolodorf, Bipindi, Dimako and their surroundings, and Londji, Kon-Yambetta and Babete of Cameroon to drive off insects, especially mosquitoes. Most of the interviewed informants were men (86.8%) against 13.2% of women, similar to the pattern previously reported by Cheikhyoussef et al. \[[@CR16]\], Bekalo et al. \[[@CR17]\], and Okello and Segawa \[[@CR18]\]. Besides, a meta-analysis on the continental level recently indicated significant differences in the knowledge of men and women in Africa when it comes to ethnobotanical knowledge. Additionally, women in rural Africa frequently collect plants for firewood purpose as one of the domestic activities they are responsible for \[[@CR19]\]. The records showed that the local inhabitants (78.6%) had knowledge about the mosquito-repellent property of plants and their use. Most of the plants recorded (50%) were reported to be burnt to produce smokes inside the house, aligning with the claims of Pålsson and Jaenson \[[@CR20]\] that indicated that burning plants might be effective in repelling insect. Similar results were equally obtained in Ethiopia and Tanzania, supporting this approach across communities as consistent to drive away mosquitoes \[[@CR21], [@CR22]\]. Biran et al, \[[@CR23]\] equally reported mixed evidence especially for using firewood smoke to protect against mosquito bite. Despite the scarcity of data on how repellent smokes or their constituents act, the repellent activity of burned plants appears to be due to the release of specific volatile compounds, either already present in the fresh/dried plant or created during the combustion process \[[@CR23], [@CR24]\]. Such compounds like ß-ocimene have been previously showed to exert mosquito repellent activity \[[@CR24]\]. Moreover, the smoke could act by disguising human kairomone cues targeted by insects and disrupting the convection currents essential for mosquito host location \[[@CR22], [@CR23]\]. Smashed plant samples for topical application (31.2%) and hung plants leaves inside the house (18.8%) were also recorded during our study as important modes of administration. These modes of repellent plants administration have been exploited for thousands of years by man, and are still in wide use throughout the developing countries \[[@CR4]\]. Plants have been used for centuries in the form of crude fumigants where they are burnt to drive away mosquitoes and later on as formulations that are applied to the skin or clothes as firstly recorded in writings by ancient Greek \[[@CR25]\], Roman \[[@CR26]\], and Indian scholars \[[@CR27]\]. Among the repellent plants species recorded in this survey, the main parts used were reported to be the leaf (52.9%) and the bark (17.6%). These findings are consistent with those of Zorloni et al. \[[@CR28]\] who reported plants leaves and bark as predominantly used for tick control in West Ethiopia. The prominent application of leaves might be owing to the ready availability of their active constituents that are more likely volatile compounds. Indeed, plants that are commonly used for repellent properties are mostly those that contain essential oils, and when crushed or brushed against, leaves release strong odours of which some are pleasant, and some not so pleasant to insects. The traditional use of plants or plant products against biting insects is a common practice in Africa \[[@CR29]\]. Their insect-repellent effect observed may be attributed to their chemical composition \[[@CR30]\]. Various active constituents with insects' repellent activity have been previously reported from some of the checked plants species. Apart from *Glossocalyx longicuspis* and *Greenwayodendron suaveolens* for which very little has been reported as repellent properties, the other listed species or related species have previously been reported to display mosquito/insect repellent or insecticidal activities. Within this scope, the insecticidal activity of the bark extract of *Erythrophleum ivorense* was previously reported in the Ashanti region of Ghana \[[@CR31]\]. Besides*, Erythrophleum ivorense* is resistant to fungi, dry wood borers and termites \[[@CR32]\]. This denotes repellency/insecticidal properties that might be explained by the presence of pharmacologically active alkaloids in the bark and seed such as cassaine, cassaidine and erythrophleguine. However, it should be noted that high doses of the bark extract are extremely strong, rapid-acting cardiac poison in warm-blooded animals causing shortness of breath, seizures and cardiac arrest in a few minutes. Furthermore, the seeds are reported to be more toxic due to a strong haemolytic saponin which acts synergistically with the alkaloids \[[@CR32]\]. Fresh bark of this plant was reported to be burnt by Mbamda (Bafia) people to repel in-house mosquitoes. Given the presence of toxic alkaloids in the bark, the resulting smokes are highly likely to be equally poisonous to insects and human, stressing the fact that it should be used with caution or simply discontinued. *Canarium schweinfurthii* (Burseraceae), commonly known as African elemi or canarium, is a species of large tree native to tropical Africa*.* The African elemi tree is one of several sources of the economically useful oleoresin known elemi. In West Africa this resin is traditionally burned for fumigating dwellings and mixed with oil for body paint \[[@CR33]\], and might likely exhibit insect repellent activity. *C. schweinfurthii* was the more cited by informants (FC \> 32 when recorded) and was also previously reported to have insecticidal activity against *Callosobruchus maculates,* a major pest of cowpeas, green gram and lentils \[[@CR34]\]. The authors assumed that this activity might be due to the presence of saponnins that have otherwise been found to affect the respiratory system of insect and to cause emetic effect by their detergent action. In addition, tannic acid found in this plant species was reported by David \[[@CR35]\] to act as toxin and feeding deterrent to insects. *Chromolaena odorata* (Compositae) is a tropical and subtropical species of flowering shrub in the sunflower family. It is native to North America, from Florida and Texas to Mexico and the Caribbean \[[@CR36]\], and has been introduced to South America, tropical Asia, West Africa, and parts of Australia \[[@CR37], [@CR38]\]. It is sometimes grown as a medicinal and ornamental plant. It is used as a traditional medicine in Indonesia, Thailand, Malaysia and parts of Africa. In traditional medicine of Thailand, the plant is used for the treatment of wounds, rashes, diabetes, and as insect repellent. It has antifungal and antibacterial properties. It has previously been reported to have insecticidal properties against adult stage of *Periplaneta americana,* an omnivorous and opportunistic feeder which is the largest common species of pest cockroach especially in the tropics and subtropics \[[@CR39]\]. Moreover, a survey undertaken in the Ashanti Region of Ghana revealed the use of *C. odorata* leaf as repellent with insecticidal properties \[[@CR31]\]. *C. odorata* leaf and root have been shown to contain alkaloids, phenols, flavonoids, saponins, cardenolides, anthraquinones and tannins \[[@CR40]\] that might elicit the insecticidal activity. Furthermore, this plant contains carcinogenic pyrrolizidine alkaloids that are toxic to cattle and can also cause allergic reactions \[[@CR38], [@CR41]\]. Previous findings also indicate that this class of alkaloids (retronecine type alkaloids) can elicit insecticidal activity \[[@CR42]\]. This activity could also be linked to constituents such as flavonoids which are a class of phenolic compounds occurring naturally in this plant \[[@CR43]\] and that have anti-feeding and attracting deterrent properties, thereby exerting toxic effects to insects, fungi, bacteria, nematodes and weeds \[[@CR40]\]. *Citrus limon*, otherwise called lemon, is a small tree of the Rutaceae family that originated in Asia and is now grown commercially worldwide in tropical, semi-tropical, and warm temperate countries for the fruit, which is used fresh and in beverages and cooking, and is also used as a preservative due to its anti-oxidant properties. Lemon oil, obtained from the peel, is used as a wood cleaner and polish, and as a non-toxic pesticide. Traditional medicinal uses for the fruit, peels, and oil obtained from the seeds include treating fever and colic, and as an astringent and diuretic \[[@CR44]\]. The topical repellent property of extracts from peels of *C. limon* against mosquitoes was previously reported by Effiom et al. \[[@CR45]\]. In ancient medicine, Lemon citrus (*C. limon*) has long been used as natural insect repellent. Moreover, *C. limon* essential oil showed to be effective against mosquito larva, and to be also repellent against malaria vector, *Anopheles stephensi* in laboratory animal and human \[[@CR46]\]. Kazembe and Chaibva \[[@CR47]\] also showed that the whole extract of fruit peel and volatile oils had mosquito repellency against *Aedes aegypti*. On the specific compositional scale, citronellol, the most prominent component of *C. limon* essential oil, and linalool have been shown to be the main active ingredients of lemon in the distillate, and have also been identified among the main active ingredients of other botanical repellents such as citrosa and eucalyptus \[[@CR46]\]. *Cymbopogon* (Poaceae), better known as lemongrass, is a genus of Asian, African, Australian, and Tropical Island plants in the grass family \[[@CR48], [@CR49]\]. Some species (particularly *C. citratus*) are commonly cultivated as culinary and medicinal herbs because of their scent, resembling that of lemons (*Citrus limon*). Common names include lemon grass, lemongrass, barbed wire grass, silky heads, citronella grass, fever grass, amongst many others. Research has shown that lemongrass oil has antifungal properties \[[@CR50]\], but also the ability to repel some insects, such as mosquitoes. However, its oil is commonly used as a kind of trap to attract honey bees, working conveniently as honeybee\'s attractant pheromones (Beekeeping/Guide to Essential Oils). Essential oils from lemon grass principally contain geraniol and citronellol that are antiseptics, hence their use in household disinfectants and soaps. *Cymbopogon* spp. have a long history of use to repel mosquitoes and are as effective as the chemical insect repellent N,N-diethyl-meta- toluamide (DEET) \[[@CR51]\]. *C. nardus* cultivation was introduced by Religious nuns at the Monastery Saint-Benoit in Mbouda-Cameroon and is used as mosquito repellent. Historically, it was used by the Indian Army to repel mosquitoes at the beginning of the 20^th^ century \[[@CR52]\]. Essential oils and extracts from *C. nardus* and globally from the *Citronella* genus are commonly used as ingredients of plant-based mosquito repellents in commercial preparations \[[@CR53]\]. Moreover, attempts to scientifically demonstrate the repellent efficacy of Citronella grass essential oils and formulations against mosquitoes and arthropods have been previously undertaken \[[@CR53], [@CR54]\]. Trongtokit et al. \[[@CR55]\] also demonstrated that 100% essential oil of Citronella protects against *Aedes aegypti*, *Culex quinquefasciatus* and *Anopheles dirus* and related this activity to the presence of citronellol and related compounds. *Saccharum officinarum* (Poaceae) or sugarcane, is a large, strong-growing species of grass in the genus *Saccharum*. It is cultivated in tropical and subtropical countries worldwide for the production of sugar and other products. *S. officinarum* has been shown to act as repellent, toxicant, and anti-feeding against a number of Coleoptera that attack stored food crops. The insecticidal property of *S. officinarum* bagasse-based lignin may be due to the presence of phenolic and alcoholic compounds \[[@CR56]\]. Also, the efficacy of application of the stem juice of this plant on the skin as traditional personal protection method was evaluated against mosquito bites and general nuisance in Bolifamba, a rural setting of the Mount Cameroon area. The field evaluation of the skin application of the juice showed significant protection against mosquito bites though, less than the commonly used commercial diethyltoluamide \[[@CR5]\]. *Ocimum gratissimum*, (Lamiaceae) commonly called clove basil, African basil is a native species of Africa, southern Asia, and the Bismarck Archipelago \[[@CR57]\]. *O. gratissimum* is a common culinary herb in West Africa. Its essential oil contains eugenol and shows some evidence of antibacterial activity \[[@CR58], [@CR59]\]. This compound has otherwise showed anti-insect activity, therefore supporting the use of the plant as insect repellent. The authors alluded that this activity of eugenol was dependent on the structure of its phenolic hydroxyl \[[@CR60]\]. The essential oil has potential for use as a food preservative \[[@CR61]\]. An ethnobotanical investigation carried out in Bamenda, Cameroon reported *O. gratissimum* to be used as insect repellent \[[@CR62]\]. This mosquito repellent plant is cultivated around houses for such purpose \[[@CR63]\]. Also, people from many parts of Tanzania burn this plant to release smokes or hang it in houses to drive mosquitoes away \[[@CR64]\]. Moreover, many other reports describe the insect repellent and insecticidal properties of *O. gratissimum* in field and laboratory trials, particularly against the main vectors of malaria and lymphatic filariasis \[[@CR63], [@CR65]--[@CR67]\]. As asserted above, other studies also suggested that the insecticidal activities might be due to the presence of eugenol found in *O. gratissimum* \[[@CR68], [@CR69]\]. *Musa paradisiaca L.* and *Musa sapientum L.* (Musaceae) are mainly grown in the tropical and subtropical countries and are widely used for their nutritional values all over the world. The fruits as well as the other parts of the plant are used to treat different diseases in human in traditional medicines \[[@CR70]\]. Our records indicated that *M. paradisiaca* and *Musa sp*. are used as insect repellents. This information was corroborated by previous reports showing that the dried leaf and stem of *M. paradisiaca* are burnt by the Ayta people of Porac, Pampanga province, Philippines, to drive off insects especially mosquitoes. In addition, laboratory investigations revealed the remarkable mosquitocidal activity of the petroleum ether root extract of this plant against *Ae. aegypti*, *An. stephensi* and *Cx. Quinquefasciatus* \[[@CR71]\]. As well, 10% (w/v) concentration of *M. paradisiaca* leaf extract showed repellent protection of *Pterygota excelsa* wood against termite, *Odontotermes obesus* \[[@CR72], [@CR73]\]. Also, this extract exerted insecticidal activity against malaria vector *Anopheles stephensi* with 90% lethal effect after 24 h exposure \[[@CR74]\]. Besides, the phytochemical studies of many parts of *M. paradisiaca* and *M. sapientum*, including leaf, fruit, peeled fruit, fruit pulp, fruit peel, flower, bracts, and scape previously revealed the presence of many chemical classes of components such as anthocyanins, Catecholamines, tryptophan, indole compounds, pectin, flavonoids and related compounds (Leucocyanidin, quercetin and its 3-Ogalactoside, 3-O-glucoside, and 3-O-rhamnosyl glucoside), tannins, Acyl steryl glycosides such as sitoindoside-I, sitoindoside-II, sitoindoside-III, sitoindoside-IV and steryl glycosides such as sitosterol gentiobioside, sitosterol myo-inosityl- β-D-glucoside, triterpenes such as cyclomusalenol, cyclomusalenone, 24- methylenecycloartanol, stigmast-7-methylenecycloartanol, stigmast -7-en-3-ol, lanosterol and β -amyrin \[[@CR70], [@CR75]\]. Among all these chemical classes, there are some that might elicit biological activities sustaining the repellent feature of *Musa* sp. For example, compounds such as anthocyanins have been suspected to act in a vast array of plant/animal interactions, including attraction of pollinators and frugivores, as well as the repellence of herbivores and parasites \[[@CR76]\]. Also, saponins that derive from a sugar moiety glycosidically linked to a hydrophobic aglycone which may be a triterpene or a steroid have been shown to trigger plant resistance against insects \[[@CR77]\]. The mechanisms underlying the action of these compounds against insects are based on their various biological properties. They have membrane-permeabilising, haemolytic, antioxidant, anti-inflammatory, immunostimulant and anticarcinogenic activities, and can affect feed intake, growth and reproduction in animals, and can be used as fungicides, molluscicides and pesticides, as well as against some bacteria and viruses. They can also increase mortality levels, lower food intake, weight reduction, retardation and disturbances in development, and decrease reproduction in insects \[[@CR78]\]. The main hypotheses are therefore that saponins could either make the food less attractive to eat (repellent/deterrent activity), bear digestive problems, causing moulting defects or having toxic effects on insects. Another class of compounds such as flavonoids are small molecular secondary metabolites synthesized by plants with various biological activities. Due to their physical and biochemical properties, they are capable of participating in interactions with other organisms. Both flavonoids and isoflavonoids protect plants against insect pests by influencing their behavior, growth and development \[[@CR79], [@CR80]\]. In this line, Naringenin, hesperetin-7-O-rutinoside and quercetin-3-O-rutinoside were reported to stimulate oviposition in swallowtail butterfly Papilio on young leaves of citrus plants \[[@CR81]\]. Also, Chrysin, Kampferol, and 3,7- dimethylether quercetin were found to exert remarkable repellent action against house flies (*Musca domestica*) \[[@CR82]\]. Another interesting example is the tannins that were assessed as repellents in other studies. Tannins have a strong deleterious effect on phytophagous insects and affect the insect growth and development by binding to the proteins, reduce nutrient absorption efficiency, and cause midgut lesions \[[@CR83]\]. *Elaeis guineensis* (Arecaceae) commonly called African oil palm or macaw-fat is the principal source of palm oil. It is native to west and southwest Africa, specifically the area between Angola and the Gambia. An innovative phytodrug (API-PALU) was recently formulated from the crude alkaloids extract of this plant and is extensively used in West Africa to treat malaria (http://www.africanews.com/2016/06/24/beninese-wins-100000-for-innovation-of-anti-malaria-drug/). The innovator recently won many prizes including the 1^st^ IPA 2016 prize (http://www.africanews.com/2016/06/24/beninese-wins-100000-for-innovation-of-anti-malaria-drug/). *E. guineensis* was one of the most cited species recorded in our survey and has been reported in Africa as reducing mosquito biting activity when used as repellent \[[@CR84]\]. In fact, smokes from burned infructescences of *E. guineensis* were reported to reduce the numbers of mosquitoes indoors at night. Also, a field experiment using these smokes showed 69% repellent activity \[[@CR20]\]. As well, *E. guineensis* palm nut (lotions and creams) and oil were shown to reduce significantly the number of bites by *Simulium damnosum* and *Anopheles gambiae* \[[@CR85], [@CR86]\]. Moreover, flowers of *E. guineensis* were recorded in this study to be the primary ingredient to combine with the leaves of *Chromolaena odorata, Saccharum officinarum, Musa paradisiaca, Musa sp.,* and the bark of *Cylicodiscus gabunensis* to repel insects and mosquitoes. The phytochemical composition of the leaf and flower of *E. guineensis* could justify the repellent activity elicited by this plant. Indeed, a recent study recently revealed the presence of phenolic compounds, flavonoids, tannins, coumarins, alkaloids, saponins, terpenoids and steroids, and carbohydrates in the leaf of *E. guineensis* \[[@CR87]\]. Another previous study identified p-methoxyallylbenzene (estragole) as the predominant (\~95%) constituent of essential oils from both male and female flowers of *E. guineensis* \[[@CR88]\]. Many of the secondary metabolites classes identified in the leaf have already been reported as having deleterious effect on insects \[[@CR40], [@CR42], [@CR43], [@CR56], [@CR77], [@CR79]--[@CR81]\]. From recent studies, it was demonstrated that the main component of *E. guineensis* flowers essential oils (estragole) has strong repellent effect to three of the major grain pest insect species, *viz. Rhyzopertha dominica, Sitophilus zeamais,* and *Tribolium confusum.* Interestingly, estragole and (E)-anethole (a related phenolic compound also found in essential oils) showed a strong synergistic co-repellent effect against *R. dominica* \[[@CR89]\]. These latter findings further emphasize the potential of components of single and/or mixed plants to act synergistically to repel insects. *Cylicodiscus gabunensis* (Leguminosae) is a large tree, common in the rainforests of West and Central Africa. The stem bark is used to treat jaundice and malaria among other diseases. It contains triterpene saponins, cylicodiscus acid, a dihydroxy-pentacyclic triterpene carboxylic acid and cyclodione, a dimeric diterpene \[[@CR90]\]. There are however, no references in the literature describing the repellent application of the stem bark of this plant despite its strong odour \[[@CR91]\]. Nevertheless, *C. gabunensis* wood is very resistant to worm and insect attack \[[@CR92]\]. However, the strong odour exhaled by *C. gabunensis* stem bark is likely elicited by its terpenoids content, and might otherwise justify its repellent effect against insects. Our study also reported *Piper umbellatum* (Piperaceae) to be used as mosquitoes/insects repellent. It is commonly known as English cow-foot leaf (Sierra Leone), *Fula-Pulaar* (Guinea); Poponidagui *(*Sierra Leone). It is an upright shrub to 2 m high in moist shady places occurring from Guinea to Cameroon, and widespread throughout the tropics where it is used as condiment, and also considered a fetish as well as a medicine to treat a vast array of ailments including but not limited to pain, arthritis, rheumatism, fever, diarrhoea, dysentery, venereal diseases, hemorrhoids \[[@CR93]\]. Previous studies have reported the repellent and insecticidal activities of the essential oil of *P. umbellatum* when investigated against grain storage pest insects, bean weevil (*Callosbruchus maculatus*) and rice weevil (*Sitophilus oryzae*), suggesting its suitability for insect pest control \[[@CR31], [@CR94]\]. As well, when *P. umbellatum* fresh leaves are crushed and rubbed on the skin there is an effective though transient (1-2 h) repellent effect, particularly against mosquitos. The active agent was found to be the essential oil consisting of aldehydes, ketones, and phenols in addition to the principal constituents, *viz.* cadinene, caryophyllene, and phallandrene. The repellent effect may be prolonged by mixing the crushed leaves with oils or glycerin-alcohol \[[@CR94], [@CR95]\]. On another hand, extracts from closely related species *Piper nigrum*, *Piper guineense*, and *Piper tuberculatum* contain isobutyl amides, and other plant secondary metabolites that act as neurotoxins and showed repellent activity against insects \[[@CR96]\]. *Premna angolensis* (Lamiaceae) has also been reported to have repellency potential \[[@CR97]\]. The various *Premna* species are well known for their medicinal properties and their further phytochemical investigations resulted in the isolation of secondary metabolites including iridoids and their glycosides, diterpenoids, sesquiterpenoids, triterpenoids, flavonoids, isoflavones, lignans, xanthones and other classes of compounds \[[@CR97]\] that might be involved in the repellency properties. Many plants from Lamiaceae family have been found to be effective against a variety of mosquito vectors. For instance, the crude aqueous, chloroform and methanol extracts and essential oils from the leaves of *P. latifolia* showed mosquito larvicidal efficacy against the fourth instar larvae of *Aedes albopictus* Skuse (Diptera: Culicidae) \[[@CR98]\]. *P. angolensis* and *P. quadrifolia* leaves are burned and used as fumigant in the attics of cereals against pests. Besides, their essential oils displayed insecticidal and repellent effects against *Sitotroga cerealella*, an insect pest of rice stocks \[[@CR99]\]. Conclusion {#Sec15} ========== This study that aimed at documenting insects repellent plant species used by indigenous populations of 6 localities of East, South, West and Centre regions of Cameroon has successfully identified indigenous knowledge on insects' repellent plants. There was consistency in documented information with previous reports for most of the plants cited, and the results achieved have potential as baseline for further scientific investigation to strengthen the concept of plant-based mosquito repellents. These findings will also provide validation to the ethnobotanical knowledge of the targeted communities. Besides, it was recorded that the bark of *Erythrophleum ivorense* that contains toxic alkaloids is burnt to repel mosquitoes in houses. Given its toxicity, this plant should be used with caution or simply discontinued. Further laboratory investigations are ongoing and will contribute to establish the cidal or static effects of extracts from selected plants against mosquitoes in particular and insect pests in general. Additional file {#Sec16} =============== Additional file 1:Ethnobotanical survey of insect/mosquito repellent plants. Interview respondents were identified and further questioned face-to-face using a semi-structured questionnaire. Responses to all questions were recorded following a sequential guideline. (DOC 53 kb) FC : Frequency of citation **Electronic supplementary material** The online version of this article (doi:10.1186/s13002-017-0155-x) contains supplementary material, which is available to authorized users. The authors acknowledge the invaluable collaboration of populations and herbalists from the study areas: Dimako (East region), Kon-Yambetta (Centre region), Lolodorf, Bipindi, and Kribi-Londji (South region), and Mbouda-Babete (West region). Funding {#FPar1} ======= This research received logistic support from the Laboratory for Phytobiochemistry and Medicinal Plants Studies, University of Yaoundé 1, Cameroon. Availability of data and materials {#FPar2} ================================== Supplementary material: Semi-structure questionnaire. Authors' contributions {#FPar3} ====================== The study was designed and supervised by FFB; RDFY, PVTF, EZM, RK, IB-V, and VN carried out the survey and collected the data during the study; VN identified the plants (species and family), collected the plants samples, and archived the voucher specimens at the National Herbarium of Cameroon, Yaoundé. FFB, RDFY, PVTF, EZM, RK, and IB-V contributed to the drafting and revision of the manuscript and have agreed to its submission to the Journal of Etnobiology and Ethnomedicine. All authors read and approved the final manuscript. Competing interests {#FPar4} =================== The authors declare that they have no competing interest. Consent for publication {#FPar5} ======================= Not applicable. Ethics approval and consent to participate {#FPar6} ========================================== This ethnobotanical survey was performed according to the current legislation and the status of the biodiversity rights of rural communities in Cameroon \[[@CR100]\] and the provisions of the United Nations Framework Convention on Biodiversity, Brazil in 1992. Prior informed consent to record and publish their traditional knowledge was obtained verbally from informants before they were interviewed. Informants were free to opt out of the study at any time without any consequences. Publisher's Note {#FPar7} ================ Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

1. Introduction {#sec1-ijms-18-01408} =============== Kidney disease associated with diabetes is the leading cause of chronic kidney disease (CKD) and end-stage kidney disease worldwide and nearly one-third of patients with diabetes develop nephropathy \[[@B1-ijms-18-01408]\]. As the incidence of both types 1 and 2 diabetes rises worldwide, diabetic nephropathy (DN) is likely become a significant health and economic burden for society \[[@B2-ijms-18-01408]\]. Current therapy for diabetic nephropathy includes glycemic optimization using antidiabetics and blood pressure control with blockade of the renin-angiotensin system \[[@B3-ijms-18-01408]\]. However, these strategies are slow but cannot reverse or at least stop the disease progression \[[@B4-ijms-18-01408]\]. Although several clinical trials are currently in progress, there are still no drugs approved for the treatment of DN. Among these ongoing phase 3 clinical trials, atrasentan is still in progress, while bardoxolone methyl and paricalcitol failed to meet the primary endpoint or was terminated on safety concerns \[[@B4-ijms-18-01408],[@B5-ijms-18-01408]\]. Recently, there has been renewed interest in the development of direct β1-selective Adenosine monophosphate-activated protein kinase (AMPK) activators that have the potential to treat diabetic nephropathy \[[@B6-ijms-18-01408]\]. AMPK is master sensor of cellular energy and plays a critical role in the regulation of metabolic homeostasis \[[@B7-ijms-18-01408]\]. AMPK is a heterotrimeric kinase comprised of a highly conserved catalytic α subunit and two regulatory subunits (β and γ) \[[@B8-ijms-18-01408]\]. The α subunit possess a N-terminal serine/threonine catalytic kinase domain (KD) that is followed by an autoinhibitory domain (AID) and a C-terminal β subunit-binding domain \[[@B9-ijms-18-01408]\]. The β subunit serves as a scaffold to bridge α and γ subunits that contains a glycogen binding domain (GBD) and a C-terminal domain \[[@B10-ijms-18-01408]\]. The γ subunit is composed of a β subunit-binding region and two Bateman domains \[[@B11-ijms-18-01408]\]. These seven subunits (α1, α2, β1, β2, γ1, γ2, and γ3) are encoded by separate genes, resulting in 12 different αβγ AMPK heterotrimers \[[@B12-ijms-18-01408]\]. The distinct physiological functions of each AMPK isoforms are not fully understood, but derive from differential expression patterns among different tissues \[[@B13-ijms-18-01408]\]. For instance, the α1 subunit appears to be relatively evenly expressed in kidney, rat heart, liver, brain, lung and skeletal muscle tissues, while the α2 subunit is mainly expressed in skeletal muscle, heart, and liver tissues \[[@B14-ijms-18-01408]\]. Among the two known β subunits, β1 subunit is highly abundant in kidney as suggested by mRNA levels \[[@B6-ijms-18-01408]\]. More recently, a direct AMPK activator PF-06409577 was reported to activate α1β1γ1 and α2β1γ1 AMPK isoforms with EC~50~ of 7.0 nM and 6.8 nM but was much less active against α1β2γ1/α2β2γ1/α2β2γ3 AMPK isoforms with EC50 greater 4000 nM \[[@B6-ijms-18-01408]\]. Besides, compound PF-06409577 exhibited efficacy in a preclinical model of diabetic nephropathy. Compounds A-769662 and 991 possessed similar potency toward AMPK heterotrimers containing a β1 subunit as PF-96409577 \[[@B15-ijms-18-01408]\]. On the other hand, an allosteric site of AMPK has been named allosteric drug and metabolite site (ADaM site) \[[@B16-ijms-18-01408]\], which was constructed by the catalytic kinase domain (KD) of α subunit and the regulatory carbohydrate-binding module (CBM) of β subunit \[[@B13-ijms-18-01408],[@B17-ijms-18-01408]\]. The three known direct AMPK activators (PF-06409577 \[[@B6-ijms-18-01408]\], A-769662 \[[@B18-ijms-18-01408]\], and 991 \[[@B19-ijms-18-01408]\], [Figure 1](#ijms-18-01408-f001){ref-type="fig"}) all bound to the allosteric site and showed better potency for isoforms that contain the β1 subunit. This implies that the allosteric site can be used to design the selective activators of AMPK containing the β1 subunit. The present study aims to investigate details of the domain structure and identify new potential β1-selective AMPK activators. Hence, sequence alignment and structural comparison were used to identify the key amino acids that are involved in the interaction with activators and structure difference between different subunits. Furthermore, molecular docking was performed for virtual screening to discover direct β1-selective AMPK activators. The screened retrieved hits were then subjected to several filters such as estimated activity and quantitative estimation of drug-likeness (QED) \[[@B20-ijms-18-01408],[@B21-ijms-18-01408]\]. Finally, 12 compounds with diverse scaffolds were selected as potential hit compounds for the design of novel β1-selective AMPK activators. These findings provided a useful molecular basis for the design and development of novel β1-selective AMPK activators. 2. Results and Discussion {#sec2-ijms-18-01408} ========================= 2.1. Sequence Alignment and Structural Comparison {#sec2dot1-ijms-18-01408} ------------------------------------------------- To reveal the possible molecular mechanism for the selective potency of activators against the β1-isoform of AMPK, sequence and secondary structure elements comparison between carbohydrate-binding module of β1 and β2 subunits were investigated. As shown in [Figure 2](#ijms-18-01408-f002){ref-type="fig"}, the sequences that were boxed blue were located within the range of 5 Å of active site. Sequence alignment reveals that β1 and β2 subunits shares 77.1% sequence identity. As shown in [Figure 3](#ijms-18-01408-f003){ref-type="fig"}, superposition with the two subunits reveals a deflexion of sheet1 in β2 subunit as compared with β1 subunit. The Phe-82 of β1 subunit corresponded to Ile-81 in β2 subunit, as well as the Thr-85 to Ser-84, Gly-86 to Glu-85, which may account for the deflexion of sheet1 in β2 subunit. The large aromatic Phe residues and small Thr and Gly presented a binding surface more capable of accommodating ligand. The sheet 2 torsion may attribute to the amino acid sequence differences of the sites of 106 and 107. The most notable is supposed to the Ser108 (red and stick), an autophosphorylation site, phosphorylated serine (pSer108) formed hydrogen bonds with Thr-21, Lys-29, Lys-31, His-109′, and Asn-111′ enhancing the ADaM site stabilization \[[@B22-ijms-18-01408]\], and the phosphate group contributed to the binding of activators \[[@B23-ijms-18-01408]\]. The Gln-109′ and Asn-111′ were mutated to His-109′ and Asp-111′, which abolished original hydrogen bonds and generated a large conformational change. We speculated that the above differences between β1 and β2 may affect the binding of activators to AMPk isoforms. 2.2. Parameter Setting for Molecular Docking {#sec2dot2-ijms-18-01408} -------------------------------------------- Docking parameters, which exert an important influence on molecular docking-based virtual screening, were optimized in advance. The crystal structure of PF-06409577 bound to the α1β1γ1 AMPK isoform (PDB ID: 5KQ5) and A-769662 bound to the α2β1γ1 AMPK isoform (PDB ID: 4CFF) were chosen as the reference, the docking parameters were adjusted until the docked poses were as close as possible to the original crystallized structures. The ring flexibility was mainly considered in final optimized docking parameters according to the default settings. The overlay of the original ligand from X-ray crystal (stick and magenta) and the conformation from Surflex-Dock results (stick and green) were shown in [Figure 4](#ijms-18-01408-f004){ref-type="fig"}, in which the indole moiety of PF-06409577 and terminal benzene ring of A-769662 generated a little deflection and there was no effect on the interaction between compounds and the active site. The hydrogen bond interactions appeared consistent with the original ligands and the root mean square deviation (RMSD) between these two conformations are 0.53 and 0.56 Å, respectively. The molecular docking results indicated that the Surflex-Dock was reliable and could be used for the further virtual screening. 2.3. High-Throughput Virtual Screening Procedure {#sec2dot3-ijms-18-01408} ------------------------------------------------ To identify new potent activators of AMPK, virtual screening was performed on the active sites as mentioned previously. A chemical library containing with 1,500,000 commercially available compounds (ChemDiv database) was docked to the molecular models of α1β1γ1 and α2β1γ1 AMPK isoforms in silico, respectively. Prior to docking, the ChemDiv database was split into eight subsets for molecular docking. About 600 top ranked compounds with high total-scores were screened and subsequently checked for their binding modes and interactions with the active site, especially the hydrogen bonds formed with the residuals of Asp-88, Lys-29, Lys-31, and Gly-19. Then the potential hit compounds were evaluated for their drug-likeness model scores using Lipinski's rule of five ([Table 1](#ijms-18-01408-t001){ref-type="table"}). Finally, six potential hits with new scaffolds could serve as activators for α1β1γ1 AMPK isoform and six for α2β1γ1 AMPK isoform were visually chosen from the top potential hits. 2.4. Analysis of Binding Mode of Activators for α1β1γ1 AMPK Isoform {#sec2dot4-ijms-18-01408} ------------------------------------------------------------------- The structures of retrieved hits as activators of α1β1γ1 AMPK isoform are shown in [Figure 5](#ijms-18-01408-f005){ref-type="fig"}. Although these compounds possess different chemical scaffolds, they exhibit similar binding modes at the active site. Among these compounds, compounds F064-1335 and M5653-1884 possess higher docking scores, compounds M8006-4303 and F264-3019 have perfect drug-likeness model scores. As shown in [Figure 6](#ijms-18-01408-f006){ref-type="fig"}A, the compound F064-1335 with the highest docking score (10.50) formed several hydrogen bonds with active site residues. The two oxygen atoms of sulfonamide established a hydrogen bond network with the side chain of Lys-31, Lys-29, and Asn-111′. The carbonyl oxygen atom of the ester group formed two hydrogen bonds with the main chain of Lys-29, the anther oxygen atom of the ester group was bound to the main chain of Asn-48 by a hydrogen bond, which made the alkoxy group trend into a hydrophobic pocket formed by the Lys-51, Ile-52, Val-62, and Leu-47. In addition, the carbonyl oxygen of benzoxazolone ring formed a hydrogen bond with the side chain of Arg-83′. The compound M5653-1884 with a considerable docking score (10.24) and the bind mode is shown in [Figure 6](#ijms-18-01408-f006){ref-type="fig"}B. Four hydrogen bonds were observed between the compound and the active site residues. One carbonyl oxygen atom of 1,3-indandione formed hydrogen bond with the side chain of Lys-31, another carbonyl oxygen atom formed a hydrogen bond with the main chain of Val-11. The carbonyl oxygen atom of the amide group showed hydrogen bond interactions with the side chain of Lys-29 and Asn-111. In addition, there was a hydrophobic effect with the side chain of Ile-46, Asn-48, Asp-88, and Phe-88. As shown in [Figure 6](#ijms-18-01408-f006){ref-type="fig"}C, the compound of M8006-4303 exhibited similar binding mode as PF-06409577. The ethanol group attached to the piperazine group participated in two hydrogen bond interactions with the side chain of Gly-19 and Lys-31. The carbonyl oxygen atoms of pyrrolidine-2,5-dione formed a hydrogen bond interaction with the side chain of Lys-29. In addition, the oxygen atom of oxygen butyl associated with the benzene ring accepted a hydrogen bond from the main chain of Asn-48. Within the cavity of the active site, Ile-47, Asn-48, Lys-51, and Ile-52 probably generated a hydrophobic effect. The binding mode of compound F264-3091 with a prefect drug-likeness score (1.00) was shown in [Figure 6](#ijms-18-01408-f006){ref-type="fig"}D. The oxgen atom of an oxyethyl group on the benzene ring participated in a hydrogen bond with the main chain of Val-11. The carbonyl oxygen atom of the amide group showed two hydrogen bonds with the main chain of Gln-19 and side chain of Lys-31. In addition, one hydrogen bond was formed between the side chain of Asn-48 and the oxyethyl group connected with flavone B-ring while the B-ring showed a stacked cation-π interaction with the side chain of Val-83′. 2.5. Analysis of Binding Mode of Activators for α2β1γ1 AMPK Isoform {#sec2dot5-ijms-18-01408} ------------------------------------------------------------------- The chemical structures of six compounds as activators of α2β1γ1 AMPK isoform are shown in [Figure 7](#ijms-18-01408-f007){ref-type="fig"}. The molecular docking results indicated that all the compounds possess higher docking scores than A-769662 and 991. The binding modes of the representative compound M2958-7438 and M5050-0116 in the active site of α2β1γ1 AMPK isoform are shown in [Figure 8](#ijms-18-01408-f008){ref-type="fig"}. As shown in [Figure 8](#ijms-18-01408-f008){ref-type="fig"}A, six hydrogen bonds were formed between the compound M2958-7438 and active site residues, in which the barbituric acid ring formed three hydrogen bonds with the side chain of Asp-88, making prominent contributions to the high docking score (10.04). The oxygen atom of the anisole associated with the barbituric acid ring accepted a hydrogen bond from the side chain of Lys-29, and two oxygen atoms in the linker participated in two hydrogen bonds with the side chain of Lys-31. In addition, the barbituric acid ring generated a stacked cation-π interaction with the side chain of Arg-83′. The compound M5050-0116 with a docking score of 9.27 and formed four hydrogen bonds with Val-11, Leu-18, Lys-29, and Asn-111′. As shown in [Figure 8](#ijms-18-01408-f008){ref-type="fig"}B, the Lys-29 of α subunit and Asn-111′ of β subunit, simultaneously coordinated the oxygen atom of dibenzofuran with hydrogen bonds. The hydroxyl group attached on pyrimidine anchored in a suitable geometry and formed two hydrogen bonds with the main chain of Glu-19 and the side chain of Val-11. Additionally, the compound M5050-0116 exhibited hydrophobic interactions with several residues, which formed a hydrophobic pocket including Ile-46, Leu-47, Asn-48, Asp-88, and Phe-90. 2.6. Biological Activities {#sec2dot6-ijms-18-01408} -------------------------- The six screened compounds based on α1β1γ1 AMPK isoform were evaluated for activities against AMPK (α1β1γ1 isoform) at a dosages of 2 µM. A-769662, a known β1-selective AMPK activator, was used as a control. The preliminary in vitro assay ([Figure 9](#ijms-18-01408-f009){ref-type="fig"}) indicated that most of the selected α1β1γ1 AMPK activators displayed promising activation potency against α1β1γ1 AMPK isoform. Compounds D454-0135 and F264-3019 displayed comparable activation activity against AMPK in the comparison with the known β1-selective AMPK activator A-769662. Morever, compounds M8006-4303 and F264-3019 showed stronger activation activities against α1β1γ1 AMPK than A-769662. Compound M563-1884 with a higher logP value and showed a relatively low activation activity among the assayed compounds. This implies that the lipophilicity may play an important role in the bioavailability for the compound. 3. Materials and Methods {#sec3-ijms-18-01408} ======================== 3.1. Sequence Alignment and Structural Comparison {#sec3dot1-ijms-18-01408} ------------------------------------------------- Sequence alignment is an essential method for similarity/dissimilarity analysis of protein, DNA, or RNA sequences \[[@B24-ijms-18-01408]\]. The software used for sequence alignment tasks include HAlign, BioEdit, EMBL-EBI, T-Coffee, and CLUSTAL \[[@B25-ijms-18-01408],[@B26-ijms-18-01408]\]. The crystal structure data of AMPK (α1β1γ1: 5KQ5, 4QFG; α1β2γ1: 4REW and α2β1γ1: 4CFF) were obtained from RCSB Protein Data Bank \[[@B6-ijms-18-01408],[@B8-ijms-18-01408],[@B13-ijms-18-01408],[@B19-ijms-18-01408]\], as well as the amino acid sequences of carbohydrate-binding module. The amino acid sequences of carbohydrate-binding module (CBM) on β subunits were used to study the differences. The sequence alignment between α1β1γ1 isoform (PDB: 4QFG) and α1β2γ1 (PDB:4REW) isoform was performed and edited using BioEdit software (version 7.1.8) \[[@B27-ijms-18-01408]\], which is a user-friendly biological sequence alignment editor and analysis program. The crystallographic structures of AMPK for molecular docking studying were added to the hydrogen atoms and the charge was given to the Gasteiger-Huckel. The crystal structures comparison was conducted by Sybyl X 2.1 (Tripos Associates Inc., S.H. R.: St. Louis, MO, USA.) \[[@B28-ijms-18-01408]\] and the binding modes were generated by PyMOL V0.99 (Schrödinger, New York, NY, USA.) \[[@B29-ijms-18-01408]\]. The polar hydrogen atoms were added to the crystal structures of the AMPK via the biopolymer module and the Gasteiger-Huckel charges were loaded on the atoms of proteins. The protein peptide backbones were shown in cartoon and colored by different colors, the side chains of the nonconservative amino acids were shown in line and colored by chain. 3.2. Molecular Docking {#sec3dot2-ijms-18-01408} ---------------------- The virtual screening and molecular docking studies were performed using Surflex docking module in Sybyl X 2.1. There were still some deficiencies due to the fact that the receptor was regarded as a rigid structure. Therefore, it was essential to optimize the docking parameters, the co-crystallized ligand was extracted and re-docked into the active site of the AMPK with the varied parameters, and then the conformation of the original ligand and the re-docking ligand were compared. The binding site was defined as a sphere containing the residues that stay within 5 Å from the co-ligand. The maximum conformations per fragment and maximum number of rotatable bonds per molecule were 20 and 100, respectively. Furthermore, the options for pre-dock minimization and post-dock minimization of molecules were omitted, while other parameters were set as default options. The top 20 conformational poses were selected according to the docking score. Dock scores were evaluated by Consensus Score (CScore), which integrates the strengths of individual scoring functions combine to rank the affinity of ligands bound to the active site of a receptor. 3.3. High-Throughput Virtual Screening {#sec3dot3-ijms-18-01408} -------------------------------------- High-throughput virtual screening was regarded as an important tool to identify novel lead compounds suitable for specific protein targets \[[@B30-ijms-18-01408]\], and the screened compounds can be easily obtained from commercial sources for biological evaluation as well \[[@B31-ijms-18-01408]\]. The ChemDiv database was supplied by Topscience Co. (Shanghai, China), which includes 1,500,000 compounds was employed for virtual screening through Surflex docking module in Sybyl X 2.1. To accelerate virtual screening, the maximum quantity of conformations was reduced from 20 to 10, the maximum quantity of rotatable bonds was decreased from 100 to 50, and the top six conformations were collected. The same as the molecular docking studies, the default optimization of molecules before and after the docking was canceled. Other parameters were kept as default values. Compounds PF-06409577 and A-769662 were severed as reference molecules, respectively. The compounds with the docking score (≥8.0) were extracted for further analyzing the interactions between ligand and active site, to this end, 100 compounds were collected to calculate the drug-likeness model score. Drug-likeness model scores were computed for hit compounds using the MolSoft software (MolSoft, San Diego, CA, USA) \[[@B32-ijms-18-01408]\]. 3.4. The In Vitro Activation Assay {#sec3dot4-ijms-18-01408} ---------------------------------- The in vitro preliminary kinase assays human α1β1γ1 AMPK were carried out according to the previous experimental method \[[@B33-ijms-18-01408]\]. The screened compounds and α1β1γ1 AMPK isoform were provided by Topscience Co. (Shanghai, China) and Huawei Pharmaceutical Co. Ltd. (Shanghai, China), respectively. Generally, each of the evaluated compounds was dissolved in 10% Dimethyl sulphoxide at 10 μM and diluted to a required concentration with buffer solution. Then, 5 μL of the dilution was added to a 30 μL kinase assay buffer and 5 μL AMPK isoform per well. The solution was mixed at 0 °C for 30 min. Next, 5 μL of AMARA petide and 5 μL Adenosine triphosphate (ATP) were added to the well. The enzymatic reactions were conducted at 30 °C for 30 min. The AMPK activity was determined by quantifying the amount of ATP remaining in assay solution with Kinase-Glo Plus luminescent kinase assay kit (Promega, Madison, WI, USA). The luminescent signal is correlated with the amount of ATP present, while inversely correlated with the kinase activity. The mean values from three independent experiments were used for the expression of relative activities. A-769662, a β1-selective AMPK activator reported by Abbott laboratories, was used as a control. 4. Conclusions {#sec4-ijms-18-01408} ============== In summary, the sequence alignment and structural comparison was performed to identify the AMPK domain structure detail, which provides a molecular basis of selective AMPK activators on β1-containing isoforms. The key amino acid residues (Phe/Ile82, Thr/Ser85, Gly/Glu86, Thr/Ile106, Arg/Lys107, Gln/His109, Asn/Asp111) may contribute to the selectivity and provide a foundation for structure-based design of new direct β1-selective AMPK activators. Furthermore, the structure-based virtual screening workflow for the identification of selective activators of AMPK (α1β1γ1 and α2β1γ1) was established and six potential hit compounds for α1β1γ1 isoform and α2β1γ1 isoform were obtained, respectively. The preliminary assay indicated that most of the selected α1β1γ1 AMPK activators displayed promising activation potency. Overall, these findings revealed extensive interactions of activators and AMPK for rational design of novel selective AMPK activators. Further in vitro testing of retrieved hits is still in progress in our laboratory. This work was supported by the Postdoctoral Science Foundation funded project (2017M611916), Natural Science Foundation of Jiangsu Province (Grants No. BK20140225), Science and Technology Plan Projects of Xuzhou (Grants No. KC16SG249), Scientific Research Foundation for Talented Scholars of Xuzhou Medical College (No. D2014008), Xuzhou Medical University School of Pharmacy Graduate Student Scientific Research Innovation Projects (No. 2015YKYCX014). Tonghui Huang and Jie Sun performed the sequence alignment and structural comparison; Shanshan Zhou and Jian Gao performed virtual screening study; Yi Liu and Tonghui Huang analyzed the data; Tonghui Huang wrote the paper. The authors declare no conflict of interest. {#ijms-18-01408-f001} {#ijms-18-01408-f002} {#ijms-18-01408-f003} {#ijms-18-01408-f004} {#ijms-18-01408-f005} {#ijms-18-01408-f006} {#ijms-18-01408-f007} {#ijms-18-01408-f008} {#ijms-18-01408-f009} ijms-18-01408-t001_Table 1 ###### The docking scores and drug-likeness model scores of selected activators for AMPK (α1β1γ1 and α2β1γ1). Isoforms Compound No. Total-Score Crash Polar Similarity Number of HBA/HBD MolLog P Drug-Likeness Model Score -------------------------- -------------- ------------- ------- ------- ------------ ------------------- ---------- --------------------------- AMPK (α1β1γ1) activators F064-1335 10.50 −2.35 3.54 0.44 6/1 3.79 −0.16 M5653-1884 10.24 −1.43 2.90 0.50 5/1 6.07 0.22 D454-0135 10.20 −1.58 3.41 0.44 6/2 4.03 −0.31 M8006-4303 10.07 −1.83 4.29 0.58 6/1 1.01 1.02 F264-3019 9.93 −1.39 3.19 0.46 6/1 5.44 1.00 F377-1213 10.03 −1.43 1.32 0.44 6/1 4.12 0.16 PF-06409577 7.29 −0.07 3.08 0.93 3/3 3.80 0.71 AMPK (α2β1γ1) activators L267-1138 10.96 −2.46 2.78 0.52 4/1 6.05 −0.08 F684-0053 10.60 −2.77 4.31 0.54 7/3 2.04 0.54 C804-0412 10.15 −3.27 3.39 0.54 5/2 2.53 1.00 M5976-1661 9.46 −0.92 1.32 0.46 6/0 4.84 0.62 M039-0295 9.35 −1.61 1.48 0.50 6/1 2.66 −0.20 M5050-0116 9.27 −1.35 2.82 0.49 7/2 5.13 0.38 A-769662 7.44 −1.46 1.26 0.93 5/3 3.46 0.30 991 8.38 −0.96 4.08 0.73 4/2 5.38 0.41

1. Introduction {#sec1-ijms-19-00773} =============== MicroRNAs (miRNAs) are endogenous, small (\~22 nucleotides), and single-stranded noncoding RNAs. The role of different miRNAs in biological systems is well established. They are generally regarded as negative regulators of gene expression, as they bind to the 3′ untranslated region (3′UTR) of messengerRNAs (mRNAs), leading to mRNA degradation and/or suppression of mRNA translation \[[@B1-ijms-19-00773],[@B2-ijms-19-00773],[@B3-ijms-19-00773]\]. Currently, thousands of miRNAs have been identified as participating in a number of biological processes, such as cellular growth, proliferation, development, and metabolism \[[@B4-ijms-19-00773]\]. Based on Solexa sequencing, the expression of microRNA-25 (miR-25) was higher in the longissimus dorsi muscle of Large White pigs (a lean type) than in those of Tongcheng pigs (a Chinese indigenous fatty pig). Because skeletal muscle plays a vital role in whole-body metabolism \[[@B5-ijms-19-00773]\], we speculated that miR-25 could play a regulatory role in metabolism. Previous studies have reported that miR-25 plays an important role in many biological processes. The expression of miR-25-3p was significantly increased in the plasma of thyroid papillary carcinoma, as compared with patients with benign tumors or healthy individuals \[[@B6-ijms-19-00773]\]. miR-25 expression was higher in ovarian epithelial tissue, gastric cancer, lung adenocarcinoma, and many other tumors, and miR-25 expression levels were also closely related to tumor stage and lymph node metastasis \[[@B7-ijms-19-00773],[@B8-ijms-19-00773],[@B9-ijms-19-00773],[@B10-ijms-19-00773]\]. Inhibition of miR-25 markedly improved cardiac contractility in the failing heart \[[@B11-ijms-19-00773]\]. miR-25 could protect cardiomyocytes against oxidative damage by downregulating the mitochondrial calcium uniporter (MCU) \[[@B12-ijms-19-00773]\]. Variations in miR-25 expression influenced the severity of diabetic kidney disease \[[@B13-ijms-19-00773]\]. However, to our knowledge, the role of miR-25 in metabolism has not been reported, and its transcriptional regulatory mechanism is not clear. Thus, in this study, we first investigated whether miR-25 was involved in metabolism by gain-of-function and loss-of-function assays. Then, the target gene of miR-25, AKT serine/threonine kinase 1 (*Akt1*), which is related to metabolism, was predicted and verified using bioinformatics software and experiments. Finally, the core promoter of miR-25 was identified, and the binding of the transcription factor activator protein-2α (AP-2α) to the core promoter was shown to promote the transcriptional activity of miR-25 and downregulate *Akt1* expression. 2. Results {#sec2-ijms-19-00773} ========== 2.1. miR-25 Is Highly Conserved in Mammals {#sec2dot1-ijms-19-00773} ------------------------------------------ Clustal Omega (Available online: <https://www.ebi.ac.uk/Tools/msa/clustalo/>) \[[@B14-ijms-19-00773]\] was used to build the phylogenetic tree of pre-miRNA of miR-25. The results show that compared with other species selected in this study, the genetic relationship between mice and humans, cattle and goats, and gorillas and rhesus monkeys is closer ([Figure 1](#ijms-19-00773-f001){ref-type="fig"}A). The mature sequences of miR-25 are highly conserved in mammals, including pigs, mice, humans, goats, rats, hamsters, gorillas, chimpanzees, cattle, and rhesus monkeys. The "seed" sequences of miR-25 are identical, although there is a base deletion at the end of the chimpanzee sequence (ptr) ([Figure 1](#ijms-19-00773-f001){ref-type="fig"}B). 2.2. Effects of miR-25 on the Metabolism of C2C12 Cells {#sec2dot2-ijms-19-00773} ------------------------------------------------------- To investigate the role of miR-25-3p in metabolism, miR-25-3p mimics/negative control (NC) or inhibitors/NC were respectively transfected into growing C2C12 cells (mouse muscle myoblasts). The abundance of miR-25-3p was detected, which was \~3300-fold (*p* \< 0.01) higher as compared with another microRNA ([Figure S1](#app1-ijms-19-00773){ref-type="app"}). The mRNA and protein expression levels of the metabolism-related gene *PI3K* were repressed by miR-25-3p overexpression, while the levels of *PI3K* were upregulated in the inhibitor group, as compared with the negative controls ([Figure 2](#ijms-19-00773-f002){ref-type="fig"}A,B). In addition, the overexpression of miR-25-3p decreased levels of triglyceride (TG), whereas the knockdown of miR-25-3p increased them ([Figure 2](#ijms-19-00773-f002){ref-type="fig"}C). Conversely, the overexpression of miR-25-3p increased ATP and ROS levels, and the knockdown of miR-25-3p decreased their levels ([Figure 2](#ijms-19-00773-f002){ref-type="fig"}D,E).These data indicate that miR-25-3p plays a role in metabolism. 2.3. miR-25-3p Directly Targets Akt1 {#sec2dot3-ijms-19-00773} ------------------------------------ To explore the molecular mechanism of miR-25-3p effects on metabolism, the possible targets for miR-25-3p were predicted using TargetScan, and a putative binding site for miR-25-3p was predicted in the 3′UTR of *Akt1* mRNA. miR-25-3p targeting elements in the *Akt1*-3′UTR were relatively conserved in many mammals, including mice, humans, chimpanzees, rhesus monkeys, and rats ([Figure 3](#ijms-19-00773-f003){ref-type="fig"}A). To validate whether miR-25-3p directly targets *Akt1*, a luciferase reporter containing a 250 bp fragment from the *Akt1* 3′UTR was tested in vitro. Additionally, we generated a mutated version of the above mentioned reporter, in which five nucleotides of the predicted binding site were changed in order to abolish the putative interaction between miR-25-3p and *Akt1* mRNA ([Figure 3](#ijms-19-00773-f003){ref-type="fig"}B). The *Akt1* 3′UTR and mutant luciferase plasmid were cotransfected with mimics or NC into growing C2C12 cells. Twenty-four hours after transfection, analyses of luciferase activity revealed that miR-25-3p mimics significantly decreased the luciferase activity of the wild reporter plasmid as compared with NC, while there was no significant effect on the mutant plasmids ([Figure 3](#ijms-19-00773-f003){ref-type="fig"}C). These results revealed that miR-25-3p directly targets the 3′UTR of *Akt1* in vitro. To directly test the validity of the putative target, we transfected miR-25-3p mimics and miR-25-3p inhibitors into growing C2C12 cells. We found that the overexpression of miR-25-3p repressed *Akt1* expression, as measured by qRT-PCR (*p* \< 0.01) and Western blotting, whereas the knockdown of miR-25-3p derepressed it ([Figure 3](#ijms-19-00773-f003){ref-type="fig"}D,E). These results demonstrate that *Akt1* was a target of miR-25-3p. 2.4. Identification and Characterization of the Mouse miR-25-3p Promoter {#sec2dot4-ijms-19-00773} ------------------------------------------------------------------------ To further identify the promoter region and regulatory elements of mouse miR-25-3p, we used luciferase assays to analyze a series of deletions in the potential promoter region, as predicted by neural network promoter prediction (NNPP) online software ([Figure 4](#ijms-19-00773-f004){ref-type="fig"}A).The plasmids containing the various lengths of the miR-25-3p promoter were transiently transfected into growing BHK and C2C12 cells. Analyses of luciferase activity revealed that miR-25-3p-P9 (−119/+144) showed the greatest transcriptional activity, and the longer fragment showed lower transcriptional activity ([Figure 4](#ijms-19-00773-f004){ref-type="fig"}B), indicating that the region from −1870 to −119 contains one or more *cis*-acting elements that can repress miR-25-3p expression. The result demonstrates that this 263 bp-long sequence was the core promoter of mouse miR-25-3p. 2.5. The Transcription Factor AP-2α Binds to the Core Promoter of Mouse miR-25-3p {#sec2dot5-ijms-19-00773} --------------------------------------------------------------------------------- To further search the transcription factors that bind to the core promoter of mouse miR-25-3p, AliBaba 2.1 and Genomatix software programs were utilized to analyze the putative transcription factors. As shown in [Figure S2](#app1-ijms-19-00773){ref-type="app"}, AP-2α was found to be able to bind to the core promoter of mouse miR-25-3p. To examine whether AP-2α influences the activity of the mouse miR-25-3p promoter, an AP-2α overexpression plasmid (pc-AP-2α) was generated and cotransfected with the miR-25-3p-P9 plasmid into growing C2C12 cells. Twenty-four hours after transfection, analyses of luciferase activity showed that pc-AP-2α significantly increased miR-25-3p promoter transcriptional activity ([Figure 4](#ijms-19-00773-f004){ref-type="fig"}C). To determine the functional importance of the AP-2α binding site, we mutated the AP-2α binding site at −109 to −102, by using the wild-type miR-25-3p-P9 plasmid as the template. The mutant was constructed and transfected into growing C2C12 cells. As shown in [Figure 4](#ijms-19-00773-f004){ref-type="fig"}D, the luciferase activity of the mutant was significantly decreased as compared with the wild-type miR-25-3p-P9 construct. These results indicated that transcription factor AP-2α may induce transcriptional activity by directly binding to the core promoter of mouse miR-25-3p. To further verify whether transcription factor AP-2α binds to the core promoter of mouse miR-25-3p, ChIP was performed in growing C2C12 cells. Chromatin was immunoprecipitated using the AP-2α antibody, and PCR amplification was performed, using the DNA fragment of the expected size as a template. The ChIP-Q-PCR assay showed that AP-2α interacted with the miR-25-3p promoter within the binding site ([Figure 4](#ijms-19-00773-f004){ref-type="fig"}E). These results confirmed that the transcription factor AP-2αis capable of binding to the AP-2α binding site in the mouse miR-25-3p promoter region, and induces miR-25-3p transcription. 2.6. AP-2α Regulates miR-25-3p and Akt1 Expression {#sec2dot6-ijms-19-00773} -------------------------------------------------- Because *Akt1* was identified as a direct target of miR-25-3p, and the transcription factor AP-2α could upregulate miR-25-3p transcription, the effect of AP-2α on *Akt1* expression was further appraised by the overexpression or knockdown of AP-2α in growing C2C12 cells. As AP-2α mRNA expression was significantly decreased by doublestranded short interfering AP-2α RNA ( si-AP-2α-1) and si-AP-2α-2, and the inhibitory effect of si-AP-2α-2 was greater than that of si-AP-2α-1 ([Figure S3](#app1-ijms-19-00773){ref-type="app"}), si-AP-2α-2 was chosen for subsequent experiments. pc-AP-2α or si-AP-2α was transfected into growing C2C12 cells, respectively. Fourty-eight hours after transfection, RNA and protein were isolated. The overexpression of AP-2α significantly increased miR-25-3p expression, while the knockdown of AP-2α resulted in the significant suppression of miR-25-3p expression ([Figure 5](#ijms-19-00773-f005){ref-type="fig"}A). Conversely, the mRNA and protein expression of Akt1 were significantly suppressed by AP-2α overexpression, and were increased by si-AP-2α ([Figure 5](#ijms-19-00773-f005){ref-type="fig"}B--D). These results indicate that AP-2α activated mature miR-25 expression, and downregulated the expression of *Akt1*. 3. Discussion {#sec3-ijms-19-00773} ============= Increasing evidence shows that miR-25, a member of the miR-106b-25 cluster, is involved in many biological processes. For instance, miR-25 inhibits human gastric adenocarcinoma cell apoptosis \[[@B15-ijms-19-00773]\], promotes glioblastoma cell proliferation and invasion \[[@B16-ijms-19-00773]\], and regulates human ovarian cancer apoptosis \[[@B17-ijms-19-00773]\]. The miR-106b-25 cluster regulates adult neural stem/progenitor cell proliferation, migration, and differentiation \[[@B18-ijms-19-00773],[@B19-ijms-19-00773]\]. miR-25 plays an important role in heart disease \[[@B11-ijms-19-00773],[@B12-ijms-19-00773]\] and diabetic kidney disease \[[@B13-ijms-19-00773]\]. In addition, numerous studies have demonstrated that miRNAs are implicated in metabolism \[[@B20-ijms-19-00773],[@B21-ijms-19-00773],[@B22-ijms-19-00773],[@B23-ijms-19-00773]\]. However, miR-25 has not been functionally related to metabolism until now. In this study, miR-25 was identified as a novel regulator of metabolism. The gain-of-function and loss-of-function assays showed that miR-25-3p inhibited the expression of *PI3K* and reduced levels of triglyceride (TG), while levels of ATP and ROS were increased. PI3K has been implicated in insulin-regulated glucose metabolism \[[@B24-ijms-19-00773]\], and PI3K signaling has a role in many cellular processes, such as metabolic control, immunity, and cardiovascular homeostasis \[[@B25-ijms-19-00773],[@B26-ijms-19-00773],[@B27-ijms-19-00773]\]. It is well-known that triglycerides (TG) are a component of lipids, and participate in lipid metabolism. ATP is the most direct source of energy in an organism, and takes part in many metabolic processes. ROS, a class of single electron radicals of oxygen, comprise superoxide anions (O~2~^−^), hydrogen peroxide (H~2~O~2~), and hydroxyl radicals (**^·^**OH) \[[@B28-ijms-19-00773]\], and are closely related to adipogenesis and myogenesis \[[@B28-ijms-19-00773],[@B29-ijms-19-00773],[@B30-ijms-19-00773],[@B31-ijms-19-00773]\]. These data indicate that miR-25-3p indeed participates in metabolism in mice. To further understand the molecular mechanism by which miR-25-3p regulates metabolism, we searched for potential target genes of miR-25-3p via TargetScan. Fortunately, the 3′UTR of *Akt1* contained a 7 nucleotides perfect match site complementary to the miR-25-3p seed region ([Figure 3](#ijms-19-00773-f003){ref-type="fig"}B). The serine-threonine kinase ATK, also known as protein kinase B (PKB), is an important effector for PI3K signaling as initiated by numerous growth factors and hormones \[[@B32-ijms-19-00773]\]. *Akt* can control glucose uptake by regulating GLUT4 in cells, thereby reducing blood sugar and promoting glycogen synthesis \[[@B32-ijms-19-00773],[@B33-ijms-19-00773],[@B34-ijms-19-00773]\]. *Akt* usually promotes glycogen synthase kinase-3 alpha (GSK3α) phosphorylation and inhibits its activity \[[@B35-ijms-19-00773]\], and then activates glycogen synthesis \[[@B36-ijms-19-00773]\]. A previous study has demonstrated that overexpression of miR-25-3p downregulates *Akt* expression and inactivates Akt phosphorylation in the tongue squamous cell carcinoma cell line Tca8113 \[[@B37-ijms-19-00773]\]. Consequently, we deduced that the role of miR-25-3p in metabolism may arise from its inhibition of *Akt1*. First, the dual luciferase reporter assay demonstrated that *Akt1* was a direct target of miR-25-3p, shown by the steady decrease luciferase activity of the pmirGLO-Akt1-wt vector; but not the mutant form ([Figure 3](#ijms-19-00773-f003){ref-type="fig"}C). Meanwhile, qRT-PCR and Western blotting results showed that the expression of *Akt1* was inhibited by the miR-25-3p mimics, and that this inhibition was reversed by the miR-25-3p inhibitors ([Figure 3](#ijms-19-00773-f003){ref-type="fig"}D,E). These results suggested that the effect of miR-25-3p in metabolism was due, at least in part, to the suppression of *Akt1*. An increasing number of studies have shown that transcription factors are capable of binding to miRNA promoter elements and modulating miRNA transcription \[[@B38-ijms-19-00773],[@B39-ijms-19-00773],[@B40-ijms-19-00773]\]. Therefore, we analyzed the transcriptional mechanism of miR-25-3p in this study. Nine fragments of 5′-flanking sequences of mouse miR-25-3p were isolated. Subsequently, a series of experiments, including dual luciferase, site-directed mutagenesis, and ChIP assays, confirmed that AP-2α bound to the miR-25-3p promoter region and promoted its transcription activity ([Figure 4](#ijms-19-00773-f004){ref-type="fig"}). Moreover, qRT-PCR and Western blotting results showed that overexpression of AP-2α resulted in the upregulation of miR-25-3p and downregulation of *Akt1*, and that the knockdown of AP-2α reversed these results ([Figure 5](#ijms-19-00773-f005){ref-type="fig"}). The AP-2 family of transcription factors consists of five members, in humans and mice: AP-2α, AP-2β, AP-2γ, AP-2δ, and AP-2ε; which play important roles in several cellular processes, such as apoptosis, migration, and differentiation \[[@B41-ijms-19-00773],[@B42-ijms-19-00773]\]. AP-2α was first identified by its ability to bind to the enhancer regions of SV40 and human metallothionein IIA \[[@B43-ijms-19-00773]\]. Subsequently, numerous studies have demonstrated that AP-2α can regulate gene expression. For instance, AP-2α binding to the *C/EBPα* promoter results in decreased *C/EBPα* expression \[[@B44-ijms-19-00773]\], and AP-2α can bind to the *TACE* promoter and decrease its expression in dendritic cells \[[@B45-ijms-19-00773]\]. Furthermore, Qiao et al. \[[@B46-ijms-19-00773]\] reported that there was an AP-2α binding site in the *DEK* core promoter, and overexpression of AP-2α upregulated *DEK* expression. In this study, we identified that AP-2α binds to the miR-25-3p promoter region and promotes its transcription activity. In conclusion, our results demonstrate that miR-25-3p acts as a positive regulator of the metabolism of growing C2C12 cells, by affecting *Akt1* gene expression through directly binding to its 3′UTR. Moreover, the transcription factor AP-2α is able to bind to the core promoter of mouse miR-25-3p, activating mature miR-25 expression and downregulating the expression of *Akt1* ([Figure 6](#ijms-19-00773-f006){ref-type="fig"}). 4. Materials and Methods {#sec4-ijms-19-00773} ======================== 4.1. miRNA, Small RNA Oligonucleotide Synthesis, and Plasmid Construction {#sec4dot1-ijms-19-00773} ------------------------------------------------------------------------- The miR-25-3p oligonucleotides (miR-25-3p mimics, NC, miR-25-3p inhibitors, and inhibitor-NC) and double-stranded short interfering RNAs (siRNAs) targeting AP-2α were designed and synthesized by RiboBio (Guangzhou, China).The oligonucleotides are listed in [Table S1](#app1-ijms-19-00773){ref-type="app"}. To construct the AP-2α overexpression vector pc-AP-2α, the AP-2α coding sequence (1314 bp) was amplified from mouse C2C12 cells cDNA using the following primers: forward: 5′-CCCAAGCTTGCCACCATGCTTTGGAAACTGACGGA-3′; reverse: 5′-CCGCTCGAGTCACTTTCTGTGTTTCTCTT-3′. The PCR product was subcloned into the *Hind*III/*Xho*I sites of the pcDNA3.1(+) vector (Invitrogen, Carlsbad, CA, USA). The potential target site of miR-25-3p, localized in the 3′UTR of *Akt1* mRNA, was predicted by TargetScan (Available online: <http://www.targetscan.org/>) \[[@B47-ijms-19-00773]\]. The *Akt1* 3′UTR was amplified from C2C12 cell cDNA and inserted into the *Pme*I/*Xho*I sites of the pmirGLO vector (Promega, Madison, WI, USA). Point mutations in the seed region of the predicted miR-25-3p sites within the 3′UTR of *Akt1* were generated using overlap-extension PCR \[[@B48-ijms-19-00773]\]. The corresponding primers are listed in [Table S2](#app1-ijms-19-00773){ref-type="app"}. The potential promoter regions of miR-25-3p was predicted by using the neural network promoter prediction (NNPP) software (Available online: <http://www.fruitfly.org/seq_tools/promoter.html>) \[[@B49-ijms-19-00773]\]. Nine miR-25-3p promoter deletion fragments were amplified from the mouse genome via PCR with the primers listed in [Table S3](#app1-ijms-19-00773){ref-type="app"}.The nine purified PCR products were ligated into the *Kpn*I/*Hind*III sites of the pGL3-Basic vector (Promega). AliBaba2.1 (Available online: <http://www.gene-regulation.com/>) \[[@B50-ijms-19-00773]\] and MatInspector (Available online: <http://www.genomatix.de/online_help/help_matinspector/matinspector_help.html>) \[[@B49-ijms-19-00773]\] were used to predict the potential transcription factor binding sites. The AP-2α transcription factor binding sites of the miR-25-3p promoter region were also mutated by overlap-extension PCR. The primers are provided in [Table S3](#app1-ijms-19-00773){ref-type="app"}. 4.2. Cell Culture and Luciferase Reporter Assays {#sec4dot2-ijms-19-00773} ------------------------------------------------ C2C12 (mouse muscle myoblast) and BHK (baby hamster kidney) cells were cultured in DMEM (Gibco, Gaithersburg, MD, USA) containing 10% fetal bovine serum (FBS) (Gibco) at 5% CO~2~ and 37 °C. For luciferase reporter assays, growing C2C12 or BHK cells were seeded in 48-well plates. After 12--16 h, the plated cells were transfected with a recombinant plasmid using Lipofectamine 2000 (Invitrogen). To verify the miR-25-3p targeting *Akt1* 3′UTR, 1 μL miR-25-3p mimics/NC was cotransfected with 0.1 μg *Akt1* 3′UTR/mutant plasmid into C2C12 cells. For the miR-25-3p promoter luciferase reporter assay, 0.4 μg pGL3-Basic or recombinant plasmids and 20 ng pRL-TK vector were transfected. For cotransfection luciferase assays, each well contained 0.2 μg pGL3-(Basic, miR-25-3p-P9 and AP-2α-mut), 20 ng pRL-TK, and 0.2 μg pc-AP-2α. Empty pcDNA-3.1(+) cotransfected with pGL3-(Basic, miR-25-3p-P9 and AP-2α-mut) was used as the control. After 24 h of incubation, luciferase activity was measured using a PerkinElmer 2030 Multilabel Reader (PerkinElmer, Norwalk, CT, USA). 4.3. Triglyceride Content, ATP, and Reactive Oxygen Species (ROS) Assays {#sec4dot3-ijms-19-00773} ------------------------------------------------------------------------ For detecting the concentrations of triglyceride (TG), ATP, and ROS, growing C2C12 cells were seeded in 24-well plates the day before transfection. miR-25-3p mimic, NC, miR-25-3p inhibitor, and inhibitor-NC were transfected into confluent (\~80%) cells, respectively, at a concentration of 12 nM with Lipofectamine 2000 (Invitrogen). After 24--48 h, the concentrations of TG and ATP in the lysates of cells were measured with commercial kits (Applygen (Beijing, China) and Beyotime (Shanghai, China), respectively) following the manufacturer's instructions, and normalized to the protein content (μmol/mg protein) using the BCA assay kit (Thermo Scientific, Waltham, MA, USA). ROS were measured using the reactive oxygen species assay kit (Beyotime) following the manufacturer's protocol. 4.4. Chromatin Immunoprecipitation (ChIP) {#sec4dot4-ijms-19-00773} ----------------------------------------- ChIP assays were performed to assess the binding of endogenous AP-2α to the miR-25-3p promoter in C2C12 cells using the EZ-ChIP™ Kit (Millipore, Boston, MA, USA), following a previously described method \[[@B49-ijms-19-00773]\]. Precleared chromatin was incubated with the AP-2α antibody (Santa Cruz Biotechnology, Dallas, TX, USA) or normal mouse IgG (Millipore) antibodies (control) overnight at 4 °C. Purified DNA from the samples and the input controls were analyzed for the presence of miR-25-3p promoter sequences containing putative AP-2α response elements using qPCR. The primers used here are listed in [Table S4](#app1-ijms-19-00773){ref-type="app"}. 4.5. RNA Isolation and qRT-PCR {#sec4dot5-ijms-19-00773} ------------------------------ For quantifying the mRNA expression of genes, growing C2C12 cells were seeded in 6-well plates. miR-25-3p mimic, NC, miR-25-3p inhibitor, inhibitor-NC, si-AP-2α, and NC were transfected into confluent (\~80%) cells, respectively, at a concentration of 50 nM with Lipofectamine 2000 (Invitrogen). After 48 h, total RNA was isolated using a HP Total RNA Kit (Omega, Norcross, GA, USA) according to the manufacturer's protocol. The cDNA was synthesized using a PrimeScript™RT reagent Kit with gDNA Eraser (Takara, Osaka, Japan) according to the manufacturer's protocol. The qRT-PCR was performed in triplicate with iQSYBR green Supermix (Bio-Rad, Hercules, CA, USA) in a LightCycler 480 Realtime PCR machine (Roche, Basel, Switzerland). The mRNA levels of target genes were reported relative to those of the house keeping gene β-actin by using the 2^−ΔΔ^*^C^*^t^ method. The qRT-PCR primers are listed in [Table S5](#app1-ijms-19-00773){ref-type="app"}. 4.6. Protein Isolation and Western Blotting {#sec4dot6-ijms-19-00773} ------------------------------------------- For detecting the protein expression of PI3K and Akt1, growing C2C12 cells were seeded in6-well plates. miR-25-3p mimic, NC, miR-25-3p inhibitor, inhibitor-NC, si-AP-2α, and NC were transfected into confluent (\~80%) cells, respectively, at a concentration of 50 nM with Lipofectamine 2000 (Invitrogen). After 48 h, total protein was isolated using RIPA Lysis Buffer (Beyotime). The cells were washed briefly with cold phosphate-buffered saline (PBS), 150 μL RIPA Lysis Buffer (containing 1 mM PMSF) was added, incubated for 1 min at room temperature, and then centrifuged at 12,000× *g* for 5 min. The supernatant extract was used for Western blot analysis. Western blot analysis was performed to analyze the expression levels of Akt1 (Affinity Biosciences, Cincinnati, OH, USA) andPI3K (Abclonal, Wuhan, China) according to the methods of Huang et al. \[[@B47-ijms-19-00773]\]. β-actin (Santa Cruz Biotechnology) served as the loading control. 4.7. Statistical Analysis {#sec4dot7-ijms-19-00773} ------------------------- All the results are presented as the means ± SD. Student's *t*-test was used for statistical comparisons. A *p* value of \< 0.05 was considered to be statistically significant. \*\* *p* \< 0.01; \* *p* \< 0.05; NS, not significant. This research wassupported by the China Postdoctoral Science Foundation (2017M610465), the Open Project of Key Laboratory of Animal Embryo Engineering and Molecular Breeding of Hubei Province (KLAEMB201602),the Postdoctoral Innovation Post of Hubei Province (2016), the National Natural Science Foundationof China (31472075, 31402051 and 31501932), and Natural Science Foundation of Hubei Province key projects of technical innovation (2016ABA117). The following are available online at <http://www.mdpi.com/1422-0067/19/3/773/s1>. ###### Click here for additional data file. Mingxin Chen and Siwen Jiang conceived and supervised the study; Feng Zhang and Kun Chen designed experiments; Tingting Kang and Qianhui Zeng performed experiments; Hu Tao and Qi Xiong analysed data; Feng Zhang wrote the manuscript; Yang Liu made manuscript revisions. The authors declare no conflict of interest. {#ijms-19-00773-f001} {#ijms-19-00773-f002} {#ijms-19-00773-f003} {#ijms-19-00773-f004} {#ijms-19-00773-f005} {#ijms-19-00773-f006}

1. Introduction {#sec1} =============== Radiotherapy (RT) has played an indispensable role in the modern treatment of head and neck malignancies. Among the most devastating complications of RT is osteoradionecrosis (ORN) of the jaw. The most recent large cohort study reported the prevalence of ORN to be 6.2% in patients who underwent RT for oral cancer \[[@B1]\]. ORN can predispose patients to recurrent infection, orocutaneous fistula, pathological fracture, and injury to inferior alveolar nerve (IAN). Patients with these complications experience several symptoms including trismus, severe pain, chronic drainage, disfigurement, and insufficient nutrition, resulting in serious impairment in quality of life \[[@B2], [@B3]\]. Several classifications have been proposed in literature to facilitate the diagnosis of ORN \[[@B2]--[@B10]\]. The only reliable treatment option for advanced ORN (resorption of the inferior border of the mandible, orocutaneous fistula, and pathological fracture) is surgical debridement and free flap reconstruction \[[@B3], [@B11]--[@B13]\]. To achieve a successful outcome for this surgery, both adequate bone resection and survival of the transferred flap are essential. The extent of resection is generally determined by presence of bleeding at resected edges; however, ORN can recur at viable margins \[[@B14]\]. To understand the underlying pathological mechanisms of ORN, a histopathological study is needed. Previous studies performed histological analysis of ORN \[[@B15]--[@B18]\]; however the bone specimens analyzed were obtained during sequestrectomy or decortication. These studies lacked detailed information regarding location of obtained bone specimens. In the only histological study analyzing the resection margins after segmental mandibulectomy by Zaghi et al. \[[@B14]\], the anteroposterior location of the bone specimens and differences between cortical and cancellous bone were not described. This histopathological study analyzed the anteroposterior margins after segmental mandibulectomy for advanced ORN. The purpose of this study was to evaluate the necrotic changes in cortical and cancellous bones at resection margins and proximal areas of bone destruction. 2. Materials and Methods {#sec2} ======================== We enrolled eleven consecutive patients who underwent segmental mandibulectomy for surgical debridement and simultaneous free fibula osteocutaneous flap transfer for treatment of advanced mandibular ORN in our department between July 2013 and August 2016. No patients who underwent surgical intervention for ORN were excluded. ORN was defined as a nonhealing exposure of bone of at least 6 months\' duration \[[@B19]\]. All patients had various symptoms, including pain, infection, trismus, and difficulty eating, and were resistant to conservative therapy. Severe pain (e.g., lightning pain causing sleep deprivation) caused by damage to the IAN was regarded as the most important finding guiding the decision to intervene surgically. The extent of mandibulectomy with adequate safety margins (more than 10 mm) from apparent osteolytic areas was determined by preoperative thin-slice computed tomographic imaging. The Medical Ethics Committee of Kobe University Hospital approved this study. All subjects gave written informed consent to release clinical information and bone samples for the study. The following epidemiological data were gathered retrospectively from medical charts: age, sex, pathological diagnosis, primary tumor sites, types of RT, radiation dose, chemotherapy, surgery for primary tumor, time interval between the end of RT and surgery for ORN, the existence of pathological fracture, location of ORN, and surgical findings (the extent of segmental mandibulectomy and bleeding at the resection margins). Segmental mandibulectomy defects were classified according to the "CAT classification" used in our previous reports \[[@B20], [@B21]\]. Briefly, defects were classified on the basis of three anatomical landmarks: the mental tubercle \[T\], mandibular angle \[A\], and condyle \[C\]. The lesions that did not include tubercle, angle, and condyle were classified as mandibular "body." 3. Histopathological Analysis {#sec3} ============================= All bone specimens were decalcified and fixed in formalin but not frozen. The details of decalcification method were as follows: formic acid (98%) (Wako, Osaka, Japan) was diluted to 10% by distilled water. The bone specimens were immersed in 10% formic acid with ion exchange resin and treated by ultrasonic histoprocessor Histra-DC (Jokoh, Tokyo, Japan) for several days to several weeks. Thin sections were obtained from paraffin blocks and stained with hematoxylin and eosin for light microscopy. To analyze the differences in necrotic changes among the anteroposterior locations, thin sections from the anterior and posterior margins of segmental mandibulectomies were used for observations, as shown in [Figure 1](#fig1){ref-type="fig"}. The cross sections of the anterior and posterior margins were prepared separately from the true resection margins to avoid artifact caused by the heat of the surgical saw ([Figure 1(a)](#fig1){ref-type="fig"}). To analyze the differences between cortical and cancellous bone, we examined cancellous bone within the bone marrow cavity at the middle level of the mandible (Figures [1(c)](#fig1){ref-type="fig"} and [1(g)](#fig1){ref-type="fig"}) in each section. To analyze the differences of cortices among the craniocaudal locations, we examined cortical bone at the middle level of the mandible and near the inferior border of the mandible in each thin section at the anterior and posterior margins (Figures [1(c)](#fig1){ref-type="fig"} and [1(g)](#fig1){ref-type="fig"}). To analyze viability near the center of lesion, we examined cancellous bone near the most advanced area of bone destruction (Figures [1(b)](#fig1){ref-type="fig"} and [1(k)](#fig1){ref-type="fig"}). A previous study analyzed bone viability based on microscopic findings of osteocyte nuclei within lacunae and presence of viable blood vessels within Haversian canals \[[@B14]\]. To confirm the validity of differentiation between viable and necrotic cortical bone, we performed a preliminary microscopic analysis of randomly selected non-irradiated bone specimens obtained from patients who underwent segmental mandibulectomy for benign tumors or oral malignancies, but did not receive RT. As shown in [Figure 2(a)](#fig2){ref-type="fig"}, empty lacunae were frequently found in non-irradiated viable cortical bone with obvious viable blood vessels in Haversian canals. Osteocyte number varies with age \[[@B22]\]. Therefore, this study focused on the presence of blood vessels and red blood cells within Haversian canals rather than the number of osteocyte nuclei within lacunae. Cortical bone specimens with complete obstruction of Haversian canals as shown in [Figure 2(d)](#fig2){ref-type="fig"} were regarded as "necrotic". Conversely, the viability of cancellous bone within bone marrow cavity was determined by the presence of osteocyte nuclei within lacunae. Image acquisition of whole bone specimens (×4) was performed with a BZ-X 700 (Keyence, Osaka, Japan). Analysis of necrotic changes was independently performed by four observers (MA, SW, JK, and MS). MA and JK are oral and maxillofacial surgeons with more than ten years of experience, SW is a graduate fellow in our department, and MS is an oral pathologist with more than ten years of experience. There were discrepant results among the four observers in some specimens, as detailed below. In specimens with discrepant results, a mixture of viable and necrotic bone was found. Those specimens were defined as "heterogeneously necrotic" in this study. 4. Results {#sec4} ========== Clinical characteristics of patients in this study are shown in [Table 1](#tab1){ref-type="table"}. Eleven consecutive patients underwent surgical resection and simultaneous reconstruction with free fibula osteocutaneous flap for advanced mandibular ORN. Ten were male, and median age of the eleven patients was 65 years (range, 58--80 years). The median radiation dose was 66 Gy (range, 60--81 Gy). The median time interval between the end of RT and surgery for ORN was 84 months (range, 6--152 months). One patient (number 5) required surgical intervention after having developed a cutaneous fistula and subsequent mandibular pathological fracture shortly after completion of intensity-modulated radiotherapy. The location of ORN was ipsilateral to the primary tumor in four patients (36%), and contralateral in seven (64%). In operative findings, bleeding after segmental mandibulectomy was noted. There was one total flap loss in patient number 4. The remaining ten flaps survived. [Table 2](#tab2){ref-type="table"} shows the histopathological results of bone samples. The concordance rate of the differentiation of viable and necrotic bone among four observers was 73--100%. Cancellous bones at the anterior margins were viable in all specimens. In contrast, cortical bone at the middle level of the mandible of the anterior margin was viable in only four of eleven specimens (36%). Cortical bone near the inferior border of the mandible at the anterior margin was "necrotic" or "heterogeneously necrotic" in ten of eleven specimens (91%). Cancellous bones at the posterior margins were viable in eight of eleven specimens (73%). Evidence of viability was observed in five of eleven specimens in cortical bones at the middle level (45%) and four of eleven specimens near the inferior border of the mandible (36%) at the posterior margin. Cancellous bone near the most advanced area of bone destruction was viable in seven of eleven specimens (64%). Representative clinical and histopathological images are shown in [Figure 1](#fig1){ref-type="fig"}. Figures [1(a)](#fig1){ref-type="fig"} and [1(b)](#fig1){ref-type="fig"} show a bone sample from patient number 10 who received concomitant chemoradiotherapy for oropharyngeal carcinoma 7 years prior to surgery. Bone destruction around osseointegrated dental implants extended to the inferior border of the mandible. Color change due to necrosis was found along the cortical bone of the inferior border of the mandible as well as in alveolar bone ([Figure 1(b)](#fig1){ref-type="fig"}). Specimens of cancellous bone at the anterior ([Figure 1(d)](#fig1){ref-type="fig"}) and posterior ([Figure 1(h)](#fig1){ref-type="fig"}) margins were viable. Although cortical bone was viable at the middle level of the mandible at the anterior ([Figure 1(e)](#fig1){ref-type="fig"}) and posterior ([Figure 1(i)](#fig1){ref-type="fig"}) margins, as well as near the inferior border of the mandible of the posterior margin ([Figure 1(j)](#fig1){ref-type="fig"}), cortical bone near the inferior border of the mandible was necrotic ([Figure 1(f)](#fig1){ref-type="fig"}). Cancellous bone near the most advanced area of bone destruction was viable ([Figure 1(k)](#fig1){ref-type="fig"}). [Figure 3](#fig3){ref-type="fig"} shows bone specimens from patient number 9. Cancellous bone was viable surrounding the mental nerve at the anterior margin ([Figure 3(d)](#fig3){ref-type="fig"}) and the mandibular canal at the posterior margin ([Figure 3(g)](#fig3){ref-type="fig"}). However, cortical bone near the inferior border of the mandible at the anterior margin was necrotic ([Figure 3(e)](#fig3){ref-type="fig"}). Cortical bone near the inferior border of the mandible at the posterior margin was "heterogeneously necrotic" ([Figure 3(h)](#fig3){ref-type="fig"}). Bone near the osteolytic area was viable ([Figure 3(i)](#fig3){ref-type="fig"}). [Figure 4](#fig4){ref-type="fig"} shows bone specimens from patient number 8. During surgery, bleeding from bone marrow both at anterior ([Figure 4(b)](#fig4){ref-type="fig"}) and posterior ([Figure 4(c)](#fig4){ref-type="fig"}) margins was identified. Histopathological examination revealed obvious hypocellular fibrosis within the bone marrow cavity (Figures [4(d)](#fig4){ref-type="fig"} and [4(f)](#fig4){ref-type="fig"}). Except for viable cancellous bone at the anterior margin ([Figure 4(d)](#fig4){ref-type="fig"}), bone samples of cortical bone at the anterior margin ([Figure 4(e)](#fig4){ref-type="fig"}) and cancellous bone at the posterior margin ([Figure 4(g)](#fig4){ref-type="fig"}) were necrotic. Cancellous bone near the inferior border of the mandible at the posterior margin ([Figure 4(f)](#fig4){ref-type="fig"}) was "heterogeneously necrotic". Viable bone was observed near the most advanced area of bone destruction ([Figure 4(h)](#fig4){ref-type="fig"}), whereas bone in the center of osteolysis was necrotic ([Figure 4(i)](#fig4){ref-type="fig"}). The median follow-up period was 18 months (range, 11--48 months). The progression of necrosis after surgery occurred at the posterior viable margin only in one patient (number 3), but ceased in a few months. In contrast, there were cases in which good bone union between fibula flaps and the resection margins diagnosed histologically as necrotic. Overall, the incidence of progression of bone resorption arising from resection margins after surgery was 9%. 5. Discussion {#sec5} ============= This study analyzed the anterior and posterior margins of segmental mandibulectomies and identified the differences in necrotic changes between cortical and cancellous bone in advanced mandibular ORN. To our knowledge, there have been no prior studies analyzing the resection margins at both ends, or analyses of the differences in necrotic changes between cortical and cancellous bone in advanced ORN. This study illustrates the heterogeneity of necrotic changes in advanced mandibular ORN. Necrotic change is more prevalent in cortical bone than in cancellous bone. We report a rate of 91% necrotic change in cortical bone at the inferior border of the anterior margin. By contrast, all cancellous bones at the anterior margins were viable. The percentage of viable cancellous bones near the most advanced area of bone destruction was unexpectedly high: 64%. We also note that necrotic changes of cancellous bone at the posterior margins were found in three bone specimens (27%). The heterogeneity of bone viability between cortical and cancellous bone was probably due to differences in blood supply. The facial artery is the major extraosseous source of blood to mandibular body. Therefore, "swing" osteotomy, ligation of the facial artery, and RT may all affect blood flow to the mandible \[[@B23]\]. Considering the importance of blood supply to bone repair in the mandibular cortex, Saka et al. \[[@B24]\] performed a study of blood flow in human mandibles. They divided the mandible into three anatomical zones as follows: Zone I: mandibular body, beginning in the symphysis and ending at the connecting line between the retromolar area and the mandibular angle; Zone II: the caudal part of the mandibular ramus, located dorsally and cranially to Zone I, extending to the condylar base; and Zone III: the condyle (i.e., the condylar process with the mandibular head, located cranially to Zone II). The main blood supply to the cortices in Zone I is periosteal, deriving from the mental, submental, and sublingual arteries. Collateral supply is endosteal and periosteal, deriving from the inferior alveolar and mental arteries. In this study, almost all resected specimens were classified as Zone I. Necrotic changes, especially in the cortices near the inferior border of the mandible at the anterior margins, are probably due to reduction of periosteal blood flow caused by RT. We did not find a greater occurrence of necrotic changes of cortices in patients who underwent ipsilateral neck dissection for primary carcinomas, as compared with other patients. This suggests that necrotic changes of cortices are caused by damage to the microcirculation throughout the entire periosteum, rather than the absence of one dominant nutrient artery (i.e., the facial artery). Poor outcome in ORN is associated with minimal surgical debridement alone \[[@B25]\]. The refractoriness of ORN to minimal debridement may be because necrotic changes occur mostly in cortical bone rather than in cancellous bone. We unexpectedly found that cancellous bone near the site of bone destruction was viable at a high frequency. This finding suggests that surgeons may find bleeding from the bone marrow in surgical margins of ORN during surgery, irrespective of surgical extent. A review by Jacobson et al. \[[@B3]\] classified ORN into early, intermediate, and advanced stages. Early ORN should be managed conservatively with local wound care and the administration of antibiotics. Advanced ORN requires surgical management with wide extirpation of disease and simultaneous free flap reconstruction \[[@B3], [@B13]\]. Appropriate definitive treatment of intermediate stage ORN remains uncertain \[[@B3]\]. Jacobson et al. \[[@B3]\] recommend that, for intermediate ORN, all necrotic bone be debrided transorally up to the bleeding edge, and the wound should be primarily closed. We mostly agree with their recommendation; however the results of our study suggest that repeated minimal debridement in intermediate ORN creates the risk of advancing fragility of the residual bone. This is because necrotic changes are dominant in cortices near the inferior border of the mandible. The "hypoxic-hypocellular-hypovascular" theory of ORN pathophysiology was proposed by Marx in 1983 \[[@B10]\]. In 1990, Bras et al. \[[@B26]\] analyzed sequestrectomy specimens and reported that the dominant feature of mandibular ORN was ischemic necrosis due to obliteration of the inferior alveolar artery by RT. Revascularization by branches of the facial artery was disturbed by vascular disease and periosteal damage caused by RT \[[@B26]\]. They also proposed that the most vulnerable part of the mandible was the buccal cortex of the premolar, molar, and retromolar regions \[[@B26]\]. Imbalance of bone remodeling is another pathogenic feature of ORN. Bone remodeling involves a balance between osteoclast resorption and osteoblast deposition \[[@B27]\]. Local external irradiation causes a decrease in numbers of osteocytes and osteoblasts \[[@B27]\]. A study using mini pigs by Xu et al. \[[@B27]\] suggested that the decrease of local blood flow due to microvessel damage may be an initiating factor for ORN, followed by secondary induction of an imbalance in bone remodeling. Another model for pathogenesis of ORN is the "fibroatrophic" theory \[[@B28]\]. This theory proposes three clinical and histopathological phases: a prefibrotic specific inflammatory phase, a constitutive fibrotic cellular phase, and a matrix densification and remodeling phase, possibly ending in terminal tissue necrosis \[[@B28]\]. A histopathological study of human bone samples of ORN obtained from sequestrectomy samples suggested that ORN is characterized by increased collagen deposition (fibrosis) \[[@B17]\]. In our study, we found that the severity of fibrosis within the bone marrow cavity was different in each patient. Further analysis of fibrosis in advanced ORN is needed. The determination of resection extent for ORN remains an unresolved issue of importance. One serious postoperative complication of ORN surgery is residual, or recurrent ORN. Suh et al. \[[@B29]\] found a 25% rate of recurrent ORN after segmental mandibulectomy. Zaghi et al. \[[@B14]\] found that the presence of residual necrotic bone at resection margins of segmental mandibulectomies did not correlate with the recurrence of ORN. This was because all recurrence of ORN occurred in viable resection margins, and there was no progression of ORN at the resection margins in residual nonviable bone \[[@B14]\]. Our study suggests that complete extirpation of necrotic bone is sometimes impossible, despite resection of the destroyed bone area with a wide safety margin ([Figure 4](#fig4){ref-type="fig"}). Recurrence of ORN may be related to factors (e.g., infection) other than the presence of residual necrotic bone at the resection margin. The limitation of this study is the relatively short period of follow-up (median 18 months). Therefore, our study could not evaluate the recurrence and bone union between the residual mandible and transferred fibula flap through long-term observation. In the future we hope to perform a long-term follow-up study looking at the rate of ORN recurrence and union of bone junctions. 6. Conclusion {#sec6} ============= In ORN cases requiring surgical intervention, cortical necrosis, especially near the inferior border of the mandible, was more common than necrosis of cancellous bone. Cortical necrosis was probably due to a decrease in periosteal blood supply. Cancellous bone, even near the area of bone destruction, was viable in some cases. In cases of severe fibrotic ORN, complete extirpation of necrotic bone may be difficult even though segmental mandibulectomy with wide safety margins is performed. Conflicts of Interest ===================== The authors declare that there are no conflicts of interest to disclose. {#fig1} {#fig2} {#fig3} {#fig4} ###### Clinical characteristics of patients. ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- Case Sex Age Pathological diagnosis Primary site Types of RT/dose (Gy) Chemotherapy Surgery for\ Time interval\ Pathological\ Lesion location^*∗∗*^ Extent of\ primary tumor (months)^*∗*^ fracture SM^*∗∗∗*^ ------ ----- ----- ------------------------ ------------------------ ----------------------- -------------- ------------------ ---------------- --------------- ----------------------- ------------ 1 M 71 SCC Oral cavity Conventional/61.5 --- Tumor resection\ 120 \+ Ipsilateral Body biND/RF 2 M 58 SCC Oropharynx Conventional/70 CDDP uniND 104 − Contralateral Body 3 M 70 SCC Neck (unknown primary) Conventional/66 CDDP uniND 75 − Contralateral Body 4 M 62 SCC Oropharynx Conventional/70 CDDP/5-FU --- 78 \+ Contralateral A 5 F 80 SCC Oral cavity IMRT/60 --- Tumor resection\ 6 \+ Ipsilateral AT uniND/RF 6 M 66 AC Neck (unknown primary) Conventional/60 --- uniND 121 \+ Contralateral A 7 M 65 SCC Oropharynx Conventional/60 CDDP/NDP Tumor resection\ 137 − Ipsilateral A uniND/RF 8 M 64 SCC Neck (unknown primary) Conventional/81 CDDP/5-FU --- 152 \+ Ipsilateral AT 9 M 63 SCC Nasopharynx Conventional/70 CDDP/5-FU --- 56 − Contralateral Body 10 M 63 SCC Oropharynx Conventional/70 CDDP --- 84 \+ Contralateral Body 11 M 74 SCC Oropharynx Conventional/66 CDDP/5-FU Tumor resection\ 71 − Contralateral Body biND/RAMC ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- RT, radiotherapy; ORN, osteoradionecrosis; SCC, squamous cell carcinoma; AC, adenocarcinoma; IMRT, intensity-modulated radiotherapy; CDDP, cisplatin; 5-FU, fluorouracil; NDP, nedaplatin; biND, bilateral neck dissection; uniND, unilateral neck dissection; RF, radial forearm free flap; RAMC, rectus abdominis myocutaneous free flap; SM, segmental mandibulectomy. ^*∗*^Time interval between the end of RT and the day of surgical debridement and fibula flap reconstruction. ^*∗∗*^Ipsilateral: ORN occurred at the same side of primary tumor exposed to radiation. Contralateral: ORN occurred at contralateral side of primary tumor exposed to radiation. ^*∗∗∗*^Extent of segmental mandibulectomy was classified according to the CAT classification, described in detail in the text. ###### Histopathological results of bone specimens. Case Anterior margin Medial area  Posterior margin ---------------------------- ----------------- -------------- ------------------ -------- -------- -------- -------- 1 *NNNN* *VVVV* *VVVV* *VVVV* *VVVV* *VVVV* *VVVV* 2 *NNNN* *NNNN* *VVVV* *VVVV* *NNNN* *NVNN* *VVVV* 3 NVNV *VVVV* *VVVV* VNVV *VVVV* *VVVV* *VVVV* 4 *NNNN* VNNN *VVVV* *NNNN* *NNNN* *NNNN* NVNN 5 *NNNN* *NNNN* *VVVV* NVNV NVNN *VVVV* *VVVV* 6 *NNNN* NVNN *VVVV* *VVVV* *NNNN* *NNNN* *VVVV* 7 *VVVV* *VVVV* *VVVV* *VVVV* *VVVV* *VVVV* *VVVV* 8 *NNNN* *NNNN* *VVVV* *VVVV* NVNN *NNNN* *NNNN* 9 *NNNN* *NNNN* *VVVV* *VVVV* NVNN *NNNN* *VVVN* 10 *NNNN* *VVVV* *VVVV* *VVVV* *VVVV* *VVVV* *VVVV* 11 *NNNN* *NNNN* *VVVV* VVNN *NNNN* *NNNN* VVNN Concordance rate (%)^*∗∗*^ 91 82 100 73 73 100 82 ^*∗*^Cortical bone was histopathologically analyzed at two levels (near the inferior border and the middle level of the mandible).   ^*∗∗*^Independently evaluated by four observers. V, viable; N, necrotic. [^1]: Academic Editor: Li Wu Zheng

Introduction ============ Mechanical shunting of cerebrospinal fluid (CSF) is an effective treatment for non-obstructive hydrocephalus. In particular, ventriculoperitoneal (VP) shunting has become a preferred method in most clinical centers. Despite the wide acceptance of VP shunting, there are important complications associated with this technique. Common problems include obstruction, mechanical shunt failure and infections \[[@B1]\] while CSF ascites and thoracic complications, such as CSF hydrothorax, are less frequently observed sequelae \[[@B2]\]. Most cases of the lattermost complication, CSF hydrothorax, occur secondary to intrathoracic shunt tip migration \[[@B3]\]. However, a small number of cases, mostly in the pediatric population and secondary to massive ascites, have been reported with normal shunt position \[[@B2]\]. We report a 14-month-old Caucasian boy with a symptomatic pleural effusion that developed two and a half months post-VP shunt insertion for hydrocephalus of unknown etiology. We will review potential mechanisms behind the development of this complication and discuss the use of beta-2-transferrin for the identification of CSF hydrothorax in children with VP shunts. Case presentation ================= A 14-month-old Caucasian boy with idiopathic non-obstructive hydrocephalus and a VP shunt presented to the emergency room with a 1-day history of mild respiratory distress without cough or fever. The patient had presented at 11 months of age with a history of enlarging head circumference and developmental delay. A computed tomography (CT) scan of his head at 11 months of age revealed enlargement of the lateral, third and fourth ventricles and a bulky choroid plexus. A right-sided programmable valve VP shunt was inserted at that time. On examination, the 14-month-old patient was found to be afebrile and tachypneic (34 breaths/minute) with mildly increased work of breathing and decreased respiratory sounds in the distal aspect of the right lung. The cardiovascular exam was within normal limits. An examination of the abdomen revealed distension but no hepatomegaly or signs of peritoneal irritation. A chest radiograph (Figures [1](#F1){ref-type="fig"} and [2](#F2){ref-type="fig"}) revealed a large right-sided pleural effusion, which was confirmed by chest ultrasonography to be a fluid collection measuring 12.3 × 9.2cm. A shunt series demonstrated the VP shunt to be in place without signs of discontinuity or leakage. {#F1} {#F2} At the time of admission, venous blood gas, serum electrolytes and creatinine were within normal limits. Blood urea nitrogen and white blood count were mildly elevated at 5.1mmol/L and 12.5 × 10^9^/L, respectively. Further blood work demonstrated an alanine aminotransferase (ALT) level of 46U/L, an aspartate aminotransferase (AST) level of 51U/L, a serum calcium level of 2.61mmol/L and an alkaline phosphatase (ALP) level of 1403U/L. Subsequently, ALP steadily decreased to 498U/L but remained elevated throughout the patient\'s admission. A thoracentesis was performed and a chest tube was inserted which drained \>300cc/day of clear, yellow fluid. Around the time of chest tube insertion, the patient was noted to have an increasing abdominal girth, sizeable positive fluid balance and weight gain. An abdominal/pelvic ultrasound revealed marked ascites. Structurally normal major abdominal and pelvic viscera as well as normal vena caval and hepatic venous flows were reported. A head CT showed no change from a previous scan carried out at 11 months of age. A parallel pleural fluid and CSF analysis was also performed which is detailed in Table [1](#T1){ref-type="table"}. Secondary to the discordance between the white cell counts of the CSF and pleural fluid analyses, a sample of pleural fluid was sent for *β*~2~-transferrin assay and was found to be positive. Subsequent to this finding, the VP shunt was externalized followed by a dramatic decrease in chest tube drainage. A repeat chest radiograph revealed resolution of the pleural fluid collection and the chest tube was removed. The patient\'s abdominal girth decreased and a repeat abdominal/pelvic ultrasound demonstrated resolution of ascites. The patient\'s externalized ventricular drain was subsequently converted to a ventriculoarterial shunt. ###### Pleural and cerebrospinal fluid analysis Biochemical measure Pleural fluid Cerebrospinal fluid ---------------------- ---------------------------------- ---------------------------------- Glucose 5.1g/L 4.1g/L Total protein ≤10g/L \<0.1g/L Triglyceride \<0.11g/L White blood count 440 38 Red blood count 20 0 Lymphocytes 50 56 Mesothelial cells 13 Gram stain & culture No organisms seen; sterile fluid No organisms seen; sterile fluid Discussion ========== We report the use of *β*~2~-transferrin for the diagnosis of CSF hydrothorax in a child with a transudative pleural effusion. Although used previously in other applications, to the best of our knowledge, this is the first reported use of *β*~2~-transferrin to detect a CSF pleural effusion in a child with a VP shunt. A variety of techniques to investigate ascites and/or pleural effusion in patients with VP shunts have been reported in the literature. Among the described methods is the use of radiopaque contrast. This particular procedure involves the injection of radioactive dye at a point along the shunt pathway followed by subsequent imaging studies to trace the migration of contrast-injected CSF \[[@B4]\]. An alternative diagnostic strategy in cases of CSF ascites or hydrothorax involves repositioning of the distal shunt, usually through relocation of the catheter tip within the peritoneal cavity or into the right atrium, with subsequent resolution of CSF fluid collections \[[@B5]\]. A novel diagnostic strategy was utilized in the case of our patient. A sample of fluid, drained via tube thoracostomy, was sent for beta-2-transferrin level. This desialated isoform of transferrin is almost exclusively found in the CSF with only minimal amounts present in cochlear perilymph and in the aqueous and vitreous humor of the eye. Multiple studies have validated the use of beta-2-transferrin as a specific marker for CSF leakage with sensitivity and specificity approaching 100% and 95%, respectively \[[@B6]\]. Furthermore, Huggins and Sahn (2003) report the use of beta-2-transferrin to identify the presence of a CSF pleural effusion in an elderly patient with a duro-pleural fistula \[[@B7]\]. To our knowledge, however, beta-2-transferrin has never been previously utilized to identity CSF hydrothorax in children with VP shunts. Following externalization of the distal VP tip, our patient\'s ascites and pleural effusion promptly resolved. This finding, in combination with a positive beta-2-transferrin assay, was highly suggestive of a CSF hydrothorax. The findings of CSF hydrothorax and ascites, as demonstrated in our patient, are rare complications of VP shunts. Reported to occur anywhere from days to years postoperatively, CSF ascites usually develops within the first 2 years after VP shunt placement \[[@B8]\]. In greater than 50% of these cases, clinical and/or laboratory investigations fail to reveal an underlying abdominal or other disease process to account for CSF ascites \[[@B9]\]. We propose a pathological intraperitoneal process, such as subclinical peritonitis, to account for the development of CSF ascites in our patient. Inflammation of the peritoneal membrane may lead to intraperitoneal accumulation of fluid through the impairment of lymphatic flow \[[@B9]\] and/or by increasing intra-abdominal pressure and volume \[[@B10]\]. As evidence to date suggests that modern shunt materials are inert, alternative sources of peritoneal irritation need to be considered. For example, an immune reaction to a foreign protein such as a vaccine may be enough to precipitate ascites in patients with VP shunts \[[@B5]\]. As the timing of our patient\'s most recent immunization in relation to his presentation is unknown, this possibility cannot be eliminated. Furthermore, although our patient\'s CSF and pleural fluid cultures were sterile, it is not possible to definitely refute the presence of an infectious process. For example, a subclinical viral infection may have been sufficient to cause peritoneal inflammation in our patient with resultant CSF ascites \[[@B11]\]. Furthermore, a strengthened immune response in infants, coupled with age-related decreased peritoneal absorptive capacity, makes subclinical peritonitis an important consideration in our patient \[[@B8]\]. Alternative explanations for CSF ascites in the literature such as excess CSF production, surgical procedures, and high CSF protein are less plausible explanations for our patient\'s CSF ascites. Firstly, it has been suggested that CSF ascites can result from excessive CSF production, with the excess fluid overwhelming the absorptive capacity of the peritoneum \[[@B12]\]. Despite an undiagnosed etiology for hydrocephalus in our patient, the absence of an interval change on CT scan makes augmented CSF overproduction an unlikely precipitant for his ascites. Secondly, a history of shunt revision or previous abdominal surgery has also been reported in patients with CSF ascites \[[@B5]\]. Although our patient underwent a shunt surgery two and a half months before presentation, there was no history of shunt revisions or abdominal surgeries. Lastly, high CSF protein has been proposed as a potential cause of CSF ascites \[[@B9]\]. According to this supposition, an elevated CSF protein content may lead to reversed osmosis at the level of the blood-brain barrier with the resultant accumulation of excessive CSF fluid within the peritoneal cavity \[[@B13]\]. However, numerous cases of high CSF protein without resultant ascites have been reported \[[@B9]\]. Furthermore, as demonstrated in a previous case report, CSF ascites can occur despite normal levels of protein in the CSF \[[@B10]\]. The finding of CSF ascites in our patient can be used to explain the coincident finding of pleural effusion. Taub and Lavyne (2004) suggest three mechanisms to account for the development of CSF hydrothorax post-VP shunt placement \[[@B3]\]. Firstly, they propose that an error in surgical shunt placement, with resultant intrathoracic trauma, may account for some cases of CSF hydrothorax. Secondly, Taub and Lavyne suggest that pleural effusion may develop secondary to supra- or trans-diaphragmatic migration of the shunt tip into the thorax. Considering that no surgical complications during VP shunt placement were reported in our patient and that his shunt series at the time of admission was normal, an alternative explanation is required. Thus, it is the third mechanism proposed by Taub and Lavyne, the development of hydrothorax secondary to CSF ascites that offers the most promising explanation for our patient\'s pleural effusion \[[@B3]\]. In order for CSF ascites to result in pleural effusion, some degree of open communication must exist between the peritoneal cavity and the pleural space. Congenital diaphragmatic defects or weak points in the diaphragm such as the Foramen of Morgagni or Foramen of Bochdalek provide potential conduits. Furthermore, one or more microscopic diaphragmatic communications is likely to be sufficient for the transdiaphragmatic movement of peritoneal fluid \[[@B2]\]. These posited congenital or acquired fenestrations enable peritoneal fluid to pass into the pleural space along a unidirectional pressure gradient \[[@B14]\]. This posited pressure differential results from a cyclic negative intrathoracic pressure, created during inspiration, coupled with increased hydrostatic intra-abdominal pressure \[[@B2]\]. In addition to causing a build-up of ascitic fluid, sub-clinical inflammation likely contributes to this process by facilitating the transudation of fluid through diaphragmatic capillary and lymphatic channels \[[@B4]\]. Fluid subsequently accumulates in the pleural space secondary to high volume transdiaphragmatic flow and the overwhelming of pleural absorptive abilities \[[@B14]\]. As the pleural surface remains pathology free with its sieve characteristics consequently intact, the cell and protein content of the ensuing pleural effusion should remain low \[[@B15]\]. These pleural fluid characteristics, along with a predominance of lymphocytes and mesothelial cells, suggest a transudative pleural effusion. Although no gross diaphragmatic defect was identified in our patient, the presence of microscopic diaphragmatic communications may have been present. Notably, the tapped pleural fluid characteristics of our patient were found to be in keeping with a transudative effusion. The collected pleural fluid also demonstrated a marginally higher protein concentration than in the patient\'s CSF fluid. Differences in the solute composition of ventricular-extracted CSF and pleural fluid have been previously demonstrated in patients with pleural effusion secondary to VP shunts \[[@B2]\]. The finding of a slightly higher protein concentration in the pleural verses peritoneal fluid, the latter having been derived from CSF, can be explained by the preferential absorption of water over protein across the visceral pleura \[[@B15]\]. With this in mind, inconsistent solute characteristics, such as protein concentration, between pleural effusion and pure CSF should not be used to refute the possibility of CSF hydrothorax in patients with VP shunts. Conclusion ========== In summary, we present the case of a 14-month-old boy with CSF hydrothorax secondary to VP shunt. Pleural effusion following VP shunt insertion is a rare and potentially life-threatening event and hence requires prompt diagnosis and correction. We report a novel and non-invasive technique by which to identify the presence of CSF in pleural fluid. Beta-2-transferrin assay may serve as a useful new means for the identification of CSF hydrothorax in the context of patients with VP shunts. Consent ======= Written informed consent was obtained from the patient\'s mother for publication of this case report and any accompanying images. A copy of the written consent is available for review by the Editor-in-Chief of this journal. Competing interests =================== The authors declare that they have no competing interests. Authors\' contributions ======================= Both authors contributed to the conception of this case report. Dr. Jennifer Smith drafted the report while Dr. Eyal Cohen provided critical feedback. Both authors have reviewed and approve of the final version. Acknowledgements ================ The authors would like to thank the patient and his mother for making this manuscript possible.

Introduction ============ Choices between uncertain rewards require decision-makers to evaluate each option along multiple dimensions. At the very least, a decision-maker needs to simultaneously consider the reward\'s quantity and probability of occurrence if he is to evaluate its attractiveness in relation to other choice prospects. The von Neumann and Morgenstern Expected Utility (EU) theory was the first axiomatic model of rational behavior capable of describing people\'s choices in these situations ([@B37]). EU theory rigorously introduced the concept of utility as a representation of a decision-maker\'s subjective value for an objective reward quantity. Through the metric of utility, EU theory was able to describe different risk attitudes, such as the risk-seeking behavior of a gambler or the risk aversion of an insurance buyer; it was, however, soon challenged by the various experimental results of behavioral economics ([@B38]; for review, see [@B24]; [@B32]). Attempts to resolve some of these challenges led to the development of several generalized expected utility theories, many of which (notably prospect theory, rank-dependent utility theory, and cumulative prospect theory) incorporated the concept of probability distortion ([@B20]; [@B31]; [@B35]). While maintaining the nonlinear relationship between subjective utility and objective reward magnitudes, these theories made use of subjective probability weightings, or probability distortions, to account for the idea that reward probabilities were also treated nonlinearly during choice. Experimental measures of probability distortion in humans and monkeys typically show that, whereas small probabilities tend to be overweighted by decision-makers, large probabilities are instead underweighted ([@B20]; [@B11]; [@B33]). There is, however, dramatic variation in this pattern of distortion across both different subjects ([@B11]; [@B5]; [@B6]) and between different task contexts ([@B17]; [@B40]; [@B9]). While the causes of such variability have yet to be identified, differences in probability distortions could relate to the way in which probability information is presented to decision-makers ([@B17]), or the way in which probability knowledge is acquired and stored by the decision-maker ([@B8]). Some studies suggested that prospect theory might altogether be incapable of explaining differences in risk attitudes across these contexts ([@B21]). Here we investigated the role of choice context, specifically sequence structure, as a possible source of probability distortion variability in rhesus macaques, animals known to show quantifiable and reproducible probability distortions ([@B33]). To achieve this, we first measured the certainty equivalents (CEs) of specific gambles, defined as the amount of reward for which the animal was choice-indifferent with regards to said gambles; the CE therefore indicated the subjective value of the gamble in the "currency" of the safe reward. We then simultaneously estimated the contributions of utility and probability distortion to these subjective values, allowing us to model the shape of the monkeys\' probability distortion independently from utility. We used this technique to investigate the possible influence of trial sequence structure on the shape of the probability distortion in two different task situations: randomly intermixing the trials required for the CE measurements of all gambles, or determining the CEs of different gambles via separate blocks of trials. We performed an out-of-sample test to validate and extend the results of our main task, and investigated the contribution of trial history as a possible correlate of probability distortion variance. Our data showed that a change in the presentation order of probability information indeed altered the observed probability distortion pattern, inducing a reversal in probability distortion shape. Materials and Methods ===================== ### #### Animals and experimental setup. Two male rhesus macaques (*Macaca mulatta*) were used in this study (11.2 and 13.2 kg). During experiments, the monkey sat in a primate chair (Crist Instruments) and made choices between two rewarding stimuli presented on a computer monitor positioned 30 cm in front of them. The animals reported their choices between options with a left-right motion joystick (Biotronix Workshop). Joystick position and task event times were sampled and stored at 1 kHz on a Windows 7 computer running custom-made software written in MATLAB (The MathWorks) using Psychtoolbox (version 3.0.11). All experimental protocols were assessed and approved by the Home Office of the United Kingdom. #### Experimental design. We trained the monkeys to associate visual stimuli with specific juice rewards that varied along two dimensions: the quantity of juice delivered (reward magnitude, *m*) and the delivery probability of the reward (reward probability, *p*). To capture both dimensions descriptively, the visual stimuli consisted of a horizontal bar or of a pair of horizontal bars framed between two vertical lines. The vertical position of the horizontal bars signaled the magnitude of juice delivered; the width of the bar signaled the probability of their delivery from no bar (no reward) to touching the frame on both side (certain reward). To ensure that the bar\'s edge position relative to the frame was not used as a cue for the gamble\'s mathematical expected value (EV; i.e., the product of *m* and *p*), the bars were randomly shifted horizontally on each trial. This guaranteed that magnitude and probability information were independently presented and used to make choices. Multiple partial bars found between the vertical frames signaled gambles between "risky" rewards, whereas a singular, full-width horizontal bar signaled a safe, riskless reward. Across all trials, monkeys experienced rewards ranging from 0 ml to 0.5 ml in 0.05 ml increments, and gamble probabilities varying between 0.1 and 1 in decimal increments (0.1). The animals learned to associate rewards and magnitudes with the visual stimuli schema through \>10,000 single-outcome, or imperative, trials. In these trials, only one option was presented on either side of the screen. To obtain the cued reward, the animals were required to select the side on which the reward was presented. All reward options were repeated on both the left and right sides of the computer screen, alternating pseudorandomly to control for any side preference. Following imperative training, we presented the animals with a binary choice paradigm where they had to choose one of two reward options presented simultaneously. Most binary choice trials pitted a safe reward against a gamble. All gambles consisted of two probabilistic rewards: the monkey could either get a fixed 0.5 ml of juice with probability *p*, or 0 ml of juice with probability 1 − *p*. Safe options varied in terms of reward magnitude only. In separate sets of trials, we presented the animals with choices between two gambles with two outcomes each (possible outcomes: 0, 0.25, 0.5 ml). In these trials, one of the gambles could have two non-zero outcomes (0.25 and 0.5 ml). In all cases, reward was delivered probabilistically, matching the probabilities cued by each stimulus. Trials began with a white cross at the center of a black screen, followed by the appearance of a joystick-driven cursor. The cursor had to be moved to the center cross in order for a trial to begin. After successfully maintaining the cursor on the central cross for 0.5--1 s, two visual option cues appeared left and right of the central cross (see [Fig. 1](#F1){ref-type="fig"}*a*). In the case of imperative trials, only one option appeared while the other side remained dark. The animal had 3 s to convey his decision by moving the joystick to the selected side, after which the unselected option would disappear. The animal\'s response time (RT; i.e., the time interval between the cues appearance and the beginning of the joystick movement) was collected for individual trials. Reward delivery occurred after the holding time (0.1--0.2 s), and the selected option lingered on the screen for 1 s postreward delivery to reinforce stimulus--reward associations with visual feedback. A variable intertrial period of 1--1.5 s (blank screen) led to the next trial onset. Unsuccessful central hold, side selection hold, or trials where no choices were made resulted in a 6 s timeout for the animal, after which the trial would be repeated. #### Psychometric elicitation of CEs. The likelihood of a monkey choosing a specific, individual gamble over different safe options was assessed through the binary choice paradigm (see [Fig. 1](#F1){ref-type="fig"}*b*). The resulting choice ratios were then used to fit a logistic sigmoid function, or psychometric curve, to estimate choice likelihoods for every possible safe-gamble pairing within the tested reward range as follows: These psychometric curves captured the likelihood of choosing a safe option over a gamble through two free parameters: x~0~, measuring the *x* position of the curve\'s inflection point, and σ, the function\'s temperature parameter, reflecting the steepness of the curve. Importantly, only sequences that contained choices between a gamble and a minimum of three different safe options (repeated at least 4 times) were used in the analysis. The point of choice indifference between gamble and safe options, corresponding to the inflection point x~0~ of the resulting model, represented a gamble\'s CE: the certain safe reward that was of equal subjective value to the gamble. CEs could then be used to categorize behavior. Gambles where the CEs were of greater value than the predicted EV signaled risk-seeking behavior for that gamble\'s probability value. Gambles with CEs lower than their EVs indicated risk-averse behavior for that option. For cases where CEs were equal to EVs, the animals were seen as being risk-neutral. To explore the role of task structure on the variability of one\'s probability distortion pattern, we measured CEs in one of two elicitation conditions: MIXED or REPEATED trial sequences (see [Fig. 1](#F1){ref-type="fig"}*c--e*). In the case of MIXED sequences, multiple CEs were elicited through single blocks of randomized choice trials involving different gambles and safe options. Such blocks were repeated until each gamble-safe pair had been presented a minimum of 4 times each. In the case of REPEATED sequences, CEs were elicited using blocks of trials that contained a unique gamble. These REPEATED trial blocks pitted multiple safe options against a single gamble for the elicitation sequence. Other than these sequence designs, everything from visual cues to timescales was identical. The only difference between elicitation conditions was the number of different probabilities of reward (gambles) experienced within a trial block. Testing for each elicitation condition was done consecutively over multiple days, with each monkey receiving imperative training before their daily elicitation sessions. We collected on average 172.95 ± 20.24 (SEM) trials per daily session over 56 sessions for Monkey A (22 REPEATED and 34 MIXED sessions, in consecutive days), and 414.63 ± 27.87 trials over 59 sessions for Monkey B (31 REPEATED and 28 MIXED sessions, in consecutive days). #### Analysis of behavioral data. All data were collected, stored, and analyzed using custom MATLAB and Python (SciPy 1.1.0) ([@B29]) software. Analyses were run on trial-by-trial choice data, and on the CEs elicited psychometrically from these trial-by-trial choices. The data were stored and analyzed separately for the 2 animals. Before any comparative analyses, the use of visual stimuli to guide the monkeys\' decision behavior was verified through analyzing all CE elicitation trials (excluding error trials where the animals made no choices) in a logistic regression model as follows: The dependent variable took a value of *y* = 1 if the gamble was chosen and *y* = 0 if the safe option was chosen instead. As had been previously done ([@B33]), we fitted four independent variables: option values (*EV~gamble~*, *EV~safe~*) were defined as the EVs of gamble and safe rewards; gamble position (*Position~LR~*) as 0 for left, 1 for right screen side; and the outcome\'s risk value (*Risk*) was defined as $\sqrt{p*\left( {1 - p} \right)}$, a proportional representation of probabilistic variance. We fitted individual testing days separately, fully standardizing the β coefficients and then testing for statistical significance (one-sample *t* test, *p* \< 0.05) to identify relevant decision variables. Positive regression coefficients indicated an increase in the likelihood of choosing a gamble over a safe option with increasing independent variable value; negative regression coefficients indicated a decrease in the likelihood of choosing the gamble. Once the use of onscreen stimuli to guide choices had been confirmed, CEs were measured using the aforementioned psychometric fit. CEs gathered in the MIXED condition were compared with CEs gathered under the REPEATED condition using a two-factor ANOVA with gamble probability and elicitation condition as main factors. The ANOVA also captured any interaction between the two factors, highlighting any condition effects present at a sequence level. We used trial-by-trial choices to parametrically model the respective effects of utility and probability distortion on single choices, and more generally, on the subjective value of gambles (CEs). For each daily testing session, we simultaneously estimated both the utility and probability distortion functions from within a standard discrete choice model. Functional parameters that best described choices between gamble-safe pairs were elicited in this way, capturing the individual effects of nonlinear utility and probability distortion. The model ran on trial-by-trial choice data, with data binned into several sets containing one gamble and all safe options presented against it on the day (CE elicitation sequence). The discrete choice (softmax) function returned the probability of choosing the gamble option based on the subjective value of both the gamble (*V~G~*) and the safe reward presented (*V~S~*). The softmax parameter, λ, defined the likeliness of choosing the better prospect; each option\'s value (*V*) was defined according to prospect theory ([@B20]), as the product of utility (*u*) and probability distortion (*w*) outputs as follows: Utility was modeled through a power function as follows: where ρ \> 1 captured risk-seeking choice behavior, ρ \< 1 captured risk-averse choice behavior (ρ \< 1), and ρ = 0 implied risk neutrality ([@B19]). Magnitude values were divided by 0.5 ml (*m~max~*), such that the maximal reward a monkey could get was anchored at 1 unit of utility. We compared four functional models of probability distortion in an attempt to best capture changes in probability distortion across conditions. Of these classical models, two had a single fitting parameter: the one-parameter Prelec function ([Eq. 6](#FD6){ref-type="disp-formula"}, *Prelec-1*, parameter: α) and the Kahneman and Tversky probability weighting function ([Eq. 7](#FD7){ref-type="disp-formula"}, *Tversky,* parameter: ε); the others had two fitting parameters: the two-parameter Prelec function ([Eq. 8](#FD8){ref-type="disp-formula"}, *Prelec-2*, parameters: α, β) and the Gonzalez and Wu log-odds model ([Eq. 9](#FD9){ref-type="disp-formula"}, *Gonzalez*, parameters: γ, δ). Formally: Using a maximum likelihood estimation (MLE) method, we simultaneously estimated the functional parameters (Θ) from the experimental data. We defined the log-likelihood function as follows: The log-likelihood function was defined on all trials in a session (*n*), the trial number (*i*), and the choice outcome variable for the gambles and safe options (*y* and *y′*, respectively). The outcome variables took a value of 1 if their respective option was chosen; 0 otherwise. We used an unconstrained Nelder--Mead search algorithm (MATLAB: fminsearch) to compute the functional parameters that minimized the negative log-likelihood (−*LL*). This MLE approach allowed for the simultaneous estimation of the model\'s free parameters, placing no constraints on their values ([@B1]; [@B30]; [@B33]). The algorithm identified the best fitting softmax, utility, and probability distortion parameters with respect to each monkey\'s daily choices in CE elicitation sequences. Four complete models were parametrized, accounting for the different probability distortion functions investigated. From these, we calculated the Bayesian Information Criterion (BIC) to pinpoint the probability distortion function most reliable in capturing behavior. Four sets of parameters and their BIC were estimated for every testing day, independently for each model. We selected a single model for further analysis, based on the flexibility of the functional model, its comparative BIC score (one-factor ANOVA with repeated measures, Greenhouse-Geisser--corrected *p* values: *pGGc*), and the deviance between the model\'s predicted CEs and the experimental ones (one-factor ANOVA with repeated measures, Greenhouse-Geisser--corrected *p* values) ([@B12]). We further validated the parameter estimation procedure by running 10 simulated choice datasets within the fitting algorithm. Datasets used for testing were generated by fixing the utility parameter (ρ) and varying the probability distortion parameter (α), or vice versa. The softmax temperature parameter was kept constant (λ = 10), as we specifically wanted to test the robustness of the estimation procedure in relation to variability in the utility and probability parameters. These fixed models were used to simulate individual trial choices. We simulated 6 trials for every gamble-safe pair (safe magnitude levels: 0--0.5 ml in steps of 0.05 ml). Five datasets varied in terms of utility (ρ = 0.20, 0.50, 1.00, 1.50, 3.00), and five in terms of probability distortion (α = 0.33, 0.67, 1.00, 1.50, 3.00). We measured estimation accuracy as the 95% CI on estimated parameters from Monte Carlo simulations on the parameter-derived datasets. The final estimated parameters were first log-transformed to account for the asymmetric distribution of the utility and probability distortion parameters (ranging from 0 to ∞, with a value of 1 defining the linear case). We then compared the parameter estimates via one-way MANOVA with elicitation condition as main factor. From this multivariate analysis, we identified any significant effect of individual decision functions while recognizing the collective role all three parameters in capturing risk preference. More specifically, the MANOVA identified which model function parameters (choice softmax, utility, or probability distortion) differed significantly between CE elicitation conditions. In the REPEATED condition, the gamble option did not change for long sequences of trials and could, theoretically, be ignored. To test the possibility of an attentional shift toward the safe option in this condition, we defined a model with different weights applied to the two options\' values as follows: The weight parameter (*k*) captured the attentional shift toward one option, if significantly \>0.5. The options\' values (*V~G~*, *V~S~*) were computed, as in the previous model, using the power utility function and the selected probability distortion function (*Prelec-1*). #### Evaluation of probability distortion in the Marschak--Machina triangle. We introduced the Marschak--Machina triangle ([@B25]; [@B23]) to compare the choice behavior between the MIXED and REPEATED conditions in an out-of-sample test, and to evaluate the theoretical predictions of the discrete choice model vis-à-vis utility and probability distortions. The Marschak--Machina triangle defines a 2D space where any probabilistic combination of three fixed reward magnitudes *m*~1~ \< *m*~2~ \< *m*~3~ can be represented (for details, see Results). The *x* and *y* axes correspond to the probability of obtaining the lowest (*p*~1~) reward *m*~1~ and the highest (*p*~3~) reward m~3~, respectively. The probability of the middle magnitude is not explicitly represented in the diagram, but it can be readily obtained as *p*~2~ = 1 − (*p*~1~ + *p*~3~). Points on the horizontal axis therefore correspond to gambles with outcomes *m*~1~ and *m*~2~, whereas points on the vertical axis identify gambles with *m*~2~ and *m*~3~ as possible outcomes; the hypotenuse comprises all gambles containing outcomes *m*~1~ and *m*~3~ only. In our experiment, we set the fixed magnitude levels to *m*~1~ = 0 ml, *m*~2~ = 0.25 ml, and *m*~3~ = 0.5 ml. We characterized Monkey A\'s behavior within the Marschak--Machina triangle, by defining indifference lines between points on the triangle edges as follows: we presented choices between a fixed gamble (A), defined on one of the axes, and a set of gambles (B~i~) located on the triangle\'s hypotenuse; by fitting a psychometric curve to the ratio of B~i~ and A choices, we identified the indifference point on the hypotenuse as the probability *p*~3~ corresponding to a choice ratio of 0.5. We then defined an indifference line as the segment connecting the fixed gamble on the axis with its corresponding indifference point. This procedure was repeated for four fixed gambles on the *x* axis (*p*~1~ = 0.2, 0.4, 0.6, 0.8) and for another four fixed gambles on the *y* axis (*p*~3~ = 0.2, 0.4, 0.6, 0.8), resulting in 8 indifference lines. Such indifference lines characterized points on the triangle edges (two-outcome gambles): they did not represent complete indifference curves within the Marschak--Machina triangle (three-outcome gambles). Nevertheless, the slopes of the indifference lines univocally identified a directional property a monkey\'s risk preference pattern: a gradual change in the slope (fanning-in or fanning-out) of indifference lines has been extensively used in the economic literature to characterize choice behavior, particularly in relation to the predictions of generalized expected utility theories. This property allowed us to quantify behavioral changes across elicitation conditions and to compare the observed data with predictions from the theoretical economic model. Crucially, gambles resting on the two axes were never used in the elicitation of CEs, representing an out-of-sample test. As a consequence, the choice behavior observed in the Marschak--Machina triangle could be used as independent validation for our previous results. We computed the theoretical indifference lines by calculating, for each of the eight fixed gambles defined above, the probability *p*~3~ for which the theoretical subjective value of the fixed gamble was equal to that of the gamble on the hypotenuse. The subjective value of a two-outcome gamble was defined according to cumulative prospect theory as follows: where *m*~3~ and *m*~1~ represent the magnitude of the highest and lowest outcome, respectively, *p*~3~ the probability of occurrence of the highest outcome, *u* the power utility function, and *w* the *Prelec-1* probability distortion function. The indifference point was defined as the point on the hypotenuse with subjective value (*u*(*m*~3~) · *w*(*p*~3~)) equal to the subjective value of the fixed gamble (*V*(*gamble*)). Thus, knowing the value of the fixed gamble, one could identify the indifference point as the probability *p*~3~ satisfying the equation *u*(*m*~3~) · *w(p*~3~) = *V*(*gamble*) as follows: where *w*^−*1*^ represents the inverse of the probability distortion function: that is, *w*^−*1*^ = *exp(*−*(*−*ln(w))*^*1*/α^*).* Each daily set of indifference points was elicited after CE elicitation sequences, for both the MIXED and REPEATED CE elicitation sessions. This resulted in two sets of indifference lines, distinctly associated with the REPEATED and MIXED conditions. Both datasets were obtained using intermingled gamble sequences, so any difference in the pattern of indifference lines could only be attributed to the effect of the previous block of trials (i.e., REPEATED or MIXED CE elicitation). The directional pattern of the indifference lines was characterized by a measure of the "fanning" direction, corresponding to a gradual change in the slopes of indifference lines. When moving from the lower right to the top left corner of the Marschak--Machina triangle, indifference lines decreasing their slope would fan-in, whereas indifference lines increasing their slope would fan-out, much like the structural slats of a folding fan. A linear regression analysis on the indifference line slopes was used to statistically characterize the fanning pattern. A positive regression coefficient identified fanning-out of the indifference lines, whereas a negative regression coefficient identified fanning-in. It should be noted that the relation between the slopes of the indifference lines, as we defined them, was not expected to be linear, but the linear regression served as a reasonable description of the expected theoretical pattern and was then used to characterize the measured behavior. To statistically compare the predicted and observed sequence effects on the steepness of the indifference lines, we first calculated the shift of indifference points (change in *p*~3~ value) between the REPEATED and MIXED conditions; we did this for each of the eight indifference lines, for both the measured data and the model\'s predicted lines. We then performed a correlation analysis on the modeled and measured shifts. #### Trial history effects. Because gamble presentation order was the only difference between the MIXED and REPEATED elicitation sequences, we sought to categorize the effects of said order on the subjective distortion of probabilities. Using past gamble EVs as a quantitative measure of past experiences (specific to probabilities) we compared the distribution and use of previous gamble EVs across elicitation condition. We first compared the variability of consecutive gamble probabilities in both conditions using a two-sample *t* test. We used the absolute value of consecutive gamble EV differences to contrast order in an unsigned manner, as signed differences would amount to 0 in both cases. We then assessed the use of past gamble EVs using the following logistic regression: Again, the dependent variable took a value of *y* = 1 if the gamble was chosen and *y* = 0 if the safe option was chosen instead. The EV of both the current gamble and safe (*EV~gamble~*, *EV~safe~*), as well as the gamble EV of up to 8 trials in the past (*EV*~*gamble*−*n*~), served as independent variables. Trials that did not have a minimum of 8 previous trials, in individual sessions, were removed for this analysis. We again standardized regression coefficients and identified how many past gamble EVs had a significant impact on current choice (one-sample *t* test, *p* \< 0.05). Refining the analysis to a singular preceding trial, we investigated the use of a win-stay/lose-shift (WSLS) strategy by the animals. A common strategy for human and nonhuman primates alike, a WSLS choice pattern involves repeating a "winning" choice until it results in a "loss," after which one would shift and try their luck on another choice option. Because choice options in the CE elicitation sequences involved many different values for both the gamble and the safe options, we instead explored a more relaxed WSLS model as follows: If the previous choice had been that of a gamble, and that gamble had won (i.e., resulted in a 0.5 ml reward), the third independent variable (*Outcome~past~*) took a value of 1; if the past chosen gamble had instead been unsuccessful, *Outcome~past~* was 0. By including current *EV~Gamble~*, *EV~Safe~*, and *Position~LR~*, we could identify the relative effect of a previous gamble\'s outcome on current choice. The logistic regression analysis was only applied to trials in which the previous trial\'s gamble was chosen. A positive regression coefficient for *Outcome~past~* implied a greater likelihood of picking the gamble after a "win", regardless of its value. A negative coefficient would, instead, capture a decrease in the likelihood of picking the gamble, whatever it may be, after a "loss." To compare the performance of this model with the previously defined model ([Eq. 2](#FD2){ref-type="disp-formula"}), which did not include the contribution of past trials, we computed the BIC scores of the two models only in trials in which the previous gamble was chosen. After this trial selection, we removed 5 sessions in Monkey A\'s data, as they had fewer than 4 trials per gamble-safe pair. To further investigate the effect of past outcomes on the risk patterns, we defined a reinforcement learning model, in which each gamble value was updated, starting from its EV, by adding or removing a fixed amount following a win or a loss, respectively. Formally, choices were evaluated according to the discrete choice model defined earlier ([Eq. 2](#FD2){ref-type="disp-formula"}), in which the safe value (*V~S~*) was the certain option\'s magnitude (linear coding of magnitudes), whereas the gamble value (*V~G~*) was updated on each trial according to the following rule: Where *pre~Win~* and *pre~Loss~* are variables encoding the last trial\'s outcome: *pre~Win~* = 1 if a gamble was won in the previous trial, 0 otherwise, and vice versa for *pre~Loss~*. The value-updating parameter η represents the amount of "value" (in milliliters) added or removed to the gamble value based on the previous outcome. According to this model, the gamble value was not updated if the safe option had been chosen on the previous trial. We retrieved the η parameter value using MLE, and used the resulting average value to simulate choices and compute the resulting CEs. The simulation was run on MIXED and REPEATED sequences separately, to compare the effect of a value-updating model on the CEs in the two sequence conditions. #### Statistical analysis. We used MATLAB and/or Python for all statistical analyses. Logistic regressions were computed per session, and results were standardized by multiplying each coefficient with the ratio of the corresponding independent variable\'s SD over the SD of the predicted variable ([@B27]). Standardized regression coefficients were tested for statistical significance through one-sample *t* test. Two-factor ANOVA, one-factor MANOVA, linear regression, and *t* test results were considered significant at *p* \< 0.05, whereas one-way repeated-measures ANOVAs were Greenhouse--Geisser corrected (degrees of freedom adjustment) to account for sphericity violations (Mauchly\'s test *p* \< 0.05; [@B12]). *Post hoc* analysis with Bonferroni--Holm correction for multiple comparisons was applied to ANOVA results. Cohen\'s *d* values were used as a measure of effect sizes. In all analyses of data from single sessions, we reported mean ± SEM across sessions. Results ======= Design ------ We tested whether the shape of the probability distortion would be influenced by the order in which probability information is presented in a sequence of decisions. Once the animals had been extensively trained with the reward-predicting stimuli (\>10,000 trials), we presented them with sequences of binary choices between different probabilistic rewards (or gambles) and safe rewards ([Fig. 1](#F1){ref-type="fig"}). We then used the choice ratios to measure the value of gambles relative to certain rewards, pinpointing the certain rewards that were subjectively equivalent to gambles, or a gamble\'s CE. This procedure revealed the animals\' attitude toward risky choices: gamble CEs larger than said gamble\'s objective EV reflected risk-seeking behavior; risk-aversion was characterized instead by gamble CEs smaller than the gamble\'s EV. {#F1} By simultaneously estimating the individual contributions of utility and probability distortion to these measures of risk attitudes, we could model the shape of the monkeys\' probability distortion regardless of the utility function. Basic behavioral performance ---------------------------- A logistic regression analysis demonstrated that the monkeys used the information from the visual stimuli to guide their decisions on all daily testing sessions ([Fig. 2](#F2){ref-type="fig"}*a*). A positive regression coefficient for gamble EV (one-sample *t* test, Monkey A: *t*~(55)~ = 29.41, *p* = 2.5 × 10^−35^; Monkey B: *t*~(58)~ = 30.16, *p* = 3.9 × 10^−37^) indicated that animals were more likely to choose higher probability gambles than lower probability ones; conversely, the negative coefficient for safe reward EV (Monkey A: *t*~(55)~ = −44.65, *p* = 6.8 × 10^−45^; Monkey B: *t*~(58)~ = −58.61, *p* = 2.6 × 10^−53^) indicated that monkeys chose the safe option more frequently when its value was of higher magnitude. Both animals preferred gambles of higher over lower probabilistic variance (i.e., they preferred gambles that were more uncertain, regardless of the outcome) (positive coefficient for risk; Monkey A: *t*~(55)~ = 4.58, *p* = 2.7 × 10^−5^; Monkey B: *t*~(58)~ = 7.79, *p* = 1.4 × 10^−10^). Monkey A, but not Monkey B, showed a side bias (positive coefficient for the position variable), which was taken into account by balancing the positions of gambles and safe rewards: every option was presented the same number of times on each side of the computer monitor. ![Basic choice behavior and estimation of CEs. ***a***, Logistic regression of choice behavior. Four task variables (gamble EV, safe EV \[magnitude\], risk variance, gamble position) were used as regressors for the gamble choice. Positive standardized coefficients for gamble EV and risk indicated that monkeys preferred gambles with higher EV to gambles with lower EV, and more risky gambles to less risky ones. Negative coefficient for safe EV confirmed that monkeys preferred higher reward magnitudes to lower ones. The positive position factor for 1 monkey indicated a side bias, which was taken into account by repeating all choice pairs with inverted left-right positions.\* Significant regression coefficient (one-sample *t* test). ***b***, Psychometric estimation of CEs. CEs of two example gambles with probabilities 0.1 (top) and 0.8 (bottom), estimated in the two different elicitation sequences: MIXED (red) and REPEATED (blue) sequences. The percentages of safe choices as a function of safe magnitude (circles) were fitted to softmax functions (curves). Vertical lines indicate the gambles EVs (dashed lines). Filled circles represent the CEs. In both monkeys, low probability gambles (top) had a lower CE in the REPEATED condition than in the MIXED condition, where the CEs were consistently higher than the EVs, indicating risk seeking behavior. High probability gambles (bottom) showed the inverse pattern, indicating risk seeking behavior only in the REPEATED condition. ***c***, Pattern of CEs across the two elicitation sequences (MIXED vs REPEATED). Single session CEs (small data points) and average CEs across sessions (large data points) plotted as a function of gamble EV, with cubic spline interpolated curves. The full pattern of CEs shows a smooth transition from low to high probability gambles in terms of CE difference across the two elicitation sequences. For low probability gambles, both monkeys showed higher CEs in the MIXED than in the REPEATED conditions; when increasing gamble probabilities, the CE difference across conditions gradually reduced and inverted, resulting in lower CEs in the MIXED than in the REPEATED condition for high reward probabilities. Single session data points were shifted horizontally (REPEATED condition: left; MIXED condition: right) for visualization purposes. ***d***, Mean RT (± SEM across sessions) in the two elicitation conditions. RTs for Monkey A were similar in the two conditions (RT difference = 3.0 ms, *t*~(9088)~ = −0.59, *p* = 0.56); Monkey B showed faster response in the MIXED condition compared with the REPEATED condition (RT difference = 30.0 ms, *t*~(22233)~ = −15.88, *p* = 1.77 × 10^−56^) (for RT as a function of the options\' EV, see [Figure 2-1](#fig2-1){ref-type="supplementary-material"}). \*Significant RT difference between conditions (two-sample *t* test).](zns9991915170002){#F2} 10.1523/JNEUROSCI.1454-18.2018.f2-1 Response time vs EV. Top: Mean RT (± SEM across sessions) as a function of EV difference between the two presented options (gamble EV - safe magnitude). Choices between options with similar EV produced higher RT. Bottom: Mean RT (± SEM across sessions) as a function of the EV of the chosen option. Faster RTs were associated to higher EV of the chosen option, while slower RTs corresponded to choices where a low EV option was selected. Download Figure 2-1, TIF file Estimation of subjective values using different sequence structures ------------------------------------------------------------------- We used a binary choice paradigm to estimate the monkeys\' subjective valuation of specific gambles. We measured the choice ratios between different safe rewards and gambles ranging in probabilities from *p* = 0.1 to *p* = 0.9. Fitting a softmax curve to each of these gamble-safe groups allowed us to estimate the CEs corresponding to different gamble probabilities (see Materials and Methods). These CEs served as a measure of subjective value for unique probabilities and provided us with a precise measure of an animal\'s risk preference over the range of probabilities tested. We elicited CEs in both monkeys using two different elicitation conditions: MIXED and REPEATED gamble sequences ([Fig. 2](#F2){ref-type="fig"}*b*). In the MIXED condition, we estimated CEs from sequences of binary choices containing several different gambles pitted against safe rewards. All gamble and safe options presented were randomly intermixed, and multiple CEs were estimated from these sequences, one for each gamble. In the REPEATED condition, we elicited CEs from blocks of trials that contained a single, unique gamble versus different safe rewards. In this way, we elicited a unique gamble\'s CE for each given block. Importantly, the two conditions used the same visual stimuli; any difference between estimated CEs would therefore be due to the elicitation sequence in which CEs were estimated. We aggregated the daily CEs of individual monkeys, for both conditions, to determine the risk-preference pattern derived from the CEs measured in each elicitation sequence. The risk-preference pattern was therefore directly inferred from the relation between the CEs and the respective EVs, as opposed to being theoretically derived from the shape of utility and probability distortion functions. We found a significant difference between the distribution of CE values elicited in REPEATED versus those elicited in MIXED sequences (two-way ANOVA, factors: gamble probability, elicitation condition). As expected, we found a significant main effect of reward probability on a gamble\'s CE: higher probability gambles had a higher CE in both animals (Monkey A: *F*~(8,237)~ = 444.12, *p* = 5.2 × 10^−138^; Monkey B: *F*~(8,337)~ = 241.14, *p* = 1.4 × 10^−134^). We also saw a main effect of elicitation conditions (Monkey A: *F*~(1,237)~ = 7.69, *p* = 0.006; Monkey B: *F*~(1,337)~ = 20.21, *p* = 9.6 × 10^−6^), where CEs elicited in the MIXED condition were significantly different from those in the REPEATED condition. Adding to this effect, we observed a significant interaction effect between probability and condition (Monkey A: *F*~(8,237)~ = 7.73, *p* = 3.3 × 10^−9^; Monkey B: *F*~(8,337)~ = 12.56, *p* = 8.5 × 10^−16^), suggesting that the different elicitation sequences had a more complex effect on CE values than a mere monotonic increase or decrease. This effect was readily observable from the condition-specific CE distributions ([Fig. 2](#F2){ref-type="fig"}*c*), where the concave pattern of the MIXED-condition CEs contrasts with the S-shaped distribution of the REPEATED-condition CEs. Analysis of the RTs showed no significant difference across conditions for Monkey A, while monkey B responded faster in the MIXED than in the REPEATED condition ([Fig. 2](#F2){ref-type="fig"}*d*). In general, monkeys showed a consistent RT pattern ([Fig. 2-1](#fig2-1){ref-type="supplementary-material"}): shorter RTs when choosing higher EV compared to lower EV options, and longer RTs for smaller EV differences between options. Sequence-dependent changes in probability distortion ---------------------------------------------------- Because CE elicitation rested on reward options that varied in both magnitude and probability, any risk-preference changes could be attributed to nonlinear utility, probability distortion, or a combination of both. To better understand the role of these decision variables in shaping a gamble\'s subjective value, we simultaneously estimated the shape of both functions from the monkeys\' daily binary choices. Using a standard discrete choice model ([Eq. 3](#FD3){ref-type="disp-formula"}), we elicited functional parameters that best explained each animal\'s choices between gamble-safe choice pairs on individual days, assuming nonlinear utility and probability distortion. The estimation procedure allowed parameters to take on any value, imposing no constraints beyond the functional forms of the discrete choice softmax, probability distortion, and utility curves. We defined the value of each reward option as the product of its subjective probability and utility, consistent with prospect theory and other modern decision theories ([@B20]; [@B35]). As is traditionally done, we modeled utility through a one-parameter power function. The simple power function accounted well for risk-seeking (ρ \> 1), risk-averse (ρ \< 1), or risk neutral attitude (ρ = 1) for the range of reward magnitudes. We tested only one model for utility, as magnitude presentations did not differ across conditions. Instead, we sought to optimize our choice model with regards to subjective probability because CE elicitation sequences differed in terms of the order in which gamble probabilities were experienced. We tested four classical models of probability distortion to maximize the reliability of our model in capturing real choices; two of these functions had one free parameter, and the others had two. Finally, we defined cumulative log-likelihood functions for each of these models and estimated the best-fitting parameters for each decision function through MLE (see Materials and Methods). Across all testing sessions, the BIC scores of the Prelec curves were consistently lower than the one-parameter Tversky and lower than the Gonzalez models in at least monkeys ([Fig. 3](#F3){ref-type="fig"}*a*). However, while the two-parameter Prelec had a marginally lower BIC score in both animals, the one-parameter Prelec showed had a marginally lower sum of squared errors between predicted and average experimental CEs (one-factor ANOVA with repeated measures, Monkey A: *F*~(3,144)~ = 6.166, *pGGc* = 5.7 × 10^4^; Monkey B: *F*~(3,168)~ = 3.699, *pGGc* = 1.3 × 10^−2^). We ultimately selected the one-parameter Prelec due to this lower sum of squared errors, lower parameter count, and because of its ease of interpretation: for the curvature parameter α \> 1, the function underweighted low probabilities and overweighted high ones, for α \< 1, low probabilities were overweighted and high ones were underweighted. With an α = 1, probabilities were treated linearly. Monte Carlo simulations from predefined parameters confirmed the reliability of the MLE method for the selected model: we recovered accurate parameters for both the utility ([Fig. 3](#F3){ref-type="fig"}*b*) and probability distortion ([Fig. 3](#F3){ref-type="fig"}*c*) functions. {ref-type="supplementary-material"}). Estimated choice percentages were computed using the discrete choice model with the MLE-recovered parameters ([Eq. 3](#FD3){ref-type="disp-formula"}, using the *Prelec-1* probability weighting function). Red and blue points represent estimated CEs. Vertical dashed lines indicate EVs. The estimated psychometric functions closely approximated the measured data points, and differences in estimated CEs across conditions are compatible with the observed data for both low and high probabilities ([Fig. 2](#F2){ref-type="fig"}*b*).](zns9991915170003){#F3} 10.1523/JNEUROSCI.1454-18.2018.f3-1 Modeled vs measured choice behavior. Comparison of estimated (curves) and measured (circles) percentage of safe choices as a function of safe magnitude. Conventions and symbols as in Fig. 3d. Thin lines represent differences between estimated and experimental data percentages, with the horizontal line (at 0.5 on the y axis) corresponding to perfect estimate (difference=0). Download Figure 3-1, TIF file Having selected the one-parameter Prelec as the best-fitting probability distortion function, we estimated the functional parameters of our choice model ([Eq. 3](#FD3){ref-type="disp-formula"}) using the MLE method. The model was able to capture the characteristic pattern of risk attitudes observed in our experimental data: CEs of low probability gambles resulted in larger than the respective EVs in the MIXED condition, whereas CEs of high probability gambles were larger than their EVs in the REPEATED condition ([Fig. 3](#F3){ref-type="fig"}*d*, see [Fig. 3-1](#fig3-1){ref-type="supplementary-material"} for the full dataset), in accordance with the measured behavior ([Fig. 2](#F2){ref-type="fig"}*b*). We compared daily estimated parameters across CE elicitation conditions for utility and probability distortion ([Fig. 4](#F4){ref-type="fig"}*a*). Both animals exhibited convex utility (ρ \> 1) in the tested range of 0--0.5 ml accounting for risk-seeking behavior, with linearity only in the case of Monkey B\'s REPEATED condition. Importantly, probability distortions inverted across elicitation condition. In the MIXED elicitation condition, both animals overweighted low probabilities and underweighted high ones (α \> 1), whereas they instead underweighted low probabilities and overweighted high ones within the REPEATED condition (α \< 1) ([Fig. 4](#F4){ref-type="fig"}*b*). MANOVA confirmed the impact of the different elicitation sequences on both animals\' choice pattern (Monkey A: *F*~(1,54)~ = 24.96, Wilks\'s λ = 0.41, *p* = 3.85 × 10^−10^, η^2^ = 0.59; Monkey B: *F*~(1,57)~ = 40.78, Wilk\'s λ = 0.31, *p* = 5.2 × 10^−14^, η^2^ = 0.69) with only the probability distortion parameter (α) consistently different across conditions ([Fig. 4](#F4){ref-type="fig"}*a*,*c*). The change in risk-attitude between the two conditions could therefore, at least in the case of gamble-safe choices, be reduced to a reversal in the probability distortion function. {#F4} The REPEATED condition was a much less complex decision situation compared with the MIXED one, theoretically allowing for a simpler choice strategy: it would have been sufficient to evaluate the certain option, ignoring the gamble option in the majority of trials, to make choices. We tested for this possibility by fitting a model with an attentional parameter to the choice data ([Eq. 11](#FD11){ref-type="disp-formula"}). We found that there was no significant difference in attention given to the safe compared with the gamble option (the weight parameter was not significantly different from 0.5; Monkey A: *t*~(21)~ = −2.01, *p* = 5.7 × 10^−2^ (*t* test); Monkey B: *t*~(30)~ = −1.25; *p* = 2.2 × 10^−1^), suggesting that both options were fully considered when making choices in the REPEATED condition. Furthermore, shorter RTs in the REPEATED condition, expected if the monkeys ignored the gamble option, were not observed ([Fig. 2](#F2){ref-type="fig"}*d*). Reversal of probability distortion in the Marschak--Machina triangle -------------------------------------------------------------------- To extend our findings beyond gamble-safe choices, we characterized the choice behavior of 1 monkey in a different set of gambles using the Marschak--Machina triangle. This diagram was first introduced as a way of "organizing" a series of anomalies observed in risky choices, most notably the common ratio and common consequence effects, which violated the independence axiom of EU theory ([@B3]). Several economic theories were developed to explain these apparent paradoxes. Each theory predicted indifference curves with distinctive shapes in the Marschak--Machina triangle, making it an ideal framework to evaluate and compare the alternative theories ([@B23]). The use of this diagram, which makes it possible to represent a more general class of choice options (i.e., gambles with three fixed outcomes of varying probabilities) ([Fig. 5](#F5){ref-type="fig"}*a*), allowed us to extend our results to a wider range of problems. We did this to test the robustness of the parametric modeling (out-of-sample test) and, most importantly, to investigate the effect of elicitation condition from a different perspective: by looking at the change in direction of indifference lines, which connected points of the triangle edges (specific two-outcome gambles) for which the animal expressed choice indifference ([Fig. 5](#F5){ref-type="fig"}*b*), we could quantify the effects of elicitation condition that were specifically dependent on changes in probability distortion, and independent of changes in the shape of the utility function. {#F5} One of the theoretical consequences of probability distortions in the Marschak--Machina triangle is that indifference lines would not be parallel to each other, as in the case of linear probability weighting, but would instead fan-out or fan-in depending on the probability distortion ([Fig. 5](#F5){ref-type="fig"}*c*): an inverse S-shaped probability distortion would induce fanning-out, whereas an S-shaped one would result in indifference lines fanning-in. Fanning-out would indeed correspond to an increase in the steepness of the indifference lines when shifting "probability mass" from worse to better outcomes. As steeper lines correlate with more risk-seeking behavior, fanning-out would imply an inverse S-shaped probability distortion. The opposite would happen with fanning-in indifference lines, then corresponding to an S-shaped probability distortion function ([@B7]). Crucially, because the outcome magnitudes used in the Marschak--Machina triangle are fixed, the fanning direction is independent of the utility function and is therefore solely determined by the shape of the probability distortion. In that sense, any observed change in the fanning direction of the indifference lines with a change in elicitation sequence could only be due to a change in the probability weighting function ([Fig. 5](#F5){ref-type="fig"}*c*). We used the previously recovered parameters for utility and probability distortion to estimate the expected pattern of indifference lines in the two experimental conditions: MIXED and REPEATED sequences. We then compared the predicted directions of the indifference lines with the measured ones. As expected, the theoretical indifference lines, modeled using the previously elicited parameters, showed a slight fanning-out pattern for the MIXED condition, where a weakly inverse S-shaped probability distortion was measured. Conversely, we saw a fanning-in pattern in the REPEATED condition, for which we had observed an S-shaped probability distortion ([Fig. 6](#F6){ref-type="fig"}*a*, left). {#F6} The direct experimental measure of indifference lines was performed by presenting the animals with binary choices between a gamble represented by a fixed point on the triangle edge and one of several points on the triangle\'s hypotenuse. The indifference line was defined as the segment connecting the fixed point with the point corresponding to choice indifference on the hypotenuse. This procedure resulted in a directional pattern of indifference lines compatible with the theoretically predicted one, with no clear fanning direction of indifference lines in the MIXED condition, and clear fanning-in in the REPEATED condition ([Fig. 6](#F6){ref-type="fig"}*a*, right). We quantified this directional pattern of indifference lines using a measure for the fanning direction. The fanning of indifference lines corresponds to a gradual change in the slope of indifference lines: when moving from the lower right corner of the probability triangle to the upper left corner, an increasing slope would produce fanning-out, whereas a decreasing slope would produce fanning-in. Following this principle, we statistically assessed the fanning direction of the indifference lines by computing a linear regression on the slopes of the indifference lines. Results show no significant regression slope in the MIXED condition (*R*^2^ = 0.08, *p* = 0.50), suggesting no fanning of indifference curves, whereas in the REPEATED condition a significant linear regression (*R*^2^ = 0.98, *p* = 4.4 × 10^−6^) indicated fanning-out of the indifference lines. These results are consistent with predictions from the modeled indifference lines, which show a similar pattern of fanning directions ([Fig. 6](#F6){ref-type="fig"}*b*). We statistically compared the measured and predicted patterns of indifference lines by calculating the shift in the location of indifference points across conditions, corresponding to changes in the slope of indifference lines. A significant correlation between the predicted and measured shifts (Pearson\'s correlation coefficient *r* = 0.78, *p* = 4.0 × 10^−3^) confirmed that the experimental data complied with our theoretical predictions ([Fig. 6](#F6){ref-type="fig"}*c*) and supported the finding that probability distortion drove the change in risk attitude between REPEATED and MIXED conditions. The effect of trial history on the probability distortion --------------------------------------------------------- Because the structure of elicitation sequences appeared to affect probability distortions specifically, we investigated whether the differences in choice behavior could be explained in relation to past experiences, or trial history. One key difference between elicitation sequences was the order of the probabilities presented on the screen. In the MIXED sequences, the monkeys were much more likely to have experienced different gambles in their immediate past than in trials from REPEATED sequences, where the same gamble was repeated numerous times. Consequently, while the range of probabilities, magnitudes, and safe outcomes was identical in both conditions, the variability of past gambles was significantly different between the two conditions ([Fig. 1](#F1){ref-type="fig"}*d*,*e*). Because human and nonhuman primates, much like rodents, often base part of their risky decisions on recent experiences ([@B28]; [@B4]; [@B26]; [@B13]), we again ran a logistic regression on the probability of choosing the gamble option: this time to verify whether the EV of past gambles had any impact on the animals\' decisions ([Eq. 14](#FD14){ref-type="disp-formula"}). We found that, in the MIXED condition, both monkeys made use of at least one past gamble to make their decision ([Fig. 7](#F7){ref-type="fig"}*a*). The monkeys appeared to bias their choices in favor of the gamble (positive regression coefficient) when the prior gamble\'s EV was higher. In game-theoric terms, and taking the gamble\'s EV as a proxy for its "win rate," monkeys seemed to follow a WSLS strategy, whereby receiving a reward from a risky choice option increased the likelihood of choosing a similar option again; the opposite was true for choices where the risky option resulted in a loss (no reward). To validate this hypothesis, we applied a WSLS-compatible model ([Eq. 15](#FD15){ref-type="disp-formula"}) on the immediate trial history of both monkeys, looking at their propensity to choose a gamble over a safe outcome when they had previously chosen a gamble and won ([Fig. 7](#F7){ref-type="fig"}*b*). As expected, we found a significant effect of both the current gamble\'s EV (one-sample *t* test, Monkey A: *t*~(50)~ = 29.41, *p* = 3.19 × 10^−33^; Monkey B: *t*~(58)~ = 32.28, *p* = 9.38 × 10^−39^) and the current safe outcome\'s EV on the likelihood of choosing a gamble (one-sample *t* test, Monkey A: *t*~(50)~ = −38.71, *p* = 6.05 × 10^−39^; Monkey B: *t*~(58)~ = −46.19, *p* = 1.9 × 10^−47^). Both monkeys had a small but significant side bias (one-sample *t* test, Monkey A: *t*~(50)~ = −4.59, *p* = 2.97 × 10^−5^; Monkey B: *t*~(58)~ = −2.55, *p* = 1.3 × 10^−2^). More importantly, there was a significant positive effect of "winning" the preceding gamble on the likelihood of selecting the gamble option again, regardless of value (one-sample *t* test, Monkey A: *t*~(50)~ = 10.75, *p* = 1.3 × 10^−14^; Monkey B: *t*~(58)~ = 8.32, *p* = 1.76 × 10^−11^). In other words, receiving a reward from a risky gamble made the next gamble more attractive relative to the safe outcome. {ref-type="disp-formula"}) was at least as good at describing the choice data compared with the model with no past trials\' influence ([Eq. 2](#FD2){ref-type="disp-formula"}) (Monkey A: BIC~2~ = 84.2, BIC~14~ = 82.3, *t* test: *p* = 0.14; Monkey B: BIC~2~ = 221.4, BIC~14~ = 215.8, *t* test: *p* = 5.8 × 10^−5^). ***c***, Effect of past outcomes on the utility and probability distortion functions. The utility function appeared more convex following a gamble-win trial (0.5 ml reward) than following a loss (no reward), suggesting that gamble outcomes had an influence on the relative value of gamble and safe options on the next trial. The utility parameter estimates following win and loss trials are indicated as αW and αL, respectively, whereas probability distortion parameter as ρW and ρL, respectively. Arrows indicate the change in the utility parameter between loss and win trials. Error bars indicate the 95% CIs of the parameter estimates. ***d***, Simulated effect of EV update mechanism based on past outcomes. Mean ± SEM across simulated sessions (*N* = 50) of the CE resulting from choices simulated using the learning model ([Eq. 16](#FD16){ref-type="disp-formula"}) in MIXED and REPEATED conditions. The parameters used in the simulation were recovered from the MLE procedure with the same model separately for each monkey. Linear probability weighting and linear magnitude coding were used in the simulation, demonstrating that an EV update mechanism interacting with the local trial structure could explain the observed change in risk attitudes across conditions without explicitly introducing a nonlinear probability distortion.](zns9991915170007){#F7} We investigated this effect further, by estimating separate utility and probability distortion parameters in trials where a past gamble had been selected and "won" and in trials where the past selected gamble had been "lost." Due to lower trial counts per session after this trial selection, all sessions were pooled for each condition. In both animals, the utility function estimated from the former class of trials was more convex than the utility estimated from unrewarded trials ([Fig. 7](#F7){ref-type="fig"}*c*). Probability distortions, however, were not consistently different between these two classes of trials, maintaining their respective inverse-S and S shapes for MIXED and REPEATED conditions. Much like in the logistic regression, these results suggested a tendency to choose the gamble option more often after rewarded (win) trials, compared with unrewarded trials (a more convex utility function corresponding to stronger risk-seeking behavior). What it also highlighted, however, was a change in the relative value distribution between gambles and safe options: one that varies with past experience. In other words, gambles following a rewarded trial would be of higher relative value for the monkeys than those following unrewarded trials, at least in terms of safe rewards. Past win or lost effects on subjective value could account for some of the gap in probability distortion observed across our two conditions. A MIXED sequence of gambles would drive subjective value estimates in an opposing pattern to that of a REPEATED elicitation sequence simply due to task structure. In the case of MIXED sequences, the random distribution of gamble probabilities would indeed result in an inverse-S probability distortion. Gambles with probabilities \>0.5 would, more often than not, follow a gamble of lower EV; the monkey would then, on average, be less likely to pick said gamble due to the decrease in subjective value estimate following lower past returns. This would drive down the CE value of high probability gambles. In the case of low probability gambles, high past returns would drive CEs up. From this, we would expect an opposing distortion pattern in a REPEATED condition. For any gamble, the CE value would be distorted in a way proportional to its own probability: a low probability gamble would be driven down in value by repeated experience, whereas a high probability gamble would see its value go up. A change in gamble value, rather than a simple WSLS strategy, might also have longer lasting effects and could explain the persistence of sequence-type effects when looking at choices in the Marschak--Machina triangle paradigm. To test this hypothesis directly, we developed a simple reinforcement learning model in which gamble values were updated based on the previous trial\'s outcome: the value of a gamble increased by a fixed amount after a win and decreased by the same amount after a loss ([Eq. 16](#FD16){ref-type="disp-formula"}). Importantly, in the choice model, the gambles\' starting values were the respective objective EVs, which were compared with the objective safe magnitudes to make choices. No utility or probability distortion was included, only the previous choice softmax function, and we made no distinction between parameters estimated in repeated or mixed sequences. We again estimated the model parameters through MLE on each session\'s trial-by-trial choice data and retrieved a significant, mean value-updating parameter for both monkeys (Monkey A: η = 4.5 × 10^−3^ ± 9.0 × 10^−4^ SEM; *t*~(55)~ = 4.96, *p* = 7.1 × 10^−6^; Monkey B: η = 4.1 × 10^−3^ ± 5.8 × 10^−4^ SEM; *t*~(58)~ = 7.1, *p* = 2.0 × 10^−9^). The value of η corresponded to the fixed amount of value being added to or removed from the gamble\'s subjective value estimate following "win" and "lose" trials, respectively. After running the estimation procedure on all sessions, we tested whether the average observed value-updating parameter could explain the different CE distributions seen across the MIXED and REPEATED conditions. We computed CEs from simulated choices using the learning model defined above ([Eq. 16](#FD16){ref-type="disp-formula"}), using the mean recovered softmax and value-updating parameters, still holding utility and probability weights linear. The resulting pattern of simulated CEs ([Fig. 7](#F7){ref-type="fig"}*d*) followed the experimental pattern. In particular, it captured the clear separation between the two CE elicitation sequences. Although this model appeared to have a higher BIC score than the "classical" prospect theory model ([Eq. 3](#FD3){ref-type="disp-formula"}) (Monkey A: BIC~Eq16~ = 160.7, BIC~Eq3~ = 137.5, *t*~(55)~ = 6.92, *p* = 5.01 × 10^−9^; Monkey B: BIC~Eq16~ = 419.8, BIC~Eq3~ = 392.7, *t*~(58)~ = 4.69, *p* = 1.70 × 10^−5^), it accounted for the change in the pattern of CEs across both conditions using a single set of parameters. Conversely, two different sets of parameters were necessary for the prospect theory counterpart to capture the different CE patterns. Together, these results suggest that a simple value-updating mechanism that modifies gamble values based on the previous outcomes, applied to different elicitation sequences, would be sufficient to induce a reversal in the observed probability distortion patterns of monkeys during choice. Discussion ========== This study demonstrated that the shape of the probability weighting function guiding value-based choices in monkeys depended largely on the task\'s sequence structure. When deriving CEs from sequences in which different probabilistic rewards pseudorandomly alternated (MIXED), we found that monkeys overweighted low probability rewards and underweighted high probability ones. Conversely, the same CE elicitation method yielded the opposite choice pattern (underweighting of low probabilities and overweighting of high ones) when choice sequences consisted of trial blocks each containing a unique, REPEATED gamble. By simultaneously eliciting utility and probability weighting functions from each of these elicitation conditions, we showed that the two opposing choice patterns we observed could be explained by a reversal of the standard inverse S-shaped probability distortion function, seen when gambles were MIXED, to an S-shaped distortion when identical gambles were REPEATED. We confirmed and extended these results by comparing choice indifference lines in the Marschak--Machina triangle representations of the two elicitation conditions. The triangle\'s indifference maps were compatible with the observed inversion of probability distortions, preserving the weighting patterns in trials where no safe options were presented. Finally, by analyzing both sequence structure and monkeys\' choices in relation to previous trials, we showed that a past-driven update of subjective values could partially explain the observed reversal in probability distortion. Modern economic theories of choice under risk introduced distorted probability weightings to account for biases and departures from expected utility theory\'s predictions ([@B37]; [@B3]; [@B20]). Since then, the typical finding has been that humans overweighted low probabilities, all the while underweighting high ones ([@B22]; [@B11]; [@B1]; [@B34]): an inverse-S probability distortion ([@B20]). This shape has also been replicated in monkeys ([@B33]), where human-ported tasks resulted in a reliable inverse-S probability distortion. The current study ties in with these findings, using a coherent set of visual stimuli for both gambles and safe reward options to control for any bias introduced by the different visual representations of the two option types. Our results, in addition to reliability capturing macaque behavior using modern economic choice theories, further characterize the effects of sequence structure on utility and probability distortion. In contrast to the generally reported inverse-S-shaped probability distortion, a growing number of studies on human and animal subjects have highlighted the variability in probability distortion shapes, both across subjects and between task conditions ([@B18]; [@B5]; [@B9]). Recent work by [@B9]) emphasized the flexibility of probability weights in adapting to contextual changes, after finding that S-shaped and linear probability distortions could be elicited in monkeys when performing different tasks. Our experimental data confirmed this high level of behavioral flexibility in monkeys, whereby directly manipulating the order of presented gambles in a single task produced opposing patterns of probability distortion. Other findings from human experiments suggest that the way in which probability information is presented could account for the reported variability in subjects\' risk attitudes. For example, when reward probabilities are explicitly described (choice from description) to human subjects, they act as if overweighting the probability of rare events, but when probabilities are learned from experience (choice from experience), subjects choose as if underweighting the probability of rare events. This effect has been aptly referred to as the description-experience (DE) gap ([@B17]) and appears to extend to other primates. Indeed, monkeys have been shown to be more risk-seeking for experienced than for described gambles, suggesting a similar DE gap effect in nonhuman primates ([@B14]). Whereas some authors have called for two separate theories explaining choices from description and choices from experience ([@B15]; [@B2]), others have suggested that prospect theory could effectively describe choice in the two situations when allowing for a change in the probability distortion function between the two settings ([@B36]; [@B10]). While the dichotomous choice patterns we observed are comparable with those described in the DE gap studies, here the cues representing reward probabilities were identical in the two sequence conditions. In both MIXED and REPEATED sequences, probabilities were described explicitly through cues, learned from experience by the animals; the conditions only differed in the presentation order of the probability information. While the task design was different from previous human DE studies in this respect, the repeated sampling of outcomes typically used to "learn" the value of risky prospects in choices from experience (for review, see [@B39]) echoes the repetitive structure of our REPEATED sequence; conversely, described prospects are typically presented in a less structured, randomized sequence, analogous to our MIXED condition. While a direct comparison remains to be made, findings in both the DE gap experiments and in the present study suggest that past trial outcomes play a role in shaping the subjective perception of reward probabilities. Sampling bias has been identified as a source of variability in probability distortions, particularly in relation to the DE gap. Indeed, sampling bias is particularly problematic in "experienced" conditions due to the limited number of trials used in learning the options\' values: with small sample sizes, low probability gambles are often rewarded less frequently than would be prescribed by their nominal probability ([@B15]; [@B16]; [@B8]). The use of identical descriptive cues and elicitation procedures in the present study ensured that similar sampling sizes were applied, and indeed required, to estimate CEs for every gamble. Any bias would therefore affect the two conditions in a similar manner. With no obvious sampling biases, our data suggest that the DE gap could be modeled on the probability distortion changes we observed across task conditions, and that much like in the present study, the observed changes in risk-preferences from described to experienced reward probabilities, might result from differences in the task\'s presentation order of probability information. A final source of variability we considered was that the REPEATED condition was a much less complex decision situation than the MIXED one: one could ignore the gamble in long, repeated sequences. However, we found that the animals neither differentially weighed the options nor made choices faster in the REPEATED condition, indicating that they were not using widely differing valuation strategies. The Marschak--Machina triangle, a diagram widely used in the economics literature, allows for the intuitive representation of choices between two- and three-outcome gambles, serving as an ideal framework for investigating complex economic choice problems ([@B24]; [@B7]). In the current experiment, we elicited indifference points in the Marschak--Machina triangle representation of the monkeys\' behavior, which crucially provided a link between animal and human studies. Although full indifference curves within the Marschak--Machina triangle remain to be tested, we showed that indifference points on the triangle edges complied with economic theories of choice, and confirmed the reversal of probability distortion across conditions, this time with probabilistic rewards only. Consequently, we demonstrated the possibility of rigorous behavioral characterization in nonhuman primates, paving the way for future investigations into the neurophysiological basis of advanced economic constructs, such as probability distortion, specific economic axioms, or the neural counterparts of alternative economic theories. In conclusion, our results demonstrated the effect of a task\'s sequence structure on the shape of a monkey\'s elicited probability distortion, and highlighted the potential influence of past rewards on subjective value. Moreover, and perhaps most significantly, these adaptive effects extended through time: the patterns of indifference lines observed in the Marschak--Machina triangle after a session of MIXED or REPEATED sequences were compatible with the probability distortion shapes measured in the preceding CE elicitation session, even though the paradigm used in the Marschak--Machina triangle was always randomized. In this sense, the CE elicitation sequences preceding the Marschak--Machina triangle paradigm might have driven and reinforced a gap between the subjective values of identical probabilities, one that influenced choices between gambles in the Marschak--Machina triangle. The reinforcement learning model we used supports this hypothesis, implying that each experienced outcome could reinforce and update the subjective value of probabilities, leading to a flexible and contextually driven judgment of probabilistic information. More sophisticated models, such as the addition of a standard Rescorla--Wagner learning rule or a nonlinear transformation of safe magnitudes to the current value updating mechanism, could be more biologically plausible and successful in explaining the choice mechanism, and remain to be explored. It should be noted that the monkeys\' initial learning/association phase was not analyzed here in reinforcement learning terms, as it was performed with imperative trials. A better understanding of probability learning, and the permanence of subjective values reinforced across different conditions, could shed light on the core elements of prospect theory and on the undeniably adaptive nature of utility and probability distortions. This work was supported by Wellcome Grants WT 095495 and WT 204811 and European Research Council Advanced Grant 293549. The authors declare no competing financial interests. [^1]: Author contributions: S.F.-T. and P.M.B. wrote the first draft of the paper; S.F.-T., P.M.B., and W.S. edited the paper; S.F.-T. and W.S. designed research; S.F.-T. and P.M.B. performed research; S.F.-T. and P.M.B. analyzed data; S.F.-T. and P.M.B. wrote the paper. [^2]: \*S.F.-T. and P.M.B. contributed equally to this work.